1 2018-02-07 Iain Sandoe <iain@codesourcery.com>
3 * config/rs6000/altivec.md (*restore_world): Remove LR use.
4 * config/rs6000/predicates.md (restore_world_operation): Adjust op
7 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
10 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
11 Convert from define_expand to be define_insn_and_split. Rework
12 float/double/_Float128 conversions to QI/HI/SImode to work with
13 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
14 conversions to QI/HImode types did a store and then a load to
15 truncate the value. For conversions to VSX registers, don't split
16 the insn, instead emit the code directly. Use the code iterator
17 any_fix to combine signed and unsigned conversions.
18 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
19 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
20 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
21 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
22 (fix_<mode>di2_hw): Likewise.
23 (fixuns_<mode>di2_hw): Likewise.
24 (fix_<mode>si2_hw): Likewise.
25 (fixuns_<mode>si2_hw): Likewise.
26 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
27 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
28 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
29 fix<uns>_trunc<SFDF:mode>si2_p8.
30 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
32 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
33 (fix<uns>_<mode>_mem): Likewise.
34 (fctiw<u>z_<mode>_mem): Likewise.
35 (fix<uns>_<mode>_mem): Likewise.
36 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
37 the register allocator from doing a direct move to the GPRs to do
38 a store, and instead use the ISA 3.0 store byte/half-word from
39 vector register instruction. For IEEE 128-bit floating point,
40 also optimize stores of 32-bit ints.
41 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
43 2018-02-07 Alan Hayward <alan.hayward@arm.com>
45 * genextract.c (push_pathstr_operand): New function to support
47 (walk_rtx): Call push_pathstr_operand.
48 (print_path): Support [a-zA-Z].
50 2018-02-07 Richard Biener <rguenther@suse.de>
52 PR tree-optimization/84037
53 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
54 (cse_and_gimplify_to_preheader): Declare.
55 (vect_get_place_in_interleaving_chain): Likewise.
56 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
58 (_loop_vec_info::~_loop_vec_info): Delete it.
59 (cse_and_gimplify_to_preheader): New function.
60 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
61 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
62 (vectorizable_load): Likewise. For grouped stores always base
63 the IV on the first element.
64 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
65 condition before gimplifying.
67 2018-02-07 Jakub Jelinek <jakub@redhat.com>
69 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
70 *DIV_EXPR and *MOD_EXPR.
72 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
75 * config/i386/i386.c (ix86_option_override_internal): Mask out
76 the CF_SET bit when checking -fcf-protection.
78 2018-02-07 Tom de Vries <tom@codesourcery.com>
81 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
84 2018-02-07 Richard Biener <rguenther@suse.de>
86 PR tree-optimization/84204
87 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
90 PR tree-optimization/84205
91 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
92 special-case isl_ast_op_zdiv_r.
94 PR tree-optimization/84223
95 * graphite-scop-detection.c (gather_bbs::before_dom_children):
96 Only add conditions from within the region.
97 (gather_bbs::after_dom_children): Adjust.
99 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
102 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
103 * config/avr/avr.md: Only post-reload split REG-REG moves if
104 either register is REGERAL_REG_P.
106 2018-02-07 Jakub Jelinek <jakub@redhat.com>
108 PR tree-optimization/84235
109 * tree-ssa-scopedtables.c
110 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
111 if the subtraction is performed in floating point type where NaNs are
112 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
113 build 1. Formatting fix.
115 2018-02-06 Jakub Jelinek <jakub@redhat.com>
118 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
119 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
120 and skip it regardless of bb boundaries. Use CALL_P macro,
121 don't test INSN_P (insn) together with CALL_P or JUMP_P check
122 unnecessarily, formatting fix.
124 2018-02-06 Michael Collison <michael.collison@arm.com>
126 * config/arm/thumb2.md:
127 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
128 (*thumb_mov_notscc): Ditto.
130 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
133 * config/rs6000/rs6000.md (su code attribute): Use "u" for
134 unsigned_fix, not "s".
136 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
138 * configure.ac (gcc_fn_eh_frame_ro): New function.
139 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
140 correct .eh_frame permissions.
141 * configure: Regenerate.
143 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
145 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
148 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
150 * config/rs6000/rs6000.c (rs6000_option_override_internal):
151 Display warning message for -mno-speculate-indirect-jumps.
153 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
155 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
157 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
159 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
161 PR tree-optimization/84225
162 * tree-eh.c (find_trapping_overflow): Only call
163 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
165 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
168 * config/i386/i386.c: Reimplement the check of possible options
169 -mibt/-mshstk conbination. Change error messages.
170 * doc/invoke.texi: Fix a typo: remove extra '='.
172 2018-02-06 Marek Polacek <polacek@redhat.com>
174 PR tree-optimization/84228
175 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
177 2018-02-06 Tamar Christina <tamar.christina@arm.com>
180 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
181 emitted arch directives.
182 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
183 __ARM_FEATURE_COPROC before changing architectures.
185 2018-02-06 Richard Biener <rguenther@suse.de>
187 * config/i386/i386.c (print_reg): Fix typo.
188 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
190 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
192 * configure: Regenerate.
194 2018-02-05 Martin Sebor <msebor@redhat.com>
196 PR tree-optimization/83369
197 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
200 2018-02-05 Martin Liska <mliska@suse.cz>
202 * doc/invoke.texi: Cherry-pick upstream r323995.
204 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
206 * ira.c (ira_init_register_move_cost): Adjust comment.
208 2018-02-05 Martin Liska <mliska@suse.cz>
210 PR gcov-profile/84137
211 * doc/gcov.texi: Fix typo in documentation.
213 2018-02-05 Martin Liska <mliska@suse.cz>
215 PR gcov-profile/83879
216 * doc/gcov.texi: Document necessity of --dynamic-list-data when
217 using dlopen functionality.
219 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
221 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
222 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
223 _mm_maskz_range_ss, _mm_mask_range_round_ss,
224 _mm_maskz_range_round_ss): New intrinsics.
225 (__builtin_ia32_rangesd128_round)
226 (__builtin_ia32_rangess128_round): Remove.
227 (__builtin_ia32_rangesd128_mask_round,
228 __builtin_ia32_rangess128_mask_round): New builtins.
229 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
230 __builtin_ia32_rangess128_round): Remove.
231 (__builtin_ia32_rangesd128_mask_round,
232 __builtin_ia32_rangess128_mask_round): New builtins.
233 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
234 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
235 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
236 "<round_saeonly_constraint>")): Changed to ...
237 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
238 "<round_saeonly_scalar_constraint>")): ... this.
239 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
240 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
241 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
242 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
243 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
245 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
247 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
249 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
250 Remove all values except native, 8540 and 8548.
252 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
254 * config/i386/i386.c (ix86_output_function_return): Pass
255 INVALID_REGNUM, instead of -1, as invalid register number to
256 indirect_thunk_name and output_indirect_thunk.
258 2018-02-02 Julia Koval <julia.koval@intel.com>
260 * config.gcc: Add -march=icelake.
261 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
262 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
263 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
264 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
265 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
266 (processor_target_table): Add icelake.
267 (ix86_option_override_internal): Handle new PTAs.
268 (get_builtin_code_for_version): Handle icelake.
269 (M_INTEL_COREI7_ICELAKE): New.
270 (fold_builtin_cpu): Handle icelake.
271 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
272 * doc/invoke.texi: Add -march=icelake.
274 2018-02-02 Julia Koval <julia.koval@intel.com>
276 * config/i386/i386.c (ix86_option_override_internal): Change flags type
278 * wide-int-bitmask.h: New.
280 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
283 * config/i386/i386.md: Replace Pmode with word_mode in
284 builtin_setjmp_setup and builtin_longjmp to support x32.
286 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
290 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
292 (rs6000_supported_cpu_names): New static variable.
293 (linux_cpu_translation_table): Likewise.
294 (elf_platform) <cpu>: Define new static variable and use it.
295 Translate kernel AT_PLATFORM name to canonical name if needed.
296 Error if platform name is unknown.
298 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
301 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
303 2018-02-01 Jeff Law <law@redhat.com>
306 * config/i386/i386.c (release_scratch_register_on_entry): Add new
307 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
308 the scratch if RELEASE_VIA_POP is false.
309 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
310 If we have to save a temporary register, decrement SIZE appropriately.
311 Pass new arguments to release_scratch_register_on_entry.
312 (ix86_adjust_stack_and_probe): Likewise.
313 (ix86_emit_probe_stack_range): Pass new arguments to
314 release_scratch_register_on_entry.
316 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
318 PR rtl-optimization/84157
319 * combine.c (change_zero_ext): Use REG_P predicate in
320 front of HARD_REGISTER_P predicate.
322 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
324 * config/avr/avr.c (avr_option_override): Move disabling of
325 -fdelete-null-pointer-checks to...
326 * common/config/avr/avr-common.c (avr_option_optimization_table):
329 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
331 PR tree-optimization/81635
332 * tree-data-ref.c (split_constant_offset_1): For types that
333 wrap on overflow, try to use range info to prove that wrapping
336 2018-02-01 Renlin Li <renlin.li@arm.com>
339 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
341 (aarch64_register_move_cost): Likewise.
342 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
344 (REG_CLASS_NAMES): Likewise.
345 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
346 TAILCALL_ADDR_REGS. Remove IP registers.
347 * config/aarch64/aarch64.md (Ucs): Update register constraint.
349 2018-02-01 Richard Biener <rguenther@suse.de>
351 * domwalk.h (dom_walker::dom_walker): Add additional constructor
352 for specifying RPO order and allow NULL for that.
353 * domwalk.c (dom_walker::dom_walker): Likewise.
354 (dom_walker::walk): Handle NULL RPO order.
355 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
357 (rewrite_update_dom_walker): Likewise.
358 (mark_def_dom_walker): Likewise.
360 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
362 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
363 (aarch64_maybe_expand_sve_subreg_move): Declare.
364 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
365 * config/aarch64/predicates.md (aarch64_any_register_operand): New
367 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
368 that are semantically a reverse operation.
369 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
370 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
371 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
373 (aarch64_can_change_mode_class): For big-endian, forbid changes
374 between two SVE modes if they have different element sizes.
376 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
378 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
379 the TImode handling for big-endian targets.
381 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
383 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
384 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
386 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
387 Remove BSWAP handing for big-endian targets and use the form of
388 LD1RQ appropariate for the mode.
390 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
392 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
393 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
396 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
399 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
400 check for operands that need to go through aarch64_sve_reload_be.
402 2018-02-01 Jakub Jelinek <jakub@redhat.com>
404 PR tree-optimization/81661
405 PR tree-optimization/84117
406 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
407 * tree-eh.c: Include gimplify.h.
408 (find_trapping_overflow, replace_trapping_overflow,
409 rewrite_to_non_trapping_overflow): New functions.
410 * tree-vect-loop.c: Include tree-eh.h.
411 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
412 * tree-data-ref.c: Include tree-eh.h.
413 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
415 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
417 PR rtl-optimization/84123
418 * combine.c (change_zero_ext): Check if hard register satisfies
419 can_change_dest_mode before calling gen_lowpart_SUBREG.
421 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
424 * ira.c (ira_init_register_move_cost): Remove assert.
426 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
428 PR rtl-optimization/84071
429 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
430 * doc/tm.texi: Regenerate.
432 2018-01-31 Richard Biener <rguenther@suse.de>
434 PR tree-optimization/84132
435 * tree-data-ref.c (analyze_miv_subscript): Properly
436 check whether evolution_function_is_affine_multivariate_p
437 before calling gcd_of_steps_may_divide_p.
439 2018-01-31 Julia Koval <julia.koval@intel.com>
442 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
443 * config/i386/i386.md (rdpid_rex64) New.
444 (rdpid): Make 32bit only.
446 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
449 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
450 an IDENTIFIER_NODE for FUNCTION_TYPE's.
452 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
455 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
457 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
459 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
461 PR rtl-optimization/84071
462 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
463 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
465 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
467 * config/arc/arc.c (arc_handle_aux_attribute): New function.
468 (arc_attribute_table): Add 'aux' attribute.
469 (arc_in_small_data_p): Consider aux like variables.
470 (arc_is_aux_reg_p): New function.
471 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
472 (arc_get_aux_arg): New function.
473 (prepare_move_operands): Handle aux-register access.
474 (arc_handle_aux_attribute): New function.
475 * doc/extend.texi (ARC Variable attributes): Add subsection.
477 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
479 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
480 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
481 (arc_attribute_table): Add 'uncached' attribute.
482 (arc_print_operand): Print '.di' flag for uncached memory
484 (arc_in_small_data_p): Do not consider for small data the uncached
486 (arc_is_uncached_mem_p): New function.
487 * config/arc/predicates.md (compact_store_memory_operand): Check
488 for uncached memory accesses.
489 (nonvol_nonimm_operand): Likewise.
490 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
492 2018-01-31 Jakub Jelinek <jakub@redhat.com>
495 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
496 falign-loops=): Add Optimization flag.
498 2018-01-30 Jeff Law <law@redhat.com>
501 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
502 INT_REGISTERS_SAVED. Check it prior to calling
503 get_scratch_register_on_entry.
504 (ix86_adjust_stack_and_probe): Similarly.
505 (ix86_emit_probe_stack_range): Similarly.
506 (ix86_expand_prologue): Corresponding changes.
508 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
511 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
512 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
514 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
517 * lra-constraints.c (curr_insn_transform): Process AND in the
520 2018-01-30 Jakub Jelinek <jakub@redhat.com>
522 PR rtl-optimization/83986
523 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
524 dependence against last_pending_memory_flush in addition to
527 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
529 PR tree-optimization/81611
530 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
533 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
535 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
538 2018-01-30 Richard Biener <rguenther@suse.de>
539 Jakub Jelinek <jakub@redhat.com>
541 PR tree-optimization/84111
542 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
543 inner loops added during recursion, as they don't have up-to-date
546 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
549 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
550 (can_inline_edge_by_limits_p): ... here.
551 (can_early_inline_edge_p, check_callers,
552 update_caller_keys, update_callee_keys, recursive_inlining,
553 add_new_edges_to_heap, speculation_useful_p,
554 inline_small_functions,
555 inline_small_functions, flatten_function,
556 inline_to_all_callers_1): Update.
558 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
560 * profile-count.c (profile_count::combine_with_ipa_count): Handle
563 2018-01-30 Richard Biener <rguenther@suse.de>
565 PR tree-optimization/83008
566 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
567 invariant and constant vector uses in stmts when they need
570 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
573 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
574 * configure: Regenerate.
576 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
578 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
580 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
581 Use gen_rtx_REG rather than gen_lowpart.
583 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
585 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
586 rather than 0 when creating partial subregs.
588 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
590 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
593 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
596 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
597 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
598 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
599 flags. This restores the settings used before the 2017-07-24.
600 Turning off pre increment/decrement/modify allows IVOPTS to
601 optimize DF/SF loops where the index is an int.
603 2018-01-29 Richard Biener <rguenther@suse.de>
604 Kelvin Nilsen <kelvin@gcc.gnu.org>
607 * tree-vect-stmts.c (vectorizable_call): Don't call
608 targetm.vectorize_builtin_md_vectorized_function if callee is
611 2018-01-22 Carl Love <cel@us.ibm.com>
613 * doc/extend.tex: Fix typo in second arg in
614 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
616 2018-01-29 Richard Biener <rguenther@suse.de>
618 PR tree-optimization/84086
619 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
620 (flush_ssaname_freelist): When SSA names were released reset
623 2018-01-29 Richard Biener <rguenther@suse.de>
625 PR tree-optimization/84057
626 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
627 removed paths when removing edges.
629 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
631 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
632 -mfunction-return=@var{choice}.
634 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
637 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
639 (layout::print_source_line): Likewise.
640 (test_get_line_width_without_trailing_whitespace): Add test cases.
642 2018-01-27 Jakub Jelinek <jakub@redhat.com>
645 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
648 2018-01-26 Jim Wilson <jimw@sifive.com>
650 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
652 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
655 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
657 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
658 and CMP + SUB-immediate -> SUBS.
660 2018-01-26 Martin Sebor <msebor@redhat.com>
662 PR tree-optimization/83896
663 * tree-ssa-strlen.c (get_string_len): Rename...
664 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
665 Avoid assuming length is constant.
666 (handle_char_store): Use HOST_WIDE_INT for string length.
668 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
671 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
672 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
674 2018-01-26 Richard Biener <rguenther@suse.de>
676 PR rtl-optimization/84003
677 * dse.c (record_store): Only record redundant stores when
678 the earlier store aliases at least all accesses the later one does.
680 2018-01-26 Jakub Jelinek <jakub@redhat.com>
682 PR rtl-optimization/83985
683 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
684 REG_CFA_RESTORE insns.
685 (delete_unmarked_insns): Don't ignore separate shrink wrapping
686 REG_CFA_RESTORE insns here.
689 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
690 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
692 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
694 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
695 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
696 (arc_init): Likewise.
697 (arc_override_options): Likewise.
698 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
700 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
702 * config/arc/arc.h (TARGET_DBNZ): Define.
703 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
704 properly set the tune attribute.
705 (dbnz): Use TARGET_DBNZ guard.
706 * config/arc/arc.opt (mtune): Add core3 option.
708 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
710 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
711 recognize new pic like addresses.
712 (arc_delegitimize_address): Clean up.
714 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
716 * config/arc/arc-arches.def: Option mrf16 valid for all
718 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
719 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
720 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
721 * config/arc/arc-tables.opt: Regenerate.
722 * config/arc/arc.c (arc_conditional_register_usage): Handle
723 reduced register file case.
724 (arc_file_start): Set must have build attributes.
725 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
727 * config/arc/arc.opt (mrf16): Add new option.
728 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
729 * config/arc/genmultilib.awk: Handle new mrf16 option.
730 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
731 * config/arc/t-multilib: Regenerate.
732 * doc/invoke.texi (ARC Options): Document mrf16 option.
734 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
736 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
737 * config/arc/arc.c (arc_handle_secure_attribute): New function.
738 (arc_attribute_table): Add 'secure_call' attribute.
739 (arc_print_operand): Print secure call operand.
740 (arc_function_ok_for_sibcall): Don't optimize tail calls when
742 (arc_is_secure_call_p): New function. * config/arc/arc.md
743 (call_i): Add support for sjli instruction.
744 (call_value_i): Likewise.
745 * config/arc/constraints.md (Csc): New constraint.
747 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
748 John Eric Martin <John.Martin@emmicro-us.com>
750 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
751 * config/arc/arc.c (_arc_jli_section): New struct.
752 (arc_jli_section): New type.
753 (rc_jli_sections): New static variable.
754 (arc_handle_jli_attribute): New function.
755 (arc_attribute_table): Add jli_always and jli_fixed attribute.
756 (arc_file_end): New function.
757 (TARGET_ASM_FILE_END): Define.
758 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
759 (arc_add_jli_section): New function.
760 (jli_call_scan): Likewise.
761 (arc_reorg): Call jli_call_scan.
762 (arc_output_addsi): Remove 'S' from printing asm operand.
763 (arc_is_jli_call_p): New function.
764 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
766 (movhi_insn): Likewise.
767 (movsi_insn): Likewise.
768 (movsi_set_cc_insn): Likewise.
769 (loadqi_update): Likewise.
770 (load_zeroextendqisi_update): Likewise.
771 (load_signextendqisi_update): Likewise.
772 (loadhi_update): Likewise.
773 (load_zeroextendhisi_update): Likewise.
774 (load_signextendhisi_update): Likewise.
775 (loadsi_update): Likewise.
776 (loadsf_update): Likewise.
777 (movsicc_insn): Likewise.
778 (bset_insn): Likewise.
779 (bxor_insn): Likewise.
780 (bclr_insn): Likewise.
781 (bmsk_insn): Likewise.
782 (bicsi3_insn): Likewise.
783 (cmpsi_cc_c_insn): Likewise.
784 (movsi_ne): Likewise.
785 (movsi_cond_exec): Likewise.
786 (clrsbsi2): Likewise.
797 (call_i): Remove 'S' asm letter, add jli instruction.
798 (call_value_i): Likewise.
799 * config/arc/arc.op (mjli-always): New option.
800 * config/arc/constraints.md (Cji): New constraint.
801 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
803 (subsf3_fpx): Likewise.
804 (mulsf3_fpx): Likewise.
805 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
807 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
809 * doc/invoke.texi (ARC): Document mjli-always option.
811 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
813 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
814 and use incw and decw where possible
815 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
817 2018-01-26 Richard Biener <rguenther@suse.de>
819 PR tree-optimization/81082
820 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
821 association if it requires casting to unsigned.
822 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
823 from fold_plusminus_mult_expr to catch important cases late when
824 range info is available.
826 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
828 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
829 * configure.ac (hidden_linkonce): New test.
830 * configure: Regenerate.
831 * config.in: Regenerate.
833 2018-01-26 Julia Koval <julia.koval@intel.com>
835 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
836 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
837 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
838 _mm_mask_bitshuffle_epi64_mask): Fix type.
839 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
840 USI_FTYPE_V4DI_V4DI_USI): Remove.
841 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
842 __builtin_ia32_vpshufbitqmb256_mask,
843 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
844 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
845 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
847 2018-01-26 Alan Modra <amodra@gmail.com>
850 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
851 UNSPEC_VBPERMQ. Sort other unspecs.
853 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
855 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
857 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
860 * predict.c (drop_profile): Do not push/pop cfun; update also
862 (handle_missing_profiles): Fix logic looking for zero profiles.
864 2018-01-25 Jakub Jelinek <jakub@redhat.com>
867 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
868 on functions with #pragma omp declare simd or functions with simd
870 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
871 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
872 Remove trailing \n from warning_at calls.
874 2018-01-25 Tom de Vries <tom@codesourcery.com>
877 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
878 for neutered workers.
880 2018-01-24 Joseph Myers <joseph@codesourcery.com>
883 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
884 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
886 2018-01-24 Jeff Law <law@redhat.com>
889 * i386.c (get_probe_interval): Move to earlier point.
