1 2018-02-08 Richard Biener <rguenther@suse.de>
3 PR tree-optimization/84278
4 * tree-vect-stmts.c (vectorizable_store): When looking for
5 smaller vector types to perform grouped strided loads/stores
6 make sure the mode is supported by the target.
7 (vectorizable_load): Likewise.
9 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
11 * config/aarch64/aarch64.c (aarch64_components_for_bb):
12 Increase LDP/STP opportunities by adding adjacent callee-saves.
14 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
16 PR rtl-optimization/84068
17 PR rtl-optimization/83459
18 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
20 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
22 PR tree-optimization/84224
23 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
24 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
27 2018-02-07 Iain Sandoe <iain@codesourcery.com>
29 * config/rs6000/altivec.md (*restore_world): Remove LR use.
30 * config/rs6000/predicates.md (restore_world_operation): Adjust op
31 count, remove one USE.
33 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
35 * doc/install.texi (Configuration): Document the
36 --with-long-double-format={ibm,ieee} PowerPC configuration
40 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
41 Convert from define_expand to be define_insn_and_split. Rework
42 float/double/_Float128 conversions to QI/HI/SImode to work with
43 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
44 conversions to QI/HImode types did a store and then a load to
45 truncate the value. For conversions to VSX registers, don't split
46 the insn, instead emit the code directly. Use the code iterator
47 any_fix to combine signed and unsigned conversions.
48 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
49 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
50 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
51 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
52 (fix_<mode>di2_hw): Likewise.
53 (fixuns_<mode>di2_hw): Likewise.
54 (fix_<mode>si2_hw): Likewise.
55 (fixuns_<mode>si2_hw): Likewise.
56 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
57 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
58 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
59 fix<uns>_trunc<SFDF:mode>si2_p8.
60 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
62 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
63 (fix<uns>_<mode>_mem): Likewise.
64 (fctiw<u>z_<mode>_mem): Likewise.
65 (fix<uns>_<mode>_mem): Likewise.
66 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
67 the register allocator from doing a direct move to the GPRs to do
68 a store, and instead use the ISA 3.0 store byte/half-word from
69 vector register instruction. For IEEE 128-bit floating point,
70 also optimize stores of 32-bit ints.
71 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
73 2018-02-07 Alan Hayward <alan.hayward@arm.com>
75 * genextract.c (push_pathstr_operand): New function to support
77 (walk_rtx): Call push_pathstr_operand.
78 (print_path): Support [a-zA-Z].
80 2018-02-07 Richard Biener <rguenther@suse.de>
82 PR tree-optimization/84037
83 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
84 (cse_and_gimplify_to_preheader): Declare.
85 (vect_get_place_in_interleaving_chain): Likewise.
86 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
88 (_loop_vec_info::~_loop_vec_info): Delete it.
89 (cse_and_gimplify_to_preheader): New function.
90 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
91 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
92 (vectorizable_load): Likewise. For grouped stores always base
93 the IV on the first element.
94 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
95 condition before gimplifying.
97 2018-02-07 Jakub Jelinek <jakub@redhat.com>
99 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
100 *DIV_EXPR and *MOD_EXPR.
102 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
105 * config/i386/i386.c (ix86_option_override_internal): Mask out
106 the CF_SET bit when checking -fcf-protection.
108 2018-02-07 Tom de Vries <tom@codesourcery.com>
111 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
114 2018-02-07 Richard Biener <rguenther@suse.de>
116 PR tree-optimization/84204
117 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
120 PR tree-optimization/84205
121 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
122 special-case isl_ast_op_zdiv_r.
124 PR tree-optimization/84223
125 * graphite-scop-detection.c (gather_bbs::before_dom_children):
126 Only add conditions from within the region.
127 (gather_bbs::after_dom_children): Adjust.
129 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
132 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
133 * config/avr/avr.md: Only post-reload split REG-REG moves if
134 either register is REGERAL_REG_P.
136 2018-02-07 Jakub Jelinek <jakub@redhat.com>
138 PR tree-optimization/84235
139 * tree-ssa-scopedtables.c
140 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
141 if the subtraction is performed in floating point type where NaNs are
142 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
143 build 1. Formatting fix.
145 2018-02-06 Jakub Jelinek <jakub@redhat.com>
148 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
149 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
150 and skip it regardless of bb boundaries. Use CALL_P macro,
151 don't test INSN_P (insn) together with CALL_P or JUMP_P check
152 unnecessarily, formatting fix.
154 2018-02-06 Michael Collison <michael.collison@arm.com>
156 * config/arm/thumb2.md:
157 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
158 (*thumb_mov_notscc): Ditto.
160 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
163 * config/rs6000/rs6000.md (su code attribute): Use "u" for
164 unsigned_fix, not "s".
166 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
168 * configure.ac (gcc_fn_eh_frame_ro): New function.
169 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
170 correct .eh_frame permissions.
171 * configure: Regenerate.
173 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
175 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
178 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
180 * config/rs6000/rs6000.c (rs6000_option_override_internal):
181 Display warning message for -mno-speculate-indirect-jumps.
183 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
185 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
187 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
189 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
191 PR tree-optimization/84225
192 * tree-eh.c (find_trapping_overflow): Only call
193 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
195 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
198 * config/i386/i386.c: Reimplement the check of possible options
199 -mibt/-mshstk conbination. Change error messages.
200 * doc/invoke.texi: Fix a typo: remove extra '='.
202 2018-02-06 Marek Polacek <polacek@redhat.com>
204 PR tree-optimization/84228
205 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
207 2018-02-06 Tamar Christina <tamar.christina@arm.com>
210 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
211 emitted arch directives.
212 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
213 __ARM_FEATURE_COPROC before changing architectures.
215 2018-02-06 Richard Biener <rguenther@suse.de>
217 * config/i386/i386.c (print_reg): Fix typo.
218 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
220 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
222 * configure: Regenerate.
224 2018-02-05 Martin Sebor <msebor@redhat.com>
226 PR tree-optimization/83369
227 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
230 2018-02-05 Martin Liska <mliska@suse.cz>
232 * doc/invoke.texi: Cherry-pick upstream r323995.
234 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
236 * ira.c (ira_init_register_move_cost): Adjust comment.
238 2018-02-05 Martin Liska <mliska@suse.cz>
240 PR gcov-profile/84137
241 * doc/gcov.texi: Fix typo in documentation.
243 2018-02-05 Martin Liska <mliska@suse.cz>
245 PR gcov-profile/83879
246 * doc/gcov.texi: Document necessity of --dynamic-list-data when
247 using dlopen functionality.
249 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
251 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
252 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
253 _mm_maskz_range_ss, _mm_mask_range_round_ss,
254 _mm_maskz_range_round_ss): New intrinsics.
255 (__builtin_ia32_rangesd128_round)
256 (__builtin_ia32_rangess128_round): Remove.
257 (__builtin_ia32_rangesd128_mask_round,
258 __builtin_ia32_rangess128_mask_round): New builtins.
259 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
260 __builtin_ia32_rangess128_round): Remove.
261 (__builtin_ia32_rangesd128_mask_round,
262 __builtin_ia32_rangess128_mask_round): New builtins.
263 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
264 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
265 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
266 "<round_saeonly_constraint>")): Changed to ...
267 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
268 "<round_saeonly_scalar_constraint>")): ... this.
269 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
270 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
271 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
272 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
273 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
275 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
277 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
279 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
280 Remove all values except native, 8540 and 8548.
282 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
284 * config/i386/i386.c (ix86_output_function_return): Pass
285 INVALID_REGNUM, instead of -1, as invalid register number to
286 indirect_thunk_name and output_indirect_thunk.
288 2018-02-02 Julia Koval <julia.koval@intel.com>
290 * config.gcc: Add -march=icelake.
291 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
292 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
293 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
294 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
295 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
296 (processor_target_table): Add icelake.
297 (ix86_option_override_internal): Handle new PTAs.
298 (get_builtin_code_for_version): Handle icelake.
299 (M_INTEL_COREI7_ICELAKE): New.
300 (fold_builtin_cpu): Handle icelake.
301 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
302 * doc/invoke.texi: Add -march=icelake.
304 2018-02-02 Julia Koval <julia.koval@intel.com>
306 * config/i386/i386.c (ix86_option_override_internal): Change flags type
308 * wide-int-bitmask.h: New.
310 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
313 * config/i386/i386.md: Replace Pmode with word_mode in
314 builtin_setjmp_setup and builtin_longjmp to support x32.
316 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
320 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
322 (rs6000_supported_cpu_names): New static variable.
323 (linux_cpu_translation_table): Likewise.
324 (elf_platform) <cpu>: Define new static variable and use it.
325 Translate kernel AT_PLATFORM name to canonical name if needed.
326 Error if platform name is unknown.
328 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
331 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
333 2018-02-01 Jeff Law <law@redhat.com>
336 * config/i386/i386.c (release_scratch_register_on_entry): Add new
337 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
338 the scratch if RELEASE_VIA_POP is false.
339 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
340 If we have to save a temporary register, decrement SIZE appropriately.
341 Pass new arguments to release_scratch_register_on_entry.
342 (ix86_adjust_stack_and_probe): Likewise.
343 (ix86_emit_probe_stack_range): Pass new arguments to
344 release_scratch_register_on_entry.
346 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
348 PR rtl-optimization/84157
349 * combine.c (change_zero_ext): Use REG_P predicate in
350 front of HARD_REGISTER_P predicate.
352 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
354 * config/avr/avr.c (avr_option_override): Move disabling of
355 -fdelete-null-pointer-checks to...
356 * common/config/avr/avr-common.c (avr_option_optimization_table):
359 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
361 PR tree-optimization/81635
362 * tree-data-ref.c (split_constant_offset_1): For types that
363 wrap on overflow, try to use range info to prove that wrapping
366 2018-02-01 Renlin Li <renlin.li@arm.com>
369 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
371 (aarch64_register_move_cost): Likewise.
372 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
374 (REG_CLASS_NAMES): Likewise.
375 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
376 TAILCALL_ADDR_REGS. Remove IP registers.
377 * config/aarch64/aarch64.md (Ucs): Update register constraint.
379 2018-02-01 Richard Biener <rguenther@suse.de>
381 * domwalk.h (dom_walker::dom_walker): Add additional constructor
382 for specifying RPO order and allow NULL for that.
383 * domwalk.c (dom_walker::dom_walker): Likewise.
384 (dom_walker::walk): Handle NULL RPO order.
385 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
387 (rewrite_update_dom_walker): Likewise.
388 (mark_def_dom_walker): Likewise.
390 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
392 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
393 (aarch64_maybe_expand_sve_subreg_move): Declare.
394 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
395 * config/aarch64/predicates.md (aarch64_any_register_operand): New
397 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
398 that are semantically a reverse operation.
399 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
400 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
401 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
403 (aarch64_can_change_mode_class): For big-endian, forbid changes
404 between two SVE modes if they have different element sizes.
406 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
408 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
409 the TImode handling for big-endian targets.
411 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
413 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
414 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
416 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
417 Remove BSWAP handing for big-endian targets and use the form of
418 LD1RQ appropariate for the mode.
420 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
422 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
423 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
426 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
429 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
430 check for operands that need to go through aarch64_sve_reload_be.
432 2018-02-01 Jakub Jelinek <jakub@redhat.com>
434 PR tree-optimization/81661
435 PR tree-optimization/84117
436 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
437 * tree-eh.c: Include gimplify.h.
438 (find_trapping_overflow, replace_trapping_overflow,
439 rewrite_to_non_trapping_overflow): New functions.
440 * tree-vect-loop.c: Include tree-eh.h.
441 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
442 * tree-data-ref.c: Include tree-eh.h.
443 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
445 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
447 PR rtl-optimization/84123
448 * combine.c (change_zero_ext): Check if hard register satisfies
449 can_change_dest_mode before calling gen_lowpart_SUBREG.
451 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
454 * ira.c (ira_init_register_move_cost): Remove assert.
456 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
458 PR rtl-optimization/84071
459 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
460 * doc/tm.texi: Regenerate.
462 2018-01-31 Richard Biener <rguenther@suse.de>
464 PR tree-optimization/84132
465 * tree-data-ref.c (analyze_miv_subscript): Properly
466 check whether evolution_function_is_affine_multivariate_p
467 before calling gcd_of_steps_may_divide_p.
469 2018-01-31 Julia Koval <julia.koval@intel.com>
472 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
473 * config/i386/i386.md (rdpid_rex64) New.
474 (rdpid): Make 32bit only.
476 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
479 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
480 an IDENTIFIER_NODE for FUNCTION_TYPE's.
482 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
485 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
487 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
489 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
491 PR rtl-optimization/84071
492 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
493 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
495 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
497 * config/arc/arc.c (arc_handle_aux_attribute): New function.
498 (arc_attribute_table): Add 'aux' attribute.
499 (arc_in_small_data_p): Consider aux like variables.
500 (arc_is_aux_reg_p): New function.
501 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
502 (arc_get_aux_arg): New function.
503 (prepare_move_operands): Handle aux-register access.
504 (arc_handle_aux_attribute): New function.
505 * doc/extend.texi (ARC Variable attributes): Add subsection.
507 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
509 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
510 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
511 (arc_attribute_table): Add 'uncached' attribute.
512 (arc_print_operand): Print '.di' flag for uncached memory
514 (arc_in_small_data_p): Do not consider for small data the uncached
516 (arc_is_uncached_mem_p): New function.
517 * config/arc/predicates.md (compact_store_memory_operand): Check
518 for uncached memory accesses.
519 (nonvol_nonimm_operand): Likewise.
520 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
522 2018-01-31 Jakub Jelinek <jakub@redhat.com>
525 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
526 falign-loops=): Add Optimization flag.
528 2018-01-30 Jeff Law <law@redhat.com>
531 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
532 INT_REGISTERS_SAVED. Check it prior to calling
533 get_scratch_register_on_entry.
534 (ix86_adjust_stack_and_probe): Similarly.
535 (ix86_emit_probe_stack_range): Similarly.
536 (ix86_expand_prologue): Corresponding changes.
538 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
541 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
542 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
544 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
547 * lra-constraints.c (curr_insn_transform): Process AND in the
550 2018-01-30 Jakub Jelinek <jakub@redhat.com>
552 PR rtl-optimization/83986
553 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
554 dependence against last_pending_memory_flush in addition to
557 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
559 PR tree-optimization/81611
560 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
563 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
565 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
568 2018-01-30 Richard Biener <rguenther@suse.de>
569 Jakub Jelinek <jakub@redhat.com>
571 PR tree-optimization/84111
572 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
573 inner loops added during recursion, as they don't have up-to-date
576 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
579 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
580 (can_inline_edge_by_limits_p): ... here.
581 (can_early_inline_edge_p, check_callers,
582 update_caller_keys, update_callee_keys, recursive_inlining,
583 add_new_edges_to_heap, speculation_useful_p,
584 inline_small_functions,
585 inline_small_functions, flatten_function,
586 inline_to_all_callers_1): Update.
588 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
590 * profile-count.c (profile_count::combine_with_ipa_count): Handle
593 2018-01-30 Richard Biener <rguenther@suse.de>
595 PR tree-optimization/83008
596 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
597 invariant and constant vector uses in stmts when they need
600 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
603 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
604 * configure: Regenerate.
606 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
608 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
610 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
611 Use gen_rtx_REG rather than gen_lowpart.
613 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
615 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
616 rather than 0 when creating partial subregs.
618 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
620 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
623 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
626 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
627 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
628 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
629 flags. This restores the settings used before the 2017-07-24.
630 Turning off pre increment/decrement/modify allows IVOPTS to
631 optimize DF/SF loops where the index is an int.
633 2018-01-29 Richard Biener <rguenther@suse.de>
634 Kelvin Nilsen <kelvin@gcc.gnu.org>
637 * tree-vect-stmts.c (vectorizable_call): Don't call
638 targetm.vectorize_builtin_md_vectorized_function if callee is
641 2018-01-22 Carl Love <cel@us.ibm.com>
643 * doc/extend.tex: Fix typo in second arg in
644 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
646 2018-01-29 Richard Biener <rguenther@suse.de>
648 PR tree-optimization/84086
649 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
650 (flush_ssaname_freelist): When SSA names were released reset
653 2018-01-29 Richard Biener <rguenther@suse.de>
655 PR tree-optimization/84057
656 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
657 removed paths when removing edges.
659 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
661 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
662 -mfunction-return=@var{choice}.
664 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
667 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
669 (layout::print_source_line): Likewise.
670 (test_get_line_width_without_trailing_whitespace): Add test cases.
672 2018-01-27 Jakub Jelinek <jakub@redhat.com>
675 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
678 2018-01-26 Jim Wilson <jimw@sifive.com>
680 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
682 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
685 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
687 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
688 and CMP + SUB-immediate -> SUBS.
690 2018-01-26 Martin Sebor <msebor@redhat.com>
692 PR tree-optimization/83896
693 * tree-ssa-strlen.c (get_string_len): Rename...
694 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
695 Avoid assuming length is constant.
696 (handle_char_store): Use HOST_WIDE_INT for string length.
698 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
701 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
702 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
704 2018-01-26 Richard Biener <rguenther@suse.de>
706 PR rtl-optimization/84003
707 * dse.c (record_store): Only record redundant stores when
708 the earlier store aliases at least all accesses the later one does.
710 2018-01-26 Jakub Jelinek <jakub@redhat.com>
712 PR rtl-optimization/83985
713 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
714 REG_CFA_RESTORE insns.
715 (delete_unmarked_insns): Don't ignore separate shrink wrapping
716 REG_CFA_RESTORE insns here.
719 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
720 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
722 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
724 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
725 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
726 (arc_init): Likewise.
727 (arc_override_options): Likewise.
728 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
730 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
732 * config/arc/arc.h (TARGET_DBNZ): Define.
733 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
734 properly set the tune attribute.
735 (dbnz): Use TARGET_DBNZ guard.
736 * config/arc/arc.opt (mtune): Add core3 option.
738 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
740 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
741 recognize new pic like addresses.
742 (arc_delegitimize_address): Clean up.
744 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
746 * config/arc/arc-arches.def: Option mrf16 valid for all
748 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
749 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
750 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
751 * config/arc/arc-tables.opt: Regenerate.
752 * config/arc/arc.c (arc_conditional_register_usage): Handle
753 reduced register file case.
754 (arc_file_start): Set must have build attributes.
755 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
757 * config/arc/arc.opt (mrf16): Add new option.
758 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
759 * config/arc/genmultilib.awk: Handle new mrf16 option.
760 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
761 * config/arc/t-multilib: Regenerate.
762 * doc/invoke.texi (ARC Options): Document mrf16 option.
764 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
766 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
767 * config/arc/arc.c (arc_handle_secure_attribute): New function.
768 (arc_attribute_table): Add 'secure_call' attribute.
769 (arc_print_operand): Print secure call operand.
770 (arc_function_ok_for_sibcall): Don't optimize tail calls when
772 (arc_is_secure_call_p): New function. * config/arc/arc.md
773 (call_i): Add support for sjli instruction.
774 (call_value_i): Likewise.
775 * config/arc/constraints.md (Csc): New constraint.
777 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
778 John Eric Martin <John.Martin@emmicro-us.com>
780 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
781 * config/arc/arc.c (_arc_jli_section): New struct.
782 (arc_jli_section): New type.
783 (rc_jli_sections): New static variable.
784 (arc_handle_jli_attribute): New function.
785 (arc_attribute_table): Add jli_always and jli_fixed attribute.
786 (arc_file_end): New function.
787 (TARGET_ASM_FILE_END): Define.
788 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
789 (arc_add_jli_section): New function.
790 (jli_call_scan): Likewise.
791 (arc_reorg): Call jli_call_scan.
792 (arc_output_addsi): Remove 'S' from printing asm operand.
793 (arc_is_jli_call_p): New function.
794 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
796 (movhi_insn): Likewise.
797 (movsi_insn): Likewise.
798 (movsi_set_cc_insn): Likewise.
799 (loadqi_update): Likewise.
800 (load_zeroextendqisi_update): Likewise.
801 (load_signextendqisi_update): Likewise.
802 (loadhi_update): Likewise.
803 (load_zeroextendhisi_update): Likewise.
804 (load_signextendhisi_update): Likewise.
805 (loadsi_update): Likewise.
806 (loadsf_update): Likewise.
807 (movsicc_insn): Likewise.
808 (bset_insn): Likewise.
809 (bxor_insn): Likewise.
810 (bclr_insn): Likewise.
811 (bmsk_insn): Likewise.
812 (bicsi3_insn): Likewise.
813 (cmpsi_cc_c_insn): Likewise.
814 (movsi_ne): Likewise.
815 (movsi_cond_exec): Likewise.
816 (clrsbsi2): Likewise.
827 (call_i): Remove 'S' asm letter, add jli instruction.
828 (call_value_i): Likewise.
829 * config/arc/arc.op (mjli-always): New option.
830 * config/arc/constraints.md (Cji): New constraint.
831 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
833 (subsf3_fpx): Likewise.
834 (mulsf3_fpx): Likewise.
835 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
837 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
839 * doc/invoke.texi (ARC): Document mjli-always option.
841 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
843 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
844 and use incw and decw where possible
845 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
847 2018-01-26 Richard Biener <rguenther@suse.de>
849 PR tree-optimization/81082
850 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
851 association if it requires casting to unsigned.
852 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
853 from fold_plusminus_mult_expr to catch important cases late when
854 range info is available.
856 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
858 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
859 * configure.ac (hidden_linkonce): New test.
860 * configure: Regenerate.
861 * config.in: Regenerate.
863 2018-01-26 Julia Koval <julia.koval@intel.com>
865 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
866 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
867 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
868 _mm_mask_bitshuffle_epi64_mask): Fix type.
869 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
870 USI_FTYPE_V4DI_V4DI_USI): Remove.
871 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
872 __builtin_ia32_vpshufbitqmb256_mask,
873 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
874 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
875 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
877 2018-01-26 Alan Modra <amodra@gmail.com>
880 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
881 UNSPEC_VBPERMQ. Sort other unspecs.
883 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
885 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
887 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
890 * predict.c (drop_profile): Do not push/pop cfun; update also
892 (handle_missing_profiles): Fix logic looking for zero profiles.
894 2018-01-25 Jakub Jelinek <jakub@redhat.com>
897 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
898 on functions with #pragma omp declare simd or functions with simd
900 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
901 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
902 Remove trailing \n from warning_at calls.
904 2018-01-25 Tom de Vries <tom@codesourcery.com>
907 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
908 for neutered workers.
910 2018-01-24 Joseph Myers <joseph@codesourcery.com>
913 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
914 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
916 2018-01-24 Jeff Law <law@redhat.com>
919 * i386.c (get_probe_interval): Move to earlier point.
920 (ix86_compute_frame_layout): If -fstack-clash-protection and
921 the frame is larger than the probe interval, then use pushes
922 to save registers rather than reg->mem moves.
923 (ix86_expand_prologue): Remove conditional for int_registers_saved
926 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
929 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
930 min/max for never referenced object.
932 2018-01-24 Jakub Jelinek <jakub@redhat.com>
935 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
937 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
938 attributes from DECL_ATTRIBUTES (decl) without affecting
939 DECL_ATTRIBUTES (current_function_decl).
940 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
941 functions with non-NULL DECL_ABSTRACT_ORIGIN.
943 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
945 PR tree-optimization/83979
946 * fold-const.c (fold_comparison): Use constant_boolean_node
947 instead of boolean_{true,false}_node.
949 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
951 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
954 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
956 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
957 Simplify the clause that sets the length attribute.
958 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
959 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
960 clause that sets the length attribute.
961 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
963 2018-01-24 Tom de Vries <tom@codesourcery.com>
966 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
967 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
968 Add strict parameter.
969 (prevent_branch_around_nothing): Insert dummy insn between branch to
970 label and label with no ptx insn inbetween.
971 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
973 2018-01-24 Tom de Vries <tom@codesourcery.com>
976 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
977 for neutered threads in warp.
978 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
980 2018-01-24 Richard Biener <rguenther@suse.de>
982 PR tree-optimization/83176
983 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
986 2018-01-24 Richard Biener <rguenther@suse.de>
988 PR tree-optimization/82819
989 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
990 code generating pluses that are no-ops in the target precision.
992 2018-01-24 Richard Biener <rguenther@suse.de>
995 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
997 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
999 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1000 to merge probabilities.
1001 * predict.c (probably_never_executed): Also mark as cold functions
1002 with global 0 profile and guessed local profile.
1003 * profile-count.c (profile_probability::combine_with_count): New
1005 * profile-count.h (profile_probability::operator*,
1006 profile_probability::operator*=, profile_probability::operator/,
1007 profile_probability::operator/=): Reduce precision to adjusted
1008 and set value to guessed on contradictory divisions.
1009 (profile_probability::combine_with_freq): Remove.
1010 (profile_probability::combine_wiht_count): Declare.
1011 (profile_count::force_nonzero):: Set to adjusted.
1012 (profile_count::probability_in):: Set quality to adjusted.
1013 * tree-ssa-tail-merge.c (replace_block_by): Use
1016 2018-01-23 Andrew Waterman <andrew@sifive.com>
1017 Jim Wilson <jimw@sifive.com>
1019 * config/riscv/riscv.c (riscv_stack_boundary): New.
1020 (riscv_option_override): Set riscv_stack_boundary. Handle
1021 riscv_preferred_stack_boundary_arg.
1022 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1023 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1024 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1025 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1026 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1027 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1029 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1032 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1033 of struct ix86_frame.
1034 (ix86_expand_epilogue): Likewise. Add a local variable for
1035 the reg_save_offset field in struct ix86_frame.
1037 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1039 PR tree-optimization/82604
1040 * tree-loop-distribution.c (enum partition_kind): New enum item
1041 PKIND_PARTIAL_MEMSET.
1042 (partition_builtin_p): Support above new enum item.
1043 (generate_code_for_partition): Ditto.
1044 (compute_access_range): Differentiate cases that equality can be
1045 proven at all loops, the innermost loops or no loops.
1046 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1047 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1048 (finalize_partitions, distribute_loop): Don't fuse partition of
1049 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1050 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1053 2018-01-23 Martin Liska <mliska@suse.cz>
1055 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1056 order to ignore the predictor.
1057 (PRED_POLYMORPHIC_CALL): Likewise.
1058 (PRED_RECURSIVE_CALL): Likewise.
1060 2018-01-23 Martin Liska <mliska@suse.cz>
1062 * tree-profile.c (tree_profiling): Print function header to
1063 aware reader which function we are working on.
1064 * value-prof.c (gimple_find_values_to_profile): Do not print
1065 not interesting value histograms.
1067 2018-01-23 Martin Liska <mliska@suse.cz>
1069 * profile-count.h (enum profile_quality): Add
1070 profile_uninitialized as the first value. Do not number values
1071 as they are zero based.
1072 (profile_count::verify): Update sanity check.
1073 (profile_probability::verify): Likewise.
1075 2018-01-23 Nathan Sidwell <nathan@acm.org>
1077 * doc/invoke.texi (ffor-scope): Deprecate.
1079 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1081 PR tree-optimization/83510
1082 * domwalk.c (set_all_edges_as_executable): New function.
1083 (dom_walker::dom_walker): Convert bool param
1084 "skip_unreachable_blocks" to enum reachability. Move setup of
1085 edge flags to set_all_edges_as_executable and only do it when
1086 reachability is REACHABLE_BLOCKS.
1087 * domwalk.h (enum dom_walker::reachability): New enum.
1088 (dom_walker::dom_walker): Convert bool param
1089 "skip_unreachable_blocks" to enum reachability.
1090 (set_all_edges_as_executable): New decl.
1091 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1092 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1094 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1095 but converting true to REACHABLE_BLOCKS.
1096 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1098 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1099 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1100 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1102 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1103 if check_all_array_refs will be called.
1105 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1107 * tree.c (selftest::test_location_wrappers): Add more test
1110 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1112 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1113 (selftest::test_bit_in_range): Likewise.
1115 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1118 * doc/sourcebuild.texi (vect_float): Say that the selector
1119 only describes the situation when -funsafe-math-optimizations is on.
1120 (vect_float_strict): Document.
1122 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1124 PR tree-optimization/83965
1125 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1126 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1127 instead of checking only for a reduction.
1128 (vect_recog_widen_sum_pattern): Likewise.
1130 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1132 * predict.c (probably_never_executed): Only use precise profile info.
1133 (compute_function_frequency): Skip after inlining hack since we now
1134 have quality checking.
1136 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1138 * profile-count.h (profile_probability::very_unlikely,
1139 profile_probability::unlikely, profile_probability::even): Set
1140 precision to guessed.
1142 2018-01-23 Richard Biener <rguenther@suse.de>
1144 PR tree-optimization/83963
1145 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1146 Properly terminate dominator walk when crossing the exit edge not
1147 when visiting its source block.
1149 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1152 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1153 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1155 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1157 PR tree-optimization/83957
1158 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1159 semicolon after for body surrounded by braces.
1161 PR tree-optimization/83081
1162 * profile-count.h (profile_probability::split): New method.
1163 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1164 Use profile_probability::split.
1165 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1166 when splitting a single conditional jump into 2.
1168 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1170 PR tree-optimization/69452
1171 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1174 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1176 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1177 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1178 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1180 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1182 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1183 * config/rl78/rl78.md: New define_expand "movdi"
1184 * config/rl78/rl78.c: New function definition rl78_split_movdi
1186 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1189 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1191 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1192 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1193 128-bit to produce an UNSPEC move to get the double word with the
1194 signbit and then a shift directly to do signbit.
1195 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1196 implementation with a new version that just does either a direct
1197 move or a regular move. Move memory interface to separate insns.
1198 Move insns so they are next to the expander.
1199 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1200 with signbit move. Split big and little endian case.
1201 (signbit<mode>2_dm_mem_le): Likewise.
1202 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1203 (signbit<mode>2_dm2): Likewise.
1205 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1207 * config/rl78/rl78.md: New define_expand "anddi3".
1209 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1211 * config/rl78/rl78.md: New define_expand "umindi3".
1213 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1215 * config/rl78/rl78.md: New define_expand "smindi3".
1217 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1219 * config/rl78/rl78.md: New define_expand "smaxdi3".
1221 2018-01-22 Carl Love <cel@us.ibm.com>
1223 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1224 LVX_V1TI): Add macro expansion.
1225 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1226 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1227 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1228 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1229 Change check to determine if the instruction is a byte reversing
1230 entry. Fix typo in comment.
1231 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1232 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1233 Add def_builtin calls for new builtins.
1234 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1235 Add define_insn expansion.
1237 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1239 * config/rl78/rl78.md: New define_expand "umaxdi3".
1241 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1243 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1244 for non-QImode registers
1246 2018-01-22 Richard Biener <rguenther@suse.de>
1248 PR tree-optimization/83963
1249 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1250 including the loop exit block.
1251 (scop_detection::merge_sese): Likewise.
1252 (scop_detection::add_scop): Do it here instead.
1254 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1256 * doc/sourcebuild.texi (arm_softfloat): Document.
1258 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1261 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1262 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1263 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1265 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1266 David Edelsohn <dje.gcc@gmail.com>
1269 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1270 Change "crset eq" to "crset 2".
1271 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1272 (*call_indirect_aix<mode>_nospec): Likewise.
1273 (*call_value_indirect_aix<mode>_nospec): Likewise.
1274 (*call_indirect_elfv2<mode>_nospec): Likewise.
1275 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1276 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1277 change assembly output from . to $.
1278 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1279 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1280 (*tablejump<mode>_internal1_nospec): Likewise.
1282 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1285 * config/sh/sh_optimize_sett_clrt.cc:
1286 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1288 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1290 PR tree-optimization/83940
1291 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1292 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1293 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1294 use it to pass back the definition type.
1295 (vect_check_store_rhs): Likewise rhs_dt_out.
1296 (vect_build_gather_load_calls): Add a mask_dt argument and use
1297 it instead of a call to vect_is_simple_use.
1298 (vectorizable_store): Update calls to vect_check_load_store_mask
1299 and vect_check_store_rhs. Use the dt returned by the latter instead
1300 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1301 instead of calls to vect_is_simple_use. Pass the scalar rather
1302 than the vector operand to vect_is_simple_use when handling
1303 second and subsequent copies of an rhs value.
1304 (vectorizable_load): Update calls to vect_check_load_store_mask
1305 and vect_build_gather_load_calls. Use the cached mask_dt and
1306 gs_info.offset_dt instead of calls to vect_is_simple_use.
1308 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1311 * tree-emutls.c: Include gimplify.h.
1312 (lower_emutls_2): New function.
1313 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1314 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1315 it before further processing.
1318 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1319 UINTVAL (trueop1) instead of INTVAL (op1).
1321 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1325 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1326 INCOMING_FRAME_SP_OFFSET if not defined.
1327 (scan_trace): Add ENTRY argument. If true and
1328 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1329 emit a note to adjust the CFA offset.
1330 (create_cfi_notes): Adjust scan_trace callers.
1331 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1332 INCOMING_FRAME_SP_OFFSET in the CIE.
1333 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1334 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1336 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1337 * doc/tm.texi: Regenerated.
1339 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1341 PR rtl-optimization/83147
1342 * lra-constraints.c (remove_inheritance_pseudos): Use
1343 lra_substitute_pseudo_within_insn.
1345 2018-01-19 Tom de Vries <tom@codesourcery.com>
1346 Cesar Philippidis <cesar@codesourcery.com>
1349 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1351 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1354 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1355 spaces for function labels.
1357 2018-01-19 Martin Liska <mliska@suse.cz>
1359 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1360 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1361 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1362 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1363 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1364 (PRED_CONST_RETURN): Change from 69 to 65.
1365 (PRED_NULL_RETURN): Change from 91 to 71.
1366 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1367 (PRED_LOOP_GUARD): Change from 66 to 73.
1369 2018-01-19 Martin Liska <mliska@suse.cz>
1371 * predict.c (predict_insn_def): Add new assert.
1372 (struct branch_predictor): Change type to signed integer.
1373 (test_prediction_value_range): Amend test to cover
1375 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1376 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1377 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1378 (PRED_LOOP_IV_COMPARE): Likewise.
1379 * predict.h (PROB_UNINITIALIZED): Define new constant.
1381 2018-01-19 Martin Liska <mliska@suse.cz>
1383 * predict.c (dump_prediction): Add new format for
1384 analyze_brprob.py script which is enabled with -details
1386 * profile-count.h (precise_p): New function.
1388 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1390 PR tree-optimization/83922
1391 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1392 there are no statements that need masking.
1393 (vect_active_double_reduction_p): New function.
1394 (vect_analyze_loop_operations): Use it when handling phis that
1395 are not in the loop header.
1397 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1399 PR tree-optimization/83914
1400 * tree-vect-loop.c (vectorizable_induction): Don't convert
1401 init_expr or apply the peeling adjustment for inductions
1402 that are nested within the vectorized loop.
1404 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1406 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1409 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1413 * function.h (gimplify_parameters): Add gimple_seq * argument.
1414 * function.c: Include gimple.h and options.h.
1415 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1416 for the added local temporaries if needed.
1417 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1418 if there are any parameter cleanups, wrap whole body into a
1419 try/finally with the cleanups.
1421 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1424 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1425 Use GET_MODE_CLASS for scalar floating point.
1427 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1431 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1432 Fix call of call_cgraph_insertion_hooks.
1434 2018-01-18 Martin Sebor <msebor@redhat.com>
1436 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1438 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1441 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1444 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1447 * common.opt: (-ffile-prefix-map): New option.
1448 * opts.c (common_handle_option): Defer it.
1449 * opts-global.c (handle_common_deferred_options): Handle it.
1450 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1451 * file-prefix-map.h: New file.
1452 (remap_debug_filename, add_debug_prefix_map): ...here.
1453 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1454 * final.c (debug_prefix_map, add_debug_prefix_map
1455 remap_debug_filename): Move to...
1456 * file-prefix-map.c: New file.
1457 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1458 generalize, get rid of alloca(), use strrchr() instead of strchr().
1459 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1460 Implement in terms of add_prefix_map().
1461 (remap_macro_filename, remap_debug_filename): Implement in term of
1463 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1464 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1465 * dbxout.c: Include file-prefix-map.h.
1466 * varasm.c: Likewise.
1467 * vmsdbgout.c: Likewise.
1468 * xcoffout.c: Likewise.
1469 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1470 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1471 * doc/invoke.texi (-ffile-prefix-map): Document.
1472 (-fdebug-prefix-map): Update description.
1474 2018-01-18 Martin Liska <mliska@suse.cz>
1476 * config/i386/i386.c (indirect_thunk_name): Document that also
1478 (output_indirect_thunk): Document why both instructions
1479 (pause and lfence) are generated.
1481 2018-01-18 Richard Biener <rguenther@suse.de>
1483 PR tree-optimization/83887
1484 * graphite-scop-detection.c
1485 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1486 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1487 (scop_detection::merge_sese): Re-implement with a flood-fill
1488 algorithm that properly finds a SESE region if it exists.
1490 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1493 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1494 pointer_diff optimizations use view_convert instead of convert.
1496 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1498 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1499 Generate different code for -mno-speculate-indirect-jumps.
1500 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1501 (*call_indirect_aix<mode>): Disable for
1502 -mno-speculate-indirect-jumps.
1503 (*call_indirect_aix<mode>_nospec): New define_insn.
1504 (*call_value_indirect_aix<mode>): Disable for
1505 -mno-speculate-indirect-jumps.
1506 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1507 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1508 -mno-speculate-indirect-jumps.
1509 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1511 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1513 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1514 long double type, set the flags for noting the default long double
1515 type, even if we don't pass or return a long double type.
1517 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1520 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1523 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1525 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1526 support for merge[hl].
1527 (fold_mergehl_helper): New helper function.
1528 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1529 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1530 (altivec_vmrglw_direct): Add xxmrglw insn.
1532 2018-01-17 Andrew Waterman <andrew@sifive.com>
1534 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1535 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1537 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1540 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1541 call the lto_location_cache before reading the
1542 DECL_SOURCE_LOCATION of the types.
1544 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1545 Richard Sandiford <richard.sandiford@linaro.org>
1547 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1548 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1549 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1550 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1551 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1553 * config/aarch64/constraints.md (aarch64_movti_operand):
1555 * config/aarch64/predicates.md (Uti): Add new constraint.
1557 2018-01-17 Carl Love <cel@us.ibm.com>
1558 * config/rs6000/vsx.md (define_expand xl_len_r,
1559 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1560 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1562 (define_expand, define_insn): Move the shift left from the
1563 define_insn to the define_expand for lxvl and stxvl instructions.
1564 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1565 and XL_LEN_R definitions to PURE.
1567 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1569 * config/i386/i386.c (indirect_thunk_name): Declare regno
1570 as unsigned int. Compare regno with INVALID_REGNUM.
1571 (output_indirect_thunk): Ditto.
1572 (output_indirect_thunk_function): Ditto.
1573 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1574 in the call to output_indirect_thunk_function.
1576 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1579 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1580 rather than the size of inner_type to determine the stack slot size
1581 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1583 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1586 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1589 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1591 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1592 endian Linux systems to optionally enable multilibs for selecting
1593 the long double type if the user configured an explicit type.
1594 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1595 have no long double multilibs if not defined.
1596 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1597 warn if the user used -mabi={ieee,ibm}longdouble and we built
1598 multilibs for long double.
1599 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1600 appropriate multilib option.
1601 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1603 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1604 for building long double multilibs.
1605 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1607 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1609 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1612 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1614 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1617 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1620 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1623 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1625 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1626 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1628 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1630 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1631 different rtl trees depending on TARGET_64BIT.
1632 (rs6000_gen_lvx): Likewise.
1634 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1636 * config/visium/visium.md (nop): Tweak comment.
1637 (hazard_nop): Likewise.
1639 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1641 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1642 -mspeculate-indirect-jumps.
1643 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1644 for -mno-speculate-indirect-jumps.
1645 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1646 (*call_value_indirect_elfv2<mode>): Disable for
1647 -mno-speculate-indirect-jumps.
1648 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1649 (indirect_jump): Emit different RTL for
1650 -mno-speculate-indirect-jumps.
1651 (*indirect_jump<mode>): Disable for
1652 -mno-speculate-indirect-jumps.
1653 (*indirect_jump<mode>_nospec): New define_insn.
1654 (tablejump): Emit different RTL for
1655 -mno-speculate-indirect-jumps.
1656 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1657 (tablejumpsi_nospec): New define_expand.
1658 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1659 (tablejumpdi_nospec): New define_expand.
1660 (*tablejump<mode>_internal1): Disable for
1661 -mno-speculate-indirect-jumps.
1662 (*tablejump<mode>_internal1_nospec): New define_insn.
1663 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1666 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1668 * caller-save.c (insert_save): Drop unnecessary parameter. All
1671 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1672 Richard Biener <rguenth@suse.de>
1675 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1676 return early, inline manually is_gimple_sizepos. Make sure if we
1677 call gimplify_expr we don't end up with a gimple constant.
1678 * tree.c (variably_modified_type_p): Don't return true for
1679 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1680 * gimplify.h (is_gimple_sizepos): Remove.
1682 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1684 PR tree-optimization/83857
1685 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1686 vectorizable_live_operation for pure SLP statements.
1687 (vectorizable_live_operation): Handle PHIs.
1689 2018-01-16 Richard Biener <rguenther@suse.de>
1691 PR tree-optimization/83867
1692 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1693 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1695 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1698 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1699 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1700 If off is not INTEGER_CST, issue a may not be aligned warning
1701 rather than isn't aligned. Use isn%'t rather than isn't.
1702 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1704 <case MULT_EXPR>: Improve the case when bottom and one of the
1705 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1706 operand, in that case check if the other operand is multiple of
1707 bottom divided by the INTEGER_CST operand.
1709 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1712 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1713 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1714 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1715 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1716 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1717 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1718 (pa_function_arg_size): New function.
1720 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1722 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1723 in a separate statement.
1725 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1727 PR tree-optimization/83847
1728 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1729 group gathers and scatters.
1731 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1733 PR rtl-optimization/86620
1734 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1736 PR rtl-optimization/83213
1737 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1738 to last if both are JUMP_INSNs.
1740 PR tree-optimization/83843
1741 * gimple-ssa-store-merging.c
1742 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1743 store_immediate_info for bswap/nop orig_stores.
1745 2018-01-15 Andrew Waterman <andrew@sifive.com>
1747 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1749 <UDIV>: Increase cost if !TARGET_DIV.
1751 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1753 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1754 (define_attr "cr_logical_3op"): New.
1755 (cceq_ior_compare): Adjust.
1756 (cceq_ior_compare_complement): Adjust.
1757 (*cceq_rev_compare): Adjust.
1758 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1759 (is_cracked_insn): Adjust.
1760 (insn_must_be_first_in_group): Adjust.
1761 * config/rs6000/40x.md: Adjust.
1762 * config/rs6000/440.md: Adjust.
1763 * config/rs6000/476.md: Adjust.
1764 * config/rs6000/601.md: Adjust.
1765 * config/rs6000/603.md: Adjust.
1766 * config/rs6000/6xx.md: Adjust.
1767 * config/rs6000/7450.md: Adjust.
1768 * config/rs6000/7xx.md: Adjust.
1769 * config/rs6000/8540.md: Adjust.
1770 * config/rs6000/cell.md: Adjust.
1771 * config/rs6000/e300c2c3.md: Adjust.
1772 * config/rs6000/e500mc.md: Adjust.
1773 * config/rs6000/e500mc64.md: Adjust.
1774 * config/rs6000/e5500.md: Adjust.
1775 * config/rs6000/e6500.md: Adjust.
1776 * config/rs6000/mpc.md: Adjust.
1777 * config/rs6000/power4.md: Adjust.
1778 * config/rs6000/power5.md: Adjust.
1779 * config/rs6000/power6.md: Adjust.
1780 * config/rs6000/power7.md: Adjust.
1781 * config/rs6000/power8.md: Adjust.
1782 * config/rs6000/power9.md: Adjust.
1783 * config/rs6000/rs64.md: Adjust.
1784 * config/rs6000/titan.md: Adjust.
1786 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1788 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1789 ix86_indirect_branch_register logic.