890 (ix86_compute_frame_layout): If -fstack-clash-protection and
891 the frame is larger than the probe interval, then use pushes
892 to save registers rather than reg->mem moves.
893 (ix86_expand_prologue): Remove conditional for int_registers_saved
896 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
899 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
900 min/max for never referenced object.
902 2018-01-24 Jakub Jelinek <jakub@redhat.com>
905 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
907 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
908 attributes from DECL_ATTRIBUTES (decl) without affecting
909 DECL_ATTRIBUTES (current_function_decl).
910 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
911 functions with non-NULL DECL_ABSTRACT_ORIGIN.
913 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
915 PR tree-optimization/83979
916 * fold-const.c (fold_comparison): Use constant_boolean_node
917 instead of boolean_{true,false}_node.
919 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
921 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
924 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
926 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
927 Simplify the clause that sets the length attribute.
928 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
929 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
930 clause that sets the length attribute.
931 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
933 2018-01-24 Tom de Vries <tom@codesourcery.com>
936 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
937 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
938 Add strict parameter.
939 (prevent_branch_around_nothing): Insert dummy insn between branch to
940 label and label with no ptx insn inbetween.
941 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
943 2018-01-24 Tom de Vries <tom@codesourcery.com>
946 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
947 for neutered threads in warp.
948 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
950 2018-01-24 Richard Biener <rguenther@suse.de>
952 PR tree-optimization/83176
953 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
956 2018-01-24 Richard Biener <rguenther@suse.de>
958 PR tree-optimization/82819
959 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
960 code generating pluses that are no-ops in the target precision.
962 2018-01-24 Richard Biener <rguenther@suse.de>
965 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
967 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
969 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
970 to merge probabilities.
971 * predict.c (probably_never_executed): Also mark as cold functions
972 with global 0 profile and guessed local profile.
973 * profile-count.c (profile_probability::combine_with_count): New
975 * profile-count.h (profile_probability::operator*,
976 profile_probability::operator*=, profile_probability::operator/,
977 profile_probability::operator/=): Reduce precision to adjusted
978 and set value to guessed on contradictory divisions.
979 (profile_probability::combine_with_freq): Remove.
980 (profile_probability::combine_wiht_count): Declare.
981 (profile_count::force_nonzero):: Set to adjusted.
982 (profile_count::probability_in):: Set quality to adjusted.
983 * tree-ssa-tail-merge.c (replace_block_by): Use
986 2018-01-23 Andrew Waterman <andrew@sifive.com>
987 Jim Wilson <jimw@sifive.com>
989 * config/riscv/riscv.c (riscv_stack_boundary): New.
990 (riscv_option_override): Set riscv_stack_boundary. Handle
991 riscv_preferred_stack_boundary_arg.
992 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
993 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
994 (STACK_BOUNDARY): Set to riscv_stack_boundary.
995 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
996 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
997 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
999 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1002 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1003 of struct ix86_frame.
1004 (ix86_expand_epilogue): Likewise. Add a local variable for
1005 the reg_save_offset field in struct ix86_frame.
1007 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1009 PR tree-optimization/82604
1010 * tree-loop-distribution.c (enum partition_kind): New enum item
1011 PKIND_PARTIAL_MEMSET.
1012 (partition_builtin_p): Support above new enum item.
1013 (generate_code_for_partition): Ditto.
1014 (compute_access_range): Differentiate cases that equality can be
1015 proven at all loops, the innermost loops or no loops.
1016 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1017 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1018 (finalize_partitions, distribute_loop): Don't fuse partition of
1019 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1020 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1023 2018-01-23 Martin Liska <mliska@suse.cz>
1025 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1026 order to ignore the predictor.
1027 (PRED_POLYMORPHIC_CALL): Likewise.
1028 (PRED_RECURSIVE_CALL): Likewise.
1030 2018-01-23 Martin Liska <mliska@suse.cz>
1032 * tree-profile.c (tree_profiling): Print function header to
1033 aware reader which function we are working on.
1034 * value-prof.c (gimple_find_values_to_profile): Do not print
1035 not interesting value histograms.
1037 2018-01-23 Martin Liska <mliska@suse.cz>
1039 * profile-count.h (enum profile_quality): Add
1040 profile_uninitialized as the first value. Do not number values
1041 as they are zero based.
1042 (profile_count::verify): Update sanity check.
1043 (profile_probability::verify): Likewise.
1045 2018-01-23 Nathan Sidwell <nathan@acm.org>
1047 * doc/invoke.texi (ffor-scope): Deprecate.
1049 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1051 PR tree-optimization/83510
1052 * domwalk.c (set_all_edges_as_executable): New function.
1053 (dom_walker::dom_walker): Convert bool param
1054 "skip_unreachable_blocks" to enum reachability. Move setup of
1055 edge flags to set_all_edges_as_executable and only do it when
1056 reachability is REACHABLE_BLOCKS.
1057 * domwalk.h (enum dom_walker::reachability): New enum.
1058 (dom_walker::dom_walker): Convert bool param
1059 "skip_unreachable_blocks" to enum reachability.
1060 (set_all_edges_as_executable): New decl.
1061 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1062 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1064 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1065 but converting true to REACHABLE_BLOCKS.
1066 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1068 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1069 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1070 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1072 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1073 if check_all_array_refs will be called.
1075 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1077 * tree.c (selftest::test_location_wrappers): Add more test
1080 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1082 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1083 (selftest::test_bit_in_range): Likewise.
1085 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1088 * doc/sourcebuild.texi (vect_float): Say that the selector
1089 only describes the situation when -funsafe-math-optimizations is on.
1090 (vect_float_strict): Document.
1092 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1094 PR tree-optimization/83965
1095 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1096 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1097 instead of checking only for a reduction.
1098 (vect_recog_widen_sum_pattern): Likewise.
1100 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1102 * predict.c (probably_never_executed): Only use precise profile info.
1103 (compute_function_frequency): Skip after inlining hack since we now
1104 have quality checking.
1106 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1108 * profile-count.h (profile_probability::very_unlikely,
1109 profile_probability::unlikely, profile_probability::even): Set
1110 precision to guessed.
1112 2018-01-23 Richard Biener <rguenther@suse.de>
1114 PR tree-optimization/83963
1115 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1116 Properly terminate dominator walk when crossing the exit edge not
1117 when visiting its source block.
1119 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1122 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1123 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1125 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1127 PR tree-optimization/83957
1128 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1129 semicolon after for body surrounded by braces.
1131 PR tree-optimization/83081
1132 * profile-count.h (profile_probability::split): New method.
1133 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1134 Use profile_probability::split.
1135 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1136 when splitting a single conditional jump into 2.
1138 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1140 PR tree-optimization/69452
1141 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1144 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1146 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1147 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1148 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1150 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1152 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1153 * config/rl78/rl78.md: New define_expand "movdi"
1154 * config/rl78/rl78.c: New function definition rl78_split_movdi
1156 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1159 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1161 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1162 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1163 128-bit to produce an UNSPEC move to get the double word with the
1164 signbit and then a shift directly to do signbit.
1165 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1166 implementation with a new version that just does either a direct
1167 move or a regular move. Move memory interface to separate insns.
1168 Move insns so they are next to the expander.
1169 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1170 with signbit move. Split big and little endian case.
1171 (signbit<mode>2_dm_mem_le): Likewise.
1172 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1173 (signbit<mode>2_dm2): Likewise.
1175 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1177 * config/rl78/rl78.md: New define_expand "anddi3".
1179 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1181 * config/rl78/rl78.md: New define_expand "umindi3".
1183 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1185 * config/rl78/rl78.md: New define_expand "smindi3".
1187 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1189 * config/rl78/rl78.md: New define_expand "smaxdi3".
1191 2018-01-22 Carl Love <cel@us.ibm.com>
1193 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1194 LVX_V1TI): Add macro expansion.
1195 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1196 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1197 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1198 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1199 Change check to determine if the instruction is a byte reversing
1200 entry. Fix typo in comment.
1201 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1202 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1203 Add def_builtin calls for new builtins.
1204 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1205 Add define_insn expansion.
1207 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1209 * config/rl78/rl78.md: New define_expand "umaxdi3".
1211 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1213 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1214 for non-QImode registers
1216 2018-01-22 Richard Biener <rguenther@suse.de>
1218 PR tree-optimization/83963
1219 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1220 including the loop exit block.
1221 (scop_detection::merge_sese): Likewise.
1222 (scop_detection::add_scop): Do it here instead.
1224 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1226 * doc/sourcebuild.texi (arm_softfloat): Document.
1228 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1231 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1232 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1233 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1235 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1236 David Edelsohn <dje.gcc@gmail.com>
1239 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1240 Change "crset eq" to "crset 2".
1241 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1242 (*call_indirect_aix<mode>_nospec): Likewise.
1243 (*call_value_indirect_aix<mode>_nospec): Likewise.
1244 (*call_indirect_elfv2<mode>_nospec): Likewise.
1245 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1246 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1247 change assembly output from . to $.
1248 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1249 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1250 (*tablejump<mode>_internal1_nospec): Likewise.
1252 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1255 * config/sh/sh_optimize_sett_clrt.cc:
1256 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1258 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1260 PR tree-optimization/83940
1261 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1262 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1263 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1264 use it to pass back the definition type.
1265 (vect_check_store_rhs): Likewise rhs_dt_out.
1266 (vect_build_gather_load_calls): Add a mask_dt argument and use
1267 it instead of a call to vect_is_simple_use.
1268 (vectorizable_store): Update calls to vect_check_load_store_mask
1269 and vect_check_store_rhs. Use the dt returned by the latter instead
1270 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1271 instead of calls to vect_is_simple_use. Pass the scalar rather
1272 than the vector operand to vect_is_simple_use when handling
1273 second and subsequent copies of an rhs value.
1274 (vectorizable_load): Update calls to vect_check_load_store_mask
1275 and vect_build_gather_load_calls. Use the cached mask_dt and
1276 gs_info.offset_dt instead of calls to vect_is_simple_use.
1278 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1281 * tree-emutls.c: Include gimplify.h.
1282 (lower_emutls_2): New function.
1283 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1284 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1285 it before further processing.
1288 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1289 UINTVAL (trueop1) instead of INTVAL (op1).
1291 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1295 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1296 INCOMING_FRAME_SP_OFFSET if not defined.
1297 (scan_trace): Add ENTRY argument. If true and
1298 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1299 emit a note to adjust the CFA offset.
1300 (create_cfi_notes): Adjust scan_trace callers.
1301 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1302 INCOMING_FRAME_SP_OFFSET in the CIE.
1303 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1304 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1306 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1307 * doc/tm.texi: Regenerated.
1309 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1311 PR rtl-optimization/83147
1312 * lra-constraints.c (remove_inheritance_pseudos): Use
1313 lra_substitute_pseudo_within_insn.
1315 2018-01-19 Tom de Vries <tom@codesourcery.com>
1316 Cesar Philippidis <cesar@codesourcery.com>
1319 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1321 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1324 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1325 spaces for function labels.
1327 2018-01-19 Martin Liska <mliska@suse.cz>
1329 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1330 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1331 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1332 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1333 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1334 (PRED_CONST_RETURN): Change from 69 to 65.
1335 (PRED_NULL_RETURN): Change from 91 to 71.
1336 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1337 (PRED_LOOP_GUARD): Change from 66 to 73.
1339 2018-01-19 Martin Liska <mliska@suse.cz>
1341 * predict.c (predict_insn_def): Add new assert.
1342 (struct branch_predictor): Change type to signed integer.
1343 (test_prediction_value_range): Amend test to cover
1345 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1346 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1347 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1348 (PRED_LOOP_IV_COMPARE): Likewise.
1349 * predict.h (PROB_UNINITIALIZED): Define new constant.
1351 2018-01-19 Martin Liska <mliska@suse.cz>
1353 * predict.c (dump_prediction): Add new format for
1354 analyze_brprob.py script which is enabled with -details
1356 * profile-count.h (precise_p): New function.
1358 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1360 PR tree-optimization/83922
1361 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1362 there are no statements that need masking.
1363 (vect_active_double_reduction_p): New function.
1364 (vect_analyze_loop_operations): Use it when handling phis that
1365 are not in the loop header.
1367 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1369 PR tree-optimization/83914
1370 * tree-vect-loop.c (vectorizable_induction): Don't convert
1371 init_expr or apply the peeling adjustment for inductions
1372 that are nested within the vectorized loop.
1374 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1376 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1379 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1383 * function.h (gimplify_parameters): Add gimple_seq * argument.
1384 * function.c: Include gimple.h and options.h.
1385 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1386 for the added local temporaries if needed.
1387 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1388 if there are any parameter cleanups, wrap whole body into a
1389 try/finally with the cleanups.
1391 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1394 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1395 Use GET_MODE_CLASS for scalar floating point.
1397 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1401 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1402 Fix call of call_cgraph_insertion_hooks.
1404 2018-01-18 Martin Sebor <msebor@redhat.com>
1406 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1408 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1411 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1414 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1417 * common.opt: (-ffile-prefix-map): New option.
1418 * opts.c (common_handle_option): Defer it.
1419 * opts-global.c (handle_common_deferred_options): Handle it.
1420 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1421 * file-prefix-map.h: New file.
1422 (remap_debug_filename, add_debug_prefix_map): ...here.
1423 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1424 * final.c (debug_prefix_map, add_debug_prefix_map
1425 remap_debug_filename): Move to...
1426 * file-prefix-map.c: New file.
1427 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1428 generalize, get rid of alloca(), use strrchr() instead of strchr().
1429 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1430 Implement in terms of add_prefix_map().
1431 (remap_macro_filename, remap_debug_filename): Implement in term of
1433 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1434 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1435 * dbxout.c: Include file-prefix-map.h.
1436 * varasm.c: Likewise.
1437 * vmsdbgout.c: Likewise.
1438 * xcoffout.c: Likewise.
1439 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1440 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1441 * doc/invoke.texi (-ffile-prefix-map): Document.
1442 (-fdebug-prefix-map): Update description.
1444 2018-01-18 Martin Liska <mliska@suse.cz>
1446 * config/i386/i386.c (indirect_thunk_name): Document that also
1448 (output_indirect_thunk): Document why both instructions
1449 (pause and lfence) are generated.
1451 2018-01-18 Richard Biener <rguenther@suse.de>
1453 PR tree-optimization/83887
1454 * graphite-scop-detection.c
1455 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1456 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1457 (scop_detection::merge_sese): Re-implement with a flood-fill
1458 algorithm that properly finds a SESE region if it exists.
1460 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1463 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1464 pointer_diff optimizations use view_convert instead of convert.
1466 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1468 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1469 Generate different code for -mno-speculate-indirect-jumps.
1470 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1471 (*call_indirect_aix<mode>): Disable for
1472 -mno-speculate-indirect-jumps.
1473 (*call_indirect_aix<mode>_nospec): New define_insn.
1474 (*call_value_indirect_aix<mode>): Disable for
1475 -mno-speculate-indirect-jumps.
1476 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1477 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1478 -mno-speculate-indirect-jumps.
1479 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1481 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1483 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1484 long double type, set the flags for noting the default long double
1485 type, even if we don't pass or return a long double type.
1487 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1490 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1493 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1495 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1496 support for merge[hl].
1497 (fold_mergehl_helper): New helper function.
1498 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1499 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1500 (altivec_vmrglw_direct): Add xxmrglw insn.
1502 2018-01-17 Andrew Waterman <andrew@sifive.com>
1504 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1505 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1507 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1510 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1511 call the lto_location_cache before reading the
1512 DECL_SOURCE_LOCATION of the types.
1514 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1515 Richard Sandiford <richard.sandiford@linaro.org>
1517 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1518 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1519 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1520 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1521 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1523 * config/aarch64/constraints.md (aarch64_movti_operand):
1525 * config/aarch64/predicates.md (Uti): Add new constraint.
1527 2018-01-17 Carl Love <cel@us.ibm.com>
1528 * config/rs6000/vsx.md (define_expand xl_len_r,
1529 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1530 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1532 (define_expand, define_insn): Move the shift left from the
1533 define_insn to the define_expand for lxvl and stxvl instructions.
1534 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1535 and XL_LEN_R definitions to PURE.
1537 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1539 * config/i386/i386.c (indirect_thunk_name): Declare regno
1540 as unsigned int. Compare regno with INVALID_REGNUM.
1541 (output_indirect_thunk): Ditto.
1542 (output_indirect_thunk_function): Ditto.
1543 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1544 in the call to output_indirect_thunk_function.
1546 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1549 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1550 rather than the size of inner_type to determine the stack slot size
1551 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1553 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1556 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1559 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1561 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1562 endian Linux systems to optionally enable multilibs for selecting
1563 the long double type if the user configured an explicit type.
1564 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1565 have no long double multilibs if not defined.
1566 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1567 warn if the user used -mabi={ieee,ibm}longdouble and we built
1568 multilibs for long double.
1569 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1570 appropriate multilib option.
1571 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1573 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1574 for building long double multilibs.
1575 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1577 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1579 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1582 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1584 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1587 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1590 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1593 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1595 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1596 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1598 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1600 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1601 different rtl trees depending on TARGET_64BIT.
1602 (rs6000_gen_lvx): Likewise.
1604 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1606 * config/visium/visium.md (nop): Tweak comment.
1607 (hazard_nop): Likewise.
1609 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1611 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1612 -mspeculate-indirect-jumps.
1613 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1614 for -mno-speculate-indirect-jumps.
1615 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1616 (*call_value_indirect_elfv2<mode>): Disable for
1617 -mno-speculate-indirect-jumps.
1618 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1619 (indirect_jump): Emit different RTL for
1620 -mno-speculate-indirect-jumps.
1621 (*indirect_jump<mode>): Disable for
1622 -mno-speculate-indirect-jumps.
1623 (*indirect_jump<mode>_nospec): New define_insn.
1624 (tablejump): Emit different RTL for
1625 -mno-speculate-indirect-jumps.
1626 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1627 (tablejumpsi_nospec): New define_expand.
1628 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1629 (tablejumpdi_nospec): New define_expand.
1630 (*tablejump<mode>_internal1): Disable for
1631 -mno-speculate-indirect-jumps.
1632 (*tablejump<mode>_internal1_nospec): New define_insn.
1633 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1636 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1638 * caller-save.c (insert_save): Drop unnecessary parameter. All
1641 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1642 Richard Biener <rguenth@suse.de>
1645 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1646 return early, inline manually is_gimple_sizepos. Make sure if we
1647 call gimplify_expr we don't end up with a gimple constant.
1648 * tree.c (variably_modified_type_p): Don't return true for
1649 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1650 * gimplify.h (is_gimple_sizepos): Remove.
1652 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1654 PR tree-optimization/83857
1655 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1656 vectorizable_live_operation for pure SLP statements.
1657 (vectorizable_live_operation): Handle PHIs.
1659 2018-01-16 Richard Biener <rguenther@suse.de>
1661 PR tree-optimization/83867
1662 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1663 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1665 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1668 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1669 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1670 If off is not INTEGER_CST, issue a may not be aligned warning
1671 rather than isn't aligned. Use isn%'t rather than isn't.
1672 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1674 <case MULT_EXPR>: Improve the case when bottom and one of the
1675 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1676 operand, in that case check if the other operand is multiple of
1677 bottom divided by the INTEGER_CST operand.
1679 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1682 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1683 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1684 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1685 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1686 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1687 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1688 (pa_function_arg_size): New function.
1690 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1692 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1693 in a separate statement.
1695 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1697 PR tree-optimization/83847
1698 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1699 group gathers and scatters.
1701 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1703 PR rtl-optimization/86620
1704 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1706 PR rtl-optimization/83213
1707 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1708 to last if both are JUMP_INSNs.
1710 PR tree-optimization/83843
1711 * gimple-ssa-store-merging.c
1712 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1713 store_immediate_info for bswap/nop orig_stores.
1715 2018-01-15 Andrew Waterman <andrew@sifive.com>
1717 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1719 <UDIV>: Increase cost if !TARGET_DIV.
1721 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1723 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1724 (define_attr "cr_logical_3op"): New.
1725 (cceq_ior_compare): Adjust.
1726 (cceq_ior_compare_complement): Adjust.
1727 (*cceq_rev_compare): Adjust.
1728 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1729 (is_cracked_insn): Adjust.
1730 (insn_must_be_first_in_group): Adjust.
1731 * config/rs6000/40x.md: Adjust.
1732 * config/rs6000/440.md: Adjust.
1733 * config/rs6000/476.md: Adjust.
1734 * config/rs6000/601.md: Adjust.
1735 * config/rs6000/603.md: Adjust.
1736 * config/rs6000/6xx.md: Adjust.
1737 * config/rs6000/7450.md: Adjust.
1738 * config/rs6000/7xx.md: Adjust.
1739 * config/rs6000/8540.md: Adjust.
1740 * config/rs6000/cell.md: Adjust.
1741 * config/rs6000/e300c2c3.md: Adjust.
1742 * config/rs6000/e500mc.md: Adjust.
1743 * config/rs6000/e500mc64.md: Adjust.
1744 * config/rs6000/e5500.md: Adjust.
1745 * config/rs6000/e6500.md: Adjust.
1746 * config/rs6000/mpc.md: Adjust.
1747 * config/rs6000/power4.md: Adjust.
1748 * config/rs6000/power5.md: Adjust.
1749 * config/rs6000/power6.md: Adjust.
1750 * config/rs6000/power7.md: Adjust.
1751 * config/rs6000/power8.md: Adjust.
1752 * config/rs6000/power9.md: Adjust.
1753 * config/rs6000/rs64.md: Adjust.
1754 * config/rs6000/titan.md: Adjust.
1756 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1758 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1759 ix86_indirect_branch_register logic.
1761 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1763 * config/i386/constraints.md (Bs): Update
1764 ix86_indirect_branch_register check. Don't check
1765 ix86_indirect_branch_register with GOT_memory_operand.