1791 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1793 * config/i386/constraints.md (Bs): Update
1794 ix86_indirect_branch_register check. Don't check
1795 ix86_indirect_branch_register with GOT_memory_operand.
1797 * config/i386/predicates.md (GOT_memory_operand): Don't check
1798 ix86_indirect_branch_register here.
1799 (GOT32_symbol_operand): Likewise.
1801 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1803 * config/i386/predicates.md (constant_call_address_operand):
1804 Rewrite ix86_indirect_branch_register logic.
1805 (sibcall_insn_operand): Likewise.
1807 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1809 * config/i386/constraints.md (Bs): Replace
1810 ix86_indirect_branch_thunk_register with
1811 ix86_indirect_branch_register.
1813 * config/i386/i386.md (indirect_jump): Likewise.
1814 (tablejump): Likewise.
1815 (*sibcall_memory): Likewise.
1816 (*sibcall_value_memory): Likewise.
1817 Peepholes of indirect call and jump via memory: Likewise.
1818 * config/i386/i386.opt: Likewise.
1819 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1820 (GOT_memory_operand): Likewise.
1821 (call_insn_operand): Likewise.
1822 (sibcall_insn_operand): Likewise.
1823 (GOT32_symbol_operand): Likewise.
1825 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1828 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1829 type rather than type addr's type points to.
1830 (expand_omp_atomic_mutex): Likewise.
1831 (expand_omp_atomic): Likewise.
1833 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1836 * config/i386/i386.c (output_indirect_thunk_function): Use
1837 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1838 for __x86_return_thunk.
1840 2018-01-15 Richard Biener <rguenther@suse.de>
1843 * expmed.c (extract_bit_field_1): Fix typo.
1845 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1848 * config/arm/iterators.md (VF): New mode iterator.
1849 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1850 Remove integer-related logic from pattern.
1851 (neon_vabd<mode>_3): Likewise.
1853 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1856 * common.opt (fstrict-overflow): No longer an alias.
1857 (fwrapv-pointer): New option.
1858 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1859 also for pointer types based on flag_wrapv_pointer.
1860 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1861 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1862 opts->x_flag_wrapv got set.
1863 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1864 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1865 POINTER_TYPE_OVERFLOW_UNDEFINED.
1866 * match.pd: Likewise in address comparison pattern.
1867 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1869 2018-01-15 Richard Biener <rguenther@suse.de>
1872 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1873 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1874 Reset type names to their identifier if their TYPE_DECL doesn't
1875 have linkage (and thus is used for ODR and devirt).
1876 (save_debug_info_for_decl): Remove.
1877 (save_debug_info_for_type): Likewise.
1878 (add_tree_to_fld_list): Adjust.
1879 * tree-pretty-print.c (dump_generic_node): Make dumping of
1880 type names more robust.
1882 2018-01-15 Richard Biener <rguenther@suse.de>
1884 * BASE-VER: Bump to 8.0.1.
1886 2018-01-14 Martin Sebor <msebor@redhat.com>
1889 * builtins.c (check_access): Avoid warning when the no-warning bit
1892 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1894 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1895 * ira-color (allocno_hard_regs_compare): Likewise.
1897 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1900 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1901 Use .pushsection/.popsection.
1903 2018-01-14 Martin Sebor <msebor@redhat.com>
1906 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1908 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1910 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1911 entry from extra_headers.
1912 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1913 extra_headers, make the list bitwise identical to the i?86-*-* one.
1915 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1917 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1918 -mcmodel=large with -mindirect-branch=thunk,
1919 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1920 -mfunction-return=thunk-extern.
1921 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1922 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1923 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1925 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1927 * config/i386/i386.c (print_reg): Print the name of the full
1928 integer register without '%'.
1929 (ix86_print_operand): Handle 'V'.
1930 * doc/extend.texi: Document 'V' modifier.
1932 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1934 * config/i386/constraints.md (Bs): Disallow memory operand for
1935 -mindirect-branch-register.
1937 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1938 (GOT_memory_operand): Likewise.
1939 (call_insn_operand): Likewise.
1940 (sibcall_insn_operand): Likewise.
1941 (GOT32_symbol_operand): Likewise.
1942 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1943 for -mindirect-branch-register.
1944 (tablejump): Likewise.
1945 (*sibcall_memory): Likewise.
1946 (*sibcall_value_memory): Likewise.
1947 Disallow peepholes of indirect call and jump via memory for
1948 -mindirect-branch-register.
1949 (*call_pop): Replace m with Bw.
1950 (*call_value_pop): Likewise.
1951 (*sibcall_pop_memory): Replace m with Bs.
1952 * config/i386/i386.opt (mindirect-branch-register): New option.
1953 * doc/invoke.texi: Document -mindirect-branch-register option.
1955 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1957 * config/i386/i386-protos.h (ix86_output_function_return): New.
1958 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1959 set function_return_type.
1960 (indirect_thunk_name): Add ret_p to indicate thunk for function
1962 (output_indirect_thunk_function): Pass false to
1963 indirect_thunk_name.
1964 (ix86_output_indirect_branch_via_reg): Likewise.
1965 (ix86_output_indirect_branch_via_push): Likewise.
1966 (output_indirect_thunk_function): Create alias for function
1967 return thunk if regno < 0.
1968 (ix86_output_function_return): New function.
1969 (ix86_handle_fndecl_attribute): Handle function_return.
1970 (ix86_attribute_table): Add function_return.
1971 * config/i386/i386.h (machine_function): Add
1972 function_return_type.
1973 * config/i386/i386.md (simple_return_internal): Use
1974 ix86_output_function_return.
1975 (simple_return_internal_long): Likewise.
1976 * config/i386/i386.opt (mfunction-return=): New option.
1977 (indirect_branch): Mention -mfunction-return=.
1978 * doc/extend.texi: Document function_return function attribute.
1979 * doc/invoke.texi: Document -mfunction-return= option.
1981 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1983 * config/i386/i386-opts.h (indirect_branch): New.
1984 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1985 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1986 with local indirect jump when converting indirect call and jump.
1987 (ix86_set_indirect_branch_type): New.
1988 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1989 (indirectlabelno): New.
1990 (indirect_thunk_needed): Likewise.
1991 (indirect_thunk_bnd_needed): Likewise.
1992 (indirect_thunks_used): Likewise.
1993 (indirect_thunks_bnd_used): Likewise.
1994 (INDIRECT_LABEL): Likewise.
1995 (indirect_thunk_name): Likewise.
1996 (output_indirect_thunk): Likewise.
1997 (output_indirect_thunk_function): Likewise.
1998 (ix86_output_indirect_branch_via_reg): Likewise.
1999 (ix86_output_indirect_branch_via_push): Likewise.
2000 (ix86_output_indirect_branch): Likewise.
2001 (ix86_output_indirect_jmp): Likewise.
2002 (ix86_code_end): Call output_indirect_thunk_function if needed.
2003 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2005 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2006 (ix86_attribute_table): Add indirect_branch.
2007 * config/i386/i386.h (machine_function): Add indirect_branch_type
2008 and has_local_indirect_jump.
2009 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2011 (tablejump): Likewise.
2012 (*indirect_jump): Use ix86_output_indirect_jmp.
2013 (*tablejump_1): Likewise.
2014 (simple_return_indirect_internal): Likewise.
2015 * config/i386/i386.opt (mindirect-branch=): New option.
2016 (indirect_branch): New.
2019 (thunk-inline): Likewise.
2020 (thunk-extern): Likewise.
2021 * doc/extend.texi: Document indirect_branch function attribute.
2022 * doc/invoke.texi: Document -mindirect-branch= option.
2024 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2027 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2029 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2031 * ipa-inline.c (want_inline_small_function_p): Return false if
2032 inlining has already failed with CIF_FINAL_ERROR.
2033 (update_caller_keys): Call want_inline_small_function_p before
2035 (update_callee_keys): Likewise.
2037 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2039 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2041 (rs6000_quadword_masked_address_p): Likewise.
2042 (quad_aligned_load_p): Likewise.
2043 (quad_aligned_store_p): Likewise.
2044 (const_load_sequence_p): Add comment to describe the outer-most loop.
2045 (mimic_memory_attributes_and_flags): New function.
2046 (rs6000_gen_stvx): Likewise.
2047 (replace_swapped_aligned_store): Likewise.
2048 (rs6000_gen_lvx): Likewise.
2049 (replace_swapped_aligned_load): Likewise.
2050 (replace_swapped_load_constant): Capitalize argument name in
2051 comment describing this function.
2052 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2053 and stores that access quad-word aligned addresses and replace
2054 with stvx or lvx instructions when appropriate.
2055 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2056 New function prototype.
2057 (rs6000_quadword_masked_address_p): Likewise.
2058 (rs6000_gen_lvx): Likewise.
2059 (rs6000_gen_stvx): Likewise.
2060 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2061 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2062 when memory address is aligned.
2063 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2064 this split to select lvx instruction when memory address is aligned.
2065 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2066 instruction when memory address is aligned.
2067 (*vsx_le_perm_load_v16qi): Likewise.
2068 (four unnamed splitters): Modify to select the stvx instruction
2069 when memory is aligned.
2071 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2073 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2074 which appears in the queue multiple times.
2076 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2077 Alan Hayward <alan.hayward@arm.com>
2078 David Sherwood <david.sherwood@arm.com>
2080 * tree-vectorizer.h (vec_lower_bound): New structure.
2081 (_loop_vec_info): Add check_nonzero and lower_bounds.
2082 (LOOP_VINFO_CHECK_NONZERO): New macro.
2083 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2084 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2085 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2086 fields. Make seg_len the distance travelled, not including the
2088 (dr_direction_indicator): Declare.
2089 (dr_zero_step_indicator): Likewise.
2090 (dr_known_forward_stride_p): Likewise.
2091 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2093 (runtime_alias_check_p): Allow runtime alias checks with
2095 (operator ==): Compare access_size and align.
2096 (prune_runtime_alias_test_list): Rework for new distinction between
2097 the access_size and seg_len.
2098 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2100 (get_segment_min_max): New function.
2101 (create_intersect_range_checks): Use it.
2102 (dr_step_indicator): New function.
2103 (dr_direction_indicator): Likewise.
2104 (dr_zero_step_indicator): Likewise.
2105 (dr_known_forward_stride_p): Likewise.
2106 * tree-loop-distribution.c (data_ref_segment_size): Return
2107 DR_STEP * (niters - 1).
2108 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2110 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2111 (vect_preserves_scalar_order_p): New function, split out from...
2112 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2113 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2114 (vect_vfa_access_size): New function.
2115 (vect_vfa_align): Likewise.
2116 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2117 (dump_lower_bound): New function.
2118 (vect_check_lower_bound): Likewise.
2119 (vect_small_gap_p): Likewise.
2120 (vectorizable_with_step_bound_p): Likewise.
2121 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2122 depencies if the vectorization factor is 1. Convert the checks
2123 for nonzero steps into checks on the bounds of DR_STEP. Try using
2124 a bunds check for variable steps if the minimum required step is
2125 relatively small. Update calls to the dr_with_seg_len
2126 constructor and to vect_compile_time_alias.
2127 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2129 (vect_loop_versioning): Call it.
2130 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2132 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2134 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2135 Alan Hayward <alan.hayward@arm.com>
2136 David Sherwood <david.sherwood@arm.com>
2138 * doc/sourcebuild.texi (vect_scatter_store): Document.
2139 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2141 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2143 * genopinit.c (main): Add supports_vec_scatter_store and
2144 supports_vec_scatter_store_cached to target_optabs.
2145 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2146 IFN_MASK_SCATTER_STORE.
2147 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2149 * internal-fn.h (internal_store_fn_p): Declare.
2150 (internal_fn_stored_value_index): Likewise.
2151 * internal-fn.c (scatter_store_direct): New macro.
2152 (expand_scatter_store_optab_fn): New function.
2153 (direct_scatter_store_optab_supported_p): New macro.
2154 (internal_store_fn_p): New function.
2155 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2156 IFN_MASK_SCATTER_STORE.
2157 (internal_fn_mask_index): Likewise.
2158 (internal_fn_stored_value_index): New function.
2159 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2161 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2162 * optabs-query.c (supports_vec_scatter_store_p): New function.
2163 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2164 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2165 true for scatter stores.
2166 (vect_gather_scatter_fn_p): Handle scatter stores too.
2167 (vect_check_gather_scatter): Consider using scatter stores if
2168 supports_vec_scatter_store_p.
2169 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2171 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2172 internal_fn_stored_value_index.
2173 (check_load_store_masking): Handle scatter stores too.
2174 (vect_get_store_rhs): Make public.
2175 (vectorizable_call): Use internal_store_fn_p.
2176 (vectorizable_store): Handle scatter store internal functions.
2177 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2178 when deciding whether the end of the group has been reached.
2179 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2180 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2181 (mask_scatter_store<mode>): New insns.
2183 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2184 Alan Hayward <alan.hayward@arm.com>
2185 David Sherwood <david.sherwood@arm.com>
2187 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2188 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2189 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2191 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2192 Use vect_truncate_gather_scatter_offset if we can't treat the
2193 operation as a normal gather load or scatter store.
2194 (get_group_load_store_type): Take the gather_scatter_info
2195 as argument. Try using a gather load or scatter store for
2196 single-element groups.
2197 (get_load_store_type): Update calls to get_group_load_store_type
2198 and vect_use_strided_gather_scatters_p.
2200 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2201 Alan Hayward <alan.hayward@arm.com>
2202 David Sherwood <david.sherwood@arm.com>
2204 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2205 optional tree argument.
2206 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2208 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2209 but continue to use the current value as a fallback.
2210 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2211 to compare the updates.
2212 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2213 (get_load_store_type): Use it when handling a strided access.
2214 (vect_get_strided_load_store_ops): New function.
2215 (vect_get_data_ptr_increment): Likewise.
2216 (vectorizable_load): Handle strided gather loads. Always pass
2217 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2219 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2220 Alan Hayward <alan.hayward@arm.com>
2221 David Sherwood <david.sherwood@arm.com>
2223 * doc/md.texi (gather_load@var{m}): Document.
2224 (mask_gather_load@var{m}): Likewise.
2225 * genopinit.c (main): Add supports_vec_gather_load and
2226 supports_vec_gather_load_cached to target_optabs.
2227 * optabs-tree.c (init_tree_optimization_optabs): Use
2228 ggc_cleared_alloc to allocate target_optabs.
2229 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2230 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2232 * internal-fn.h (internal_load_fn_p): Declare.
2233 (internal_gather_scatter_fn_p): Likewise.
2234 (internal_fn_mask_index): Likewise.
2235 (internal_gather_scatter_fn_supported_p): Likewise.
2236 * internal-fn.c (gather_load_direct): New macro.
2237 (expand_gather_load_optab_fn): New function.
2238 (direct_gather_load_optab_supported_p): New macro.
2239 (direct_internal_fn_optab): New function.
2240 (internal_load_fn_p): Likewise.
2241 (internal_gather_scatter_fn_p): Likewise.
2242 (internal_fn_mask_index): Likewise.
2243 (internal_gather_scatter_fn_supported_p): Likewise.
2244 * optabs-query.c (supports_at_least_one_mode_p): New function.
2245 (supports_vec_gather_load_p): Likewise.
2246 * optabs-query.h (supports_vec_gather_load_p): Declare.
2247 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2248 and memory_type field.
2249 (NUM_PATTERNS): Bump to 15.
2250 * tree-vect-data-refs.c: Include internal-fn.h.
2251 (vect_gather_scatter_fn_p): New function.
2252 (vect_describe_gather_scatter_call): Likewise.
2253 (vect_check_gather_scatter): Try using internal functions for
2254 gather loads. Recognize existing calls to a gather load function.
2255 (vect_analyze_data_refs): Consider using gather loads if
2256 supports_vec_gather_load_p.
2257 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2258 (vect_get_gather_scatter_offset_type): Likewise.
2259 (vect_convert_mask_for_vectype): Likewise.
2260 (vect_add_conversion_to_patterm): Likewise.
2261 (vect_try_gather_scatter_pattern): Likewise.
2262 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2263 (vect_vect_recog_func_ptrs): Add it.
2264 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2265 internal_fn_mask_index and internal_gather_scatter_fn_p.
2266 (check_load_store_masking): Take the gather_scatter_info as an
2267 argument and handle gather loads.
2268 (vect_get_gather_scatter_ops): New function.
2269 (vectorizable_call): Check internal_load_fn_p.
2270 (vectorizable_load): Likewise. Handle gather load internal
2272 (vectorizable_store): Update call to check_load_store_masking.
2273 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2274 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2275 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2276 (aarch64_gather_scale_operand_d): New predicates.
2277 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2278 (mask_gather_load<mode>): New insns.
2280 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2281 Alan Hayward <alan.hayward@arm.com>
2282 David Sherwood <david.sherwood@arm.com>
2284 * optabs.def (fold_left_plus_optab): New optab.
2285 * doc/md.texi (fold_left_plus_@var{m}): Document.
2286 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2287 * internal-fn.c (fold_left_direct): Define.
2288 (expand_fold_left_optab_fn): Likewise.
2289 (direct_fold_left_optab_supported_p): Likewise.
2290 * fold-const-call.c (fold_const_fold_left): New function.
2291 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2292 * tree-parloops.c (valid_reduction_p): New function.
2293 (gather_scalar_reductions): Use it.
2294 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2295 (vect_finish_replace_stmt): Declare.
2296 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2297 (needs_fold_left_reduction_p): New function, split out from...
2298 (vect_is_simple_reduction): ...here. Accept reductions that
2299 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2300 (vect_force_simple_reduction): Also store the reduction type in
2301 the assignment's STMT_VINFO_REDUC_TYPE.
2302 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2303 (merge_with_identity): New function.
2304 (vect_expand_fold_left): Likewise.
2305 (vectorize_fold_left_reduction): Likewise.
2306 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2307 scalar phi in place for it. Check for target support and reject
2308 cases that would reassociate the operation. Defer the transform
2309 phase to vectorize_fold_left_reduction.
2310 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2311 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2312 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2314 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2316 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2317 call to ifc_temp_var.
2319 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2320 Alan Hayward <alan.hayward@arm.com>
2321 David Sherwood <david.sherwood@arm.com>
2323 * target.def (legitimize_address_displacement): Take the original
2324 offset as a poly_int.
2325 * targhooks.h (default_legitimize_address_displacement): Update
2327 * targhooks.c (default_legitimize_address_displacement): Likewise.
2328 * doc/tm.texi: Regenerate.
2329 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2330 as an argument, moving assert of ad->disp == ad->disp_term to...
2331 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2332 Try calling targetm.legitimize_address_displacement before expanding
2333 the address rather than afterwards, and adjust for the new interface.
2334 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2335 Match the new hook interface. Handle SVE addresses.
2336 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2339 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2341 * Makefile.in (OBJS): Add early-remat.o.
2342 * target.def (select_early_remat_modes): New hook.
2343 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2344 * doc/tm.texi: Regenerate.
2345 * targhooks.h (default_select_early_remat_modes): Declare.
2346 * targhooks.c (default_select_early_remat_modes): New function.
2347 * timevar.def (TV_EARLY_REMAT): New timevar.
2348 * passes.def (pass_early_remat): New pass.
2349 * tree-pass.h (make_pass_early_remat): Declare.
2350 * early-remat.c: New file.
2351 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2353 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2355 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2356 Alan Hayward <alan.hayward@arm.com>
2357 David Sherwood <david.sherwood@arm.com>
2359 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2360 vfm1 with a bound_epilog parameter.
2361 (vect_do_peeling): Update calls accordingly, and move the prologue
2362 call earlier in the function. Treat the base bound_epilog as 0 for
2363 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2364 this base when peeling for gaps.
2365 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2366 with fully-masked loops.
2367 (vect_estimate_min_profitable_iters): Handle the single peeled
2368 iteration in that case.
2370 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2371 Alan Hayward <alan.hayward@arm.com>
2372 David Sherwood <david.sherwood@arm.com>
2374 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2375 single-element interleaving even if the size is not a power of 2.
2376 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2377 accesses for single-element interleaving if the group size is
2380 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2381 Alan Hayward <alan.hayward@arm.com>
2382 David Sherwood <david.sherwood@arm.com>
2384 * doc/md.texi (fold_extract_last_@var{m}): Document.
2385 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2386 * optabs.def (fold_extract_last_optab): New optab.
2387 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2388 * internal-fn.c (fold_extract_direct): New macro.
2389 (expand_fold_extract_optab_fn): Likewise.
2390 (direct_fold_extract_optab_supported_p): Likewise.
2391 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2392 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2393 EXTRACT_LAST_REDUCTION.
2394 (get_initial_def_for_reduction): Do not create an initial vector
2395 for EXTRACT_LAST_REDUCTION reductions.
2396 (vectorizable_reduction): Leave the scalar phi in place for
2397 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2398 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2399 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2400 transform phase to vectorizable_condition.
2401 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2403 (vect_finish_stmt_generation): ...here.
2404 (vect_finish_replace_stmt): New function.
2405 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2406 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2408 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2410 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2411 Alan Hayward <alan.hayward@arm.com>
2412 David Sherwood <david.sherwood@arm.com>
2414 * doc/md.texi (extract_last_@var{m}): Document.
2415 * optabs.def (extract_last_optab): New optab.
2416 * internal-fn.def (EXTRACT_LAST): New internal function.
2417 * internal-fn.c (cond_unary_direct): New macro.
2418 (expand_cond_unary_optab_fn): Likewise.
2419 (direct_cond_unary_optab_supported_p): Likewise.
2420 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2421 loops using EXTRACT_LAST.
2422 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2423 (extract_last_<mode>): ...this optab.
2424 (vec_extract<mode><Vel>): Update accordingly.
2426 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2427 Alan Hayward <alan.hayward@arm.com>
2428 David Sherwood <david.sherwood@arm.com>
2430 * target.def (empty_mask_is_expensive): New hook.
2431 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2432 * doc/tm.texi: Regenerate.
2433 * targhooks.h (default_empty_mask_is_expensive): Declare.
2434 * targhooks.c (default_empty_mask_is_expensive): New function.
2435 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2436 if the target says that empty masks are expensive.
2437 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2439 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2441 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2442 Alan Hayward <alan.hayward@arm.com>
2443 David Sherwood <david.sherwood@arm.com>
2445 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2446 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2447 (vect_use_loop_mask_for_alignment_p): New function.
2448 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2449 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2450 niters_skip argument. Make sure that the first niters_skip elements
2451 of the first iteration are inactive.
2452 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2453 Update call to vect_set_loop_masks_directly.
2454 (get_misalign_in_elems): New function, split out from...
2455 (vect_gen_prolog_loop_niters): ...here.
2456 (vect_update_init_of_dr): Take a code argument that specifies whether
2457 the adjustment should be added or subtracted.
2458 (vect_update_init_of_drs): Likewise.
2459 (vect_prepare_for_masked_peels): New function.
2460 (vect_do_peeling): Skip prologue peeling if we're using a mask
2461 instead. Update call to vect_update_inits_of_drs.
2462 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2464 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2465 alignment. Do not include the number of peeled iterations in
2466 the minimum threshold in that case.
2467 (vectorizable_induction): Adjust the start value down by
2468 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2469 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2470 Take the number of skipped iterations into account when calculating
2472 * tree-vect-stmts.c (vect_gen_while_not): New function.
2474 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2475 Alan Hayward <alan.hayward@arm.com>
2476 David Sherwood <david.sherwood@arm.com>
2478 * doc/sourcebuild.texi (vect_fully_masked): Document.
2479 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2481 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2483 (vect_analyze_loop_2): ...here. Don't check the vectorization
2484 factor against the number of loop iterations if the loop is
2487 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2488 Alan Hayward <alan.hayward@arm.com>
2489 David Sherwood <david.sherwood@arm.com>
2491 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2492 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2493 (dump_groups): Update accordingly.
2494 (iv_use::mem_type): New member variable.
2495 (address_p): New function.
2496 (record_use): Add a mem_type argument and initialize the new
2498 (record_group_use): Add a mem_type argument. Use address_p.
2499 Remove obsolete null checks of base_object. Update call to record_use.
2500 (find_interesting_uses_op): Update call to record_group_use.
2501 (find_interesting_uses_cond): Likewise.
2502 (find_interesting_uses_address): Likewise.
2503 (get_mem_type_for_internal_fn): New function.
2504 (find_address_like_use): Likewise.
2505 (find_interesting_uses_stmt): Try find_address_like_use before
2506 calling find_interesting_uses_op.
2507 (addr_offset_valid_p): Use the iv mem_type field as the type
2508 of the addressed memory.
2509 (add_autoinc_candidates): Likewise.
2510 (get_address_cost): Likewise.
2511 (split_small_address_groups_p): Use address_p.
2512 (split_address_groups): Likewise.
2513 (add_iv_candidate_for_use): Likewise.
2514 (autoinc_possible_for_pair): Likewise.
2515 (rewrite_groups): Likewise.
2516 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2517 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2518 (get_alias_ptr_type_for_ptr_address): New function.
2519 (rewrite_use_address): Rewrite address uses in calls that were
2520 identified by find_address_like_use.
2522 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2523 Alan Hayward <alan.hayward@arm.com>
2524 David Sherwood <david.sherwood@arm.com>
2526 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2528 * gimple-expr.h (is_gimple_addressable: Likewise.
2529 * gimple-expr.c (is_gimple_address): Likewise.
2530 * internal-fn.c (expand_call_mem_ref): New function.
2531 (expand_mask_load_optab_fn): Use it.
2532 (expand_mask_store_optab_fn): Likewise.
2534 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2535 Alan Hayward <alan.hayward@arm.com>
2536 David Sherwood <david.sherwood@arm.com>
2538 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2539 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2540 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2541 (cond_umax@var{mode}): Document.
2542 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2543 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2544 (cond_umin_optab, cond_umax_optab): New optabs.
2545 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2546 (COND_IOR, COND_XOR): New internal functions.
2547 * internal-fn.h (get_conditional_internal_fn): Declare.
2548 * internal-fn.c (cond_binary_direct): New macro.
2549 (expand_cond_binary_optab_fn): Likewise.
2550 (direct_cond_binary_optab_supported_p): Likewise.
2551 (get_conditional_internal_fn): New function.
2552 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2553 Cope with reduction statements that are vectorized as calls rather
2555 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2556 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2557 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2558 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2559 (UNSPEC_COND_EOR): New unspecs.
2560 (optab): Add mappings for them.
2561 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2562 (sve_int_op, sve_fp_op): New int attributes.
2564 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2565 Alan Hayward <alan.hayward@arm.com>
2566 David Sherwood <david.sherwood@arm.com>
2568 * optabs.def (while_ult_optab): New optab.
2569 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2570 * internal-fn.def (WHILE_ULT): New internal function.
2571 * internal-fn.h (direct_internal_fn_supported_p): New override
2572 that takes two types as argument.
2573 * internal-fn.c (while_direct): New macro.
2574 (expand_while_optab_fn): New function.
2575 (convert_optab_supported_p): Likewise.
2576 (direct_while_optab_supported_p): New macro.
2577 * wide-int.h (wi::udiv_ceil): New function.
2578 * tree-vectorizer.h (rgroup_masks): New structure.
2579 (vec_loop_masks): New typedef.
2580 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2582 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2583 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2584 (vect_max_vf): New function.
2585 (slpeel_make_loop_iterate_ntimes): Delete.
2586 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2587 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2588 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2589 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2590 internal-fn.h, stor-layout.h and optabs-query.h.
2591 (vect_set_loop_mask): New function.
2592 (add_preheader_seq): Likewise.
2593 (add_header_seq): Likewise.
2594 (interleave_supported_p): Likewise.
2595 (vect_maybe_permute_loop_masks): Likewise.
2596 (vect_set_loop_masks_directly): Likewise.
2597 (vect_set_loop_condition_masked): Likewise.
2598 (vect_set_loop_condition_unmasked): New function, split out from
2599 slpeel_make_loop_iterate_ntimes.
2600 (slpeel_make_loop_iterate_ntimes): Rename to..
2601 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2602 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2603 (vect_do_peeling): Update call accordingly.
2604 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2606 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2607 mask_compare_type, can_fully_mask_p and fully_masked_p.
2608 (release_vec_loop_masks): New function.
2609 (_loop_vec_info): Use it to free the loop masks.
2610 (can_produce_all_loop_masks_p): New function.
2611 (vect_get_max_nscalars_per_iter): Likewise.
2612 (vect_verify_full_masking): Likewise.
2613 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2614 retries, and free the mask rgroups before retrying. Check loop-wide
2615 reasons for disallowing fully-masked loops. Make the final decision
2616 about whether use a fully-masked loop or not.
2617 (vect_estimate_min_profitable_iters): Do not assume that peeling
2618 for the number of iterations will be needed for fully-masked loops.
2619 (vectorizable_reduction): Disable fully-masked loops.
2620 (vectorizable_live_operation): Likewise.
2621 (vect_halve_mask_nunits): New function.
2622 (vect_double_mask_nunits): Likewise.
2623 (vect_record_loop_mask): Likewise.
2624 (vect_get_loop_mask): Likewise.
2625 (vect_transform_loop): Handle the case in which the final loop
2626 iteration might handle a partial vector. Call vect_set_loop_condition
2627 instead of slpeel_make_loop_iterate_ntimes.
2628 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2629 (check_load_store_masking): New function.
2630 (prepare_load_store_mask): Likewise.
2631 (vectorizable_store): Handle fully-masked loops.
2632 (vectorizable_load): Likewise.
2633 (supportable_widening_operation): Use vect_halve_mask_nunits for
2635 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2636 (vect_gen_while): New function.
2637 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2638 (aarch64_uqdec<mode>): New insn.
2640 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2641 Alan Hayward <alan.hayward@arm.com>
2642 David Sherwood <david.sherwood@arm.com>
2644 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2645 (reduc_xor_scal_optab): New optabs.
2646 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2647 (reduc_xor_scal_@var{m}): Document.
2648 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2649 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2651 * fold-const-call.c (fold_const_call): Handle them.
2652 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2653 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2654 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2655 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2656 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2657 (UNSPEC_XORV): New unspecs.
2658 (optab): Add entries for them.
2659 (BITWISEV): New int iterator.
2660 (bit_reduc_op): New int attributes.
2662 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2663 Alan Hayward <alan.hayward@arm.com>
2664 David Sherwood <david.sherwood@arm.com>
2666 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2667 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2668 * optabs.def (vec_shl_insert_optab): New optab.
2669 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2670 (duplicate_and_interleave): Likewise.
2671 * tree-vect-loop.c: Include internal-fn.h.
2672 (neutral_op_for_slp_reduction): New function, split out from
2673 get_initial_defs_for_reduction.
2674 (get_initial_def_for_reduction): Handle option 2 for variable-length
2675 vectors by loading the neutral value into a vector and then shifting
2676 the initial value into element 0.
2677 (get_initial_defs_for_reduction): Replace the code argument with
2678 the neutral value calculated by neutral_op_for_slp_reduction.
2679 Use gimple_build_vector for constant-length vectors.
2680 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2681 but the first group_size elements have a neutral value.
2682 Use duplicate_and_interleave otherwise.
2683 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2684 Update call to get_initial_defs_for_reduction. Handle SLP
2685 reductions for variable-length vectors by creating one vector
2686 result for each scalar result, with the elements associated
2687 with other scalar results stubbed out with the neutral value.
2688 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2689 Require IFN_VEC_SHL_INSERT for double reductions on
2690 variable-length vectors, or SLP reductions that have
2691 a neutral value. Require can_duplicate_and_interleave_p
2692 support for variable-length unchained SLP reductions if there
2693 is no neutral value, such as for MIN/MAX reductions. Also require
2694 the number of vector elements to be a multiple of the number of
2695 SLP statements when doing variable-length unchained SLP reductions.
2696 Update call to vect_create_epilog_for_reduction.
2697 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2698 and remove initial values.
2699 (duplicate_and_interleave): Make public.
2700 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2701 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2704 Alan Hayward <alan.hayward@arm.com>
2705 David Sherwood <david.sherwood@arm.com>
2707 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2708 (can_duplicate_and_interleave_p): New function.
2709 (vect_get_and_check_slp_defs): Take the vector of statements
2710 rather than just the current one. Remove excess parentheses.
2711 Restriction rejectinon of vect_constant_def and vect_external_def
2712 for variable-length vectors to boolean types, or types for which
2713 can_duplicate_and_interleave_p is false.
2714 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2715 (duplicate_and_interleave): New function.
2716 (vect_get_constant_vectors): Use gimple_build_vector for
2717 constant-length vectors and suitable variable-length constant
2718 vectors. Use duplicate_and_interleave for other variable-length
2719 vectors. Don't defer the update when inserting new statements.
2721 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2722 Alan Hayward <alan.hayward@arm.com>
2723 David Sherwood <david.sherwood@arm.com>
2725 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2726 min_profitable_iters doesn't go negative.
2728 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2729 Alan Hayward <alan.hayward@arm.com>
2730 David Sherwood <david.sherwood@arm.com>
2732 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2733 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2734 * optabs.def (vec_mask_load_lanes_optab): New optab.
2735 (vec_mask_store_lanes_optab): Likewise.
2736 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2737 (MASK_STORE_LANES): Likewise.
2738 * internal-fn.c (mask_load_lanes_direct): New macro.
2739 (mask_store_lanes_direct): Likewise.
2740 (expand_mask_load_optab_fn): Handle masked operations.
2741 (expand_mask_load_lanes_optab_fn): New macro.
2742 (expand_mask_store_optab_fn): Handle masked operations.
2743 (expand_mask_store_lanes_optab_fn): New macro.
2744 (direct_mask_load_lanes_optab_supported_p): Likewise.
2745 (direct_mask_store_lanes_optab_supported_p): Likewise.
2746 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2748 (vect_load_lanes_supported): Likewise.
2749 * tree-vect-data-refs.c (strip_conversion): New function.
2750 (can_group_stmts_p): Likewise.
2751 (vect_analyze_data_ref_accesses): Use it instead of checking
2752 for a pair of assignments.
2753 (vect_store_lanes_supported): Take a masked_p parameter.
2754 (vect_load_lanes_supported): Likewise.
2755 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2756 vect_store_lanes_supported and vect_load_lanes_supported.
2757 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2758 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2759 parameter. Don't allow gaps for masked accesses.
2760 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2761 and vect_load_lanes_supported.
2762 (get_load_store_type): Take a masked_p parameter and update
2763 call to get_group_load_store_type.
2764 (vectorizable_store): Update call to get_load_store_type.
2765 Handle IFN_MASK_STORE_LANES.
2766 (vectorizable_load): Update call to get_load_store_type.
2767 Handle IFN_MASK_LOAD_LANES.
2769 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2770 Alan Hayward <alan.hayward@arm.com>
2771 David Sherwood <david.sherwood@arm.com>
2773 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2775 * config/aarch64/aarch64-protos.h
2776 (aarch64_sve_struct_memory_operand_p): Declare.
2777 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2778 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2779 (VPRED, vpred): Handle SVE structure modes.
2780 * config/aarch64/constraints.md (Utx): New constraint.
2781 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2782 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2783 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2784 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2785 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2786 structure modes. Split into pieces after RA.
2787 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2788 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2790 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2791 SVE structure modes.
2792 (aarch64_classify_address): Likewise.
2793 (sizetochar): Move earlier in file.
2794 (aarch64_print_operand): Handle SVE register lists.
2795 (aarch64_array_mode): New function.
2796 (aarch64_sve_struct_memory_operand_p): Likewise.
2797 (TARGET_ARRAY_MODE): Redefine.
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 * target.def (array_mode): New target hook.
2804 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2805 * doc/tm.texi: Regenerate.
2806 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2807 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2808 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2810 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2813 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2814 Alan Hayward <alan.hayward@arm.com>
2815 David Sherwood <david.sherwood@arm.com>
2817 * fold-const.c (fold_binary_loc): Check the argument types
2818 rather than the result type when testing for a vector operation.
2820 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2822 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2823 * doc/tm.texi: Regenerate.
2825 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2826 Alan Hayward <alan.hayward@arm.com>
2827 David Sherwood <david.sherwood@arm.com>
2829 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2830 (sve): Document new AArch64 extension.
2831 * doc/md.texi (w): Extend the description of the AArch64
2832 constraint to include SVE vectors.
2833 (Upl, Upa): Document new AArch64 predicate constraints.
2834 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2836 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2837 (msve-vector-bits=): New option.
2838 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2839 SVE when these are disabled.
2840 (sve): New extension.
2841 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2842 modes. Adjust their number of units based on aarch64_sve_vg.
2843 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2844 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2845 aarch64_addr_query_type.
2846 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2847 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2848 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2849 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2850 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2851 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2852 (aarch64_simd_imm_zero_p): Delete.
2853 (aarch64_check_zero_based_sve_index_immediate): Declare.
2854 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2855 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2856 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2857 (aarch64_sve_float_mul_immediate_p): Likewise.
2858 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2860 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2861 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2862 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2863 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2864 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2865 (aarch64_regmode_natural_size): Likewise.
2866 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2867 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2869 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2870 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2871 for VG and the SVE predicate registers.
2872 (V_ALIASES): Add a "z"-prefixed alias.
2873 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2874 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2875 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2876 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2877 (REG_CLASS_NAMES): Add entries for them.
2878 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2879 and the predicate registers.
2880 (aarch64_sve_vg): Declare.
2881 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2882 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2883 (REGMODE_NATURAL_SIZE): Define.
2884 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2886 * config/aarch64/aarch64.c: Include cfgrtl.h.
2887 (simd_immediate_info): Add a constructor for series vectors,
2888 and an associated step field.
2889 (aarch64_sve_vg): New variable.
2890 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2891 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2892 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2893 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2894 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2895 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2896 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2897 (aarch64_get_mask_mode): New functions.
2898 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2899 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2900 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2901 predicate modes and predicate registers. Explicitly restrict
2902 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2903 to store a vector mode if it is recognized by
2904 aarch64_classify_vector_mode.
2905 (aarch64_regmode_natural_size): New function.
2906 (aarch64_hard_regno_caller_save_mode): Return the original mode
2908 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2909 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2910 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2911 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2913 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2914 does not overlap dest if the function is frame-related. Handle
2916 (aarch64_split_add_offset): New function.
2917 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2918 them aarch64_add_offset.
2919 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2920 and update call to aarch64_sub_sp.
2921 (aarch64_add_cfa_expression): New function.
2922 (aarch64_expand_prologue): Pass extra temporary registers to the
2923 functions above. Handle the case in which we need to emit new
2924 DW_CFA_expressions for registers that were originally saved
2925 relative to the stack pointer, but now have to be expressed
2926 relative to the frame pointer.
2927 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2929 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2930 IP0 and IP1 values for SVE frames.
2931 (aarch64_expand_vec_series): New function.
2932 (aarch64_expand_sve_widened_duplicate): Likewise.
2933 (aarch64_expand_sve_const_vector): Likewise.
2934 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2935 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2936 into the register, rather than emitting a SET directly.
2937 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2938 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2939 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2940 (offset_9bit_signed_scaled_p): New functions.
2941 (aarch64_replicate_bitmask_imm): New function.
2942 (aarch64_bitmask_imm): Use it.
2943 (aarch64_cannot_force_const_mem): Reject expressions involving
2944 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2945 (aarch64_classify_index): Handle SVE indices, by requiring
2946 a plain register index with a scale that matches the element size.
2947 (aarch64_classify_address): Handle SVE addresses. Assert that
2948 the mode of the address is VOIDmode or an integer mode.
2949 Update call to aarch64_classify_symbol.
2950 (aarch64_classify_symbolic_expression): Update call to
2951 aarch64_classify_symbol.
2952 (aarch64_const_vec_all_in_range_p): New function.
2953 (aarch64_print_vector_float_operand): Likewise.
2954 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2955 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2956 and the FP immediates 1.0 and 0.5.
2957 (aarch64_print_address_internal): Handle SVE addresses.
2958 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2959 (aarch64_regno_regclass): Handle predicate registers.
2960 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2962 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2963 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2964 (aarch64_convert_sve_vector_bits): New function.
2965 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2966 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2968 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2969 Handle SVE vector and predicate modes. Accept VL-based constants
2970 that need only one temporary register, and VL offsets that require
2971 no temporary registers.
2972 (aarch64_conditional_register_usage): Mark the predicate registers
2973 as fixed if SVE isn't available.
2974 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2975 Return true for SVE vector and predicate modes.
2976 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2977 rather than an unsigned int. Handle SVE modes.
2978 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2980 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2982 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2983 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2984 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2985 (aarch64_sve_float_mul_immediate_p): New functions.
2986 (aarch64_sve_valid_immediate): New function.
2987 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2988 Explicitly reject structure modes. Check for INDEX constants.
2989 Handle PTRUE and PFALSE constants.
2990 (aarch64_check_zero_based_sve_index_immediate): New function.
2991 (aarch64_simd_imm_zero_p): Delete.
2992 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2993 vector modes. Accept constants in the range of CNT[BHWD].
2994 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2995 ask for an Advanced SIMD mode.
2996 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2997 (aarch64_simd_vector_alignment): Handle SVE predicates.
2998 (aarch64_vectorize_preferred_vector_alignment): New function.
2999 (aarch64_simd_vector_alignment_reachable): Use it instead of
3001 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3002 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3004 (MAX_VECT_LEN): Delete.
3005 (expand_vec_perm_d): Add a vec_flags field.
3006 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3007 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3008 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3010 (aarch64_evpc_rev): Rename to...
3011 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3012 (aarch64_evpc_rev_global): New function.
3013 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3014 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3016 (aarch64_evpc_sve_tbl): New function.
3017 (aarch64_expand_vec_perm_const_1): Update after rename of
3018 aarch64_evpc_rev. Handle SVE permutes too, trying
3019 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3020 than aarch64_evpc_tbl.
3021 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3022 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3023 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3024 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3025 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3026 (aarch64_expand_sve_vcond): New functions.
3027 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3028 of aarch64_vector_mode_p.
3029 (aarch64_dwarf_poly_indeterminate_value): New function.
3030 (aarch64_compute_pressure_classes): Likewise.
3031 (aarch64_can_change_mode_class): Likewise.
3032 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3033 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3034 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3035 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3036 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3037 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3038 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3039 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3041 (Dn, Dl, Dr): Accept const as well as const_vector.
3042 (Dz): Likewise. Compare against CONST0_RTX.
3043 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3044 of "vector" where appropriate.
3045 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3046 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3047 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3048 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3049 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3050 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3051 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3052 (v_int_equiv): Extend to SVE modes.
3053 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3055 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3056 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3057 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3058 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3059 (SVE_COND_FP_CMP): New int iterators.
3060 (perm_hilo): Handle the new unpack unspecs.
3061 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3063 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3064 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3065 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3066 (aarch64_equality_operator, aarch64_constant_vector_operand)
3067 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3068 (aarch64_sve_nonimmediate_operand): Likewise.
3069 (aarch64_sve_general_operand): Likewise.
3070 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3071 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3072 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3073 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3074 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3075 (aarch64_sve_float_arith_immediate): Likewise.
3076 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3077 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3078 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3079 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3080 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3081 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3082 (aarch64_sve_float_arith_operand): Likewise.
3083 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3084 (aarch64_sve_float_mul_operand): Likewise.
3085 (aarch64_sve_vec_perm_operand): Likewise.
3086 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3087 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3088 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3089 as well as const_vector.
3090 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3091 in file. Use CONST0_RTX and CONSTM1_RTX.
3092 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3093 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3094 Use aarch64_simd_imm_zero.
3095 * config/aarch64/aarch64-sve.md: New file.
3096 * config/aarch64/aarch64.md: Include it.
3097 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3098 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3099 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3100 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3101 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3102 (sve): New attribute.
3103 (enabled): Disable instructions with the sve attribute unless
3105 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3106 aarch64_expand_mov_immediate.
3107 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3108 CNT[BHSD] immediates.
3109 (movti): Split CONST_POLY_INT moves into two halves.
3110 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3111 Split additions that need a temporary here if the destination
3112 is the stack pointer.
3113 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3114 (*add<mode>3_poly_1): New instruction.
3115 (set_clobber_cc): New expander.
3117 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3119 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3120 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3121 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3122 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3123 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3124 Change innermode from fixed_mode_size to machine_mode.
3125 (simplify_subreg): Update call accordingly. Handle a constant-sized
3126 subreg of a variable-length CONST_VECTOR.