1767 * config/i386/predicates.md (GOT_memory_operand): Don't check
1768 ix86_indirect_branch_register here.
1769 (GOT32_symbol_operand): Likewise.
1771 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1773 * config/i386/predicates.md (constant_call_address_operand):
1774 Rewrite ix86_indirect_branch_register logic.
1775 (sibcall_insn_operand): Likewise.
1777 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1779 * config/i386/constraints.md (Bs): Replace
1780 ix86_indirect_branch_thunk_register with
1781 ix86_indirect_branch_register.
1783 * config/i386/i386.md (indirect_jump): Likewise.
1784 (tablejump): Likewise.
1785 (*sibcall_memory): Likewise.
1786 (*sibcall_value_memory): Likewise.
1787 Peepholes of indirect call and jump via memory: Likewise.
1788 * config/i386/i386.opt: Likewise.
1789 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1790 (GOT_memory_operand): Likewise.
1791 (call_insn_operand): Likewise.
1792 (sibcall_insn_operand): Likewise.
1793 (GOT32_symbol_operand): Likewise.
1795 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1798 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1799 type rather than type addr's type points to.
1800 (expand_omp_atomic_mutex): Likewise.
1801 (expand_omp_atomic): Likewise.
1803 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1806 * config/i386/i386.c (output_indirect_thunk_function): Use
1807 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1808 for __x86_return_thunk.
1810 2018-01-15 Richard Biener <rguenther@suse.de>
1813 * expmed.c (extract_bit_field_1): Fix typo.
1815 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1818 * config/arm/iterators.md (VF): New mode iterator.
1819 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1820 Remove integer-related logic from pattern.
1821 (neon_vabd<mode>_3): Likewise.
1823 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1826 * common.opt (fstrict-overflow): No longer an alias.
1827 (fwrapv-pointer): New option.
1828 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1829 also for pointer types based on flag_wrapv_pointer.
1830 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1831 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1832 opts->x_flag_wrapv got set.
1833 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1834 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1835 POINTER_TYPE_OVERFLOW_UNDEFINED.
1836 * match.pd: Likewise in address comparison pattern.
1837 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1839 2018-01-15 Richard Biener <rguenther@suse.de>
1842 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1843 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1844 Reset type names to their identifier if their TYPE_DECL doesn't
1845 have linkage (and thus is used for ODR and devirt).
1846 (save_debug_info_for_decl): Remove.
1847 (save_debug_info_for_type): Likewise.
1848 (add_tree_to_fld_list): Adjust.
1849 * tree-pretty-print.c (dump_generic_node): Make dumping of
1850 type names more robust.
1852 2018-01-15 Richard Biener <rguenther@suse.de>
1854 * BASE-VER: Bump to 8.0.1.
1856 2018-01-14 Martin Sebor <msebor@redhat.com>
1859 * builtins.c (check_access): Avoid warning when the no-warning bit
1862 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1864 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1865 * ira-color (allocno_hard_regs_compare): Likewise.
1867 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1870 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1871 Use .pushsection/.popsection.
1873 2018-01-14 Martin Sebor <msebor@redhat.com>
1876 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1878 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1880 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1881 entry from extra_headers.
1882 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1883 extra_headers, make the list bitwise identical to the i?86-*-* one.
1885 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1887 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1888 -mcmodel=large with -mindirect-branch=thunk,
1889 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1890 -mfunction-return=thunk-extern.
1891 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1892 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1893 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1895 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1897 * config/i386/i386.c (print_reg): Print the name of the full
1898 integer register without '%'.
1899 (ix86_print_operand): Handle 'V'.
1900 * doc/extend.texi: Document 'V' modifier.
1902 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1904 * config/i386/constraints.md (Bs): Disallow memory operand for
1905 -mindirect-branch-register.
1907 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1908 (GOT_memory_operand): Likewise.
1909 (call_insn_operand): Likewise.
1910 (sibcall_insn_operand): Likewise.
1911 (GOT32_symbol_operand): Likewise.
1912 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1913 for -mindirect-branch-register.
1914 (tablejump): Likewise.
1915 (*sibcall_memory): Likewise.
1916 (*sibcall_value_memory): Likewise.
1917 Disallow peepholes of indirect call and jump via memory for
1918 -mindirect-branch-register.
1919 (*call_pop): Replace m with Bw.
1920 (*call_value_pop): Likewise.
1921 (*sibcall_pop_memory): Replace m with Bs.
1922 * config/i386/i386.opt (mindirect-branch-register): New option.
1923 * doc/invoke.texi: Document -mindirect-branch-register option.
1925 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1927 * config/i386/i386-protos.h (ix86_output_function_return): New.
1928 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1929 set function_return_type.
1930 (indirect_thunk_name): Add ret_p to indicate thunk for function
1932 (output_indirect_thunk_function): Pass false to
1933 indirect_thunk_name.
1934 (ix86_output_indirect_branch_via_reg): Likewise.
1935 (ix86_output_indirect_branch_via_push): Likewise.
1936 (output_indirect_thunk_function): Create alias for function
1937 return thunk if regno < 0.
1938 (ix86_output_function_return): New function.
1939 (ix86_handle_fndecl_attribute): Handle function_return.
1940 (ix86_attribute_table): Add function_return.
1941 * config/i386/i386.h (machine_function): Add
1942 function_return_type.
1943 * config/i386/i386.md (simple_return_internal): Use
1944 ix86_output_function_return.
1945 (simple_return_internal_long): Likewise.
1946 * config/i386/i386.opt (mfunction-return=): New option.
1947 (indirect_branch): Mention -mfunction-return=.
1948 * doc/extend.texi: Document function_return function attribute.
1949 * doc/invoke.texi: Document -mfunction-return= option.
1951 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1953 * config/i386/i386-opts.h (indirect_branch): New.
1954 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1955 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1956 with local indirect jump when converting indirect call and jump.
1957 (ix86_set_indirect_branch_type): New.
1958 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1959 (indirectlabelno): New.
1960 (indirect_thunk_needed): Likewise.
1961 (indirect_thunk_bnd_needed): Likewise.
1962 (indirect_thunks_used): Likewise.
1963 (indirect_thunks_bnd_used): Likewise.
1964 (INDIRECT_LABEL): Likewise.
1965 (indirect_thunk_name): Likewise.
1966 (output_indirect_thunk): Likewise.
1967 (output_indirect_thunk_function): Likewise.
1968 (ix86_output_indirect_branch_via_reg): Likewise.
1969 (ix86_output_indirect_branch_via_push): Likewise.
1970 (ix86_output_indirect_branch): Likewise.
1971 (ix86_output_indirect_jmp): Likewise.
1972 (ix86_code_end): Call output_indirect_thunk_function if needed.
1973 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1975 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1976 (ix86_attribute_table): Add indirect_branch.
1977 * config/i386/i386.h (machine_function): Add indirect_branch_type
1978 and has_local_indirect_jump.
1979 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1981 (tablejump): Likewise.
1982 (*indirect_jump): Use ix86_output_indirect_jmp.
1983 (*tablejump_1): Likewise.
1984 (simple_return_indirect_internal): Likewise.
1985 * config/i386/i386.opt (mindirect-branch=): New option.
1986 (indirect_branch): New.
1989 (thunk-inline): Likewise.
1990 (thunk-extern): Likewise.
1991 * doc/extend.texi: Document indirect_branch function attribute.
1992 * doc/invoke.texi: Document -mindirect-branch= option.
1994 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1997 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1999 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2001 * ipa-inline.c (want_inline_small_function_p): Return false if
2002 inlining has already failed with CIF_FINAL_ERROR.
2003 (update_caller_keys): Call want_inline_small_function_p before
2005 (update_callee_keys): Likewise.
2007 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2009 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2011 (rs6000_quadword_masked_address_p): Likewise.
2012 (quad_aligned_load_p): Likewise.
2013 (quad_aligned_store_p): Likewise.
2014 (const_load_sequence_p): Add comment to describe the outer-most loop.
2015 (mimic_memory_attributes_and_flags): New function.
2016 (rs6000_gen_stvx): Likewise.
2017 (replace_swapped_aligned_store): Likewise.
2018 (rs6000_gen_lvx): Likewise.
2019 (replace_swapped_aligned_load): Likewise.
2020 (replace_swapped_load_constant): Capitalize argument name in
2021 comment describing this function.
2022 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2023 and stores that access quad-word aligned addresses and replace
2024 with stvx or lvx instructions when appropriate.
2025 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2026 New function prototype.
2027 (rs6000_quadword_masked_address_p): Likewise.
2028 (rs6000_gen_lvx): Likewise.
2029 (rs6000_gen_stvx): Likewise.
2030 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2031 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2032 when memory address is aligned.
2033 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2034 this split to select lvx instruction when memory address is aligned.
2035 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2036 instruction when memory address is aligned.
2037 (*vsx_le_perm_load_v16qi): Likewise.
2038 (four unnamed splitters): Modify to select the stvx instruction
2039 when memory is aligned.
2041 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2043 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2044 which appears in the queue multiple times.
2046 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2047 Alan Hayward <alan.hayward@arm.com>
2048 David Sherwood <david.sherwood@arm.com>
2050 * tree-vectorizer.h (vec_lower_bound): New structure.
2051 (_loop_vec_info): Add check_nonzero and lower_bounds.
2052 (LOOP_VINFO_CHECK_NONZERO): New macro.
2053 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2054 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2055 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2056 fields. Make seg_len the distance travelled, not including the
2058 (dr_direction_indicator): Declare.
2059 (dr_zero_step_indicator): Likewise.
2060 (dr_known_forward_stride_p): Likewise.
2061 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2063 (runtime_alias_check_p): Allow runtime alias checks with
2065 (operator ==): Compare access_size and align.
2066 (prune_runtime_alias_test_list): Rework for new distinction between
2067 the access_size and seg_len.
2068 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2070 (get_segment_min_max): New function.
2071 (create_intersect_range_checks): Use it.
2072 (dr_step_indicator): New function.
2073 (dr_direction_indicator): Likewise.
2074 (dr_zero_step_indicator): Likewise.
2075 (dr_known_forward_stride_p): Likewise.
2076 * tree-loop-distribution.c (data_ref_segment_size): Return
2077 DR_STEP * (niters - 1).
2078 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2080 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2081 (vect_preserves_scalar_order_p): New function, split out from...
2082 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2083 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2084 (vect_vfa_access_size): New function.
2085 (vect_vfa_align): Likewise.
2086 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2087 (dump_lower_bound): New function.
2088 (vect_check_lower_bound): Likewise.
2089 (vect_small_gap_p): Likewise.
2090 (vectorizable_with_step_bound_p): Likewise.
2091 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2092 depencies if the vectorization factor is 1. Convert the checks
2093 for nonzero steps into checks on the bounds of DR_STEP. Try using
2094 a bunds check for variable steps if the minimum required step is
2095 relatively small. Update calls to the dr_with_seg_len
2096 constructor and to vect_compile_time_alias.
2097 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2099 (vect_loop_versioning): Call it.
2100 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2102 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2104 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2105 Alan Hayward <alan.hayward@arm.com>
2106 David Sherwood <david.sherwood@arm.com>
2108 * doc/sourcebuild.texi (vect_scatter_store): Document.
2109 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2111 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2113 * genopinit.c (main): Add supports_vec_scatter_store and
2114 supports_vec_scatter_store_cached to target_optabs.
2115 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2116 IFN_MASK_SCATTER_STORE.
2117 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2119 * internal-fn.h (internal_store_fn_p): Declare.
2120 (internal_fn_stored_value_index): Likewise.
2121 * internal-fn.c (scatter_store_direct): New macro.
2122 (expand_scatter_store_optab_fn): New function.
2123 (direct_scatter_store_optab_supported_p): New macro.
2124 (internal_store_fn_p): New function.
2125 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2126 IFN_MASK_SCATTER_STORE.
2127 (internal_fn_mask_index): Likewise.
2128 (internal_fn_stored_value_index): New function.
2129 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2131 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2132 * optabs-query.c (supports_vec_scatter_store_p): New function.
2133 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2134 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2135 true for scatter stores.
2136 (vect_gather_scatter_fn_p): Handle scatter stores too.
2137 (vect_check_gather_scatter): Consider using scatter stores if
2138 supports_vec_scatter_store_p.
2139 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2141 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2142 internal_fn_stored_value_index.
2143 (check_load_store_masking): Handle scatter stores too.
2144 (vect_get_store_rhs): Make public.
2145 (vectorizable_call): Use internal_store_fn_p.
2146 (vectorizable_store): Handle scatter store internal functions.
2147 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2148 when deciding whether the end of the group has been reached.
2149 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2150 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2151 (mask_scatter_store<mode>): New insns.
2153 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2154 Alan Hayward <alan.hayward@arm.com>
2155 David Sherwood <david.sherwood@arm.com>
2157 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2158 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2159 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2161 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2162 Use vect_truncate_gather_scatter_offset if we can't treat the
2163 operation as a normal gather load or scatter store.
2164 (get_group_load_store_type): Take the gather_scatter_info
2165 as argument. Try using a gather load or scatter store for
2166 single-element groups.
2167 (get_load_store_type): Update calls to get_group_load_store_type
2168 and vect_use_strided_gather_scatters_p.
2170 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2171 Alan Hayward <alan.hayward@arm.com>
2172 David Sherwood <david.sherwood@arm.com>
2174 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2175 optional tree argument.
2176 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2178 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2179 but continue to use the current value as a fallback.
2180 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2181 to compare the updates.
2182 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2183 (get_load_store_type): Use it when handling a strided access.
2184 (vect_get_strided_load_store_ops): New function.
2185 (vect_get_data_ptr_increment): Likewise.
2186 (vectorizable_load): Handle strided gather loads. Always pass
2187 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2189 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2190 Alan Hayward <alan.hayward@arm.com>
2191 David Sherwood <david.sherwood@arm.com>
2193 * doc/md.texi (gather_load@var{m}): Document.
2194 (mask_gather_load@var{m}): Likewise.
2195 * genopinit.c (main): Add supports_vec_gather_load and
2196 supports_vec_gather_load_cached to target_optabs.
2197 * optabs-tree.c (init_tree_optimization_optabs): Use
2198 ggc_cleared_alloc to allocate target_optabs.
2199 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2200 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2202 * internal-fn.h (internal_load_fn_p): Declare.
2203 (internal_gather_scatter_fn_p): Likewise.
2204 (internal_fn_mask_index): Likewise.
2205 (internal_gather_scatter_fn_supported_p): Likewise.
2206 * internal-fn.c (gather_load_direct): New macro.
2207 (expand_gather_load_optab_fn): New function.
2208 (direct_gather_load_optab_supported_p): New macro.
2209 (direct_internal_fn_optab): New function.
2210 (internal_load_fn_p): Likewise.
2211 (internal_gather_scatter_fn_p): Likewise.
2212 (internal_fn_mask_index): Likewise.
2213 (internal_gather_scatter_fn_supported_p): Likewise.
2214 * optabs-query.c (supports_at_least_one_mode_p): New function.
2215 (supports_vec_gather_load_p): Likewise.
2216 * optabs-query.h (supports_vec_gather_load_p): Declare.
2217 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2218 and memory_type field.
2219 (NUM_PATTERNS): Bump to 15.
2220 * tree-vect-data-refs.c: Include internal-fn.h.
2221 (vect_gather_scatter_fn_p): New function.
2222 (vect_describe_gather_scatter_call): Likewise.
2223 (vect_check_gather_scatter): Try using internal functions for
2224 gather loads. Recognize existing calls to a gather load function.
2225 (vect_analyze_data_refs): Consider using gather loads if
2226 supports_vec_gather_load_p.
2227 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2228 (vect_get_gather_scatter_offset_type): Likewise.
2229 (vect_convert_mask_for_vectype): Likewise.
2230 (vect_add_conversion_to_patterm): Likewise.
2231 (vect_try_gather_scatter_pattern): Likewise.
2232 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2233 (vect_vect_recog_func_ptrs): Add it.
2234 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2235 internal_fn_mask_index and internal_gather_scatter_fn_p.
2236 (check_load_store_masking): Take the gather_scatter_info as an
2237 argument and handle gather loads.
2238 (vect_get_gather_scatter_ops): New function.
2239 (vectorizable_call): Check internal_load_fn_p.
2240 (vectorizable_load): Likewise. Handle gather load internal
2242 (vectorizable_store): Update call to check_load_store_masking.
2243 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2244 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2245 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2246 (aarch64_gather_scale_operand_d): New predicates.
2247 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2248 (mask_gather_load<mode>): New insns.
2250 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2251 Alan Hayward <alan.hayward@arm.com>
2252 David Sherwood <david.sherwood@arm.com>
2254 * optabs.def (fold_left_plus_optab): New optab.
2255 * doc/md.texi (fold_left_plus_@var{m}): Document.
2256 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2257 * internal-fn.c (fold_left_direct): Define.
2258 (expand_fold_left_optab_fn): Likewise.
2259 (direct_fold_left_optab_supported_p): Likewise.
2260 * fold-const-call.c (fold_const_fold_left): New function.
2261 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2262 * tree-parloops.c (valid_reduction_p): New function.
2263 (gather_scalar_reductions): Use it.
2264 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2265 (vect_finish_replace_stmt): Declare.
2266 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2267 (needs_fold_left_reduction_p): New function, split out from...
2268 (vect_is_simple_reduction): ...here. Accept reductions that
2269 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2270 (vect_force_simple_reduction): Also store the reduction type in
2271 the assignment's STMT_VINFO_REDUC_TYPE.
2272 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2273 (merge_with_identity): New function.
2274 (vect_expand_fold_left): Likewise.
2275 (vectorize_fold_left_reduction): Likewise.
2276 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2277 scalar phi in place for it. Check for target support and reject
2278 cases that would reassociate the operation. Defer the transform
2279 phase to vectorize_fold_left_reduction.
2280 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2281 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2282 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2284 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2286 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2287 call to ifc_temp_var.
2289 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2290 Alan Hayward <alan.hayward@arm.com>
2291 David Sherwood <david.sherwood@arm.com>
2293 * target.def (legitimize_address_displacement): Take the original
2294 offset as a poly_int.
2295 * targhooks.h (default_legitimize_address_displacement): Update
2297 * targhooks.c (default_legitimize_address_displacement): Likewise.
2298 * doc/tm.texi: Regenerate.
2299 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2300 as an argument, moving assert of ad->disp == ad->disp_term to...
2301 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2302 Try calling targetm.legitimize_address_displacement before expanding
2303 the address rather than afterwards, and adjust for the new interface.
2304 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2305 Match the new hook interface. Handle SVE addresses.
2306 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2309 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2311 * Makefile.in (OBJS): Add early-remat.o.
2312 * target.def (select_early_remat_modes): New hook.
2313 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2314 * doc/tm.texi: Regenerate.
2315 * targhooks.h (default_select_early_remat_modes): Declare.
2316 * targhooks.c (default_select_early_remat_modes): New function.
2317 * timevar.def (TV_EARLY_REMAT): New timevar.
2318 * passes.def (pass_early_remat): New pass.
2319 * tree-pass.h (make_pass_early_remat): Declare.
2320 * early-remat.c: New file.
2321 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2323 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2325 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2326 Alan Hayward <alan.hayward@arm.com>
2327 David Sherwood <david.sherwood@arm.com>
2329 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2330 vfm1 with a bound_epilog parameter.
2331 (vect_do_peeling): Update calls accordingly, and move the prologue
2332 call earlier in the function. Treat the base bound_epilog as 0 for
2333 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2334 this base when peeling for gaps.
2335 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2336 with fully-masked loops.
2337 (vect_estimate_min_profitable_iters): Handle the single peeled
2338 iteration in that case.
2340 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2341 Alan Hayward <alan.hayward@arm.com>
2342 David Sherwood <david.sherwood@arm.com>
2344 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2345 single-element interleaving even if the size is not a power of 2.
2346 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2347 accesses for single-element interleaving if the group size is
2350 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2351 Alan Hayward <alan.hayward@arm.com>
2352 David Sherwood <david.sherwood@arm.com>
2354 * doc/md.texi (fold_extract_last_@var{m}): Document.
2355 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2356 * optabs.def (fold_extract_last_optab): New optab.
2357 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2358 * internal-fn.c (fold_extract_direct): New macro.
2359 (expand_fold_extract_optab_fn): Likewise.
2360 (direct_fold_extract_optab_supported_p): Likewise.
2361 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2362 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2363 EXTRACT_LAST_REDUCTION.
2364 (get_initial_def_for_reduction): Do not create an initial vector
2365 for EXTRACT_LAST_REDUCTION reductions.
2366 (vectorizable_reduction): Leave the scalar phi in place for
2367 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2368 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2369 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2370 transform phase to vectorizable_condition.
2371 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2373 (vect_finish_stmt_generation): ...here.
2374 (vect_finish_replace_stmt): New function.
2375 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2376 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2378 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
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 (extract_last_@var{m}): Document.
2385 * optabs.def (extract_last_optab): New optab.
2386 * internal-fn.def (EXTRACT_LAST): New internal function.
2387 * internal-fn.c (cond_unary_direct): New macro.
2388 (expand_cond_unary_optab_fn): Likewise.
2389 (direct_cond_unary_optab_supported_p): Likewise.
2390 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2391 loops using EXTRACT_LAST.
2392 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2393 (extract_last_<mode>): ...this optab.
2394 (vec_extract<mode><Vel>): Update accordingly.
2396 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2397 Alan Hayward <alan.hayward@arm.com>
2398 David Sherwood <david.sherwood@arm.com>
2400 * target.def (empty_mask_is_expensive): New hook.
2401 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2402 * doc/tm.texi: Regenerate.
2403 * targhooks.h (default_empty_mask_is_expensive): Declare.
2404 * targhooks.c (default_empty_mask_is_expensive): New function.
2405 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2406 if the target says that empty masks are expensive.