3128 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3129 Alan Hayward <alan.hayward@arm.com>
3130 David Sherwood <david.sherwood@arm.com>
3132 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3133 (add_offset_to_base): New function, split out from...
3134 (create_mem_ref): ...here. When handling a scale other than 1,
3135 check first whether the address is valid without the offset.
3136 Add it into the base if so, leaving the index and scale as-is.
3138 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3141 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3142 fold_for_warn before checking if arg2 is INTEGER_CST.
3144 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3146 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3147 (store_multiple_operation): Delete.
3148 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3149 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3150 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3151 guarded by TARGET_STRING.
3152 (rs6000_output_load_multiple): Delete.
3153 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3154 OPTION_MASK_STRING / TARGET_STRING handling.
3155 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3156 (const rs6000_opt_masks) <"string">: Change mask to 0.
3157 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3158 (MASK_STRING): Delete.
3159 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3161 (load_multiple): Delete.
3168 (store_multiple): Delete.
3175 (movmemsi_8reg): Delete.
3176 (corresponding unnamed define_insn): Delete.
3177 (movmemsi_6reg): Delete.
3178 (corresponding unnamed define_insn): Delete.
3179 (movmemsi_4reg): Delete.
3180 (corresponding unnamed define_insn): Delete.
3181 (movmemsi_2reg): Delete.
3182 (corresponding unnamed define_insn): Delete.
3183 (movmemsi_1reg): Delete.
3184 (corresponding unnamed define_insn): Delete.
3185 * config/rs6000/rs6000.opt (mno-string): New.
3186 (mstring): Replace by deprecation warning stub.
3187 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3189 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3191 * regrename.c (regrename_do_replace): If replacing the same
3192 reg multiple times, try to reuse last created gen_raw_REG.
3195 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3196 main to workaround a bug in GDB.
3198 2018-01-12 Tom de Vries <tom@codesourcery.com>
3201 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3203 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3205 PR rtl-optimization/80481
3206 * ira-color.c (get_cap_member): New function.
3207 (allocnos_conflict_by_live_ranges_p): Use it.
3208 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3209 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3211 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3214 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3215 (*saddl_se_1): Ditto.
3217 (*ssubl_se_1): Ditto.
3219 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3221 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3222 rather than wi::to_widest for DR_INITs.
3223 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3224 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3225 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3227 (vect_analyze_group_access_1): Note that here.
3229 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3231 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3232 polynomial type sizes.
3234 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3236 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3237 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3238 (gimple_add_tmp_var): Likewise.
3240 2018-01-12 Martin Liska <mliska@suse.cz>
3242 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3243 (gimple_alloc_sizes): Likewise.
3244 (dump_gimple_statistics): Use PRIu64 in printf format.
3245 * gimple.h: Change uint64_t to int.
3247 2018-01-12 Martin Liska <mliska@suse.cz>
3249 * tree-core.h: Use uint64_t instead of int.
3250 * tree.c (tree_node_counts): Likewise.
3251 (tree_node_sizes): Likewise.
3252 (dump_tree_statistics): Use PRIu64 in printf format.
3254 2018-01-12 Martin Liska <mliska@suse.cz>
3256 * Makefile.in: As qsort_chk is implemented in vec.c, add
3257 vec.o to linkage of gencfn-macros.
3258 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3259 passing the info to record_node_allocation_statistics.
3260 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3262 * ggc-common.c (struct ggc_usage): Add operator== and use
3263 it in operator< and compare function.
3264 * mem-stats.h (struct mem_usage): Likewise.
3265 * vec.c (struct vec_usage): Remove operator< and compare
3266 function. Can be simply inherited.
3268 2018-01-12 Martin Jambor <mjambor@suse.cz>
3271 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3272 * tree-ssa-math-opts.c: Include domwalk.h.
3273 (convert_mult_to_fma_1): New function.
3274 (fma_transformation_info): New type.
3275 (fma_deferring_state): Likewise.
3276 (cancel_fma_deferring): New function.
3277 (result_of_phi): Likewise.
3278 (last_fma_candidate_feeds_initial_phi): Likewise.
3279 (convert_mult_to_fma): Added deferring logic, split actual
3280 transformation to convert_mult_to_fma_1.
3281 (math_opts_dom_walker): New type.
3282 (math_opts_dom_walker::after_dom_children): New method, body moved
3283 here from pass_optimize_widening_mul::execute, added deferring logic
3285 (pass_optimize_widening_mul::execute): Moved most of code to
3286 math_opts_dom_walker::after_dom_children.
3287 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3288 * config/i386/i386.c (ix86_option_override_internal): Added
3289 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3291 2018-01-12 Richard Biener <rguenther@suse.de>
3294 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3295 inline instance vars.
3297 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3300 * config/rx/rx.c (rx_is_restricted_memory_address):
3303 2018-01-12 Richard Biener <rguenther@suse.de>
3305 PR tree-optimization/80846
3306 * target.def (split_reduction): New target hook.
3307 * targhooks.c (default_split_reduction): New function.
3308 * targhooks.h (default_split_reduction): Declare.
3309 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3310 target requests first reduce vectors by combining low and high
3312 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3313 (get_vectype_for_scalar_type_and_size): Export.
3314 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3315 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3316 * doc/tm.texi: Regenerate.
3317 * config/i386/i386.c (ix86_split_reduction): Implement
3318 TARGET_VECTORIZE_SPLIT_REDUCTION.
3320 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3323 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3324 in PIC mode except for TARGET_VXWORKS_RTP.
3325 * config/sparc/sparc.c: Include cfgrtl.h.
3326 (TARGET_INIT_PIC_REG): Define.
3327 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3328 (sparc_pic_register_p): New predicate.
3329 (sparc_legitimate_address_p): Use it.
3330 (sparc_legitimize_pic_address): Likewise.
3331 (sparc_delegitimize_address): Likewise.
3332 (sparc_mode_dependent_address_p): Likewise.
3333 (gen_load_pcrel_sym): Remove 4th parameter.
3334 (load_got_register): Adjust call to above. Remove obsolete stuff.
3335 (sparc_expand_prologue): Do not call load_got_register here.
3336 (sparc_flat_expand_prologue): Likewise.
3337 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3338 (sparc_use_pseudo_pic_reg): New function.
3339 (sparc_init_pic_reg): Likewise.
3340 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3341 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3343 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3345 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3346 Add item for branch_cost.
3348 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3350 PR rtl-optimization/83565
3351 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3352 not extend the result to a larger mode for rotate operations.
3353 (num_sign_bit_copies1): Likewise.
3355 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3358 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3360 Use values-Xc.o for -pedantic.
3361 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3363 2018-01-12 Martin Liska <mliska@suse.cz>
3366 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3368 (possible_polymorphic_call_targets): Use it.
3369 (ipa_devirt): Likewise.
3371 2018-01-12 Martin Liska <mliska@suse.cz>
3373 * profile-count.h (enum profile_quality): Use 0 as invalid
3374 enum value of profile_quality.
3376 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3378 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3379 -mext-string options.
3381 2018-01-12 Richard Biener <rguenther@suse.de>
3383 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3384 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3385 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3387 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3389 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3391 * configure.ac (--with-long-double-format): Add support for the
3392 configuration option to change the default long double format on
3394 * config.gcc (powerpc*-linux*-*): Likewise.
3395 * configure: Regenerate.
3396 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3397 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3398 used without modification.
3400 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3402 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3403 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3404 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3405 MISC_BUILTIN_SPEC_BARRIER.
3406 (rs6000_init_builtins): Likewise.
3407 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3409 (speculation_barrier): New define_insn.
3410 * doc/extend.texi: Document __builtin_speculation_barrier.
3412 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3415 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3416 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3417 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3419 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3420 integral modes instead of "ss" and "sd".
3421 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3422 vectors with 32-bit and 64-bit elements.
3423 (vecdupssescalarmodesuffix): New mode attribute.
3424 (vec_dup<mode>): Use it.
3426 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3429 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3430 frame if argument is passed on stack.
3432 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3435 * ree.c (combine_reaching_defs): Optimize also
3436 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3437 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3439 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3442 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3444 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3447 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3448 after they are computed.
3450 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3452 PR tree-optimization/83695
3453 * gimple-loop-linterchange.cc
3454 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3455 reset cached scev information after interchange.
3456 (pass_linterchange::execute): Remove call to scev_reset_htab.
3458 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3460 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3461 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3462 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3463 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3464 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3465 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3466 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3467 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3468 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3469 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3470 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3471 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3472 (V_lane_reg): Likewise.
3473 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3475 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3476 (vfmal_lane_low<mode>_intrinsic,
3477 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3478 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3479 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3480 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3481 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3482 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3484 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3486 * config/arm/arm-cpus.in (fp16fml): New feature.
3487 (ALL_SIMD): Add fp16fml.
3488 (armv8.2-a): Add fp16fml as an option.
3489 (armv8.3-a): Likewise.
3490 (armv8.4-a): Add fp16fml as part of fp16.
3491 * config/arm/arm.h (TARGET_FP16FML): Define.
3492 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3494 * config/arm/arm-modes.def (V2HF): Define.
3495 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3496 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3497 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3498 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3499 vfmsl_low, vfmsl_high): New set of builtins.
3500 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3501 (vfml_op): New code attribute.
3502 (VFMLHALVES): New int iterator.
3503 (VFML, VFMLSEL): New mode attributes.
3504 (V_reg): Define mapping for V2HF.
3505 (V_hi, V_lo): New mode attributes.
3506 (VF_constraint): Likewise.
3507 (vfml_half, vfml_half_selector): New int attributes.
3508 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3510 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3511 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3513 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3514 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3515 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3516 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3518 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3519 Document new effective target and option set.
3521 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3523 * config/arm/arm-cpus.in (armv8_4): New feature.
3524 (ARMv8_4a): New fgroup.
3525 (armv8.4-a): New arch.
3526 * config/arm/arm-tables.opt: Regenerate.
3527 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3528 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3529 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3530 Add matching rules for -march=armv8.4-a and extensions.
3531 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3533 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3536 * config/rx/rx.md (BW): New mode attribute.
3537 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3539 2018-01-11 Richard Biener <rguenther@suse.de>
3541 PR tree-optimization/83435
3542 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3543 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3544 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3546 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3547 Alan Hayward <alan.hayward@arm.com>
3548 David Sherwood <david.sherwood@arm.com>
3550 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3552 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3553 (aarch64_print_address_internal): Use it to check for a zero offset.
3555 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3556 Alan Hayward <alan.hayward@arm.com>
3557 David Sherwood <david.sherwood@arm.com>
3559 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3560 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3561 Return a poly_int64 rather than a HOST_WIDE_INT.
3562 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3563 rather than a HOST_WIDE_INT.
3564 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3565 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3566 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3567 final_offset from HOST_WIDE_INT to poly_int64.
3568 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3569 to_constant when getting the number of units in an Advanced SIMD
3571 (aarch64_builtin_vectorized_function): Check for a constant number
3573 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3575 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3576 attribute instead of GET_MODE_NUNITS.
3577 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3578 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3579 GET_MODE_SIZE for fixed-size registers.
3580 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3581 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3582 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3583 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3584 (aarch64_print_operand, aarch64_print_address_internal)
3585 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3586 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3587 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3588 Handle polynomial GET_MODE_SIZE.
3589 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3590 wider than SImode without modification.
3591 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3592 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3593 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3594 passing and returning SVE modes.
3595 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3596 rather than GEN_INT.
3597 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3598 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3599 (aarch64_allocate_and_probe_stack_space): Likewise.
3600 (aarch64_layout_frame): Cope with polynomial offsets.
3601 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3602 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3604 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3605 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3606 poly_int64 rather than a HOST_WIDE_INT.
3607 (aarch64_get_separate_components, aarch64_process_components)
3608 (aarch64_expand_prologue, aarch64_expand_epilogue)
3609 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3610 (aarch64_anchor_offset): New function, split out from...
3611 (aarch64_legitimize_address): ...here.
3612 (aarch64_builtin_vectorization_cost): Handle polynomial
3613 TYPE_VECTOR_SUBPARTS.
3614 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3616 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3617 number of elements from the PARALLEL rather than the mode.
3618 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3619 rather than GET_MODE_BITSIZE.
3620 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3621 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3622 (aarch64_expand_vec_perm_const_1): Handle polynomial
3623 d->perm.length () and d->perm elements.
3624 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3625 Apply to_constant to d->perm elements.
3626 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3627 polynomial CONST_VECTOR_NUNITS.
3628 (aarch64_move_pointer): Take amount as a poly_int64 rather
3630 (aarch64_progress_pointer): Avoid temporary variable.
3631 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3632 the mode attribute instead of GET_MODE.
3634 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3635 Alan Hayward <alan.hayward@arm.com>
3636 David Sherwood <david.sherwood@arm.com>
3638 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3639 x exists before using it.
3640 (aarch64_add_constant_internal): Rename to...
3641 (aarch64_add_offset_1): ...this. Replace regnum with separate
3642 src and dest rtxes. Handle the case in which they're different,
3643 including when the offset is zero. Replace scratchreg with an rtx.
3644 Use 2 additions if there is no spare register into which we can
3645 move a 16-bit constant.
3646 (aarch64_add_constant): Delete.
3647 (aarch64_add_offset): Replace reg with separate src and dest
3648 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3649 Use aarch64_add_offset_1.
3650 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3651 an rtx rather than an int. Take the delta as a poly_int64
3652 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3653 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3654 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3655 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3656 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3658 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3659 aarch64_add_constant.
3661 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3663 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3664 Use scalar_float_mode.
3666 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3668 * config/aarch64/aarch64-simd.md
3669 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3670 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3671 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3672 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3673 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3674 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3675 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3676 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3677 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3678 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3680 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3683 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3684 targ_options->x_arm_arch_string is non NULL.
3686 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3688 * config/aarch64/aarch64.h
3689 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3691 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3694 * expmed.c (emit_store_flag_force): Swap if const op0
3695 and change VOIDmode to mode of op0.
3697 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3699 PR rtl-optimization/83761
3700 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3701 than bytes to mode_for_size.
3703 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3706 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3707 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3710 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3713 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3714 when in layout mode.
3715 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3716 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3719 2018-01-10 Michael Collison <michael.collison@arm.com>
3721 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3722 * config/aarch64/aarch64-option-extension.def: Add
3723 AARCH64_OPT_EXTENSION of 'fp16fml'.
3724 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3725 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3726 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3727 * config/aarch64/constraints.md (Ui7): New constraint.
3728 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3729 (VFMLA_SEL_W): Ditto.
3732 (VFMLA16_LOW): New int iterator.
3733 (VFMLA16_HIGH): Ditto.
3734 (UNSPEC_FMLAL): New unspec.
3735 (UNSPEC_FMLSL): Ditto.
3736 (UNSPEC_FMLAL2): Ditto.
3737 (UNSPEC_FMLSL2): Ditto.
3738 (f16mac): New code attribute.
3739 * config/aarch64/aarch64-simd-builtins.def
3740 (aarch64_fmlal_lowv2sf): Ditto.
3741 (aarch64_fmlsl_lowv2sf): Ditto.
3742 (aarch64_fmlalq_lowv4sf): Ditto.
3743 (aarch64_fmlslq_lowv4sf): Ditto.
3744 (aarch64_fmlal_highv2sf): Ditto.
3745 (aarch64_fmlsl_highv2sf): Ditto.
3746 (aarch64_fmlalq_highv4sf): Ditto.
3747 (aarch64_fmlslq_highv4sf): Ditto.
3748 (aarch64_fmlal_lane_lowv2sf): Ditto.
3749 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3750 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3751 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3752 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3753 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3754 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3755 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3756 (aarch64_fmlal_lane_highv2sf): Ditto.
3757 (aarch64_fmlsl_lane_highv2sf): Ditto.
3758 (aarch64_fmlal_laneq_highv2sf): Ditto.
3759 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3760 (aarch64_fmlalq_lane_highv4sf): Ditto.
3761 (aarch64_fmlsl_lane_highv4sf): Ditto.
3762 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3763 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3764 * config/aarch64/aarch64-simd.md:
3765 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3766 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3767 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3768 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3769 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3770 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3771 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3772 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3773 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3774 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3775 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3776 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3777 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3778 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3779 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3780 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3781 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3782 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3783 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3784 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3785 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3786 (vfmlsl_low_u32): Ditto.
3787 (vfmlalq_low_u32): Ditto.
3788 (vfmlslq_low_u32): Ditto.
3789 (vfmlal_high_u32): Ditto.
3790 (vfmlsl_high_u32): Ditto.
3791 (vfmlalq_high_u32): Ditto.
3792 (vfmlslq_high_u32): Ditto.
3793 (vfmlal_lane_low_u32): Ditto.
3794 (vfmlsl_lane_low_u32): Ditto.
3795 (vfmlal_laneq_low_u32): Ditto.
3796 (vfmlsl_laneq_low_u32): Ditto.
3797 (vfmlalq_lane_low_u32): Ditto.
3798 (vfmlslq_lane_low_u32): Ditto.
3799 (vfmlalq_laneq_low_u32): Ditto.
3800 (vfmlslq_laneq_low_u32): Ditto.
3801 (vfmlal_lane_high_u32): Ditto.
3802 (vfmlsl_lane_high_u32): Ditto.
3803 (vfmlal_laneq_high_u32): Ditto.
3804 (vfmlsl_laneq_high_u32): Ditto.
3805 (vfmlalq_lane_high_u32): Ditto.
3806 (vfmlslq_lane_high_u32): Ditto.
3807 (vfmlalq_laneq_high_u32): Ditto.
3808 (vfmlslq_laneq_high_u32): Ditto.
3809 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3810 (AARCH64_FL_FOR_ARCH8_4): New.
3811 (AARCH64_ISA_F16FML): New ISA flag.
3812 (TARGET_F16FML): New feature flag for fp16fml.
3813 (doc/invoke.texi): Document new fp16fml option.
3815 2018-01-10 Michael Collison <michael.collison@arm.com>
3817 * config/aarch64/aarch64-builtins.c:
3818 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3819 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3820 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3821 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3822 (AARCH64_ISA_SHA3): New ISA flag.
3823 (TARGET_SHA3): New feature flag for sha3.
3824 * config/aarch64/iterators.md (sha512_op): New int attribute.
3825 (CRYPTO_SHA512): New int iterator.
3826 (UNSPEC_SHA512H): New unspec.
3827 (UNSPEC_SHA512H2): Ditto.
3828 (UNSPEC_SHA512SU0): Ditto.
3829 (UNSPEC_SHA512SU1): Ditto.
3830 * config/aarch64/aarch64-simd-builtins.def
3831 (aarch64_crypto_sha512hqv2di): New builtin.
3832 (aarch64_crypto_sha512h2qv2di): Ditto.
3833 (aarch64_crypto_sha512su0qv2di): Ditto.
3834 (aarch64_crypto_sha512su1qv2di): Ditto.
3835 (aarch64_eor3qv8hi): Ditto.
3836 (aarch64_rax1qv2di): Ditto.
3837 (aarch64_xarqv2di): Ditto.
3838 (aarch64_bcaxqv8hi): Ditto.
3839 * config/aarch64/aarch64-simd.md:
3840 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3841 (aarch64_crypto_sha512su0qv2di): Ditto.
3842 (aarch64_crypto_sha512su1qv2di): Ditto.
3843 (aarch64_eor3qv8hi): Ditto.
3844 (aarch64_rax1qv2di): Ditto.
3845 (aarch64_xarqv2di): Ditto.
3846 (aarch64_bcaxqv8hi): Ditto.
3847 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3848 (vsha512h2q_u64): Ditto.
3849 (vsha512su0q_u64): Ditto.
3850 (vsha512su1q_u64): Ditto.
3851 (veor3q_u16): Ditto.
3852 (vrax1q_u64): Ditto.
3854 (vbcaxq_u16): Ditto.
3855 * config/arm/types.md (crypto_sha512): New type attribute.
3856 (crypto_sha3): Ditto.
3857 (doc/invoke.texi): Document new sha3 option.
3859 2018-01-10 Michael Collison <michael.collison@arm.com>
3861 * config/aarch64/aarch64-builtins.c:
3862 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3863 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3864 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3865 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3866 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3867 (AARCH64_ISA_SM4): New ISA flag.
3868 (TARGET_SM4): New feature flag for sm4.
3869 * config/aarch64/aarch64-simd-builtins.def
3870 (aarch64_sm3ss1qv4si): Ditto.
3871 (aarch64_sm3tt1aq4si): Ditto.
3872 (aarch64_sm3tt1bq4si): Ditto.
3873 (aarch64_sm3tt2aq4si): Ditto.
3874 (aarch64_sm3tt2bq4si): Ditto.
3875 (aarch64_sm3partw1qv4si): Ditto.
3876 (aarch64_sm3partw2qv4si): Ditto.
3877 (aarch64_sm4eqv4si): Ditto.
3878 (aarch64_sm4ekeyqv4si): Ditto.
3879 * config/aarch64/aarch64-simd.md:
3880 (aarch64_sm3ss1qv4si): Ditto.
3881 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3882 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3883 (aarch64_sm4eqv4si): Ditto.
3884 (aarch64_sm4ekeyqv4si): Ditto.
3885 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3886 (sm3part_op): Ditto.
3887 (CRYPTO_SM3TT): Ditto.
3888 (CRYPTO_SM3PART): Ditto.
3889 (UNSPEC_SM3SS1): New unspec.
3890 (UNSPEC_SM3TT1A): Ditto.
3891 (UNSPEC_SM3TT1B): Ditto.
3892 (UNSPEC_SM3TT2A): Ditto.
3893 (UNSPEC_SM3TT2B): Ditto.
3894 (UNSPEC_SM3PARTW1): Ditto.
3895 (UNSPEC_SM3PARTW2): Ditto.
3896 (UNSPEC_SM4E): Ditto.
3897 (UNSPEC_SM4EKEY): Ditto.
3898 * config/aarch64/constraints.md (Ui2): New constraint.
3899 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3900 * config/arm/types.md (crypto_sm3): New type attribute.
3901 (crypto_sm4): Ditto.
3902 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3903 (vsm3tt1aq_u32): Ditto.
3904 (vsm3tt1bq_u32): Ditto.