2407 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2409 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2411 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2412 Alan Hayward <alan.hayward@arm.com>
2413 David Sherwood <david.sherwood@arm.com>
2415 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2416 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2417 (vect_use_loop_mask_for_alignment_p): New function.
2418 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2419 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2420 niters_skip argument. Make sure that the first niters_skip elements
2421 of the first iteration are inactive.
2422 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2423 Update call to vect_set_loop_masks_directly.
2424 (get_misalign_in_elems): New function, split out from...
2425 (vect_gen_prolog_loop_niters): ...here.
2426 (vect_update_init_of_dr): Take a code argument that specifies whether
2427 the adjustment should be added or subtracted.
2428 (vect_update_init_of_drs): Likewise.
2429 (vect_prepare_for_masked_peels): New function.
2430 (vect_do_peeling): Skip prologue peeling if we're using a mask
2431 instead. Update call to vect_update_inits_of_drs.
2432 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2434 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2435 alignment. Do not include the number of peeled iterations in
2436 the minimum threshold in that case.
2437 (vectorizable_induction): Adjust the start value down by
2438 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2439 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2440 Take the number of skipped iterations into account when calculating
2442 * tree-vect-stmts.c (vect_gen_while_not): New function.
2444 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2445 Alan Hayward <alan.hayward@arm.com>
2446 David Sherwood <david.sherwood@arm.com>
2448 * doc/sourcebuild.texi (vect_fully_masked): Document.
2449 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2451 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2453 (vect_analyze_loop_2): ...here. Don't check the vectorization
2454 factor against the number of loop iterations if the loop is
2457 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2458 Alan Hayward <alan.hayward@arm.com>
2459 David Sherwood <david.sherwood@arm.com>
2461 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2462 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2463 (dump_groups): Update accordingly.
2464 (iv_use::mem_type): New member variable.
2465 (address_p): New function.
2466 (record_use): Add a mem_type argument and initialize the new
2468 (record_group_use): Add a mem_type argument. Use address_p.
2469 Remove obsolete null checks of base_object. Update call to record_use.
2470 (find_interesting_uses_op): Update call to record_group_use.
2471 (find_interesting_uses_cond): Likewise.
2472 (find_interesting_uses_address): Likewise.
2473 (get_mem_type_for_internal_fn): New function.
2474 (find_address_like_use): Likewise.
2475 (find_interesting_uses_stmt): Try find_address_like_use before
2476 calling find_interesting_uses_op.
2477 (addr_offset_valid_p): Use the iv mem_type field as the type
2478 of the addressed memory.
2479 (add_autoinc_candidates): Likewise.
2480 (get_address_cost): Likewise.
2481 (split_small_address_groups_p): Use address_p.
2482 (split_address_groups): Likewise.
2483 (add_iv_candidate_for_use): Likewise.
2484 (autoinc_possible_for_pair): Likewise.
2485 (rewrite_groups): Likewise.
2486 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2487 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2488 (get_alias_ptr_type_for_ptr_address): New function.
2489 (rewrite_use_address): Rewrite address uses in calls that were
2490 identified by find_address_like_use.
2492 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2493 Alan Hayward <alan.hayward@arm.com>
2494 David Sherwood <david.sherwood@arm.com>
2496 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2498 * gimple-expr.h (is_gimple_addressable: Likewise.
2499 * gimple-expr.c (is_gimple_address): Likewise.
2500 * internal-fn.c (expand_call_mem_ref): New function.
2501 (expand_mask_load_optab_fn): Use it.
2502 (expand_mask_store_optab_fn): Likewise.
2504 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2505 Alan Hayward <alan.hayward@arm.com>
2506 David Sherwood <david.sherwood@arm.com>
2508 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2509 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2510 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2511 (cond_umax@var{mode}): Document.
2512 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2513 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2514 (cond_umin_optab, cond_umax_optab): New optabs.
2515 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2516 (COND_IOR, COND_XOR): New internal functions.
2517 * internal-fn.h (get_conditional_internal_fn): Declare.
2518 * internal-fn.c (cond_binary_direct): New macro.
2519 (expand_cond_binary_optab_fn): Likewise.
2520 (direct_cond_binary_optab_supported_p): Likewise.
2521 (get_conditional_internal_fn): New function.
2522 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2523 Cope with reduction statements that are vectorized as calls rather
2525 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2526 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2527 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2528 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2529 (UNSPEC_COND_EOR): New unspecs.
2530 (optab): Add mappings for them.
2531 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2532 (sve_int_op, sve_fp_op): New int attributes.
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 * optabs.def (while_ult_optab): New optab.
2539 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2540 * internal-fn.def (WHILE_ULT): New internal function.
2541 * internal-fn.h (direct_internal_fn_supported_p): New override
2542 that takes two types as argument.
2543 * internal-fn.c (while_direct): New macro.
2544 (expand_while_optab_fn): New function.
2545 (convert_optab_supported_p): Likewise.
2546 (direct_while_optab_supported_p): New macro.
2547 * wide-int.h (wi::udiv_ceil): New function.
2548 * tree-vectorizer.h (rgroup_masks): New structure.
2549 (vec_loop_masks): New typedef.
2550 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2552 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2553 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2554 (vect_max_vf): New function.
2555 (slpeel_make_loop_iterate_ntimes): Delete.
2556 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2557 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2558 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2559 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2560 internal-fn.h, stor-layout.h and optabs-query.h.
2561 (vect_set_loop_mask): New function.
2562 (add_preheader_seq): Likewise.
2563 (add_header_seq): Likewise.
2564 (interleave_supported_p): Likewise.
2565 (vect_maybe_permute_loop_masks): Likewise.
2566 (vect_set_loop_masks_directly): Likewise.
2567 (vect_set_loop_condition_masked): Likewise.
2568 (vect_set_loop_condition_unmasked): New function, split out from
2569 slpeel_make_loop_iterate_ntimes.
2570 (slpeel_make_loop_iterate_ntimes): Rename to..
2571 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2572 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2573 (vect_do_peeling): Update call accordingly.
2574 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2576 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2577 mask_compare_type, can_fully_mask_p and fully_masked_p.
2578 (release_vec_loop_masks): New function.
2579 (_loop_vec_info): Use it to free the loop masks.
2580 (can_produce_all_loop_masks_p): New function.
2581 (vect_get_max_nscalars_per_iter): Likewise.
2582 (vect_verify_full_masking): Likewise.
2583 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2584 retries, and free the mask rgroups before retrying. Check loop-wide
2585 reasons for disallowing fully-masked loops. Make the final decision
2586 about whether use a fully-masked loop or not.
2587 (vect_estimate_min_profitable_iters): Do not assume that peeling
2588 for the number of iterations will be needed for fully-masked loops.
2589 (vectorizable_reduction): Disable fully-masked loops.
2590 (vectorizable_live_operation): Likewise.
2591 (vect_halve_mask_nunits): New function.
2592 (vect_double_mask_nunits): Likewise.
2593 (vect_record_loop_mask): Likewise.
2594 (vect_get_loop_mask): Likewise.
2595 (vect_transform_loop): Handle the case in which the final loop
2596 iteration might handle a partial vector. Call vect_set_loop_condition
2597 instead of slpeel_make_loop_iterate_ntimes.
2598 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2599 (check_load_store_masking): New function.
2600 (prepare_load_store_mask): Likewise.
2601 (vectorizable_store): Handle fully-masked loops.
2602 (vectorizable_load): Likewise.
2603 (supportable_widening_operation): Use vect_halve_mask_nunits for
2605 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2606 (vect_gen_while): New function.
2607 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2608 (aarch64_uqdec<mode>): New insn.
2610 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2611 Alan Hayward <alan.hayward@arm.com>
2612 David Sherwood <david.sherwood@arm.com>
2614 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2615 (reduc_xor_scal_optab): New optabs.
2616 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2617 (reduc_xor_scal_@var{m}): Document.
2618 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2619 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2621 * fold-const-call.c (fold_const_call): Handle them.
2622 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2623 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2624 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2625 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2626 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2627 (UNSPEC_XORV): New unspecs.
2628 (optab): Add entries for them.
2629 (BITWISEV): New int iterator.
2630 (bit_reduc_op): New int attributes.
2632 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2633 Alan Hayward <alan.hayward@arm.com>
2634 David Sherwood <david.sherwood@arm.com>
2636 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2637 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2638 * optabs.def (vec_shl_insert_optab): New optab.
2639 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2640 (duplicate_and_interleave): Likewise.
2641 * tree-vect-loop.c: Include internal-fn.h.
2642 (neutral_op_for_slp_reduction): New function, split out from
2643 get_initial_defs_for_reduction.
2644 (get_initial_def_for_reduction): Handle option 2 for variable-length
2645 vectors by loading the neutral value into a vector and then shifting
2646 the initial value into element 0.
2647 (get_initial_defs_for_reduction): Replace the code argument with
2648 the neutral value calculated by neutral_op_for_slp_reduction.
2649 Use gimple_build_vector for constant-length vectors.
2650 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2651 but the first group_size elements have a neutral value.
2652 Use duplicate_and_interleave otherwise.
2653 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2654 Update call to get_initial_defs_for_reduction. Handle SLP
2655 reductions for variable-length vectors by creating one vector
2656 result for each scalar result, with the elements associated
2657 with other scalar results stubbed out with the neutral value.
2658 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2659 Require IFN_VEC_SHL_INSERT for double reductions on
2660 variable-length vectors, or SLP reductions that have
2661 a neutral value. Require can_duplicate_and_interleave_p
2662 support for variable-length unchained SLP reductions if there
2663 is no neutral value, such as for MIN/MAX reductions. Also require
2664 the number of vector elements to be a multiple of the number of
2665 SLP statements when doing variable-length unchained SLP reductions.
2666 Update call to vect_create_epilog_for_reduction.
2667 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2668 and remove initial values.
2669 (duplicate_and_interleave): Make public.
2670 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2671 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2674 Alan Hayward <alan.hayward@arm.com>
2675 David Sherwood <david.sherwood@arm.com>
2677 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2678 (can_duplicate_and_interleave_p): New function.
2679 (vect_get_and_check_slp_defs): Take the vector of statements
2680 rather than just the current one. Remove excess parentheses.
2681 Restriction rejectinon of vect_constant_def and vect_external_def
2682 for variable-length vectors to boolean types, or types for which
2683 can_duplicate_and_interleave_p is false.
2684 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2685 (duplicate_and_interleave): New function.
2686 (vect_get_constant_vectors): Use gimple_build_vector for
2687 constant-length vectors and suitable variable-length constant
2688 vectors. Use duplicate_and_interleave for other variable-length
2689 vectors. Don't defer the update when inserting new statements.
2691 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2692 Alan Hayward <alan.hayward@arm.com>
2693 David Sherwood <david.sherwood@arm.com>
2695 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2696 min_profitable_iters doesn't go negative.
2698 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2699 Alan Hayward <alan.hayward@arm.com>
2700 David Sherwood <david.sherwood@arm.com>
2702 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2703 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2704 * optabs.def (vec_mask_load_lanes_optab): New optab.
2705 (vec_mask_store_lanes_optab): Likewise.
2706 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2707 (MASK_STORE_LANES): Likewise.
2708 * internal-fn.c (mask_load_lanes_direct): New macro.
2709 (mask_store_lanes_direct): Likewise.
2710 (expand_mask_load_optab_fn): Handle masked operations.
2711 (expand_mask_load_lanes_optab_fn): New macro.
2712 (expand_mask_store_optab_fn): Handle masked operations.
2713 (expand_mask_store_lanes_optab_fn): New macro.
2714 (direct_mask_load_lanes_optab_supported_p): Likewise.
2715 (direct_mask_store_lanes_optab_supported_p): Likewise.
2716 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2718 (vect_load_lanes_supported): Likewise.
2719 * tree-vect-data-refs.c (strip_conversion): New function.
2720 (can_group_stmts_p): Likewise.
2721 (vect_analyze_data_ref_accesses): Use it instead of checking
2722 for a pair of assignments.
2723 (vect_store_lanes_supported): Take a masked_p parameter.
2724 (vect_load_lanes_supported): Likewise.
2725 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2726 vect_store_lanes_supported and vect_load_lanes_supported.
2727 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2728 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2729 parameter. Don't allow gaps for masked accesses.
2730 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2731 and vect_load_lanes_supported.
2732 (get_load_store_type): Take a masked_p parameter and update
2733 call to get_group_load_store_type.
2734 (vectorizable_store): Update call to get_load_store_type.
2735 Handle IFN_MASK_STORE_LANES.
2736 (vectorizable_load): Update call to get_load_store_type.
2737 Handle IFN_MASK_LOAD_LANES.
2739 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2740 Alan Hayward <alan.hayward@arm.com>
2741 David Sherwood <david.sherwood@arm.com>
2743 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2745 * config/aarch64/aarch64-protos.h
2746 (aarch64_sve_struct_memory_operand_p): Declare.
2747 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2748 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2749 (VPRED, vpred): Handle SVE structure modes.
2750 * config/aarch64/constraints.md (Utx): New constraint.
2751 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2752 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2753 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2754 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2755 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2756 structure modes. Split into pieces after RA.
2757 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2758 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2760 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2761 SVE structure modes.
2762 (aarch64_classify_address): Likewise.
2763 (sizetochar): Move earlier in file.
2764 (aarch64_print_operand): Handle SVE register lists.
2765 (aarch64_array_mode): New function.
2766 (aarch64_sve_struct_memory_operand_p): Likewise.
2767 (TARGET_ARRAY_MODE): Redefine.
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 * target.def (array_mode): New target hook.
2774 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2775 * doc/tm.texi: Regenerate.
2776 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2777 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2778 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2780 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2783 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2784 Alan Hayward <alan.hayward@arm.com>
2785 David Sherwood <david.sherwood@arm.com>
2787 * fold-const.c (fold_binary_loc): Check the argument types
2788 rather than the result type when testing for a vector operation.
2790 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2792 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2793 * doc/tm.texi: Regenerate.
2795 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2796 Alan Hayward <alan.hayward@arm.com>
2797 David Sherwood <david.sherwood@arm.com>
2799 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2800 (sve): Document new AArch64 extension.
2801 * doc/md.texi (w): Extend the description of the AArch64
2802 constraint to include SVE vectors.
2803 (Upl, Upa): Document new AArch64 predicate constraints.
2804 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2806 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2807 (msve-vector-bits=): New option.
2808 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2809 SVE when these are disabled.
2810 (sve): New extension.
2811 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2812 modes. Adjust their number of units based on aarch64_sve_vg.
2813 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2814 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2815 aarch64_addr_query_type.
2816 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2817 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2818 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2819 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2820 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2821 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2822 (aarch64_simd_imm_zero_p): Delete.
2823 (aarch64_check_zero_based_sve_index_immediate): Declare.
2824 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2825 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2826 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2827 (aarch64_sve_float_mul_immediate_p): Likewise.
2828 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2830 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2831 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2832 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2833 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2834 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2835 (aarch64_regmode_natural_size): Likewise.
2836 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2837 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2839 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2840 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2841 for VG and the SVE predicate registers.
2842 (V_ALIASES): Add a "z"-prefixed alias.
2843 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2844 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2845 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2846 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2847 (REG_CLASS_NAMES): Add entries for them.
2848 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2849 and the predicate registers.
2850 (aarch64_sve_vg): Declare.
2851 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2852 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2853 (REGMODE_NATURAL_SIZE): Define.
2854 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2856 * config/aarch64/aarch64.c: Include cfgrtl.h.
2857 (simd_immediate_info): Add a constructor for series vectors,
2858 and an associated step field.
2859 (aarch64_sve_vg): New variable.
2860 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2861 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2862 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2863 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2864 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2865 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2866 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2867 (aarch64_get_mask_mode): New functions.
2868 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2869 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2870 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2871 predicate modes and predicate registers. Explicitly restrict
2872 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2873 to store a vector mode if it is recognized by
2874 aarch64_classify_vector_mode.
2875 (aarch64_regmode_natural_size): New function.
2876 (aarch64_hard_regno_caller_save_mode): Return the original mode
2878 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2879 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2880 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2881 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2883 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2884 does not overlap dest if the function is frame-related. Handle
2886 (aarch64_split_add_offset): New function.
2887 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2888 them aarch64_add_offset.
2889 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2890 and update call to aarch64_sub_sp.
2891 (aarch64_add_cfa_expression): New function.
2892 (aarch64_expand_prologue): Pass extra temporary registers to the
2893 functions above. Handle the case in which we need to emit new
2894 DW_CFA_expressions for registers that were originally saved
2895 relative to the stack pointer, but now have to be expressed
2896 relative to the frame pointer.
2897 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2899 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2900 IP0 and IP1 values for SVE frames.
2901 (aarch64_expand_vec_series): New function.
2902 (aarch64_expand_sve_widened_duplicate): Likewise.
2903 (aarch64_expand_sve_const_vector): Likewise.
2904 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2905 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2906 into the register, rather than emitting a SET directly.
2907 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2908 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2909 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2910 (offset_9bit_signed_scaled_p): New functions.
2911 (aarch64_replicate_bitmask_imm): New function.
2912 (aarch64_bitmask_imm): Use it.
2913 (aarch64_cannot_force_const_mem): Reject expressions involving
2914 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2915 (aarch64_classify_index): Handle SVE indices, by requiring
2916 a plain register index with a scale that matches the element size.
2917 (aarch64_classify_address): Handle SVE addresses. Assert that
2918 the mode of the address is VOIDmode or an integer mode.
2919 Update call to aarch64_classify_symbol.
2920 (aarch64_classify_symbolic_expression): Update call to
2921 aarch64_classify_symbol.
2922 (aarch64_const_vec_all_in_range_p): New function.
2923 (aarch64_print_vector_float_operand): Likewise.
2924 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2925 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2926 and the FP immediates 1.0 and 0.5.
2927 (aarch64_print_address_internal): Handle SVE addresses.
2928 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2929 (aarch64_regno_regclass): Handle predicate registers.
2930 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2932 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2933 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2934 (aarch64_convert_sve_vector_bits): New function.
2935 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2936 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2938 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2939 Handle SVE vector and predicate modes. Accept VL-based constants
2940 that need only one temporary register, and VL offsets that require
2941 no temporary registers.
2942 (aarch64_conditional_register_usage): Mark the predicate registers
2943 as fixed if SVE isn't available.
2944 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2945 Return true for SVE vector and predicate modes.
2946 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2947 rather than an unsigned int. Handle SVE modes.
2948 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2950 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2952 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2953 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2954 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2955 (aarch64_sve_float_mul_immediate_p): New functions.
2956 (aarch64_sve_valid_immediate): New function.
2957 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2958 Explicitly reject structure modes. Check for INDEX constants.
2959 Handle PTRUE and PFALSE constants.
2960 (aarch64_check_zero_based_sve_index_immediate): New function.
2961 (aarch64_simd_imm_zero_p): Delete.
2962 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2963 vector modes. Accept constants in the range of CNT[BHWD].
2964 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2965 ask for an Advanced SIMD mode.
2966 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2967 (aarch64_simd_vector_alignment): Handle SVE predicates.
2968 (aarch64_vectorize_preferred_vector_alignment): New function.
2969 (aarch64_simd_vector_alignment_reachable): Use it instead of
2971 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2972 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2974 (MAX_VECT_LEN): Delete.
2975 (expand_vec_perm_d): Add a vec_flags field.
2976 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2977 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2978 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2980 (aarch64_evpc_rev): Rename to...
2981 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2982 (aarch64_evpc_rev_global): New function.
2983 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2984 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2986 (aarch64_evpc_sve_tbl): New function.
2987 (aarch64_expand_vec_perm_const_1): Update after rename of
2988 aarch64_evpc_rev. Handle SVE permutes too, trying
2989 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2990 than aarch64_evpc_tbl.
2991 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2992 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2993 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2994 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2995 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2996 (aarch64_expand_sve_vcond): New functions.
2997 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2998 of aarch64_vector_mode_p.
2999 (aarch64_dwarf_poly_indeterminate_value): New function.
3000 (aarch64_compute_pressure_classes): Likewise.
3001 (aarch64_can_change_mode_class): Likewise.
3002 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3003 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3004 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3005 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3006 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3007 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3008 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3009 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3011 (Dn, Dl, Dr): Accept const as well as const_vector.
3012 (Dz): Likewise. Compare against CONST0_RTX.
3013 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3014 of "vector" where appropriate.
3015 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3016 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3017 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3018 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3019 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3020 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3021 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3022 (v_int_equiv): Extend to SVE modes.
3023 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3025 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3026 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3027 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3028 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3029 (SVE_COND_FP_CMP): New int iterators.
3030 (perm_hilo): Handle the new unpack unspecs.
3031 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3033 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3034 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3035 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3036 (aarch64_equality_operator, aarch64_constant_vector_operand)
3037 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3038 (aarch64_sve_nonimmediate_operand): Likewise.
3039 (aarch64_sve_general_operand): Likewise.
3040 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3041 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3042 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3043 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3044 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3045 (aarch64_sve_float_arith_immediate): Likewise.
3046 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3047 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3048 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3049 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3050 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3051 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3052 (aarch64_sve_float_arith_operand): Likewise.
3053 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3054 (aarch64_sve_float_mul_operand): Likewise.
3055 (aarch64_sve_vec_perm_operand): Likewise.
3056 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3057 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3058 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3059 as well as const_vector.
3060 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3061 in file. Use CONST0_RTX and CONSTM1_RTX.
3062 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3063 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3064 Use aarch64_simd_imm_zero.
3065 * config/aarch64/aarch64-sve.md: New file.
3066 * config/aarch64/aarch64.md: Include it.
3067 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3068 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3069 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3070 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3071 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3072 (sve): New attribute.
3073 (enabled): Disable instructions with the sve attribute unless
3075 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3076 aarch64_expand_mov_immediate.
3077 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3078 CNT[BHSD] immediates.