3905 (vsm3tt2aq_u32): Ditto.
3906 (vsm3tt2bq_u32): Ditto.
3907 (vsm3partw1q_u32): Ditto.
3908 (vsm3partw2q_u32): Ditto.
3909 (vsm4eq_u32): Ditto.
3910 (vsm4ekeyq_u32): Ditto.
3911 (doc/invoke.texi): Document new sm4 option.
3913 2018-01-10 Michael Collison <michael.collison@arm.com>
3915 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3916 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3917 (AARCH64_FL_FOR_ARCH8_4): New.
3918 (AARCH64_FL_V8_4): New flag.
3919 (doc/invoke.texi): Document new armv8.4-a option.
3921 2018-01-10 Michael Collison <michael.collison@arm.com>
3923 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3924 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3925 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3926 * config/aarch64/aarch64-option-extension.def: Add
3927 AARCH64_OPT_EXTENSION of 'sha2'.
3928 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3929 (crypto): Disable sha2 and aes if crypto disabled.
3930 (crypto): Enable aes and sha2 if enabled.
3931 (simd): Disable sha2 and aes if simd disabled.
3932 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3934 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3935 (TARGET_SHA2): New feature flag for sha2.
3936 (TARGET_AES): New feature flag for aes.
3937 * config/aarch64/aarch64-simd.md:
3938 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3939 conditional on TARGET_AES.
3940 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3941 (aarch64_crypto_sha1hsi): Make pattern conditional
3943 (aarch64_crypto_sha1hv4si): Ditto.
3944 (aarch64_be_crypto_sha1hv4si): Ditto.
3945 (aarch64_crypto_sha1su1v4si): Ditto.
3946 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3947 (aarch64_crypto_sha1su0v4si): Ditto.
3948 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3949 (aarch64_crypto_sha256su0v4si): Ditto.
3950 (aarch64_crypto_sha256su1v4si): Ditto.
3951 (doc/invoke.texi): Document new aes and sha2 options.
3953 2018-01-10 Martin Sebor <msebor@redhat.com>
3955 PR tree-optimization/83781
3956 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3959 2018-01-11 Martin Sebor <msebor@gmail.com>
3960 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3962 PR tree-optimization/83501
3963 PR tree-optimization/81703
3965 * tree-ssa-strlen.c (get_string_cst): Rename...
3966 (get_string_len): ...to this. Handle global constants.
3967 (handle_char_store): Adjust.
3969 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3970 Jim Wilson <jimw@sifive.com>
3972 * config/riscv/riscv-protos.h (riscv_output_return): New.
3973 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3974 (riscv_attribute_table, riscv_output_return),
3975 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3976 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3977 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3979 (riscv_expand_prologue): Add early return for naked function.
3980 (riscv_expand_epilogue): Likewise.
3981 (riscv_function_ok_for_sibcall): Return false for naked function.
3982 (riscv_set_current_function): New.
3983 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3984 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3985 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3986 * doc/extend.texi (RISC-V Function Attributes): New.
3988 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3990 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3991 check for 128-bit long double before checking TCmode.
3992 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3993 128-bit long doubles before checking TFmode or TCmode.
3994 (FLOAT128_IBM_P): Likewise.
3996 2018-01-10 Martin Sebor <msebor@redhat.com>
3998 PR tree-optimization/83671
3999 * builtins.c (c_strlen): Unconditionally return zero for the empty
4001 Use -Warray-bounds for warnings.
4002 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4003 for non-constant array indices with COMPONENT_REF, arrays of
4004 arrays, and pointers to arrays.
4005 (gimple_fold_builtin_strlen): Determine and set length range for
4006 non-constant character arrays.
4008 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4011 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4014 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4016 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4018 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4021 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4022 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4023 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4024 indexed_or_indirect_operand predicate.
4025 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4026 (*vsx_le_perm_load_v8hi): Likewise.
4027 (*vsx_le_perm_load_v16qi): Likewise.
4028 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4029 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4030 (*vsx_le_perm_store_v8hi): Likewise.
4031 (*vsx_le_perm_store_v16qi): Likewise.
4032 (eight unnamed splitters): Likewise.
4034 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4036 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4037 * config/rs6000/emmintrin.h: Likewise.
4038 * config/rs6000/mmintrin.h: Likewise.
4039 * config/rs6000/xmmintrin.h: Likewise.
4041 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4044 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4046 * tree.c (tree_nop_conversion): Return true for location wrapper
4048 (maybe_wrap_with_location): New function.
4049 (selftest::check_strip_nops): New function.
4050 (selftest::test_location_wrappers): New function.
4051 (selftest::tree_c_tests): Call it.
4052 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4053 (maybe_wrap_with_location): New decl.
4054 (EXPR_LOCATION_WRAPPER_P): New macro.
4055 (location_wrapper_p): New inline function.
4056 (tree_strip_any_location_wrapper): New inline function.
4058 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4061 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4062 stack_realign_offset for the largest alignment of stack slot
4064 (ix86_find_max_used_stack_alignment): New function.
4065 (ix86_finalize_stack_frame_flags): Use it. Set
4066 max_used_stack_alignment if we don't realign stack.
4067 * config/i386/i386.h (machine_function): Add
4068 max_used_stack_alignment.
4070 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4072 * config/arm/arm.opt (-mbranch-cost): New option.
4073 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4076 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4079 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4080 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4082 2018-01-10 Richard Biener <rguenther@suse.de>
4085 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4086 early out so it also covers the case where we have a non-NULL
4089 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4091 PR tree-optimization/83753
4092 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4093 for non-strided grouped accesses if the number of elements is 1.
4095 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4098 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4099 * i386.h (TARGET_USE_GATHER): Define.
4100 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4102 2018-01-10 Martin Liska <mliska@suse.cz>
4105 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4106 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4108 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4109 CLEANUP_NO_PARTITIONING is not set.
4111 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4113 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4114 for vectors, as a partial revert of r254296.
4115 * rtl.h (const_vec_p): Delete.
4116 (const_vec_duplicate_p): Don't test for vector CONSTs.
4117 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4118 * expmed.c (make_tree): Likewise.
4121 * common.md (E, F): Use CONSTANT_P instead of checking for
4123 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4124 checking for CONST_VECTOR.
4126 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4129 * predict.c (force_edge_cold): Handle in more sane way edges
4132 2018-01-09 Carl Love <cel@us.ibm.com>
4134 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4136 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4137 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4138 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4139 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4140 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4141 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4142 * config/rs6000/rs6000-protos.h: Add extern defition for
4143 rs6000_generate_float2_double_code.
4144 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4146 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4147 (float2_v2df): Add define_expand.
4149 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4152 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4153 op_mode in the force_to_mode call.
4155 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4157 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4158 instead of checking each element individually.
4159 (aarch64_evpc_uzp): Likewise.
4160 (aarch64_evpc_zip): Likewise.
4161 (aarch64_evpc_ext): Likewise.
4162 (aarch64_evpc_rev): Likewise.
4163 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4164 instead of checking each element individually. Return true without
4166 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4167 whether all selected elements come from the same input, instead of
4168 checking each element individually. Remove calls to gen_rtx_REG,
4169 start_sequence and end_sequence and instead assert that no rtl is
4172 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4174 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4175 order of HIGH and CONST checks.
4177 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4179 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4180 if the destination isn't an SSA_NAME.
4182 2018-01-09 Richard Biener <rguenther@suse.de>
4184 PR tree-optimization/83668
4185 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4187 (canonicalize_loop_form): ... here, renamed from ...
4188 (canonicalize_loop_closed_ssa_form): ... this and amended to
4189 swap successor edges for loop exit blocks to make us use
4190 the RPO order we need for initial schedule generation.
4192 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4194 PR tree-optimization/64811
4195 * match.pd: When optimizing comparisons with Inf, avoid
4196 introducing or losing exceptions from comparisons with NaN.
4198 2018-01-09 Martin Liska <mliska@suse.cz>
4201 * asan.c (shadow_mem_size): Add gcc_assert.
4203 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4205 Don't save registers in main().
4208 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4209 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4210 * config/avr/avr.c (avr_set_current_function): Don't error if
4211 naked, OS_task or OS_main are specified at the same time.
4212 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4214 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4216 * common/config/avr/avr-common.c (avr_option_optimization_table):
4217 Switch on -mmain-is-OS_task for optimizing compilations.
4219 2018-01-09 Richard Biener <rguenther@suse.de>
4221 PR tree-optimization/83572
4222 * graphite.c: Include cfganal.h.
4223 (graphite_transform_loops): Connect infinite loops to exit
4224 and remove fake edges at the end.
4226 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4228 * ipa-inline.c (edge_badness): Revert accidental checkin.
4230 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4233 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4234 symbols; not inline clones.
4236 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4239 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4240 hard registers. Formatting fixes.
4242 PR preprocessor/83722
4243 * gcc.c (try_generate_repro): Pass
4244 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4245 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4248 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4249 Kito Cheng <kito.cheng@gmail.com>
4251 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4252 (riscv_leaf_function_p): Delete.
4253 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4255 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4257 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4259 (do_ifelse): New function.
4260 (do_isel): New function.
4261 (do_sub3): New function.
4262 (do_add3): New function.
4263 (do_load_mask_compare): New function.
4264 (do_overlap_load_compare): New function.
4265 (expand_compare_loop): New function.
4266 (expand_block_compare): Call expand_compare_loop() when appropriate.
4267 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4269 (-mblock-compare-inline-loop-limit): New option.
4271 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4274 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4275 Reverse order of second and third operands in first alternative.
4276 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4277 of first and second elements in UNSPEC_VPERMR vector.
4278 (altivec_expand_vec_perm_le): Likewise.
4280 2018-01-08 Jeff Law <law@redhat.com>
4282 PR rtl-optimizatin/81308
4283 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4284 (process_switch): If group_case_labels makes a change, then set
4286 (pass_convert_switch::execute): If a switch is converted, then
4287 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4289 PR rtl-optimization/81308
4290 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4293 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4295 PR target/83663 - Revert r255946
4296 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4297 generation for cases where splatting a value is not useful.
4298 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4299 across a vec_duplicate and a paradoxical subreg forming a vector
4300 mode to a vec_concat.
4302 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4304 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4305 -march=armv8.3-a variants.
4306 * config/arm/t-multilib: Likewise.
4307 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4309 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4311 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4313 (cceq_ior_compare_complement): Give it a name so I can use it, and
4314 change boolean_or_operator predicate to boolean_operator so it can
4315 be used to generate a crand.
4316 (eqne): New code iterator.
4317 (bd/bd_neg): New code_attrs.
4318 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4319 a single define_insn.
4320 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4321 decrement (bdnzt/bdnzf/bdzt/bdzf).
4322 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4323 with the new names of the branch decrement patterns, and added the
4324 names of the branch decrement conditional patterns.
4326 2018-01-08 Richard Biener <rguenther@suse.de>
4328 PR tree-optimization/83563
4329 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4332 2018-01-08 Richard Biener <rguenther@suse.de>
4335 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4337 2018-01-08 Richard Biener <rguenther@suse.de>
4339 PR tree-optimization/83685
4340 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4341 references to abnormals.
4343 2018-01-08 Richard Biener <rguenther@suse.de>
4346 * dwarf2out.c (output_indirect_strings): Handle empty
4347 skeleton_debug_str_hash.
4348 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4350 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4352 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4353 (emit_store_direct): Likewise.
4354 (arc_trampoline_adjust_address): Likewise.
4355 (arc_asm_trampoline_template): New function.
4356 (arc_initialize_trampoline): Use asm_trampoline_template.
4357 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4358 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4359 * config/arc/arc.md (flush_icache): Delete pattern.
4361 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4363 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4364 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4367 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4370 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4371 by not USED_FOR_TARGET.
4372 (make_pass_resolve_sw_modes): Likewise.
4374 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4376 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4379 2018-01-08 Richard Biener <rguenther@suse.de>
4382 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4384 2018-01-08 Richard Biener <rguenther@suse.de>
4387 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4389 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4392 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4393 basic blocks with a small number of successors.
4394 (convert_control_dep_chain_into_preds): Improve handling of
4396 (dump_predicates): Split apart into...
4397 (dump_pred_chain): ...here...
4398 (dump_pred_info): ...and here.
4399 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4400 (can_chain_union_be_invalidated_p): Improve check for invalidation
4402 (uninit_uses_cannot_happen): Avoid unnecessary if
4403 convert_control_dep_chain_into_preds yielded nothing.
4405 2018-01-06 Martin Sebor <msebor@redhat.com>
4407 PR tree-optimization/83640
4408 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4409 subtracting negative offset from size.
4410 (builtin_access::overlap): Adjust offset bounds of the access to fall
4411 within the size of the object if possible.
4413 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4415 PR rtl-optimization/83699
4416 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4417 extract_bit_field_as_subreg to cases in which the extracted
4418 value is also a vector.
4420 * lra-constraints.c (process_alt_operands): Test for the equivalence
4421 substitutions when detecting a possible reload cycle.
4423 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4426 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4427 by default if flag_selective_schedling{,2}. Formatting fixes.
4429 PR rtl-optimization/83682
4430 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4431 if it has non-VECTOR_MODE element mode.
4432 (vec_duplicate_p): Likewise.
4435 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4436 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4438 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4441 * config/i386/i386-builtin.def
4442 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4443 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4444 Require also OPTION_MASK_ISA_AVX512F in addition to
4445 OPTION_MASK_ISA_GFNI.
4446 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4447 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4448 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4449 to OPTION_MASK_ISA_GFNI.
4450 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4451 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4452 OPTION_MASK_ISA_AVX512BW.
4453 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4454 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4455 addition to OPTION_MASK_ISA_GFNI.
4456 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4457 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4458 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4459 to OPTION_MASK_ISA_GFNI.
4460 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4461 a requirement for all ISAs rather than any of them with a few
4463 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4465 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4466 bitmasks to be enabled with 3 exceptions, instead of requiring any
4467 enabled ISA with lots of exceptions.
4468 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4469 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4470 Change avx512bw in isa attribute to avx512f.
4471 * config/i386/sgxintrin.h: Add license boilerplate.
4472 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4473 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4474 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4475 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4477 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4478 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4479 temporarily sse2 rather than sse if not enabled already.
4482 * config/i386/sse.md (VI248_VLBW): Rename to ...
4483 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4484 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4485 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4486 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4487 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4488 mode iterator instead of VI248_VLBW.
4490 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4492 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4493 (record_modified): Skip clobbers; add debug output.
4494 (param_change_prob): Use sreal frequencies.
4496 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4498 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4499 punt for user-aligned variables.
4501 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4503 * tree-chrec.c (chrec_contains_symbols): Return true for
4506 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4509 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4510 of (x|y) == x for BICS pattern.
4512 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4514 PR tree-optimization/83605
4515 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4516 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4519 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4521 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4522 * config/epiphany/rtems.h: New file.
4524 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4525 Uros Bizjak <ubizjak@gmail.com>
4528 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4529 QIreg_operand instead of register_operand predicate.
4530 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4531 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4532 comments instead of -fmitigate[-_]rop.
4534 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4537 * cgraphunit.c (symbol_table::compile): Switch to text_section
4538 before calling assembly_start debug hook.
4539 * run-rtl-passes.c (run_rtl_passes): Likewise.
4542 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4544 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4545 range_int_cst_p rather than !symbolic_range_p before calling
4546 extract_range_from_multiplicative_op_1.
4548 2018-01-04 Jeff Law <law@redhat.com>
4550 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4551 redundant test in assertion.
4553 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4555 * doc/rtl.texi: Document machine_mode wrapper classes.
4557 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4559 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4562 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4564 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4565 the VEC_PERM_EXPR fold to fail.
4567 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4570 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4571 to switched_sections.
4573 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4576 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4579 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4582 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4583 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4585 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4588 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4589 is BLKmode and bitpos not zero or mode change is needed.
4591 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4594 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4597 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4600 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4601 instead of MULT rtx. Update all corresponding splitters.
4603 (*ssub<modesuffix>): Ditto.
4605 (*cmp_sadd_di): Update split patterns.
4606 (*cmp_sadd_si): Ditto.
4607 (*cmp_sadd_sidi): Ditto.
4608 (*cmp_ssub_di): Ditto.
4609 (*cmp_ssub_si): Ditto.
4610 (*cmp_ssub_sidi): Ditto.
4611 * config/alpha/predicates.md (const23_operand): New predicate.
4612 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4613 Look for ASHIFT, not MULT inner operand.
4614 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4616 2018-01-04 Martin Liska <mliska@suse.cz>
4618 PR gcov-profile/83669
4619 * gcov.c (output_intermediate_file): Add version to intermediate
4621 * doc/gcov.texi: Document new field 'version' in intermediate
4622 file format. Fix location of '-k' option of gcov command.
4624 2018-01-04 Martin Liska <mliska@suse.cz>
4627 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4629 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4631 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4633 2018-01-03 Martin Sebor <msebor@redhat.com>
4635 PR tree-optimization/83655
4636 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4637 checking calls with invalid arguments.
4639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4641 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4642 (vectorizable_mask_load_store): Delete.
4643 (vectorizable_call): Return false for masked loads and stores.
4644 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4645 instead of gimple_assign_rhs1.
4646 (vectorizable_load): Handle IFN_MASK_LOAD.
4647 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4651 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4653 (vectorizable_mask_load_store): ...here.
4654 (vectorizable_load): ...and here.
4656 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4658 * tree-vect-stmts.c (vect_build_all_ones_mask)
4659 (vect_build_zero_merge_argument): New functions, split out from...
4660 (vectorizable_load): ...here.
4662 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4664 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4666 (vectorizable_mask_load_store): ...here.
4667 (vectorizable_store): ...and here.
4669 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4671 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4673 (vectorizable_mask_load_store): ...here.
4675 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4677 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4678 (vect_model_store_cost): Take a vec_load_store_type instead of a
4680 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4681 (vect_model_store_cost): Take a vec_load_store_type instead of a
4683 (vectorizable_mask_load_store): Update accordingly.
4684 (vectorizable_store): Likewise.
4685 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4687 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4689 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4690 IFN_MASK_LOAD calls here rather than...
4691 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4693 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4694 Alan Hayward <alan.hayward@arm.com>
4695 David Sherwood <david.sherwood@arm.com>
4697 * expmed.c (extract_bit_field_1): For vector extracts,
4698 fall back to extract_bit_field_as_subreg if vec_extract
4701 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4702 Alan Hayward <alan.hayward@arm.com>
4703 David Sherwood <david.sherwood@arm.com>
4705 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4706 they are variable or constant sized.
4707 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4708 slots for constant-sized data.
4710 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4711 Alan Hayward <alan.hayward@arm.com>
4712 David Sherwood <david.sherwood@arm.com>
4714 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4715 handling COND_EXPRs with boolean comparisons, try to find a better
4716 basis for the mask type than the boolean itself.
4718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4720 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4721 is calculated and how it can be overridden.
4722 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4723 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4725 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4728 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4729 Alan Hayward <alan.hayward@arm.com>
4730 David Sherwood <david.sherwood@arm.com>
4732 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4733 Remove the mode argument.
4734 (aarch64_simd_valid_immediate): Remove the mode and inverse
4736 * config/aarch64/iterators.md (bitsize): New iterator.
4737 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4738 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4739 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4740 aarch64_simd_valid_immediate.
4741 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4742 (aarch64_reg_or_bic_imm): Likewise.
4743 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4744 with an insn_type enum and msl with a modifier_type enum.
4745 Replace element_width with a scalar_mode. Change the shift
4746 to unsigned int. Add constructors for scalar_float_mode and
4747 scalar_int_mode elements.
4748 (aarch64_vect_float_const_representable_p): Delete.
4749 (aarch64_can_const_movi_rtx_p)
4750 (aarch64_simd_scalar_immediate_valid_for_move)
4751 (aarch64_simd_make_constant): Update call to
4752 aarch64_simd_valid_immediate.
4753 (aarch64_advsimd_valid_immediate_hs): New function.
4754 (aarch64_advsimd_valid_immediate): Likewise.
4755 (aarch64_simd_valid_immediate): Remove mode and inverse
4756 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4757 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4758 and aarch64_float_const_representable_p on the result.
4759 (aarch64_output_simd_mov_immediate): Remove mode argument.
4760 Update call to aarch64_simd_valid_immediate and use of
4761 simd_immediate_info.
4762 (aarch64_output_scalar_simd_mov_immediate): Update call
4765 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4766 Alan Hayward <alan.hayward@arm.com>
4767 David Sherwood <david.sherwood@arm.com>
4769 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4770 (mode_nunits): Likewise CONST_MODE_NUNITS.
4771 * machmode.def (ADJUST_NUNITS): Document.
4772 * genmodes.c (mode_data::need_nunits_adj): New field.
4773 (blank_mode): Update accordingly.
4774 (adj_nunits): New variable.
4775 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4777 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4778 listed in adj_nunits.
4779 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4780 listed in adj_nunits. Don't emit case statements for such modes.
4781 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4782 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4783 nothing if adj_nunits is nonnull.
4784 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4785 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4786 (emit_mode_fbit): Update use of print_maybe_const_decl.
4787 (emit_move_size): Likewise. Treat the array as non-const
4789 (emit_mode_adjustments): Handle adj_nunits.
4791 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4793 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4794 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4795 (VECTOR_MODES): Use it.
4796 (make_vector_modes): Take the prefix as an argument.
4798 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4799 Alan Hayward <alan.hayward@arm.com>
4800 David Sherwood <david.sherwood@arm.com>
4802 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4803 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4804 for MODE_VECTOR_BOOL.
4805 * machmode.def (VECTOR_BOOL_MODE): Document.
4806 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4807 (make_vector_bool_mode): New function.
4808 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4810 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4811 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4813 * stor-layout.c (int_mode_for_mode): Likewise.
4814 * tree.c (build_vector_type_for_mode): Likewise.
4815 * varasm.c (output_constant_pool_2): Likewise.
4816 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4817 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4818 for MODE_VECTOR_BOOL.
4819 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4820 of mode class checks.
4821 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4822 instead of a list of mode class checks.
4823 (expand_vector_scalar_condition): Likewise.
4824 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4827 Alan Hayward <alan.hayward@arm.com>
4828 David Sherwood <david.sherwood@arm.com>
4830 * machmode.h (mode_size): Change from unsigned short to
4832 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4833 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4834 or if measurement_type is not polynomial.