3079 (movti): Split CONST_POLY_INT moves into two halves.
3080 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3081 Split additions that need a temporary here if the destination
3082 is the stack pointer.
3083 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3084 (*add<mode>3_poly_1): New instruction.
3085 (set_clobber_cc): New expander.
3087 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3089 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3090 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3091 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3092 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3093 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3094 Change innermode from fixed_mode_size to machine_mode.
3095 (simplify_subreg): Update call accordingly. Handle a constant-sized
3096 subreg of a variable-length CONST_VECTOR.
3098 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3099 Alan Hayward <alan.hayward@arm.com>
3100 David Sherwood <david.sherwood@arm.com>
3102 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3103 (add_offset_to_base): New function, split out from...
3104 (create_mem_ref): ...here. When handling a scale other than 1,
3105 check first whether the address is valid without the offset.
3106 Add it into the base if so, leaving the index and scale as-is.
3108 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3111 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3112 fold_for_warn before checking if arg2 is INTEGER_CST.
3114 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3116 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3117 (store_multiple_operation): Delete.
3118 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3119 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3120 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3121 guarded by TARGET_STRING.
3122 (rs6000_output_load_multiple): Delete.
3123 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3124 OPTION_MASK_STRING / TARGET_STRING handling.
3125 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3126 (const rs6000_opt_masks) <"string">: Change mask to 0.
3127 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3128 (MASK_STRING): Delete.
3129 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3131 (load_multiple): Delete.
3138 (store_multiple): Delete.
3145 (movmemsi_8reg): Delete.
3146 (corresponding unnamed define_insn): Delete.
3147 (movmemsi_6reg): Delete.
3148 (corresponding unnamed define_insn): Delete.
3149 (movmemsi_4reg): Delete.
3150 (corresponding unnamed define_insn): Delete.
3151 (movmemsi_2reg): Delete.
3152 (corresponding unnamed define_insn): Delete.
3153 (movmemsi_1reg): Delete.
3154 (corresponding unnamed define_insn): Delete.
3155 * config/rs6000/rs6000.opt (mno-string): New.
3156 (mstring): Replace by deprecation warning stub.
3157 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3159 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3161 * regrename.c (regrename_do_replace): If replacing the same
3162 reg multiple times, try to reuse last created gen_raw_REG.
3165 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3166 main to workaround a bug in GDB.
3168 2018-01-12 Tom de Vries <tom@codesourcery.com>
3171 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3173 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3175 PR rtl-optimization/80481
3176 * ira-color.c (get_cap_member): New function.
3177 (allocnos_conflict_by_live_ranges_p): Use it.
3178 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3179 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3181 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3184 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3185 (*saddl_se_1): Ditto.
3187 (*ssubl_se_1): Ditto.
3189 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3191 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3192 rather than wi::to_widest for DR_INITs.
3193 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3194 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3195 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3197 (vect_analyze_group_access_1): Note that here.
3199 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3201 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3202 polynomial type sizes.
3204 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3206 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3207 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3208 (gimple_add_tmp_var): Likewise.
3210 2018-01-12 Martin Liska <mliska@suse.cz>
3212 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3213 (gimple_alloc_sizes): Likewise.
3214 (dump_gimple_statistics): Use PRIu64 in printf format.
3215 * gimple.h: Change uint64_t to int.
3217 2018-01-12 Martin Liska <mliska@suse.cz>
3219 * tree-core.h: Use uint64_t instead of int.
3220 * tree.c (tree_node_counts): Likewise.
3221 (tree_node_sizes): Likewise.
3222 (dump_tree_statistics): Use PRIu64 in printf format.
3224 2018-01-12 Martin Liska <mliska@suse.cz>
3226 * Makefile.in: As qsort_chk is implemented in vec.c, add
3227 vec.o to linkage of gencfn-macros.
3228 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3229 passing the info to record_node_allocation_statistics.
3230 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3232 * ggc-common.c (struct ggc_usage): Add operator== and use
3233 it in operator< and compare function.
3234 * mem-stats.h (struct mem_usage): Likewise.
3235 * vec.c (struct vec_usage): Remove operator< and compare
3236 function. Can be simply inherited.
3238 2018-01-12 Martin Jambor <mjambor@suse.cz>
3241 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3242 * tree-ssa-math-opts.c: Include domwalk.h.
3243 (convert_mult_to_fma_1): New function.
3244 (fma_transformation_info): New type.
3245 (fma_deferring_state): Likewise.
3246 (cancel_fma_deferring): New function.
3247 (result_of_phi): Likewise.
3248 (last_fma_candidate_feeds_initial_phi): Likewise.
3249 (convert_mult_to_fma): Added deferring logic, split actual
3250 transformation to convert_mult_to_fma_1.
3251 (math_opts_dom_walker): New type.
3252 (math_opts_dom_walker::after_dom_children): New method, body moved
3253 here from pass_optimize_widening_mul::execute, added deferring logic
3255 (pass_optimize_widening_mul::execute): Moved most of code to
3256 math_opts_dom_walker::after_dom_children.
3257 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3258 * config/i386/i386.c (ix86_option_override_internal): Added
3259 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3261 2018-01-12 Richard Biener <rguenther@suse.de>
3264 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3265 inline instance vars.
3267 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3270 * config/rx/rx.c (rx_is_restricted_memory_address):
3273 2018-01-12 Richard Biener <rguenther@suse.de>
3275 PR tree-optimization/80846
3276 * target.def (split_reduction): New target hook.
3277 * targhooks.c (default_split_reduction): New function.
3278 * targhooks.h (default_split_reduction): Declare.
3279 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3280 target requests first reduce vectors by combining low and high
3282 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3283 (get_vectype_for_scalar_type_and_size): Export.
3284 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3285 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3286 * doc/tm.texi: Regenerate.
3287 * config/i386/i386.c (ix86_split_reduction): Implement
3288 TARGET_VECTORIZE_SPLIT_REDUCTION.
3290 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3293 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3294 in PIC mode except for TARGET_VXWORKS_RTP.
3295 * config/sparc/sparc.c: Include cfgrtl.h.
3296 (TARGET_INIT_PIC_REG): Define.
3297 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3298 (sparc_pic_register_p): New predicate.
3299 (sparc_legitimate_address_p): Use it.
3300 (sparc_legitimize_pic_address): Likewise.
3301 (sparc_delegitimize_address): Likewise.
3302 (sparc_mode_dependent_address_p): Likewise.
3303 (gen_load_pcrel_sym): Remove 4th parameter.
3304 (load_got_register): Adjust call to above. Remove obsolete stuff.
3305 (sparc_expand_prologue): Do not call load_got_register here.
3306 (sparc_flat_expand_prologue): Likewise.
3307 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3308 (sparc_use_pseudo_pic_reg): New function.
3309 (sparc_init_pic_reg): Likewise.
3310 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3311 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3313 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3315 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3316 Add item for branch_cost.
3318 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3320 PR rtl-optimization/83565
3321 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3322 not extend the result to a larger mode for rotate operations.
3323 (num_sign_bit_copies1): Likewise.
3325 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3328 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3330 Use values-Xc.o for -pedantic.
3331 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3333 2018-01-12 Martin Liska <mliska@suse.cz>
3336 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3338 (possible_polymorphic_call_targets): Use it.
3339 (ipa_devirt): Likewise.
3341 2018-01-12 Martin Liska <mliska@suse.cz>
3343 * profile-count.h (enum profile_quality): Use 0 as invalid
3344 enum value of profile_quality.
3346 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3348 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3349 -mext-string options.
3351 2018-01-12 Richard Biener <rguenther@suse.de>
3353 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3354 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3355 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3357 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3359 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3361 * configure.ac (--with-long-double-format): Add support for the
3362 configuration option to change the default long double format on
3364 * config.gcc (powerpc*-linux*-*): Likewise.
3365 * configure: Regenerate.
3366 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3367 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3368 used without modification.
3370 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3372 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3373 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3374 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3375 MISC_BUILTIN_SPEC_BARRIER.
3376 (rs6000_init_builtins): Likewise.
3377 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3379 (speculation_barrier): New define_insn.
3380 * doc/extend.texi: Document __builtin_speculation_barrier.
3382 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3385 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3386 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3387 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3389 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3390 integral modes instead of "ss" and "sd".
3391 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3392 vectors with 32-bit and 64-bit elements.
3393 (vecdupssescalarmodesuffix): New mode attribute.
3394 (vec_dup<mode>): Use it.
3396 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3399 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3400 frame if argument is passed on stack.
3402 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3405 * ree.c (combine_reaching_defs): Optimize also
3406 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3407 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3409 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3412 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3414 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3417 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3418 after they are computed.
3420 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3422 PR tree-optimization/83695
3423 * gimple-loop-linterchange.cc
3424 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3425 reset cached scev information after interchange.
3426 (pass_linterchange::execute): Remove call to scev_reset_htab.
3428 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3430 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3431 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3432 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3433 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3434 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3435 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3436 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3437 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3438 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3439 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3440 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3441 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3442 (V_lane_reg): Likewise.
3443 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3445 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3446 (vfmal_lane_low<mode>_intrinsic,
3447 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3448 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3449 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3450 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3451 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3452 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3454 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3456 * config/arm/arm-cpus.in (fp16fml): New feature.
3457 (ALL_SIMD): Add fp16fml.
3458 (armv8.2-a): Add fp16fml as an option.
3459 (armv8.3-a): Likewise.
3460 (armv8.4-a): Add fp16fml as part of fp16.
3461 * config/arm/arm.h (TARGET_FP16FML): Define.
3462 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3464 * config/arm/arm-modes.def (V2HF): Define.
3465 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3466 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3467 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3468 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3469 vfmsl_low, vfmsl_high): New set of builtins.
3470 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3471 (vfml_op): New code attribute.
3472 (VFMLHALVES): New int iterator.
3473 (VFML, VFMLSEL): New mode attributes.
3474 (V_reg): Define mapping for V2HF.
3475 (V_hi, V_lo): New mode attributes.
3476 (VF_constraint): Likewise.
3477 (vfml_half, vfml_half_selector): New int attributes.
3478 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3480 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3481 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3483 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3484 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3485 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3486 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3488 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3489 Document new effective target and option set.
3491 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3493 * config/arm/arm-cpus.in (armv8_4): New feature.
3494 (ARMv8_4a): New fgroup.
3495 (armv8.4-a): New arch.
3496 * config/arm/arm-tables.opt: Regenerate.
3497 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3498 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3499 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3500 Add matching rules for -march=armv8.4-a and extensions.
3501 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3503 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3506 * config/rx/rx.md (BW): New mode attribute.
3507 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3509 2018-01-11 Richard Biener <rguenther@suse.de>
3511 PR tree-optimization/83435
3512 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3513 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3514 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3516 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3517 Alan Hayward <alan.hayward@arm.com>
3518 David Sherwood <david.sherwood@arm.com>
3520 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3522 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3523 (aarch64_print_address_internal): Use it to check for a zero offset.
3525 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3526 Alan Hayward <alan.hayward@arm.com>
3527 David Sherwood <david.sherwood@arm.com>
3529 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3530 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3531 Return a poly_int64 rather than a HOST_WIDE_INT.
3532 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3533 rather than a HOST_WIDE_INT.
3534 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3535 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3536 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3537 final_offset from HOST_WIDE_INT to poly_int64.
3538 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3539 to_constant when getting the number of units in an Advanced SIMD
3541 (aarch64_builtin_vectorized_function): Check for a constant number
3543 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3545 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3546 attribute instead of GET_MODE_NUNITS.
3547 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3548 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3549 GET_MODE_SIZE for fixed-size registers.
3550 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3551 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3552 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3553 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3554 (aarch64_print_operand, aarch64_print_address_internal)
3555 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3556 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3557 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3558 Handle polynomial GET_MODE_SIZE.
3559 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3560 wider than SImode without modification.
3561 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3562 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3563 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3564 passing and returning SVE modes.
3565 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3566 rather than GEN_INT.
3567 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3568 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3569 (aarch64_allocate_and_probe_stack_space): Likewise.
3570 (aarch64_layout_frame): Cope with polynomial offsets.
3571 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3572 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3574 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3575 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3576 poly_int64 rather than a HOST_WIDE_INT.
3577 (aarch64_get_separate_components, aarch64_process_components)
3578 (aarch64_expand_prologue, aarch64_expand_epilogue)
3579 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3580 (aarch64_anchor_offset): New function, split out from...
3581 (aarch64_legitimize_address): ...here.
3582 (aarch64_builtin_vectorization_cost): Handle polynomial
3583 TYPE_VECTOR_SUBPARTS.
3584 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3586 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3587 number of elements from the PARALLEL rather than the mode.
3588 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3589 rather than GET_MODE_BITSIZE.
3590 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3591 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3592 (aarch64_expand_vec_perm_const_1): Handle polynomial
3593 d->perm.length () and d->perm elements.
3594 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3595 Apply to_constant to d->perm elements.
3596 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3597 polynomial CONST_VECTOR_NUNITS.
3598 (aarch64_move_pointer): Take amount as a poly_int64 rather
3600 (aarch64_progress_pointer): Avoid temporary variable.
3601 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3602 the mode attribute instead of GET_MODE.
3604 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3605 Alan Hayward <alan.hayward@arm.com>
3606 David Sherwood <david.sherwood@arm.com>
3608 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3609 x exists before using it.
3610 (aarch64_add_constant_internal): Rename to...
3611 (aarch64_add_offset_1): ...this. Replace regnum with separate
3612 src and dest rtxes. Handle the case in which they're different,
3613 including when the offset is zero. Replace scratchreg with an rtx.
3614 Use 2 additions if there is no spare register into which we can
3615 move a 16-bit constant.
3616 (aarch64_add_constant): Delete.
3617 (aarch64_add_offset): Replace reg with separate src and dest
3618 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3619 Use aarch64_add_offset_1.
3620 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3621 an rtx rather than an int. Take the delta as a poly_int64
3622 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3623 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3624 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3625 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3626 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3628 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3629 aarch64_add_constant.
3631 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3633 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3634 Use scalar_float_mode.
3636 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3638 * config/aarch64/aarch64-simd.md
3639 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3640 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3641 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3642 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3643 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3644 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3645 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3646 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3647 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3648 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3650 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3653 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3654 targ_options->x_arm_arch_string is non NULL.
3656 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3658 * config/aarch64/aarch64.h
3659 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3661 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3664 * expmed.c (emit_store_flag_force): Swap if const op0
3665 and change VOIDmode to mode of op0.
3667 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3669 PR rtl-optimization/83761
3670 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3671 than bytes to mode_for_size.
3673 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3676 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3677 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3680 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3683 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3684 when in layout mode.
3685 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3686 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3689 2018-01-10 Michael Collison <michael.collison@arm.com>
3691 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3692 * config/aarch64/aarch64-option-extension.def: Add
3693 AARCH64_OPT_EXTENSION of 'fp16fml'.
3694 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3695 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3696 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3697 * config/aarch64/constraints.md (Ui7): New constraint.
3698 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3699 (VFMLA_SEL_W): Ditto.
3702 (VFMLA16_LOW): New int iterator.
3703 (VFMLA16_HIGH): Ditto.
3704 (UNSPEC_FMLAL): New unspec.
3705 (UNSPEC_FMLSL): Ditto.
3706 (UNSPEC_FMLAL2): Ditto.
3707 (UNSPEC_FMLSL2): Ditto.
3708 (f16mac): New code attribute.
3709 * config/aarch64/aarch64-simd-builtins.def
3710 (aarch64_fmlal_lowv2sf): Ditto.
3711 (aarch64_fmlsl_lowv2sf): Ditto.
3712 (aarch64_fmlalq_lowv4sf): Ditto.
3713 (aarch64_fmlslq_lowv4sf): Ditto.
3714 (aarch64_fmlal_highv2sf): Ditto.
3715 (aarch64_fmlsl_highv2sf): Ditto.
3716 (aarch64_fmlalq_highv4sf): Ditto.
3717 (aarch64_fmlslq_highv4sf): Ditto.
3718 (aarch64_fmlal_lane_lowv2sf): Ditto.
3719 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3720 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3721 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3722 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3723 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3724 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3725 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3726 (aarch64_fmlal_lane_highv2sf): Ditto.
3727 (aarch64_fmlsl_lane_highv2sf): Ditto.
3728 (aarch64_fmlal_laneq_highv2sf): Ditto.
3729 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3730 (aarch64_fmlalq_lane_highv4sf): Ditto.
3731 (aarch64_fmlsl_lane_highv4sf): Ditto.
3732 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3733 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3734 * config/aarch64/aarch64-simd.md:
3735 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3736 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3737 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3738 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3739 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3740 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3741 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3742 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3743 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3744 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3745 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3746 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3747 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3748 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3749 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3750 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3751 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3752 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3753 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3754 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3755 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3756 (vfmlsl_low_u32): Ditto.
3757 (vfmlalq_low_u32): Ditto.
3758 (vfmlslq_low_u32): Ditto.
3759 (vfmlal_high_u32): Ditto.
3760 (vfmlsl_high_u32): Ditto.
3761 (vfmlalq_high_u32): Ditto.
3762 (vfmlslq_high_u32): Ditto.
3763 (vfmlal_lane_low_u32): Ditto.
3764 (vfmlsl_lane_low_u32): Ditto.
3765 (vfmlal_laneq_low_u32): Ditto.
3766 (vfmlsl_laneq_low_u32): Ditto.
3767 (vfmlalq_lane_low_u32): Ditto.
3768 (vfmlslq_lane_low_u32): Ditto.
3769 (vfmlalq_laneq_low_u32): Ditto.
3770 (vfmlslq_laneq_low_u32): Ditto.
3771 (vfmlal_lane_high_u32): Ditto.
3772 (vfmlsl_lane_high_u32): Ditto.
3773 (vfmlal_laneq_high_u32): Ditto.
3774 (vfmlsl_laneq_high_u32): Ditto.
3775 (vfmlalq_lane_high_u32): Ditto.
3776 (vfmlslq_lane_high_u32): Ditto.
3777 (vfmlalq_laneq_high_u32): Ditto.
3778 (vfmlslq_laneq_high_u32): Ditto.
3779 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3780 (AARCH64_FL_FOR_ARCH8_4): New.
3781 (AARCH64_ISA_F16FML): New ISA flag.
3782 (TARGET_F16FML): New feature flag for fp16fml.
3783 (doc/invoke.texi): Document new fp16fml option.
3785 2018-01-10 Michael Collison <michael.collison@arm.com>
3787 * config/aarch64/aarch64-builtins.c:
3788 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3789 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3790 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3791 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3792 (AARCH64_ISA_SHA3): New ISA flag.
3793 (TARGET_SHA3): New feature flag for sha3.
3794 * config/aarch64/iterators.md (sha512_op): New int attribute.
3795 (CRYPTO_SHA512): New int iterator.
3796 (UNSPEC_SHA512H): New unspec.
3797 (UNSPEC_SHA512H2): Ditto.
3798 (UNSPEC_SHA512SU0): Ditto.
3799 (UNSPEC_SHA512SU1): Ditto.
3800 * config/aarch64/aarch64-simd-builtins.def
3801 (aarch64_crypto_sha512hqv2di): New builtin.
3802 (aarch64_crypto_sha512h2qv2di): Ditto.
3803 (aarch64_crypto_sha512su0qv2di): Ditto.
3804 (aarch64_crypto_sha512su1qv2di): Ditto.
3805 (aarch64_eor3qv8hi): Ditto.
3806 (aarch64_rax1qv2di): Ditto.
3807 (aarch64_xarqv2di): Ditto.
3808 (aarch64_bcaxqv8hi): Ditto.
3809 * config/aarch64/aarch64-simd.md:
3810 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3811 (aarch64_crypto_sha512su0qv2di): Ditto.
3812 (aarch64_crypto_sha512su1qv2di): Ditto.
3813 (aarch64_eor3qv8hi): Ditto.
3814 (aarch64_rax1qv2di): Ditto.
3815 (aarch64_xarqv2di): Ditto.
3816 (aarch64_bcaxqv8hi): Ditto.
3817 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3818 (vsha512h2q_u64): Ditto.
3819 (vsha512su0q_u64): Ditto.
3820 (vsha512su1q_u64): Ditto.
3821 (veor3q_u16): Ditto.
3822 (vrax1q_u64): Ditto.
3824 (vbcaxq_u16): Ditto.
3825 * config/arm/types.md (crypto_sha512): New type attribute.
3826 (crypto_sha3): Ditto.
3827 (doc/invoke.texi): Document new sha3 option.
3829 2018-01-10 Michael Collison <michael.collison@arm.com>
3831 * config/aarch64/aarch64-builtins.c:
3832 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3833 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3834 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3835 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3836 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3837 (AARCH64_ISA_SM4): New ISA flag.
3838 (TARGET_SM4): New feature flag for sm4.
3839 * config/aarch64/aarch64-simd-builtins.def
3840 (aarch64_sm3ss1qv4si): Ditto.
3841 (aarch64_sm3tt1aq4si): Ditto.
3842 (aarch64_sm3tt1bq4si): Ditto.
3843 (aarch64_sm3tt2aq4si): Ditto.
3844 (aarch64_sm3tt2bq4si): Ditto.
3845 (aarch64_sm3partw1qv4si): Ditto.
3846 (aarch64_sm3partw2qv4si): Ditto.
3847 (aarch64_sm4eqv4si): Ditto.
3848 (aarch64_sm4ekeyqv4si): Ditto.
3849 * config/aarch64/aarch64-simd.md:
3850 (aarch64_sm3ss1qv4si): Ditto.
3851 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3852 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3853 (aarch64_sm4eqv4si): Ditto.
3854 (aarch64_sm4ekeyqv4si): Ditto.
3855 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3856 (sm3part_op): Ditto.
3857 (CRYPTO_SM3TT): Ditto.
3858 (CRYPTO_SM3PART): Ditto.
3859 (UNSPEC_SM3SS1): New unspec.