4835 (fixed_size_mode::includes_p): Check for constant-sized modes.
4836 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4837 return a poly_uint16 rather than an unsigned short.
4838 (emit_mode_size): Change the type of mode_size from unsigned short
4839 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4840 (emit_mode_adjustments): Cope with polynomial vector sizes.
4841 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4843 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4845 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4846 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4847 * caller-save.c (setup_save_areas): Likewise.
4848 (replace_reg_with_saved_mem): Likewise.
4849 * calls.c (emit_library_call_value_1): Likewise.
4850 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4851 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4852 (gen_lowpart_for_combine): Likewise.
4853 * convert.c (convert_to_integer_1): Likewise.
4854 * cse.c (equiv_constant, cse_insn): Likewise.
4855 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4856 (cselib_subst_to_values): Likewise.
4857 * dce.c (word_dce_process_block): Likewise.
4858 * df-problems.c (df_word_lr_mark_ref): Likewise.
4859 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4860 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4861 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4862 (rtl_for_decl_location): Likewise.
4863 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4864 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4865 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4866 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4867 (expand_expr_real_1): Likewise.
4868 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4869 (pad_below): Likewise.
4870 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4871 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4872 * ira.c (get_subreg_tracking_sizes): Likewise.
4873 * ira-build.c (ira_create_allocno_objects): Likewise.
4874 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4875 (ira_sort_regnos_for_alter_reg): Likewise.
4876 * ira-costs.c (record_operand_costs): Likewise.
4877 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4878 (resolve_simple_move): Likewise.
4879 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4880 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4881 (lra_constraints): Likewise.
4882 (CONST_POOL_OK_P): Reject variable-sized modes.
4883 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4884 (add_pseudo_to_slot, lra_spill): Likewise.
4885 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4886 * optabs-query.c (get_best_extraction_insn): Likewise.
4887 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4888 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4889 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4890 * recog.c (offsettable_address_addr_space_p): Likewise.
4891 * regcprop.c (maybe_mode_change): Likewise.
4892 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4893 * regrename.c (build_def_use): Likewise.
4894 * regstat.c (dump_reg_info): Likewise.
4895 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4896 (find_reloads, find_reloads_subreg_address): Likewise.
4897 * reload1.c (eliminate_regs_1): Likewise.
4898 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4899 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4900 (simplify_binary_operation_1, simplify_subreg): Likewise.
4901 * targhooks.c (default_function_arg_padding): Likewise.
4902 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4903 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4904 (verify_gimple_assign_ternary): Likewise.
4905 * tree-inline.c (estimate_move_cost): Likewise.
4906 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4907 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4908 (get_address_cost_ainc): Likewise.
4909 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4910 (vect_supportable_dr_alignment): Likewise.
4911 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4912 (vectorizable_reduction): Likewise.
4913 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4914 (vectorizable_operation, vectorizable_load): Likewise.
4915 * tree.c (build_same_sized_truth_vector_type): Likewise.
4916 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4917 * var-tracking.c (emit_note_insn_var_location): Likewise.
4918 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4919 (ADDR_VEC_ALIGN): Likewise.
4921 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4922 Alan Hayward <alan.hayward@arm.com>
4923 David Sherwood <david.sherwood@arm.com>
4925 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4927 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4928 or if measurement_type is polynomial.
4929 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4930 * combine.c (make_extraction): Likewise.
4931 * dse.c (find_shift_sequence): Likewise.
4932 * dwarf2out.c (mem_loc_descriptor): Likewise.
4933 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4934 (extract_bit_field, extract_low_bits): Likewise.
4935 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4936 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4937 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4938 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4939 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4940 * reload.c (find_reloads): Likewise.
4941 * reload1.c (alter_reg): Likewise.
4942 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4943 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4944 * tree-if-conv.c (predicate_mem_writes): Likewise.
4945 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4946 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4947 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4948 * valtrack.c (dead_debug_insert_temp): Likewise.
4949 * varasm.c (mergeable_constant_section): Likewise.
4950 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4952 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4953 Alan Hayward <alan.hayward@arm.com>
4954 David Sherwood <david.sherwood@arm.com>
4956 * expr.c (expand_assignment): Cope with polynomial mode sizes
4957 when assigning to a CONCAT.
4959 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4960 Alan Hayward <alan.hayward@arm.com>
4961 David Sherwood <david.sherwood@arm.com>
4963 * machmode.h (mode_precision): Change from unsigned short to
4965 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4967 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4968 or if measurement_type is not polynomial.
4969 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4970 in which the mode is already known to be a scalar_int_mode.
4971 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4972 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4974 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4975 for GET_MODE_PRECISION.
4976 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4977 for GET_MODE_PRECISION.
4978 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4980 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4981 (expand_field_assignment, make_extraction): Likewise.
4982 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4983 (get_last_value): Likewise.
4984 * convert.c (convert_to_integer_1): Likewise.
4985 * cse.c (cse_insn): Likewise.
4986 * expr.c (expand_expr_real_1): Likewise.
4987 * lra-constraints.c (simplify_operand_subreg): Likewise.
4988 * optabs-query.c (can_atomic_load_p): Likewise.
4989 * optabs.c (expand_atomic_load): Likewise.
4990 (expand_atomic_store): Likewise.
4991 * ree.c (combine_reaching_defs): Likewise.
4992 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4993 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4994 * tree.h (type_has_mode_precision_p): Likewise.
4995 * ubsan.c (instrument_si_overflow): Likewise.
4997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4998 Alan Hayward <alan.hayward@arm.com>
4999 David Sherwood <david.sherwood@arm.com>
5001 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5002 polynomial numbers of units.
5003 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5004 (valid_vector_subparts_p): New function.
5005 (build_vector_type): Remove temporary shim and take the number
5006 of units as a poly_uint64 rather than an int.
5007 (build_opaque_vector_type): Take the number of units as a
5008 poly_uint64 rather than an int.
5009 * tree.c (build_vector_from_ctor): Handle polynomial
5010 TYPE_VECTOR_SUBPARTS.
5011 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5012 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5013 (build_vector_from_val): If the number of units is variable,
5014 use build_vec_duplicate_cst for constant operands and
5015 VEC_DUPLICATE_EXPR otherwise.
5016 (make_vector_type): Remove temporary is_constant ().
5017 (build_vector_type, build_opaque_vector_type): Take the number of
5018 units as a poly_uint64 rather than an int.
5019 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5021 * cfgexpand.c (expand_debug_expr): Likewise.
5022 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5023 (store_constructor, expand_expr_real_1): Likewise.
5024 (const_scalar_mask_from_tree): Likewise.
5025 * fold-const-call.c (fold_const_reduction): Likewise.
5026 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5027 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5028 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5029 (fold_relational_const): Likewise.
5030 (native_interpret_vector): Likewise. Change the size from an
5031 int to an unsigned int.
5032 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5033 TYPE_VECTOR_SUBPARTS.
5034 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5035 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5036 duplicating a non-constant operand into a variable-length vector.
5037 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5038 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5039 * ipa-icf.c (sem_variable::equals): Likewise.
5040 * match.pd: Likewise.
5041 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5042 * print-tree.c (print_node): Likewise.
5043 * stor-layout.c (layout_type): Likewise.
5044 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5045 * tree-cfg.c (verify_gimple_comparison): Likewise.
5046 (verify_gimple_assign_binary): Likewise.
5047 (verify_gimple_assign_ternary): Likewise.
5048 (verify_gimple_assign_single): Likewise.
5049 * tree-pretty-print.c (dump_generic_node): Likewise.
5050 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5051 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5052 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5053 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5054 (vect_shift_permute_load_chain): Likewise.
5055 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5056 (expand_vector_condition, optimize_vector_constructor): Likewise.
5057 (lower_vec_perm, get_compute_type): Likewise.
5058 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5059 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5060 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5061 (vect_recog_mask_conversion_pattern): Likewise.
5062 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5063 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5064 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5065 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5066 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5067 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5068 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5069 (supportable_widening_operation): Likewise.
5070 (supportable_narrowing_operation): Likewise.
5071 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5073 * varasm.c (output_constant): Likewise.
5075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5076 Alan Hayward <alan.hayward@arm.com>
5077 David Sherwood <david.sherwood@arm.com>
5079 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5080 so that both the length == 3 and length != 3 cases set up their
5081 own permute vectors. Add comments explaining why we know the
5082 number of elements is constant.
5083 (vect_permute_load_chain): Likewise.
5085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5086 Alan Hayward <alan.hayward@arm.com>
5087 David Sherwood <david.sherwood@arm.com>
5089 * machmode.h (mode_nunits): Change from unsigned char to
5091 (ONLY_FIXED_SIZE_MODES): New macro.
5092 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5093 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5094 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5096 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5097 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5098 or if measurement_type is not polynomial.
5099 * genmodes.c (ZERO_COEFFS): New macro.
5100 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5102 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5103 Use ZERO_COEFFS when emitting initializers.
5104 * data-streamer.h (bp_pack_poly_value): New function.
5105 (bp_unpack_poly_value): Likewise.
5106 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5107 for GET_MODE_NUNITS.
5108 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5109 for GET_MODE_NUNITS.
5110 * tree.c (make_vector_type): Remove temporary shim and make
5111 the real function take the number of units as a poly_uint64
5113 (build_vector_type_for_mode): Handle polynomial nunits.
5114 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5115 * emit-rtl.c (const_vec_series_p_1): Likewise.
5116 (gen_rtx_CONST_VECTOR): Likewise.
5117 * fold-const.c (test_vec_duplicate_folding): Likewise.
5118 * genrecog.c (validate_pattern): Likewise.
5119 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5120 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5121 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5122 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5123 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5124 * rtlanal.c (subreg_get_info): Likewise.
5125 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5126 (vect_grouped_load_supported): Likewise.
5127 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5128 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5129 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5130 (simplify_const_unary_operation, simplify_binary_operation_1)
5131 (simplify_const_binary_operation, simplify_ternary_operation)
5132 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5133 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5134 instead of CONST_VECTOR_NUNITS.
5135 * varasm.c (output_constant_pool_2): Likewise.
5136 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5137 explicit-encoded elements in the XVEC for variable-length vectors.
5139 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5141 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5143 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5144 Alan Hayward <alan.hayward@arm.com>
5145 David Sherwood <david.sherwood@arm.com>
5147 * coretypes.h (fixed_size_mode): Declare.
5148 (fixed_size_mode_pod): New typedef.
5149 * builtins.h (target_builtins::x_apply_args_mode)
5150 (target_builtins::x_apply_result_mode): Change type to
5151 fixed_size_mode_pod.
5152 * builtins.c (apply_args_size, apply_result_size, result_vector)
5153 (expand_builtin_apply_args_1, expand_builtin_apply)
5154 (expand_builtin_return): Update accordingly.
5156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5158 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5159 * cselib.c (cselib_hash_rtx): Likewise.
5160 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5161 CONST_VECTOR encoding.
5163 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5164 Jeff Law <law@redhat.com>
5167 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5168 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5169 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5170 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5173 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5174 explicitly probe *sp in a noreturn function if there were any callee
5175 register saves or frame pointer is needed.
5177 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5180 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5181 BLKmode for ternary, binary or unary expressions.
5184 * var-tracking.c (delete_vta_debug_insn): New inline function.
5185 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5186 insns from get_insns () to NULL instead of each bb separately.
5187 Use delete_vta_debug_insn. No longer static.
5188 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5189 delete_vta_debug_insns callers.
5190 * rtl.h (delete_vta_debug_insns): Declare.
5191 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5192 instead of variable_tracking_main.
5194 2018-01-03 Martin Sebor <msebor@redhat.com>
5196 PR tree-optimization/83603
5197 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5198 arguments past the endof the argument list in functions declared
5199 without a prototype.
5200 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5201 Avoid checking when arguments are null.
5203 2018-01-03 Martin Sebor <msebor@redhat.com>
5206 * doc/extend.texi (attribute const): Fix a typo.
5207 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5208 issuing -Wsuggest-attribute for void functions.
5210 2018-01-03 Martin Sebor <msebor@redhat.com>
5212 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5213 offset_int::from instead of wide_int::to_shwi.
5214 (maybe_diag_overlap): Remove assertion.
5215 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5216 * gimple-ssa-sprintf.c (format_directive): Same.
5217 (parse_directive): Same.
5218 (sprintf_dom_walker::compute_format_length): Same.
5219 (try_substitute_return_value): Same.
5221 2018-01-03 Jeff Law <law@redhat.com>
5224 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5225 non-constant residual for zero at runtime and avoid probing in
5226 that case. Reorganize code for trailing problem to mirror handling
5229 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5231 PR tree-optimization/83501
5232 * tree-ssa-strlen.c (get_string_cst): New.
5233 (handle_char_store): Call get_string_cst.
5235 2018-01-03 Martin Liska <mliska@suse.cz>
5237 PR tree-optimization/83593
5238 * tree-ssa-strlen.c: Include tree-cfg.h.
5239 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5240 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5241 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5243 (strlen_dom_walker::before_dom_children): Call
5244 gimple_purge_dead_eh_edges. Dump tranformation with details
5246 (strlen_dom_walker::before_dom_children): Update call by adding
5247 new argument cleanup_eh.
5248 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5250 2018-01-03 Martin Liska <mliska@suse.cz>
5253 * cif-code.def (VARIADIC_THUNK): New enum value.
5254 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5257 2018-01-03 Jan Beulich <jbeulich@suse.com>
5259 * sse.md (mov<mode>_internal): Tighten condition for when to use
5260 vmovdqu<ssescalarsize> for TI and OI modes.
5262 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5264 Update copyright years.
5266 2018-01-03 Martin Liska <mliska@suse.cz>
5269 * ipa-visibility.c (function_and_variable_visibility): Skip
5270 functions with noipa attribure.
5272 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5274 * gcc.c (process_command): Update copyright notice dates.
5275 * gcov-dump.c (print_version): Ditto.
5276 * gcov.c (print_version): Ditto.
5277 * gcov-tool.c (print_version): Ditto.
5278 * gengtype.c (create_file): Ditto.
5279 * doc/cpp.texi: Bump @copying's copyright year.
5280 * doc/cppinternals.texi: Ditto.
5281 * doc/gcc.texi: Ditto.
5282 * doc/gccint.texi: Ditto.
5283 * doc/gcov.texi: Ditto.
5284 * doc/install.texi: Ditto.
5285 * doc/invoke.texi: Ditto.
5287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5289 * vector-builder.h (vector_builder::m_full_nelts): Change from
5290 unsigned int to poly_uint64.
5291 (vector_builder::full_nelts): Update prototype accordingly.
5292 (vector_builder::new_vector): Likewise.
5293 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5294 (vector_builder::operator ==): Likewise.
5295 (vector_builder::finalize): Likewise.
5296 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5297 Take the number of elements as a poly_uint64 rather than an
5299 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5300 from unsigned int to poly_uint64.
5301 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5302 (vec_perm_indices::new_vector): Likewise.
5303 (vec_perm_indices::length): Likewise.
5304 (vec_perm_indices::nelts_per_input): Likewise.
5305 (vec_perm_indices::input_nelts): Likewise.
5306 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5307 number of elements per input as a poly_uint64 rather than an
5308 unsigned int. Use the original encoding for variable-length
5309 vectors, rather than clamping each individual element.
5310 For the second and subsequent elements in each pattern,
5311 clamp the step and base before clamping their sum.
5312 (vec_perm_indices::series_p): Handle polynomial element counts.
5313 (vec_perm_indices::all_in_range_p): Likewise.
5314 (vec_perm_indices_to_tree): Likewise.
5315 (vec_perm_indices_to_rtx): Likewise.
5316 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5317 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5318 (tree_vector_builder::new_binary_operation): Handle polynomial
5319 element counts. Return false if we need to know the number
5320 of elements at compile time.
5321 * fold-const.c (fold_vec_perm): Punt if the number of elements
5322 isn't known at compile time.
5324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5326 * vec-perm-indices.h (vec_perm_builder): Change element type
5327 from HOST_WIDE_INT to poly_int64.
5328 (vec_perm_indices::element_type): Update accordingly.
5329 (vec_perm_indices::clamp): Handle polynomial element_types.
5330 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5331 (vec_perm_indices::all_in_range_p): Likewise.
5332 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5334 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5335 polynomial vec_perm_indices element types.
5336 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5337 * fold-const.c (fold_vec_perm): Likewise.
5338 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5339 * tree-vect-generic.c (lower_vec_perm): Likewise.
5340 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5341 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5342 element type to HOST_WIDE_INT.
5344 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5345 Alan Hayward <alan.hayward@arm.com>
5346 David Sherwood <david.sherwood@arm.com>
5348 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5349 rather than an int. Use plus_constant.
5350 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5351 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
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 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5358 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
5365 mode sizes. Require a constant size for BLKmode parameters
5366 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5367 forces a parameter to be padded at the lsb end in order to
5368 fill a complete number of words, require the parameter size
5369 to be ordered wrt UNITS_PER_WORD.
5371 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5372 Alan Hayward <alan.hayward@arm.com>
5373 David Sherwood <david.sherwood@arm.com>
5375 * reload1.c (spill_stack_slot_width): Change element type
5376 from unsigned int to poly_uint64_pod.
5377 (alter_reg): Treat mode sizes as polynomial.
5379 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5380 Alan Hayward <alan.hayward@arm.com>
5381 David Sherwood <david.sherwood@arm.com>
5383 * reload.c (complex_word_subreg_p): New function.
5384 (reload_inner_reg_of_subreg, push_reload): Use it.
5386 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5387 Alan Hayward <alan.hayward@arm.com>
5388 David Sherwood <david.sherwood@arm.com>
5390 * lra-constraints.c (process_alt_operands): Reject matched
5391 operands whose sizes aren't ordered.
5392 (match_reload): Refer to this check here.
5394 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5395 Alan Hayward <alan.hayward@arm.com>
5396 David Sherwood <david.sherwood@arm.com>
5398 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5399 that the mode size is in the set {1, 2, 4, 8, 16}.
5401 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5402 Alan Hayward <alan.hayward@arm.com>
5403 David Sherwood <david.sherwood@arm.com>
5405 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5406 Use plus_constant instead of gen_rtx_PLUS.
5408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5409 Alan Hayward <alan.hayward@arm.com>
5410 David Sherwood <david.sherwood@arm.com>
5412 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5413 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5414 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5415 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5416 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5417 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5418 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5419 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5420 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5421 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5423 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5424 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5425 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5426 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5427 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5428 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5429 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5430 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5431 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5432 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5434 * expr.c (emit_move_resolve_push): Treat the input and result
5435 of PUSH_ROUNDING as a poly_int64.
5436 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5437 (emit_push_insn): Likewise.
5438 * lra-eliminations.c (mark_not_eliminable): Likewise.
5439 * recog.c (push_operand): Likewise.
5440 * reload1.c (elimination_effects): Likewise.
5441 * rtlanal.c (nonzero_bits1): Likewise.
5442 * calls.c (store_one_arg): Likewise. Require the padding to be
5443 known at compile time.
5445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5446 Alan Hayward <alan.hayward@arm.com>
5447 David Sherwood <david.sherwood@arm.com>
5449 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5450 Use plus_constant instead of gen_rtx_PLUS.
5452 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5453 Alan Hayward <alan.hayward@arm.com>
5454 David Sherwood <david.sherwood@arm.com>
5456 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5459 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5460 Alan Hayward <alan.hayward@arm.com>
5461 David Sherwood <david.sherwood@arm.com>
5463 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5464 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5465 via stack temporaries. Treat the mode size as polynomial too.
5467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5468 Alan Hayward <alan.hayward@arm.com>
5469 David Sherwood <david.sherwood@arm.com>
5471 * expr.c (expand_expr_real_2): When handling conversions involving
5472 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5473 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5474 as a poly_uint64 too.
5476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5477 Alan Hayward <alan.hayward@arm.com>
5478 David Sherwood <david.sherwood@arm.com>
5480 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5482 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5483 Alan Hayward <alan.hayward@arm.com>
5484 David Sherwood <david.sherwood@arm.com>
5486 * combine.c (can_change_dest_mode): Handle polynomial
5487 REGMODE_NATURAL_SIZE.
5488 * expmed.c (store_bit_field_1): Likewise.
5489 * expr.c (store_constructor): Likewise.
5490 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5491 and polynomial REGMODE_NATURAL_SIZE.
5492 (gen_lowpart_common): Likewise.
5493 * reginfo.c (record_subregs_of_mode): Likewise.
5494 * rtlanal.c (read_modify_subreg_p): Likewise.
5496 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5497 Alan Hayward <alan.hayward@arm.com>
5498 David Sherwood <david.sherwood@arm.com>
5500 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5501 numbers of elements.
5503 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5504 Alan Hayward <alan.hayward@arm.com>
5505 David Sherwood <david.sherwood@arm.com>
5507 * match.pd: Cope with polynomial numbers of vector elements.
5509 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5510 Alan Hayward <alan.hayward@arm.com>
5511 David Sherwood <david.sherwood@arm.com>
5513 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5514 in a POINTER_PLUS_EXPR.
5516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5517 Alan Hayward <alan.hayward@arm.com>
5518 David Sherwood <david.sherwood@arm.com>
5520 * omp-simd-clone.c (simd_clone_subparts): New function.
5521 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5522 (ipa_simd_modify_function_body): Likewise.
5524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5525 Alan Hayward <alan.hayward@arm.com>
5526 David Sherwood <david.sherwood@arm.com>
5528 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5529 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5530 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5531 (expand_vector_condition, vector_element): Likewise.
5532 (subparts_gt): New function.
5533 (get_compute_type): Use subparts_gt.
5534 (count_type_subparts): Delete.
5535 (expand_vector_operations_1): Use subparts_gt instead of
5536 count_type_subparts.
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-data-refs.c (vect_no_alias_p): Replace with...
5543 (vect_compile_time_alias): ...this new function. Do the calculation
5544 on poly_ints rather than trees.
5545 (vect_prune_runtime_alias_test_list): Update call accordingly.
5547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5548 Alan Hayward <alan.hayward@arm.com>
5549 David Sherwood <david.sherwood@arm.com>
5551 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5553 (vect_schedule_slp_instance): Likewise.
5555 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5556 Alan Hayward <alan.hayward@arm.com>
5557 David Sherwood <david.sherwood@arm.com>
5559 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5560 constant and extern definitions for variable-length vectors.
5561 (vect_get_constant_vectors): Note that the number of units
5562 is known to be constant.
5564 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5565 Alan Hayward <alan.hayward@arm.com>
5566 David Sherwood <david.sherwood@arm.com>
5568 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5569 of units as polynomial. Choose between WIDE and NARROW based
5572 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5573 Alan Hayward <alan.hayward@arm.com>
5574 David Sherwood <david.sherwood@arm.com>
5576 * tree-vect-stmts.c (simd_clone_subparts): New function.
5577 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5579 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5580 Alan Hayward <alan.hayward@arm.com>
5581 David Sherwood <david.sherwood@arm.com>
5583 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5584 vectors as polynomial. Use build_index_vector for
5587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5588 Alan Hayward <alan.hayward@arm.com>
5589 David Sherwood <david.sherwood@arm.com>
5591 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5592 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5593 for variable-length vectors.
5594 (vectorizable_mask_load_store): Treat the number of units as
5595 polynomial, asserting that it is constant if the condition has
5596 already been enforced.
5597 (vectorizable_store, vectorizable_load): Likewise.
5599 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5600 Alan Hayward <alan.hayward@arm.com>
5601 David Sherwood <david.sherwood@arm.com>
5603 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5604 of units as polynomial. Punt if we can't tell at compile time
5605 which vector contains the final result.
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-loop.c (vectorizable_induction): Treat the number
5612 of units as polynomial. Punt on SLP inductions. Use an integer
5613 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5614 cast of such a series for variable-length floating-point
5617 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5618 Alan Hayward <alan.hayward@arm.com>
5619 David Sherwood <david.sherwood@arm.com>
5621 * tree.h (build_index_vector): Declare.
5622 * tree.c (build_index_vector): New function.
5623 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5624 of units as polynomial, forcibly converting it to a constant if
5625 vectorizable_reduction has already enforced the condition.
5626 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5627 to create a {1,2,3,...} vector.
5628 (vectorizable_reduction): Treat the number of units as polynomial.
5629 Choose vectype_in based on the largest scalar element size rather
5630 than the smallest number of units. Enforce the restrictions
5633 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5634 Alan Hayward <alan.hayward@arm.com>
5635 David Sherwood <david.sherwood@arm.com>
5637 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5638 number of units as polynomial.
5640 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5641 Alan Hayward <alan.hayward@arm.com>
5642 David Sherwood <david.sherwood@arm.com>
5644 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5645 * target.def (autovectorize_vector_sizes): Return the vector sizes
5646 by pointer, using vector_sizes rather than a bitmask.
5647 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5648 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5649 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5651 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5652 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5653 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5654 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5655 * omp-general.c (omp_max_vf): Likewise.
5656 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5657 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5658 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5659 * tree-vect-slp.c (vect_slp_bb): Likewise.
5660 * doc/tm.texi: Regenerate.
5661 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5663 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5664 the vector size as a poly_uint64 rather than an unsigned int.
5665 (current_vector_size): Change from an unsigned int to a poly_uint64.
5666 (get_vectype_for_scalar_type): Update accordingly.
5667 * tree.h (build_truth_vector_type): Take the size and number of
5668 units as a poly_uint64 rather than an unsigned int.
5669 (build_vector_type): Add a temporary overload that takes
5670 the number of units as a poly_uint64 rather than an unsigned int.
5671 * tree.c (make_vector_type): Likewise.
5672 (build_truth_vector_type): Take the number of units as a poly_uint64
5673 rather than an unsigned int.
5675 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5676 Alan Hayward <alan.hayward@arm.com>
5677 David Sherwood <david.sherwood@arm.com>
5679 * target.def (get_mask_mode): Take the number of units and length
5680 as poly_uint64s rather than unsigned ints.
5681 * targhooks.h (default_get_mask_mode): Update accordingly.
5682 * targhooks.c (default_get_mask_mode): Likewise.
5683 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5684 * doc/tm.texi: Regenerate.
5686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5687 Alan Hayward <alan.hayward@arm.com>
5688 David Sherwood <david.sherwood@arm.com>
5690 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5691 * omp-general.c (omp_max_vf): Likewise.
5692 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5693 (expand_omp_simd): Handle polynomial safelen.
5694 * omp-low.c (omplow_simd_context): Add a default constructor.
5695 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5696 (lower_rec_simd_input_clauses): Update accordingly.
5697 (lower_rec_input_clauses): Likewise.
5699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5700 Alan Hayward <alan.hayward@arm.com>
5701 David Sherwood <david.sherwood@arm.com>
5703 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5704 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5705 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5706 (vect_analyze_slp_cost): Likewise.
5707 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5708 (vect_model_load_cost): Likewise.
5710 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5711 Alan Hayward <alan.hayward@arm.com>
5712 David Sherwood <david.sherwood@arm.com>
5714 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5715 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5716 from an unsigned int * to a poly_uint64_pod *.
5717 (calculate_unrolling_factor): New function.
5718 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5720 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5721 Alan Hayward <alan.hayward@arm.com>
5722 David Sherwood <david.sherwood@arm.com>
5724 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5725 from an unsigned int to a poly_uint64.
5726 (_loop_vec_info::slp_unrolling_factor): Likewise.
5727 (_loop_vec_info::vectorization_factor): Change from an int
5729 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5730 (vect_get_num_vectors): New function.
5731 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5732 (vect_get_num_copies): Use vect_get_num_vectors.
5733 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5734 to an unsigned int *.
5735 (vect_analyze_data_refs): Change min_vf from an int * to a
5737 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5738 than an unsigned HOST_WIDE_INT.
5739 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5740 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5741 to an unsigned int *.
5742 (vect_analyze_data_ref_dependences): Likewise.
5743 (vect_compute_data_ref_alignment): Handle polynomial vf.
5744 (vect_enhance_data_refs_alignment): Likewise.
5745 (vect_prune_runtime_alias_test_list): Likewise.
5746 (vect_shift_permute_load_chain): Likewise.
5747 (vect_supportable_dr_alignment): Likewise.
5748 (dependence_distance_ge_vf): Take the vectorization factor as a
5749 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5750 (vect_analyze_data_refs): Change min_vf from an int * to a
5752 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5753 vfm1 as a poly_uint64 rather than an int. Make the same change
5754 for the returned bound_scalar.
5755 (vect_gen_vector_loop_niters): Handle polynomial vf.
5756 (vect_do_peeling): Likewise. Update call to
5757 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5758 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5760 * tree-vect-loop.c (vect_determine_vectorization_factor)
5761 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5762 (vect_get_known_peeling_cost): Likewise.
5763 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5764 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5765 (vect_transform_loop): Likewise. Use the lowest possible VF when
5766 updating the upper bounds of the loop.
5767 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5769 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5770 polynomial unroll factors.
5771 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5772 (vect_make_slp_decision): Likewise.
5773 (vect_supported_load_permutation_p): Likewise, and polynomial
5775 (vect_analyze_slp_cost): Handle polynomial vf.
5776 (vect_slp_analyze_node_operations): Likewise.
5777 (vect_slp_analyze_bb_1): Likewise.
5778 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5779 than an unsigned HOST_WIDE_INT.
5780 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5781 (vectorizable_load): Handle polynomial vf.
5782 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5784 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5786 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5787 Alan Hayward <alan.hayward@arm.com>
5788 David Sherwood <david.sherwood@arm.com>
5790 * match.pd: Handle bit operations involving three constants
5791 and try to fold one pair.
5793 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5795 * tree-vect-loop-manip.c: Include gimple-fold.h.
5796 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5797 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5798 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5799 Add a path that uses a step of VF instead of 1, but disable it
5801 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5802 and niters_no_overflow parameters. Update calls to
5803 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5804 Create a new SSA name if the latter choses to use a ste other
5805 than zero, and return it via niters_vector_mult_vf_var.
5806 * tree-vect-loop.c (vect_transform_loop): Update calls to
5807 vect_do_peeling, vect_gen_vector_loop_niters and
5808 slpeel_make_loop_iterate_ntimes.
5809 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5810 (vect_gen_vector_loop_niters): Update declarations after above changes.
5812 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5814 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5815 128-bit round to integer instructions.
5816 (ceil<mode>2): Likewise.
5817 (btrunc<mode>2): Likewise.
5818 (round<mode>2): Likewise.
5820 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5822 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5823 unaligned VSX load/store on P8/P9.
5824 (expand_block_clear): Allow the use of unaligned VSX
5825 load/store on P8/P9.
5827 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5829 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5831 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5832 swap associated with both a load and a store.
5834 2018-01-02 Andrew Waterman <andrew@sifive.com>
5836 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5837 * config/riscv/riscv.md (clear_cache): Use it.
5839 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5841 * web.c: Remove out-of-date comment.
5843 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5845 * expr.c (fixup_args_size_notes): Check that any existing
5846 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5847 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5848 (emit_single_push_insn): ...here.
5850 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5852 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5853 (const_vector_encoded_nelts): New function.
5854 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5855 (const_vector_int_elt, const_vector_elt): Declare.
5856 * emit-rtl.c (const_vector_int_elt_1): New function.
5857 (const_vector_elt): Likewise.
5858 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5859 of CONST_VECTOR_ELT.
5861 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5863 * expr.c: Include rtx-vector-builder.h.
5864 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5865 directly on the tree encoding.
5866 (const_vector_from_tree): Likewise.
5867 * optabs.c: Include rtx-vector-builder.h.
5868 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5869 sequence of "u" values.
5870 * vec-perm-indices.c: Include rtx-vector-builder.h.
5871 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5872 directly on the vec_perm_indices encoding.
5874 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5876 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5877 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5878 * rtx-vector-builder.h: New file.
5879 * rtx-vector-builder.c: Likewise.
5880 * rtl.h (rtx_def::u2): Add a const_vector field.
5881 (CONST_VECTOR_NPATTERNS): New macro.
5882 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5883 (CONST_VECTOR_DUPLICATE_P): Likewise.
5884 (CONST_VECTOR_STEPPED_P): Likewise.
5885 (CONST_VECTOR_ENCODED_ELT): Likewise.
5886 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5887 (unwrap_const_vec_duplicate): Likewise.
5888 (const_vec_series_p): Check for a non-duplicated vector encoding.
5889 Say that the function only returns true for integer vectors.
5890 * emit-rtl.c: Include rtx-vector-builder.h.
5891 (gen_const_vec_duplicate_1): Delete.
5892 (gen_const_vector): Call gen_const_vec_duplicate instead of
5893 gen_const_vec_duplicate_1.
5894 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5895 (gen_const_vec_duplicate): Use rtx_vector_builder.
5896 (gen_const_vec_series): Likewise.
5897 (gen_rtx_CONST_VECTOR): Likewise.
5898 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5899 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5900 Build a new vector rather than modifying a CONST_VECTOR in-place.
5901 (handle_special_swappables): Update call accordingly.
5902 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5903 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5904 Build a new vector rather than modifying a CONST_VECTOR in-place.
5905 (handle_special_swappables): Update call accordingly.
5907 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5909 * simplify-rtx.c (simplify_const_binary_operation): Use
5910 CONST_VECTOR_ELT instead of XVECEXP.
5912 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5914 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5915 the selector elements to be different from the data elements
5916 if the selector is a VECTOR_CST.
5917 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5918 ssizetype for the selector.
5920 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5922 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5923 before testing each element individually.
5924 * tree-vect-generic.c (lower_vec_perm): Likewise.
5926 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5928 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5929 * selftest-run-tests.c (selftest::run_tests): Call it.
5930 * vector-builder.h (vector_builder::operator ==): New function.
5931 (vector_builder::operator !=): Likewise.
5932 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5933 (vec_perm_indices::all_from_input_p): New function.
5934 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5935 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5936 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5937 instead of reading the VECTOR_CST directly. Detect whether both
5938 vector inputs are the same before constructing the vec_perm_indices,
5939 and update the number of inputs argument accordingly. Use the
5940 utility functions added above. Only construct sel2 if we need to.
5942 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5944 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5945 the broadcast of the low byte.
5946 (expand_mult_highpart): Use an explicit encoding for the permutes.
5947 * optabs-query.c (can_mult_highpart_p): Likewise.
5948 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5949 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5950 (vectorizable_bswap): Likewise.
5951 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5952 explicit encoding for the power-of-2 permutes.
5953 (vect_permute_store_chain): Likewise.
5954 (vect_grouped_load_supported): Likewise.
5955 (vect_permute_load_chain): Likewise.
5957 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5959 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5960 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5961 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5962 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5963 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5964 (vect_gen_perm_mask_any): Likewise.
5966 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5968 * int-vector-builder.h: New file.
5969 * vec-perm-indices.h: Include int-vector-builder.h.
5970 (vec_perm_indices): Redefine as an int_vector_builder.
5971 (auto_vec_perm_indices): Delete.
5972 (vec_perm_builder): Redefine as a stand-alone class.
5973 (vec_perm_indices::vec_perm_indices): New function.
5974 (vec_perm_indices::clamp): Likewise.
5975 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5976 (vec_perm_indices::new_vector): New function.
5977 (vec_perm_indices::new_expanded_vector): Update for new
5978 vec_perm_indices class.
5979 (vec_perm_indices::rotate_inputs): New function.
5980 (vec_perm_indices::all_in_range_p): Operate directly on the
5981 encoded form, without computing elided elements.
5982 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5983 encoding. Update for new vec_perm_indices class.
5984 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5985 the given vec_perm_builder.
5986 (expand_vec_perm_var): Update vec_perm_builder constructor.
5987 (expand_mult_highpart): Use vec_perm_builder instead of
5988 auto_vec_perm_indices.
5989 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5990 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5991 or double series encoding as appropriate.
5992 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5993 vec_perm_indices instead of auto_vec_perm_indices.
5994 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5995 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5996 (vect_permute_store_chain): Likewise.
5997 (vect_grouped_load_supported): Likewise.
5998 (vect_permute_load_chain): Likewise.
5999 (vect_shift_permute_load_chain): Likewise.
6000 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6001 (vect_transform_slp_perm_load): Likewise.
6002 (vect_schedule_slp_instance): Likewise.
6003 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6004 (vectorizable_mask_load_store): Likewise.
6005 (vectorizable_bswap): Likewise.
6006 (vectorizable_store): Likewise.
6007 (vectorizable_load): Likewise.
6008 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6009 vec_perm_indices instead of auto_vec_perm_indices. Use
6010 tree_to_vec_perm_builder to read the vector from a tree.
6011 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6012 vec_perm_builder instead of a vec_perm_indices.
6013 (have_whole_vector_shift): Use vec_perm_builder and
6014 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6015 truncation to calc_vec_perm_mask_for_shift.
6016 (vect_create_epilog_for_reduction): Likewise.
6017 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6018 from auto_vec_perm_indices to vec_perm_indices.
6019 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6020 instead of changing individual elements.
6021 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6022 the vector in d.perm.
6023 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6024 from auto_vec_perm_indices to vec_perm_indices.
6025 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6026 instead of changing individual elements.
6027 (arm_vectorize_vec_perm_const): Use new_vector to install
6028 the vector in d.perm.
6029 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6030 Update vec_perm_builder constructor.
6031 (rs6000_expand_interleave): Likewise.
6032 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6033 (rs6000_expand_interleave): Likewise.
6035 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6037 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6038 to qimode could truncate the indices.
6039 * optabs.c (expand_vec_perm_var): Likewise.
6041 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6043 * Makefile.in (OBJS): Add vec-perm-indices.o.
6044 * vec-perm-indices.h: New file.
6045 * vec-perm-indices.c: Likewise.
6046 * target.h (vec_perm_indices): Replace with a forward class
6048 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6049 * optabs.h: Include vec-perm-indices.h.
6050 (expand_vec_perm): Delete.
6051 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6052 (expand_vec_perm_const): Declare.
6053 * target.def (vec_perm_const_ok): Replace with...
6054 (vec_perm_const): ...this new hook.
6055 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6056 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6057 * doc/tm.texi: Regenerate.
6058 * optabs.def (vec_perm_const): Delete.
6059 * doc/md.texi (vec_perm_const): Likewise.
6060 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6061 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6062 expand_vec_perm for constant permutation vectors. Assert that
6063 the mode of variable permutation vectors is the integer equivalent
6064 of the mode that is being permuted.
6065 * optabs-query.h (selector_fits_mode_p): Declare.
6066 * optabs-query.c: Include vec-perm-indices.h.
6067 (selector_fits_mode_p): New function.
6068 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6069 is defined, instead of checking whether the vec_perm_const_optab
6070 exists. Use targetm.vectorize.vec_perm_const instead of
6071 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6072 fit in the vector mode before using a variable permute.
6073 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6074 vec_perm_indices instead of an rtx.
6075 (expand_vec_perm): Replace with...
6076 (expand_vec_perm_const): ...this new function. Take the selector
6077 as a vec_perm_indices rather than an rtx. Also take the mode of
6078 the selector. Update call to shift_amt_for_vec_perm_mask.
6079 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6080 Use vec_perm_indices::new_expanded_vector to expand the original
6081 selector into bytes. Check whether the indices fit in the vector
6082 mode before using a variable permute.
6083 (expand_vec_perm_var): Make global.
6084 (expand_mult_highpart): Use expand_vec_perm_const.
6085 * fold-const.c: Includes vec-perm-indices.h.
6086 * tree-ssa-forwprop.c: Likewise.
6087 * tree-vect-data-refs.c: Likewise.
6088 * tree-vect-generic.c: Likewise.
6089 * tree-vect-loop.c: Likewise.
6090 * tree-vect-slp.c: Likewise.
6091 * tree-vect-stmts.c: Likewise.
6092 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6094 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6095 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6096 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6097 (aarch64_vectorize_vec_perm_const): ...this new function.
6098 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6099 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6100 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6101 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6102 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6103 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6104 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6106 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6107 check for NEON modes.
6108 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6109 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6110 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6111 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6113 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6114 the old VEC_PERM_CONST conditions.
6115 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6116 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6117 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6118 (ia64_vectorize_vec_perm_const_ok): Merge into...
6119 (ia64_vectorize_vec_perm_const): ...this new function.
6120 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6121 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6122 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6123 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6124 * config/mips/mips.c (mips_expand_vec_perm_const)
6125 (mips_vectorize_vec_perm_const_ok): Merge into...
6126 (mips_vectorize_vec_perm_const): ...this new function.
6127 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6128 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6129 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6130 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6131 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6132 (rs6000_expand_vec_perm_const): Delete.
6133 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6135 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6136 (altivec_expand_vec_perm_const_le): Take each operand individually.
6137 Operate on constant selectors rather than rtxes.
6138 (altivec_expand_vec_perm_const): Likewise. Update call to
6139 altivec_expand_vec_perm_const_le.
6140 (rs6000_expand_vec_perm_const): Delete.
6141 (rs6000_vectorize_vec_perm_const_ok): Delete.
6142 (rs6000_vectorize_vec_perm_const): New function.
6143 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6144 an element count and rtx array.
6145 (rs6000_expand_extract_even): Update call accordingly.
6146 (rs6000_expand_interleave): Likewise.
6147 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6148 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6149 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6150 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6151 (rs6000_expand_vec_perm_const): Delete.
6152 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6153 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6154 (altivec_expand_vec_perm_const_le): Take each operand individually.
6155 Operate on constant selectors rather than rtxes.
6156 (altivec_expand_vec_perm_const): Likewise. Update call to
6157 altivec_expand_vec_perm_const_le.
6158 (rs6000_expand_vec_perm_const): Delete.
6159 (rs6000_vectorize_vec_perm_const_ok): Delete.
6160 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6161 reference to the SPE evmerge intructions.
6162 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6163 an element count and rtx array.
6164 (rs6000_expand_extract_even): Update call accordingly.
6165 (rs6000_expand_interleave): Likewise.
6166 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6167 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6169 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6171 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6173 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6174 vector mode and that that mode matches the mode of the data
6176 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6177 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6178 directly using expand_vec_perm_1 when forcing selectors into
6180 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6182 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6184 * optabs-query.h (can_vec_perm_p): Delete.
6185 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6186 * optabs-query.c (can_vec_perm_p): Split into...
6187 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6188 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6189 particular selector is valid.
6190 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6191 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6192 (vect_grouped_load_supported): Likewise.
6193 (vect_shift_permute_load_chain): Likewise.
6194 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6195 (vect_transform_slp_perm_load): Likewise.
6196 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6197 (vectorizable_bswap): Likewise.
6198 (vect_gen_perm_mask_checked): Likewise.
6199 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6200 implementations of variable permutation vectors into account
6201 when deciding which selector to use.
6202 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6203 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6204 with a false third argument.
6205 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6206 to test whether the constant selector is valid and can_vec_perm_var_p
6207 to test whether a variable selector is valid.
6209 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6211 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6212 * optabs-query.c (can_vec_perm_p): Likewise.
6213 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6214 instead of vec_perm_indices.
6215 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6216 (vect_gen_perm_mask_checked): Likewise,
6217 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6218 (vect_gen_perm_mask_checked): Likewise,
6220 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6222 * optabs-query.h (qimode_for_vec_perm): Declare.
6223 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6224 (qimode_for_vec_perm): ...this new function.
6225 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6227 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6229 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6230 does not have a conditional at the top.
6232 2018-01-02 Richard Biener <rguenther@suse.de>
6234 * ipa-inline.c (big_speedup_p): Fix expression.
6236 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6239 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6242 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6246 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6247 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6248 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6249 cond_taken_branch_cost 3->4.
6251 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6253 PR tree-optimization/83581
6254 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6255 TODO_cleanup_cfg if any changes have been made.
6258 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6259 convert_modes if target mode has the right side, but different mode
6263 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6264 last argument when extracting from CONCAT. If either from_real or
6265 from_imag is NULL, use expansion through memory. If result is not
6266 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6267 the parts directly to inner mode, if even that fails, use expansion
6271 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6272 check for bswap in mode rather than HImode and use that in expand_unop
6275 Copyright (C) 2018 Free Software Foundation, Inc.
6277 Copying and distribution of this file, with or without modification,
6278 are permitted in any medium without royalty provided the copyright
6279 notice and this notice are preserved.