3860 (UNSPEC_SM3TT1A): Ditto.
3861 (UNSPEC_SM3TT1B): Ditto.
3862 (UNSPEC_SM3TT2A): Ditto.
3863 (UNSPEC_SM3TT2B): Ditto.
3864 (UNSPEC_SM3PARTW1): Ditto.
3865 (UNSPEC_SM3PARTW2): Ditto.
3866 (UNSPEC_SM4E): Ditto.
3867 (UNSPEC_SM4EKEY): Ditto.
3868 * config/aarch64/constraints.md (Ui2): New constraint.
3869 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3870 * config/arm/types.md (crypto_sm3): New type attribute.
3871 (crypto_sm4): Ditto.
3872 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3873 (vsm3tt1aq_u32): Ditto.
3874 (vsm3tt1bq_u32): Ditto.
3875 (vsm3tt2aq_u32): Ditto.
3876 (vsm3tt2bq_u32): Ditto.
3877 (vsm3partw1q_u32): Ditto.
3878 (vsm3partw2q_u32): Ditto.
3879 (vsm4eq_u32): Ditto.
3880 (vsm4ekeyq_u32): Ditto.
3881 (doc/invoke.texi): Document new sm4 option.
3883 2018-01-10 Michael Collison <michael.collison@arm.com>
3885 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3886 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3887 (AARCH64_FL_FOR_ARCH8_4): New.
3888 (AARCH64_FL_V8_4): New flag.
3889 (doc/invoke.texi): Document new armv8.4-a option.
3891 2018-01-10 Michael Collison <michael.collison@arm.com>
3893 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3894 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3895 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3896 * config/aarch64/aarch64-option-extension.def: Add
3897 AARCH64_OPT_EXTENSION of 'sha2'.
3898 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3899 (crypto): Disable sha2 and aes if crypto disabled.
3900 (crypto): Enable aes and sha2 if enabled.
3901 (simd): Disable sha2 and aes if simd disabled.
3902 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3904 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3905 (TARGET_SHA2): New feature flag for sha2.
3906 (TARGET_AES): New feature flag for aes.
3907 * config/aarch64/aarch64-simd.md:
3908 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3909 conditional on TARGET_AES.
3910 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3911 (aarch64_crypto_sha1hsi): Make pattern conditional
3913 (aarch64_crypto_sha1hv4si): Ditto.
3914 (aarch64_be_crypto_sha1hv4si): Ditto.
3915 (aarch64_crypto_sha1su1v4si): Ditto.
3916 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3917 (aarch64_crypto_sha1su0v4si): Ditto.
3918 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3919 (aarch64_crypto_sha256su0v4si): Ditto.
3920 (aarch64_crypto_sha256su1v4si): Ditto.
3921 (doc/invoke.texi): Document new aes and sha2 options.
3923 2018-01-10 Martin Sebor <msebor@redhat.com>
3925 PR tree-optimization/83781
3926 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3929 2018-01-11 Martin Sebor <msebor@gmail.com>
3930 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3932 PR tree-optimization/83501
3933 PR tree-optimization/81703
3935 * tree-ssa-strlen.c (get_string_cst): Rename...
3936 (get_string_len): ...to this. Handle global constants.
3937 (handle_char_store): Adjust.
3939 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3940 Jim Wilson <jimw@sifive.com>
3942 * config/riscv/riscv-protos.h (riscv_output_return): New.
3943 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3944 (riscv_attribute_table, riscv_output_return),
3945 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3946 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3947 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3949 (riscv_expand_prologue): Add early return for naked function.
3950 (riscv_expand_epilogue): Likewise.
3951 (riscv_function_ok_for_sibcall): Return false for naked function.
3952 (riscv_set_current_function): New.
3953 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3954 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3955 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3956 * doc/extend.texi (RISC-V Function Attributes): New.
3958 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3960 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3961 check for 128-bit long double before checking TCmode.
3962 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3963 128-bit long doubles before checking TFmode or TCmode.
3964 (FLOAT128_IBM_P): Likewise.
3966 2018-01-10 Martin Sebor <msebor@redhat.com>
3968 PR tree-optimization/83671
3969 * builtins.c (c_strlen): Unconditionally return zero for the empty
3971 Use -Warray-bounds for warnings.
3972 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3973 for non-constant array indices with COMPONENT_REF, arrays of
3974 arrays, and pointers to arrays.
3975 (gimple_fold_builtin_strlen): Determine and set length range for
3976 non-constant character arrays.
3978 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3981 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3984 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3986 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3988 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3991 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3992 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3993 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3994 indexed_or_indirect_operand predicate.
3995 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3996 (*vsx_le_perm_load_v8hi): Likewise.
3997 (*vsx_le_perm_load_v16qi): Likewise.
3998 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3999 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4000 (*vsx_le_perm_store_v8hi): Likewise.
4001 (*vsx_le_perm_store_v16qi): Likewise.
4002 (eight unnamed splitters): Likewise.
4004 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4006 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4007 * config/rs6000/emmintrin.h: Likewise.
4008 * config/rs6000/mmintrin.h: Likewise.
4009 * config/rs6000/xmmintrin.h: Likewise.
4011 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4014 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4016 * tree.c (tree_nop_conversion): Return true for location wrapper
4018 (maybe_wrap_with_location): New function.
4019 (selftest::check_strip_nops): New function.
4020 (selftest::test_location_wrappers): New function.
4021 (selftest::tree_c_tests): Call it.
4022 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4023 (maybe_wrap_with_location): New decl.
4024 (EXPR_LOCATION_WRAPPER_P): New macro.
4025 (location_wrapper_p): New inline function.
4026 (tree_strip_any_location_wrapper): New inline function.
4028 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4031 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4032 stack_realign_offset for the largest alignment of stack slot
4034 (ix86_find_max_used_stack_alignment): New function.
4035 (ix86_finalize_stack_frame_flags): Use it. Set
4036 max_used_stack_alignment if we don't realign stack.
4037 * config/i386/i386.h (machine_function): Add
4038 max_used_stack_alignment.
4040 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4042 * config/arm/arm.opt (-mbranch-cost): New option.
4043 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4046 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4049 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4050 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4052 2018-01-10 Richard Biener <rguenther@suse.de>
4055 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4056 early out so it also covers the case where we have a non-NULL
4059 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4061 PR tree-optimization/83753
4062 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4063 for non-strided grouped accesses if the number of elements is 1.
4065 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4068 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4069 * i386.h (TARGET_USE_GATHER): Define.
4070 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4072 2018-01-10 Martin Liska <mliska@suse.cz>
4075 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4076 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4078 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4079 CLEANUP_NO_PARTITIONING is not set.
4081 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4083 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4084 for vectors, as a partial revert of r254296.
4085 * rtl.h (const_vec_p): Delete.
4086 (const_vec_duplicate_p): Don't test for vector CONSTs.
4087 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4088 * expmed.c (make_tree): Likewise.
4091 * common.md (E, F): Use CONSTANT_P instead of checking for
4093 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4094 checking for CONST_VECTOR.
4096 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4099 * predict.c (force_edge_cold): Handle in more sane way edges
4102 2018-01-09 Carl Love <cel@us.ibm.com>
4104 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4106 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4107 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4108 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4109 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4110 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4111 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4112 * config/rs6000/rs6000-protos.h: Add extern defition for
4113 rs6000_generate_float2_double_code.
4114 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4116 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4117 (float2_v2df): Add define_expand.
4119 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4122 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4123 op_mode in the force_to_mode call.
4125 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4127 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4128 instead of checking each element individually.
4129 (aarch64_evpc_uzp): Likewise.
4130 (aarch64_evpc_zip): Likewise.
4131 (aarch64_evpc_ext): Likewise.
4132 (aarch64_evpc_rev): Likewise.
4133 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4134 instead of checking each element individually. Return true without
4136 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4137 whether all selected elements come from the same input, instead of
4138 checking each element individually. Remove calls to gen_rtx_REG,
4139 start_sequence and end_sequence and instead assert that no rtl is
4142 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4144 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4145 order of HIGH and CONST checks.
4147 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4149 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4150 if the destination isn't an SSA_NAME.
4152 2018-01-09 Richard Biener <rguenther@suse.de>
4154 PR tree-optimization/83668
4155 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4157 (canonicalize_loop_form): ... here, renamed from ...
4158 (canonicalize_loop_closed_ssa_form): ... this and amended to
4159 swap successor edges for loop exit blocks to make us use
4160 the RPO order we need for initial schedule generation.
4162 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4164 PR tree-optimization/64811
4165 * match.pd: When optimizing comparisons with Inf, avoid
4166 introducing or losing exceptions from comparisons with NaN.
4168 2018-01-09 Martin Liska <mliska@suse.cz>
4171 * asan.c (shadow_mem_size): Add gcc_assert.
4173 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4175 Don't save registers in main().
4178 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4179 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4180 * config/avr/avr.c (avr_set_current_function): Don't error if
4181 naked, OS_task or OS_main are specified at the same time.
4182 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4184 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4186 * common/config/avr/avr-common.c (avr_option_optimization_table):
4187 Switch on -mmain-is-OS_task for optimizing compilations.
4189 2018-01-09 Richard Biener <rguenther@suse.de>
4191 PR tree-optimization/83572
4192 * graphite.c: Include cfganal.h.
4193 (graphite_transform_loops): Connect infinite loops to exit
4194 and remove fake edges at the end.
4196 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4198 * ipa-inline.c (edge_badness): Revert accidental checkin.
4200 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4203 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4204 symbols; not inline clones.
4206 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4209 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4210 hard registers. Formatting fixes.
4212 PR preprocessor/83722
4213 * gcc.c (try_generate_repro): Pass
4214 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4215 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4218 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4219 Kito Cheng <kito.cheng@gmail.com>
4221 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4222 (riscv_leaf_function_p): Delete.
4223 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4225 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4227 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4229 (do_ifelse): New function.
4230 (do_isel): New function.
4231 (do_sub3): New function.
4232 (do_add3): New function.
4233 (do_load_mask_compare): New function.
4234 (do_overlap_load_compare): New function.
4235 (expand_compare_loop): New function.
4236 (expand_block_compare): Call expand_compare_loop() when appropriate.
4237 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4239 (-mblock-compare-inline-loop-limit): New option.
4241 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4244 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4245 Reverse order of second and third operands in first alternative.
4246 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4247 of first and second elements in UNSPEC_VPERMR vector.
4248 (altivec_expand_vec_perm_le): Likewise.
4250 2018-01-08 Jeff Law <law@redhat.com>
4252 PR rtl-optimizatin/81308
4253 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4254 (process_switch): If group_case_labels makes a change, then set
4256 (pass_convert_switch::execute): If a switch is converted, then
4257 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4259 PR rtl-optimization/81308
4260 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4263 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4265 PR target/83663 - Revert r255946
4266 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4267 generation for cases where splatting a value is not useful.
4268 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4269 across a vec_duplicate and a paradoxical subreg forming a vector
4270 mode to a vec_concat.
4272 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4274 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4275 -march=armv8.3-a variants.
4276 * config/arm/t-multilib: Likewise.
4277 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4279 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4281 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4283 (cceq_ior_compare_complement): Give it a name so I can use it, and
4284 change boolean_or_operator predicate to boolean_operator so it can
4285 be used to generate a crand.
4286 (eqne): New code iterator.
4287 (bd/bd_neg): New code_attrs.
4288 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4289 a single define_insn.
4290 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4291 decrement (bdnzt/bdnzf/bdzt/bdzf).
4292 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4293 with the new names of the branch decrement patterns, and added the
4294 names of the branch decrement conditional patterns.
4296 2018-01-08 Richard Biener <rguenther@suse.de>
4298 PR tree-optimization/83563
4299 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4302 2018-01-08 Richard Biener <rguenther@suse.de>
4305 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4307 2018-01-08 Richard Biener <rguenther@suse.de>
4309 PR tree-optimization/83685
4310 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4311 references to abnormals.
4313 2018-01-08 Richard Biener <rguenther@suse.de>
4316 * dwarf2out.c (output_indirect_strings): Handle empty
4317 skeleton_debug_str_hash.
4318 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4320 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4322 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4323 (emit_store_direct): Likewise.
4324 (arc_trampoline_adjust_address): Likewise.
4325 (arc_asm_trampoline_template): New function.
4326 (arc_initialize_trampoline): Use asm_trampoline_template.
4327 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4328 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4329 * config/arc/arc.md (flush_icache): Delete pattern.
4331 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4333 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4334 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4337 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4340 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4341 by not USED_FOR_TARGET.
4342 (make_pass_resolve_sw_modes): Likewise.
4344 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4346 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4349 2018-01-08 Richard Biener <rguenther@suse.de>
4352 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4354 2018-01-08 Richard Biener <rguenther@suse.de>
4357 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4359 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4362 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4363 basic blocks with a small number of successors.
4364 (convert_control_dep_chain_into_preds): Improve handling of
4366 (dump_predicates): Split apart into...
4367 (dump_pred_chain): ...here...
4368 (dump_pred_info): ...and here.
4369 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4370 (can_chain_union_be_invalidated_p): Improve check for invalidation
4372 (uninit_uses_cannot_happen): Avoid unnecessary if
4373 convert_control_dep_chain_into_preds yielded nothing.
4375 2018-01-06 Martin Sebor <msebor@redhat.com>
4377 PR tree-optimization/83640
4378 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4379 subtracting negative offset from size.
4380 (builtin_access::overlap): Adjust offset bounds of the access to fall
4381 within the size of the object if possible.
4383 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4385 PR rtl-optimization/83699
4386 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4387 extract_bit_field_as_subreg to cases in which the extracted
4388 value is also a vector.
4390 * lra-constraints.c (process_alt_operands): Test for the equivalence
4391 substitutions when detecting a possible reload cycle.
4393 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4396 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4397 by default if flag_selective_schedling{,2}. Formatting fixes.
4399 PR rtl-optimization/83682
4400 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4401 if it has non-VECTOR_MODE element mode.
4402 (vec_duplicate_p): Likewise.
4405 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4406 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4408 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4411 * config/i386/i386-builtin.def
4412 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4413 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4414 Require also OPTION_MASK_ISA_AVX512F in addition to
4415 OPTION_MASK_ISA_GFNI.
4416 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4417 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4418 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4419 to OPTION_MASK_ISA_GFNI.
4420 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4421 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4422 OPTION_MASK_ISA_AVX512BW.
4423 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4424 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4425 addition to OPTION_MASK_ISA_GFNI.
4426 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4427 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4428 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4429 to OPTION_MASK_ISA_GFNI.
4430 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4431 a requirement for all ISAs rather than any of them with a few
4433 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4435 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4436 bitmasks to be enabled with 3 exceptions, instead of requiring any
4437 enabled ISA with lots of exceptions.
4438 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4439 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4440 Change avx512bw in isa attribute to avx512f.
4441 * config/i386/sgxintrin.h: Add license boilerplate.
4442 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4443 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4444 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4445 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4447 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4448 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4449 temporarily sse2 rather than sse if not enabled already.
4452 * config/i386/sse.md (VI248_VLBW): Rename to ...
4453 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4454 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4455 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4456 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4457 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4458 mode iterator instead of VI248_VLBW.
4460 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4462 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4463 (record_modified): Skip clobbers; add debug output.
4464 (param_change_prob): Use sreal frequencies.
4466 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4468 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4469 punt for user-aligned variables.
4471 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4473 * tree-chrec.c (chrec_contains_symbols): Return true for
4476 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4479 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4480 of (x|y) == x for BICS pattern.
4482 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4484 PR tree-optimization/83605
4485 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4486 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4489 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4491 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4492 * config/epiphany/rtems.h: New file.
4494 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4495 Uros Bizjak <ubizjak@gmail.com>
4498 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4499 QIreg_operand instead of register_operand predicate.
4500 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4501 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4502 comments instead of -fmitigate[-_]rop.
4504 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4507 * cgraphunit.c (symbol_table::compile): Switch to text_section
4508 before calling assembly_start debug hook.
4509 * run-rtl-passes.c (run_rtl_passes): Likewise.
4512 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4514 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4515 range_int_cst_p rather than !symbolic_range_p before calling
4516 extract_range_from_multiplicative_op_1.
4518 2018-01-04 Jeff Law <law@redhat.com>
4520 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4521 redundant test in assertion.
4523 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4525 * doc/rtl.texi: Document machine_mode wrapper classes.
4527 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4529 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4532 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4534 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4535 the VEC_PERM_EXPR fold to fail.
4537 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4540 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4541 to switched_sections.
4543 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4546 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4549 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4552 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4553 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4555 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4558 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4559 is BLKmode and bitpos not zero or mode change is needed.
4561 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4564 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4567 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4570 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4571 instead of MULT rtx. Update all corresponding splitters.
4573 (*ssub<modesuffix>): Ditto.
4575 (*cmp_sadd_di): Update split patterns.
4576 (*cmp_sadd_si): Ditto.
4577 (*cmp_sadd_sidi): Ditto.
4578 (*cmp_ssub_di): Ditto.
4579 (*cmp_ssub_si): Ditto.
4580 (*cmp_ssub_sidi): Ditto.
4581 * config/alpha/predicates.md (const23_operand): New predicate.
4582 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4583 Look for ASHIFT, not MULT inner operand.
4584 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4586 2018-01-04 Martin Liska <mliska@suse.cz>
4588 PR gcov-profile/83669
4589 * gcov.c (output_intermediate_file): Add version to intermediate
4591 * doc/gcov.texi: Document new field 'version' in intermediate
4592 file format. Fix location of '-k' option of gcov command.
4594 2018-01-04 Martin Liska <mliska@suse.cz>
4597 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4599 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4601 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4603 2018-01-03 Martin Sebor <msebor@redhat.com>
4605 PR tree-optimization/83655
4606 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4607 checking calls with invalid arguments.
4609 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4611 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4612 (vectorizable_mask_load_store): Delete.
4613 (vectorizable_call): Return false for masked loads and stores.
4614 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4615 instead of gimple_assign_rhs1.
4616 (vectorizable_load): Handle IFN_MASK_LOAD.
4617 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4621 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4623 (vectorizable_mask_load_store): ...here.
4624 (vectorizable_load): ...and here.
4626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4628 * tree-vect-stmts.c (vect_build_all_ones_mask)
4629 (vect_build_zero_merge_argument): New functions, split out from...
4630 (vectorizable_load): ...here.
4632 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4634 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4636 (vectorizable_mask_load_store): ...here.
4637 (vectorizable_store): ...and here.
4639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4641 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4643 (vectorizable_mask_load_store): ...here.
4645 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4647 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4648 (vect_model_store_cost): Take a vec_load_store_type instead of a
4650 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4651 (vect_model_store_cost): Take a vec_load_store_type instead of a
4653 (vectorizable_mask_load_store): Update accordingly.
4654 (vectorizable_store): Likewise.
4655 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4657 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4659 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4660 IFN_MASK_LOAD calls here rather than...
4661 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4664 Alan Hayward <alan.hayward@arm.com>
4665 David Sherwood <david.sherwood@arm.com>
4667 * expmed.c (extract_bit_field_1): For vector extracts,
4668 fall back to extract_bit_field_as_subreg if vec_extract
4671 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4672 Alan Hayward <alan.hayward@arm.com>
4673 David Sherwood <david.sherwood@arm.com>
4675 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4676 they are variable or constant sized.
4677 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4678 slots for constant-sized data.
4680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4681 Alan Hayward <alan.hayward@arm.com>
4682 David Sherwood <david.sherwood@arm.com>
4684 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4685 handling COND_EXPRs with boolean comparisons, try to find a better
4686 basis for the mask type than the boolean itself.
4688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4690 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4691 is calculated and how it can be overridden.
4692 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4693 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4695 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4698 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4699 Alan Hayward <alan.hayward@arm.com>
4700 David Sherwood <david.sherwood@arm.com>
4702 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4703 Remove the mode argument.
4704 (aarch64_simd_valid_immediate): Remove the mode and inverse
4706 * config/aarch64/iterators.md (bitsize): New iterator.
4707 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4708 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4709 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4710 aarch64_simd_valid_immediate.
4711 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4712 (aarch64_reg_or_bic_imm): Likewise.
4713 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4714 with an insn_type enum and msl with a modifier_type enum.
4715 Replace element_width with a scalar_mode. Change the shift
4716 to unsigned int. Add constructors for scalar_float_mode and
4717 scalar_int_mode elements.
4718 (aarch64_vect_float_const_representable_p): Delete.
4719 (aarch64_can_const_movi_rtx_p)
4720 (aarch64_simd_scalar_immediate_valid_for_move)
4721 (aarch64_simd_make_constant): Update call to
4722 aarch64_simd_valid_immediate.
4723 (aarch64_advsimd_valid_immediate_hs): New function.
4724 (aarch64_advsimd_valid_immediate): Likewise.
4725 (aarch64_simd_valid_immediate): Remove mode and inverse
4726 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4727 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4728 and aarch64_float_const_representable_p on the result.
4729 (aarch64_output_simd_mov_immediate): Remove mode argument.
4730 Update call to aarch64_simd_valid_immediate and use of
4731 simd_immediate_info.
4732 (aarch64_output_scalar_simd_mov_immediate): Update call
4735 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4736 Alan Hayward <alan.hayward@arm.com>
4737 David Sherwood <david.sherwood@arm.com>
4739 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4740 (mode_nunits): Likewise CONST_MODE_NUNITS.
4741 * machmode.def (ADJUST_NUNITS): Document.
4742 * genmodes.c (mode_data::need_nunits_adj): New field.
4743 (blank_mode): Update accordingly.
4744 (adj_nunits): New variable.
4745 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4747 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4748 listed in adj_nunits.
4749 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4750 listed in adj_nunits. Don't emit case statements for such modes.
4751 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4752 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4753 nothing if adj_nunits is nonnull.
4754 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4755 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4756 (emit_mode_fbit): Update use of print_maybe_const_decl.
4757 (emit_move_size): Likewise. Treat the array as non-const
4759 (emit_mode_adjustments): Handle adj_nunits.
4761 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4763 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4764 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4765 (VECTOR_MODES): Use it.
4766 (make_vector_modes): Take the prefix as an argument.
4768 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4769 Alan Hayward <alan.hayward@arm.com>
4770 David Sherwood <david.sherwood@arm.com>
4772 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4773 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4774 for MODE_VECTOR_BOOL.
4775 * machmode.def (VECTOR_BOOL_MODE): Document.
4776 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4777 (make_vector_bool_mode): New function.
4778 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4780 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4781 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4783 * stor-layout.c (int_mode_for_mode): Likewise.
4784 * tree.c (build_vector_type_for_mode): Likewise.
4785 * varasm.c (output_constant_pool_2): Likewise.
4786 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4787 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4788 for MODE_VECTOR_BOOL.
4789 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4790 of mode class checks.
4791 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4792 instead of a list of mode class checks.
4793 (expand_vector_scalar_condition): Likewise.
4794 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4797 Alan Hayward <alan.hayward@arm.com>
4798 David Sherwood <david.sherwood@arm.com>
4800 * machmode.h (mode_size): Change from unsigned short to
4802 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4803 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4804 or if measurement_type is not polynomial.
4805 (fixed_size_mode::includes_p): Check for constant-sized modes.
4806 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4807 return a poly_uint16 rather than an unsigned short.
4808 (emit_mode_size): Change the type of mode_size from unsigned short
4809 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4810 (emit_mode_adjustments): Cope with polynomial vector sizes.
4811 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4813 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4815 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4816 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4817 * caller-save.c (setup_save_areas): Likewise.
4818 (replace_reg_with_saved_mem): Likewise.
4819 * calls.c (emit_library_call_value_1): Likewise.
4820 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4821 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4822 (gen_lowpart_for_combine): Likewise.
4823 * convert.c (convert_to_integer_1): Likewise.
4824 * cse.c (equiv_constant, cse_insn): Likewise.
4825 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4826 (cselib_subst_to_values): Likewise.
4827 * dce.c (word_dce_process_block): Likewise.
4828 * df-problems.c (df_word_lr_mark_ref): Likewise.
4829 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4830 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4831 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4832 (rtl_for_decl_location): Likewise.
4833 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4834 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4835 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4836 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4837 (expand_expr_real_1): Likewise.
4838 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4839 (pad_below): Likewise.
4840 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4841 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4842 * ira.c (get_subreg_tracking_sizes): Likewise.
4843 * ira-build.c (ira_create_allocno_objects): Likewise.
4844 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4845 (ira_sort_regnos_for_alter_reg): Likewise.
4846 * ira-costs.c (record_operand_costs): Likewise.
4847 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4848 (resolve_simple_move): Likewise.
4849 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4850 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4851 (lra_constraints): Likewise.
4852 (CONST_POOL_OK_P): Reject variable-sized modes.
4853 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4854 (add_pseudo_to_slot, lra_spill): Likewise.
4855 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4856 * optabs-query.c (get_best_extraction_insn): Likewise.
4857 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4858 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4859 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4860 * recog.c (offsettable_address_addr_space_p): Likewise.
4861 * regcprop.c (maybe_mode_change): Likewise.
4862 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4863 * regrename.c (build_def_use): Likewise.
4864 * regstat.c (dump_reg_info): Likewise.
4865 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4866 (find_reloads, find_reloads_subreg_address): Likewise.
4867 * reload1.c (eliminate_regs_1): Likewise.
4868 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4869 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4870 (simplify_binary_operation_1, simplify_subreg): Likewise.
4871 * targhooks.c (default_function_arg_padding): Likewise.
4872 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4873 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4874 (verify_gimple_assign_ternary): Likewise.
4875 * tree-inline.c (estimate_move_cost): Likewise.
4876 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4877 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4878 (get_address_cost_ainc): Likewise.
4879 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4880 (vect_supportable_dr_alignment): Likewise.
4881 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4882 (vectorizable_reduction): Likewise.
4883 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4884 (vectorizable_operation, vectorizable_load): Likewise.
4885 * tree.c (build_same_sized_truth_vector_type): Likewise.
4886 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4887 * var-tracking.c (emit_note_insn_var_location): Likewise.
4888 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4889 (ADDR_VEC_ALIGN): Likewise.
4891 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4892 Alan Hayward <alan.hayward@arm.com>
4893 David Sherwood <david.sherwood@arm.com>
4895 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4897 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4898 or if measurement_type is polynomial.
4899 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4900 * combine.c (make_extraction): Likewise.
4901 * dse.c (find_shift_sequence): Likewise.
4902 * dwarf2out.c (mem_loc_descriptor): Likewise.
4903 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4904 (extract_bit_field, extract_low_bits): Likewise.
4905 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4906 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4907 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4908 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4909 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4910 * reload.c (find_reloads): Likewise.
4911 * reload1.c (alter_reg): Likewise.
4912 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4913 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4914 * tree-if-conv.c (predicate_mem_writes): Likewise.
4915 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4916 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4917 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4918 * valtrack.c (dead_debug_insert_temp): Likewise.
4919 * varasm.c (mergeable_constant_section): Likewise.
4920 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4923 Alan Hayward <alan.hayward@arm.com>
4924 David Sherwood <david.sherwood@arm.com>
4926 * expr.c (expand_assignment): Cope with polynomial mode sizes
4927 when assigning to a CONCAT.
4929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4930 Alan Hayward <alan.hayward@arm.com>
4931 David Sherwood <david.sherwood@arm.com>
4933 * machmode.h (mode_precision): Change from unsigned short to
4935 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4937 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4938 or if measurement_type is not polynomial.
4939 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4940 in which the mode is already known to be a scalar_int_mode.
4941 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4942 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4944 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4945 for GET_MODE_PRECISION.
4946 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4947 for GET_MODE_PRECISION.
4948 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4950 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4951 (expand_field_assignment, make_extraction): Likewise.
4952 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4953 (get_last_value): Likewise.
4954 * convert.c (convert_to_integer_1): Likewise.
4955 * cse.c (cse_insn): Likewise.
4956 * expr.c (expand_expr_real_1): Likewise.
4957 * lra-constraints.c (simplify_operand_subreg): Likewise.
4958 * optabs-query.c (can_atomic_load_p): Likewise.
4959 * optabs.c (expand_atomic_load): Likewise.
4960 (expand_atomic_store): Likewise.
4961 * ree.c (combine_reaching_defs): Likewise.
4962 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4963 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4964 * tree.h (type_has_mode_precision_p): Likewise.
4965 * ubsan.c (instrument_si_overflow): Likewise.
4967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4968 Alan Hayward <alan.hayward@arm.com>
4969 David Sherwood <david.sherwood@arm.com>
4971 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4972 polynomial numbers of units.
4973 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4974 (valid_vector_subparts_p): New function.
4975 (build_vector_type): Remove temporary shim and take the number
4976 of units as a poly_uint64 rather than an int.
4977 (build_opaque_vector_type): Take the number of units as a
4978 poly_uint64 rather than an int.
4979 * tree.c (build_vector_from_ctor): Handle polynomial
4980 TYPE_VECTOR_SUBPARTS.
4981 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4982 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4983 (build_vector_from_val): If the number of units is variable,
4984 use build_vec_duplicate_cst for constant operands and
4985 VEC_DUPLICATE_EXPR otherwise.
4986 (make_vector_type): Remove temporary is_constant ().
4987 (build_vector_type, build_opaque_vector_type): Take the number of
4988 units as a poly_uint64 rather than an int.
4989 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4991 * cfgexpand.c (expand_debug_expr): Likewise.
4992 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4993 (store_constructor, expand_expr_real_1): Likewise.
4994 (const_scalar_mask_from_tree): Likewise.
4995 * fold-const-call.c (fold_const_reduction): Likewise.
4996 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4997 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4998 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4999 (fold_relational_const): Likewise.
5000 (native_interpret_vector): Likewise. Change the size from an
5001 int to an unsigned int.
5002 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5003 TYPE_VECTOR_SUBPARTS.
5004 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5005 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5006 duplicating a non-constant operand into a variable-length vector.
5007 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5008 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5009 * ipa-icf.c (sem_variable::equals): Likewise.
5010 * match.pd: Likewise.
5011 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5012 * print-tree.c (print_node): Likewise.
5013 * stor-layout.c (layout_type): Likewise.
5014 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5015 * tree-cfg.c (verify_gimple_comparison): Likewise.
5016 (verify_gimple_assign_binary): Likewise.
5017 (verify_gimple_assign_ternary): Likewise.
5018 (verify_gimple_assign_single): Likewise.
5019 * tree-pretty-print.c (dump_generic_node): Likewise.
5020 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5021 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5022 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5023 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5024 (vect_shift_permute_load_chain): Likewise.
5025 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5026 (expand_vector_condition, optimize_vector_constructor): Likewise.
5027 (lower_vec_perm, get_compute_type): Likewise.
5028 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5029 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5030 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5031 (vect_recog_mask_conversion_pattern): Likewise.
5032 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5033 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5034 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5035 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5036 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5037 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5038 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5039 (supportable_widening_operation): Likewise.
5040 (supportable_narrowing_operation): Likewise.
5041 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5043 * varasm.c (output_constant): Likewise.
5045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5046 Alan Hayward <alan.hayward@arm.com>
5047 David Sherwood <david.sherwood@arm.com>
5049 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5050 so that both the length == 3 and length != 3 cases set up their
5051 own permute vectors. Add comments explaining why we know the
5052 number of elements is constant.
5053 (vect_permute_load_chain): Likewise.
5055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5056 Alan Hayward <alan.hayward@arm.com>
5057 David Sherwood <david.sherwood@arm.com>
5059 * machmode.h (mode_nunits): Change from unsigned char to
5061 (ONLY_FIXED_SIZE_MODES): New macro.
5062 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5063 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5064 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5066 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5067 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5068 or if measurement_type is not polynomial.
5069 * genmodes.c (ZERO_COEFFS): New macro.
5070 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5072 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5073 Use ZERO_COEFFS when emitting initializers.
5074 * data-streamer.h (bp_pack_poly_value): New function.
5075 (bp_unpack_poly_value): Likewise.
5076 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5077 for GET_MODE_NUNITS.
5078 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5079 for GET_MODE_NUNITS.
5080 * tree.c (make_vector_type): Remove temporary shim and make
5081 the real function take the number of units as a poly_uint64
5083 (build_vector_type_for_mode): Handle polynomial nunits.
5084 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5085 * emit-rtl.c (const_vec_series_p_1): Likewise.
5086 (gen_rtx_CONST_VECTOR): Likewise.
5087 * fold-const.c (test_vec_duplicate_folding): Likewise.
5088 * genrecog.c (validate_pattern): Likewise.
5089 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5090 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5091 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5092 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5093 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5094 * rtlanal.c (subreg_get_info): Likewise.
5095 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5096 (vect_grouped_load_supported): Likewise.
5097 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5098 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5099 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5100 (simplify_const_unary_operation, simplify_binary_operation_1)
5101 (simplify_const_binary_operation, simplify_ternary_operation)
5102 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5103 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5104 instead of CONST_VECTOR_NUNITS.
5105 * varasm.c (output_constant_pool_2): Likewise.
5106 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5107 explicit-encoded elements in the XVEC for variable-length vectors.
5109 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5111 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5114 Alan Hayward <alan.hayward@arm.com>
5115 David Sherwood <david.sherwood@arm.com>
5117 * coretypes.h (fixed_size_mode): Declare.
5118 (fixed_size_mode_pod): New typedef.
5119 * builtins.h (target_builtins::x_apply_args_mode)
5120 (target_builtins::x_apply_result_mode): Change type to
5121 fixed_size_mode_pod.
5122 * builtins.c (apply_args_size, apply_result_size, result_vector)
5123 (expand_builtin_apply_args_1, expand_builtin_apply)
5124 (expand_builtin_return): Update accordingly.
5126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5128 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5129 * cselib.c (cselib_hash_rtx): Likewise.
5130 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5131 CONST_VECTOR encoding.
5133 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5134 Jeff Law <law@redhat.com>
5137 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5138 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5139 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5140 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5143 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5144 explicitly probe *sp in a noreturn function if there were any callee
5145 register saves or frame pointer is needed.
5147 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5150 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5151 BLKmode for ternary, binary or unary expressions.
5154 * var-tracking.c (delete_vta_debug_insn): New inline function.
5155 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5156 insns from get_insns () to NULL instead of each bb separately.
5157 Use delete_vta_debug_insn. No longer static.
5158 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5159 delete_vta_debug_insns callers.
5160 * rtl.h (delete_vta_debug_insns): Declare.
5161 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5162 instead of variable_tracking_main.
5164 2018-01-03 Martin Sebor <msebor@redhat.com>
5166 PR tree-optimization/83603
5167 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5168 arguments past the endof the argument list in functions declared
5169 without a prototype.
5170 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5171 Avoid checking when arguments are null.
5173 2018-01-03 Martin Sebor <msebor@redhat.com>
5176 * doc/extend.texi (attribute const): Fix a typo.
5177 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5178 issuing -Wsuggest-attribute for void functions.
5180 2018-01-03 Martin Sebor <msebor@redhat.com>
5182 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5183 offset_int::from instead of wide_int::to_shwi.
5184 (maybe_diag_overlap): Remove assertion.
5185 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5186 * gimple-ssa-sprintf.c (format_directive): Same.
5187 (parse_directive): Same.
5188 (sprintf_dom_walker::compute_format_length): Same.
5189 (try_substitute_return_value): Same.
5191 2018-01-03 Jeff Law <law@redhat.com>
5194 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5195 non-constant residual for zero at runtime and avoid probing in
5196 that case. Reorganize code for trailing problem to mirror handling
5199 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5201 PR tree-optimization/83501
5202 * tree-ssa-strlen.c (get_string_cst): New.
5203 (handle_char_store): Call get_string_cst.
5205 2018-01-03 Martin Liska <mliska@suse.cz>
5207 PR tree-optimization/83593
5208 * tree-ssa-strlen.c: Include tree-cfg.h.
5209 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5210 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5211 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5213 (strlen_dom_walker::before_dom_children): Call
5214 gimple_purge_dead_eh_edges. Dump tranformation with details
5216 (strlen_dom_walker::before_dom_children): Update call by adding
5217 new argument cleanup_eh.
5218 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5220 2018-01-03 Martin Liska <mliska@suse.cz>
5223 * cif-code.def (VARIADIC_THUNK): New enum value.
5224 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5227 2018-01-03 Jan Beulich <jbeulich@suse.com>
5229 * sse.md (mov<mode>_internal): Tighten condition for when to use
5230 vmovdqu<ssescalarsize> for TI and OI modes.
5232 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5234 Update copyright years.
5236 2018-01-03 Martin Liska <mliska@suse.cz>
5239 * ipa-visibility.c (function_and_variable_visibility): Skip
5240 functions with noipa attribure.
5242 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5244 * gcc.c (process_command): Update copyright notice dates.
5245 * gcov-dump.c (print_version): Ditto.
5246 * gcov.c (print_version): Ditto.
5247 * gcov-tool.c (print_version): Ditto.
5248 * gengtype.c (create_file): Ditto.
5249 * doc/cpp.texi: Bump @copying's copyright year.
5250 * doc/cppinternals.texi: Ditto.
5251 * doc/gcc.texi: Ditto.
5252 * doc/gccint.texi: Ditto.
5253 * doc/gcov.texi: Ditto.
5254 * doc/install.texi: Ditto.
5255 * doc/invoke.texi: Ditto.
5257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5259 * vector-builder.h (vector_builder::m_full_nelts): Change from
5260 unsigned int to poly_uint64.
5261 (vector_builder::full_nelts): Update prototype accordingly.
5262 (vector_builder::new_vector): Likewise.
5263 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5264 (vector_builder::operator ==): Likewise.
5265 (vector_builder::finalize): Likewise.
5266 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5267 Take the number of elements as a poly_uint64 rather than an
5269 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5270 from unsigned int to poly_uint64.
5271 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5272 (vec_perm_indices::new_vector): Likewise.
5273 (vec_perm_indices::length): Likewise.
5274 (vec_perm_indices::nelts_per_input): Likewise.
5275 (vec_perm_indices::input_nelts): Likewise.
5276 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5277 number of elements per input as a poly_uint64 rather than an
5278 unsigned int. Use the original encoding for variable-length
5279 vectors, rather than clamping each individual element.
5280 For the second and subsequent elements in each pattern,
5281 clamp the step and base before clamping their sum.
5282 (vec_perm_indices::series_p): Handle polynomial element counts.
5283 (vec_perm_indices::all_in_range_p): Likewise.
5284 (vec_perm_indices_to_tree): Likewise.
5285 (vec_perm_indices_to_rtx): Likewise.
5286 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5287 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5288 (tree_vector_builder::new_binary_operation): Handle polynomial
5289 element counts. Return false if we need to know the number
5290 of elements at compile time.
5291 * fold-const.c (fold_vec_perm): Punt if the number of elements
5292 isn't known at compile time.
5294 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5296 * vec-perm-indices.h (vec_perm_builder): Change element type
5297 from HOST_WIDE_INT to poly_int64.
5298 (vec_perm_indices::element_type): Update accordingly.
5299 (vec_perm_indices::clamp): Handle polynomial element_types.
5300 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5301 (vec_perm_indices::all_in_range_p): Likewise.
5302 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5304 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5305 polynomial vec_perm_indices element types.
5306 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5307 * fold-const.c (fold_vec_perm): Likewise.
5308 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5309 * tree-vect-generic.c (lower_vec_perm): Likewise.
5310 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5311 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5312 element type to HOST_WIDE_INT.
5314 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5315 Alan Hayward <alan.hayward@arm.com>
5316 David Sherwood <david.sherwood@arm.com>
5318 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5319 rather than an int. Use plus_constant.
5320 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5321 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5323 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5324 Alan Hayward <alan.hayward@arm.com>
5325 David Sherwood <david.sherwood@arm.com>
5327 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5328 a HOST_WIDE_INT to a poly_int64.
5330 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5331 Alan Hayward <alan.hayward@arm.com>
5332 David Sherwood <david.sherwood@arm.com>
5334 * calls.c (load_register_parameters): Cope with polynomial
5335 mode sizes. Require a constant size for BLKmode parameters
5336 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5337 forces a parameter to be padded at the lsb end in order to
5338 fill a complete number of words, require the parameter size
5339 to be ordered wrt UNITS_PER_WORD.
5341 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5342 Alan Hayward <alan.hayward@arm.com>
5343 David Sherwood <david.sherwood@arm.com>
5345 * reload1.c (spill_stack_slot_width): Change element type
5346 from unsigned int to poly_uint64_pod.
5347 (alter_reg): Treat mode sizes as polynomial.
5349 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5350 Alan Hayward <alan.hayward@arm.com>
5351 David Sherwood <david.sherwood@arm.com>
5353 * reload.c (complex_word_subreg_p): New function.
5354 (reload_inner_reg_of_subreg, push_reload): Use it.
5356 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5357 Alan Hayward <alan.hayward@arm.com>
5358 David Sherwood <david.sherwood@arm.com>
5360 * lra-constraints.c (process_alt_operands): Reject matched
5361 operands whose sizes aren't ordered.
5362 (match_reload): Refer to this check here.
5364 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5365 Alan Hayward <alan.hayward@arm.com>
5366 David Sherwood <david.sherwood@arm.com>
5368 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5369 that the mode size is in the set {1, 2, 4, 8, 16}.
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 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5376 Use plus_constant instead of gen_rtx_PLUS.
5378 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5379 Alan Hayward <alan.hayward@arm.com>
5380 David Sherwood <david.sherwood@arm.com>
5382 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5383 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5384 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5385 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5386 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5387 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5388 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5389 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5390 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5391 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5393 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5394 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5395 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5396 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5397 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5398 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5399 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5400 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5401 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5402 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5404 * expr.c (emit_move_resolve_push): Treat the input and result
5405 of PUSH_ROUNDING as a poly_int64.
5406 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5407 (emit_push_insn): Likewise.
5408 * lra-eliminations.c (mark_not_eliminable): Likewise.
5409 * recog.c (push_operand): Likewise.
5410 * reload1.c (elimination_effects): Likewise.
5411 * rtlanal.c (nonzero_bits1): Likewise.
5412 * calls.c (store_one_arg): Likewise. Require the padding to be
5413 known at compile time.
5415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5416 Alan Hayward <alan.hayward@arm.com>
5417 David Sherwood <david.sherwood@arm.com>
5419 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5420 Use plus_constant instead of gen_rtx_PLUS.
5422 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5423 Alan Hayward <alan.hayward@arm.com>
5424 David Sherwood <david.sherwood@arm.com>
5426 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5429 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5430 Alan Hayward <alan.hayward@arm.com>
5431 David Sherwood <david.sherwood@arm.com>
5433 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5434 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5435 via stack temporaries. Treat the mode size as polynomial too.
5437 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5438 Alan Hayward <alan.hayward@arm.com>
5439 David Sherwood <david.sherwood@arm.com>
5441 * expr.c (expand_expr_real_2): When handling conversions involving
5442 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5443 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5444 as a poly_uint64 too.
5446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5447 Alan Hayward <alan.hayward@arm.com>
5448 David Sherwood <david.sherwood@arm.com>
5450 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
5457 REGMODE_NATURAL_SIZE.
5458 * expmed.c (store_bit_field_1): Likewise.
5459 * expr.c (store_constructor): Likewise.
5460 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5461 and polynomial REGMODE_NATURAL_SIZE.
5462 (gen_lowpart_common): Likewise.
5463 * reginfo.c (record_subregs_of_mode): Likewise.
5464 * rtlanal.c (read_modify_subreg_p): Likewise.
5466 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5467 Alan Hayward <alan.hayward@arm.com>
5468 David Sherwood <david.sherwood@arm.com>
5470 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5471 numbers of elements.
5473 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5474 Alan Hayward <alan.hayward@arm.com>
5475 David Sherwood <david.sherwood@arm.com>
5477 * match.pd: Cope with polynomial numbers of vector elements.
5479 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5480 Alan Hayward <alan.hayward@arm.com>
5481 David Sherwood <david.sherwood@arm.com>
5483 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5484 in a POINTER_PLUS_EXPR.
5486 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5487 Alan Hayward <alan.hayward@arm.com>
5488 David Sherwood <david.sherwood@arm.com>
5490 * omp-simd-clone.c (simd_clone_subparts): New function.
5491 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5492 (ipa_simd_modify_function_body): Likewise.
5494 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5495 Alan Hayward <alan.hayward@arm.com>
5496 David Sherwood <david.sherwood@arm.com>
5498 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5499 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5500 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5501 (expand_vector_condition, vector_element): Likewise.
5502 (subparts_gt): New function.
5503 (get_compute_type): Use subparts_gt.
5504 (count_type_subparts): Delete.
5505 (expand_vector_operations_1): Use subparts_gt instead of
5506 count_type_subparts.
5508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5509 Alan Hayward <alan.hayward@arm.com>
5510 David Sherwood <david.sherwood@arm.com>
5512 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5513 (vect_compile_time_alias): ...this new function. Do the calculation
5514 on poly_ints rather than trees.
5515 (vect_prune_runtime_alias_test_list): Update call accordingly.
5517 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5518 Alan Hayward <alan.hayward@arm.com>
5519 David Sherwood <david.sherwood@arm.com>
5521 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5523 (vect_schedule_slp_instance): Likewise.
5525 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5526 Alan Hayward <alan.hayward@arm.com>
5527 David Sherwood <david.sherwood@arm.com>
5529 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5530 constant and extern definitions for variable-length vectors.
5531 (vect_get_constant_vectors): Note that the number of units
5532 is known to be constant.
5534 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5535 Alan Hayward <alan.hayward@arm.com>
5536 David Sherwood <david.sherwood@arm.com>
5538 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5539 of units as polynomial. Choose between WIDE and NARROW based
5542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5543 Alan Hayward <alan.hayward@arm.com>
5544 David Sherwood <david.sherwood@arm.com>
5546 * tree-vect-stmts.c (simd_clone_subparts): New function.
5547 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5549 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5550 Alan Hayward <alan.hayward@arm.com>
5551 David Sherwood <david.sherwood@arm.com>
5553 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5554 vectors as polynomial. Use build_index_vector for
5557 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5558 Alan Hayward <alan.hayward@arm.com>
5559 David Sherwood <david.sherwood@arm.com>
5561 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5562 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5563 for variable-length vectors.
5564 (vectorizable_mask_load_store): Treat the number of units as
5565 polynomial, asserting that it is constant if the condition has
5566 already been enforced.
5567 (vectorizable_store, vectorizable_load): Likewise.
5569 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5570 Alan Hayward <alan.hayward@arm.com>
5571 David Sherwood <david.sherwood@arm.com>
5573 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5574 of units as polynomial. Punt if we can't tell at compile time
5575 which vector contains the final result.
5577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5578 Alan Hayward <alan.hayward@arm.com>
5579 David Sherwood <david.sherwood@arm.com>
5581 * tree-vect-loop.c (vectorizable_induction): Treat the number
5582 of units as polynomial. Punt on SLP inductions. Use an integer
5583 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5584 cast of such a series for variable-length floating-point
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.h (build_index_vector): Declare.
5592 * tree.c (build_index_vector): New function.
5593 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5594 of units as polynomial, forcibly converting it to a constant if
5595 vectorizable_reduction has already enforced the condition.
5596 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5597 to create a {1,2,3,...} vector.
5598 (vectorizable_reduction): Treat the number of units as polynomial.
5599 Choose vectype_in based on the largest scalar element size rather
5600 than the smallest number of units. Enforce the restrictions
5603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5604 Alan Hayward <alan.hayward@arm.com>
5605 David Sherwood <david.sherwood@arm.com>
5607 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5608 number of units as polynomial.
5610 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5611 Alan Hayward <alan.hayward@arm.com>
5612 David Sherwood <david.sherwood@arm.com>
5614 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5615 * target.def (autovectorize_vector_sizes): Return the vector sizes
5616 by pointer, using vector_sizes rather than a bitmask.
5617 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5618 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5619 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5621 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5622 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5623 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5624 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5625 * omp-general.c (omp_max_vf): Likewise.
5626 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5627 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5628 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5629 * tree-vect-slp.c (vect_slp_bb): Likewise.
5630 * doc/tm.texi: Regenerate.
5631 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5633 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5634 the vector size as a poly_uint64 rather than an unsigned int.
5635 (current_vector_size): Change from an unsigned int to a poly_uint64.
5636 (get_vectype_for_scalar_type): Update accordingly.
5637 * tree.h (build_truth_vector_type): Take the size and number of
5638 units as a poly_uint64 rather than an unsigned int.
5639 (build_vector_type): Add a temporary overload that takes
5640 the number of units as a poly_uint64 rather than an unsigned int.
5641 * tree.c (make_vector_type): Likewise.
5642 (build_truth_vector_type): Take the number of units as a poly_uint64
5643 rather than an unsigned int.
5645 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5646 Alan Hayward <alan.hayward@arm.com>
5647 David Sherwood <david.sherwood@arm.com>
5649 * target.def (get_mask_mode): Take the number of units and length
5650 as poly_uint64s rather than unsigned ints.
5651 * targhooks.h (default_get_mask_mode): Update accordingly.
5652 * targhooks.c (default_get_mask_mode): Likewise.
5653 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5654 * doc/tm.texi: Regenerate.
5656 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5657 Alan Hayward <alan.hayward@arm.com>
5658 David Sherwood <david.sherwood@arm.com>
5660 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5661 * omp-general.c (omp_max_vf): Likewise.
5662 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5663 (expand_omp_simd): Handle polynomial safelen.
5664 * omp-low.c (omplow_simd_context): Add a default constructor.
5665 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5666 (lower_rec_simd_input_clauses): Update accordingly.
5667 (lower_rec_input_clauses): Likewise.
5669 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5670 Alan Hayward <alan.hayward@arm.com>
5671 David Sherwood <david.sherwood@arm.com>
5673 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5674 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5675 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5676 (vect_analyze_slp_cost): Likewise.
5677 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5678 (vect_model_load_cost): Likewise.
5680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5681 Alan Hayward <alan.hayward@arm.com>
5682 David Sherwood <david.sherwood@arm.com>
5684 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5685 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5686 from an unsigned int * to a poly_uint64_pod *.
5687 (calculate_unrolling_factor): New function.
5688 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5691 Alan Hayward <alan.hayward@arm.com>
5692 David Sherwood <david.sherwood@arm.com>
5694 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5695 from an unsigned int to a poly_uint64.
5696 (_loop_vec_info::slp_unrolling_factor): Likewise.
5697 (_loop_vec_info::vectorization_factor): Change from an int
5699 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5700 (vect_get_num_vectors): New function.
5701 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5702 (vect_get_num_copies): Use vect_get_num_vectors.
5703 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5704 to an unsigned int *.
5705 (vect_analyze_data_refs): Change min_vf from an int * to a
5707 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5708 than an unsigned HOST_WIDE_INT.
5709 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5710 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5711 to an unsigned int *.
5712 (vect_analyze_data_ref_dependences): Likewise.
5713 (vect_compute_data_ref_alignment): Handle polynomial vf.
5714 (vect_enhance_data_refs_alignment): Likewise.
5715 (vect_prune_runtime_alias_test_list): Likewise.
5716 (vect_shift_permute_load_chain): Likewise.
5717 (vect_supportable_dr_alignment): Likewise.
5718 (dependence_distance_ge_vf): Take the vectorization factor as a
5719 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5720 (vect_analyze_data_refs): Change min_vf from an int * to a
5722 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5723 vfm1 as a poly_uint64 rather than an int. Make the same change
5724 for the returned bound_scalar.
5725 (vect_gen_vector_loop_niters): Handle polynomial vf.
5726 (vect_do_peeling): Likewise. Update call to
5727 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5728 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5730 * tree-vect-loop.c (vect_determine_vectorization_factor)
5731 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5732 (vect_get_known_peeling_cost): Likewise.
5733 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5734 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5735 (vect_transform_loop): Likewise. Use the lowest possible VF when
5736 updating the upper bounds of the loop.
5737 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5739 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5740 polynomial unroll factors.
5741 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5742 (vect_make_slp_decision): Likewise.
5743 (vect_supported_load_permutation_p): Likewise, and polynomial
5745 (vect_analyze_slp_cost): Handle polynomial vf.
5746 (vect_slp_analyze_node_operations): Likewise.
5747 (vect_slp_analyze_bb_1): Likewise.
5748 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5749 than an unsigned HOST_WIDE_INT.
5750 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5751 (vectorizable_load): Handle polynomial vf.
5752 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5754 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5757 Alan Hayward <alan.hayward@arm.com>
5758 David Sherwood <david.sherwood@arm.com>
5760 * match.pd: Handle bit operations involving three constants
5761 and try to fold one pair.
5763 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5765 * tree-vect-loop-manip.c: Include gimple-fold.h.
5766 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5767 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5768 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5769 Add a path that uses a step of VF instead of 1, but disable it
5771 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5772 and niters_no_overflow parameters. Update calls to
5773 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5774 Create a new SSA name if the latter choses to use a ste other
5775 than zero, and return it via niters_vector_mult_vf_var.
5776 * tree-vect-loop.c (vect_transform_loop): Update calls to
5777 vect_do_peeling, vect_gen_vector_loop_niters and
5778 slpeel_make_loop_iterate_ntimes.
5779 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5780 (vect_gen_vector_loop_niters): Update declarations after above changes.
5782 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5784 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5785 128-bit round to integer instructions.
5786 (ceil<mode>2): Likewise.
5787 (btrunc<mode>2): Likewise.
5788 (round<mode>2): Likewise.
5790 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5792 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5793 unaligned VSX load/store on P8/P9.
5794 (expand_block_clear): Allow the use of unaligned VSX
5795 load/store on P8/P9.
5797 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5799 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5801 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5802 swap associated with both a load and a store.
5804 2018-01-02 Andrew Waterman <andrew@sifive.com>
5806 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5807 * config/riscv/riscv.md (clear_cache): Use it.
5809 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5811 * web.c: Remove out-of-date comment.
5813 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5815 * expr.c (fixup_args_size_notes): Check that any existing
5816 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5817 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5818 (emit_single_push_insn): ...here.
5820 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5822 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5823 (const_vector_encoded_nelts): New function.
5824 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5825 (const_vector_int_elt, const_vector_elt): Declare.
5826 * emit-rtl.c (const_vector_int_elt_1): New function.
5827 (const_vector_elt): Likewise.
5828 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5829 of CONST_VECTOR_ELT.
5831 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5833 * expr.c: Include rtx-vector-builder.h.
5834 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5835 directly on the tree encoding.
5836 (const_vector_from_tree): Likewise.
5837 * optabs.c: Include rtx-vector-builder.h.
5838 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5839 sequence of "u" values.
5840 * vec-perm-indices.c: Include rtx-vector-builder.h.
5841 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5842 directly on the vec_perm_indices encoding.
5844 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5846 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5847 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5848 * rtx-vector-builder.h: New file.
5849 * rtx-vector-builder.c: Likewise.
5850 * rtl.h (rtx_def::u2): Add a const_vector field.
5851 (CONST_VECTOR_NPATTERNS): New macro.
5852 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5853 (CONST_VECTOR_DUPLICATE_P): Likewise.
5854 (CONST_VECTOR_STEPPED_P): Likewise.
5855 (CONST_VECTOR_ENCODED_ELT): Likewise.
5856 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5857 (unwrap_const_vec_duplicate): Likewise.
5858 (const_vec_series_p): Check for a non-duplicated vector encoding.
5859 Say that the function only returns true for integer vectors.
5860 * emit-rtl.c: Include rtx-vector-builder.h.
5861 (gen_const_vec_duplicate_1): Delete.
5862 (gen_const_vector): Call gen_const_vec_duplicate instead of
5863 gen_const_vec_duplicate_1.
5864 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5865 (gen_const_vec_duplicate): Use rtx_vector_builder.
5866 (gen_const_vec_series): Likewise.
5867 (gen_rtx_CONST_VECTOR): Likewise.
5868 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5869 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5870 Build a new vector rather than modifying a CONST_VECTOR in-place.
5871 (handle_special_swappables): Update call accordingly.
5872 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5873 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5874 Build a new vector rather than modifying a CONST_VECTOR in-place.
5875 (handle_special_swappables): Update call accordingly.
5877 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5879 * simplify-rtx.c (simplify_const_binary_operation): Use
5880 CONST_VECTOR_ELT instead of XVECEXP.
5882 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5884 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5885 the selector elements to be different from the data elements
5886 if the selector is a VECTOR_CST.
5887 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5888 ssizetype for the selector.
5890 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5892 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5893 before testing each element individually.
5894 * tree-vect-generic.c (lower_vec_perm): Likewise.
5896 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5898 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5899 * selftest-run-tests.c (selftest::run_tests): Call it.
5900 * vector-builder.h (vector_builder::operator ==): New function.
5901 (vector_builder::operator !=): Likewise.
5902 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5903 (vec_perm_indices::all_from_input_p): New function.
5904 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5905 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5906 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5907 instead of reading the VECTOR_CST directly. Detect whether both
5908 vector inputs are the same before constructing the vec_perm_indices,
5909 and update the number of inputs argument accordingly. Use the
5910 utility functions added above. Only construct sel2 if we need to.
5912 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5914 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5915 the broadcast of the low byte.
5916 (expand_mult_highpart): Use an explicit encoding for the permutes.
5917 * optabs-query.c (can_mult_highpart_p): Likewise.
5918 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5919 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5920 (vectorizable_bswap): Likewise.
5921 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5922 explicit encoding for the power-of-2 permutes.
5923 (vect_permute_store_chain): Likewise.
5924 (vect_grouped_load_supported): Likewise.
5925 (vect_permute_load_chain): Likewise.
5927 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5929 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5930 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5931 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5932 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5933 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5934 (vect_gen_perm_mask_any): Likewise.
5936 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5938 * int-vector-builder.h: New file.
5939 * vec-perm-indices.h: Include int-vector-builder.h.
5940 (vec_perm_indices): Redefine as an int_vector_builder.
5941 (auto_vec_perm_indices): Delete.
5942 (vec_perm_builder): Redefine as a stand-alone class.
5943 (vec_perm_indices::vec_perm_indices): New function.
5944 (vec_perm_indices::clamp): Likewise.
5945 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5946 (vec_perm_indices::new_vector): New function.
5947 (vec_perm_indices::new_expanded_vector): Update for new
5948 vec_perm_indices class.
5949 (vec_perm_indices::rotate_inputs): New function.
5950 (vec_perm_indices::all_in_range_p): Operate directly on the
5951 encoded form, without computing elided elements.
5952 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5953 encoding. Update for new vec_perm_indices class.
5954 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5955 the given vec_perm_builder.
5956 (expand_vec_perm_var): Update vec_perm_builder constructor.
5957 (expand_mult_highpart): Use vec_perm_builder instead of
5958 auto_vec_perm_indices.
5959 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5960 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5961 or double series encoding as appropriate.
5962 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5963 vec_perm_indices instead of auto_vec_perm_indices.
5964 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5965 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5966 (vect_permute_store_chain): Likewise.
5967 (vect_grouped_load_supported): Likewise.
5968 (vect_permute_load_chain): Likewise.
5969 (vect_shift_permute_load_chain): Likewise.
5970 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5971 (vect_transform_slp_perm_load): Likewise.
5972 (vect_schedule_slp_instance): Likewise.
5973 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5974 (vectorizable_mask_load_store): Likewise.
5975 (vectorizable_bswap): Likewise.
5976 (vectorizable_store): Likewise.
5977 (vectorizable_load): Likewise.
5978 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5979 vec_perm_indices instead of auto_vec_perm_indices. Use
5980 tree_to_vec_perm_builder to read the vector from a tree.
5981 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5982 vec_perm_builder instead of a vec_perm_indices.
5983 (have_whole_vector_shift): Use vec_perm_builder and
5984 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5985 truncation to calc_vec_perm_mask_for_shift.
5986 (vect_create_epilog_for_reduction): Likewise.
5987 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5988 from auto_vec_perm_indices to vec_perm_indices.
5989 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5990 instead of changing individual elements.
5991 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5992 the vector in d.perm.
5993 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5994 from auto_vec_perm_indices to vec_perm_indices.
5995 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5996 instead of changing individual elements.
5997 (arm_vectorize_vec_perm_const): Use new_vector to install
5998 the vector in d.perm.
5999 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6000 Update vec_perm_builder constructor.
6001 (rs6000_expand_interleave): Likewise.
6002 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6003 (rs6000_expand_interleave): Likewise.
6005 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6007 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6008 to qimode could truncate the indices.
6009 * optabs.c (expand_vec_perm_var): Likewise.
6011 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6013 * Makefile.in (OBJS): Add vec-perm-indices.o.
6014 * vec-perm-indices.h: New file.
6015 * vec-perm-indices.c: Likewise.
6016 * target.h (vec_perm_indices): Replace with a forward class
6018 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6019 * optabs.h: Include vec-perm-indices.h.
6020 (expand_vec_perm): Delete.
6021 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6022 (expand_vec_perm_const): Declare.
6023 * target.def (vec_perm_const_ok): Replace with...
6024 (vec_perm_const): ...this new hook.
6025 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6026 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6027 * doc/tm.texi: Regenerate.
6028 * optabs.def (vec_perm_const): Delete.
6029 * doc/md.texi (vec_perm_const): Likewise.
6030 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6031 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6032 expand_vec_perm for constant permutation vectors. Assert that
6033 the mode of variable permutation vectors is the integer equivalent
6034 of the mode that is being permuted.
6035 * optabs-query.h (selector_fits_mode_p): Declare.
6036 * optabs-query.c: Include vec-perm-indices.h.
6037 (selector_fits_mode_p): New function.
6038 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6039 is defined, instead of checking whether the vec_perm_const_optab
6040 exists. Use targetm.vectorize.vec_perm_const instead of
6041 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6042 fit in the vector mode before using a variable permute.
6043 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6044 vec_perm_indices instead of an rtx.
6045 (expand_vec_perm): Replace with...
6046 (expand_vec_perm_const): ...this new function. Take the selector
6047 as a vec_perm_indices rather than an rtx. Also take the mode of
6048 the selector. Update call to shift_amt_for_vec_perm_mask.
6049 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6050 Use vec_perm_indices::new_expanded_vector to expand the original
6051 selector into bytes. Check whether the indices fit in the vector
6052 mode before using a variable permute.
6053 (expand_vec_perm_var): Make global.
6054 (expand_mult_highpart): Use expand_vec_perm_const.
6055 * fold-const.c: Includes vec-perm-indices.h.
6056 * tree-ssa-forwprop.c: Likewise.
6057 * tree-vect-data-refs.c: Likewise.
6058 * tree-vect-generic.c: Likewise.
6059 * tree-vect-loop.c: Likewise.
6060 * tree-vect-slp.c: Likewise.
6061 * tree-vect-stmts.c: Likewise.
6062 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6064 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6065 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6066 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6067 (aarch64_vectorize_vec_perm_const): ...this new function.
6068 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6069 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6070 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6071 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6072 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6073 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6074 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6076 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6077 check for NEON modes.
6078 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6079 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6080 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6081 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6083 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6084 the old VEC_PERM_CONST conditions.
6085 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6086 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6087 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6088 (ia64_vectorize_vec_perm_const_ok): Merge into...
6089 (ia64_vectorize_vec_perm_const): ...this new function.
6090 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6091 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6092 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6093 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6094 * config/mips/mips.c (mips_expand_vec_perm_const)
6095 (mips_vectorize_vec_perm_const_ok): Merge into...
6096 (mips_vectorize_vec_perm_const): ...this new function.
6097 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6098 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6099 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6100 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6101 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6102 (rs6000_expand_vec_perm_const): Delete.
6103 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6105 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6106 (altivec_expand_vec_perm_const_le): Take each operand individually.
6107 Operate on constant selectors rather than rtxes.
6108 (altivec_expand_vec_perm_const): Likewise. Update call to
6109 altivec_expand_vec_perm_const_le.
6110 (rs6000_expand_vec_perm_const): Delete.
6111 (rs6000_vectorize_vec_perm_const_ok): Delete.
6112 (rs6000_vectorize_vec_perm_const): New function.
6113 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6114 an element count and rtx array.
6115 (rs6000_expand_extract_even): Update call accordingly.
6116 (rs6000_expand_interleave): Likewise.
6117 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6118 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6119 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6120 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6121 (rs6000_expand_vec_perm_const): Delete.
6122 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6123 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6124 (altivec_expand_vec_perm_const_le): Take each operand individually.
6125 Operate on constant selectors rather than rtxes.
6126 (altivec_expand_vec_perm_const): Likewise. Update call to
6127 altivec_expand_vec_perm_const_le.
6128 (rs6000_expand_vec_perm_const): Delete.
6129 (rs6000_vectorize_vec_perm_const_ok): Delete.
6130 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6131 reference to the SPE evmerge intructions.
6132 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6133 an element count and rtx array.
6134 (rs6000_expand_extract_even): Update call accordingly.
6135 (rs6000_expand_interleave): Likewise.
6136 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6137 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6139 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6141 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6143 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6144 vector mode and that that mode matches the mode of the data
6146 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6147 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6148 directly using expand_vec_perm_1 when forcing selectors into
6150 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6152 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6154 * optabs-query.h (can_vec_perm_p): Delete.
6155 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6156 * optabs-query.c (can_vec_perm_p): Split into...
6157 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6158 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6159 particular selector is valid.
6160 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6161 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6162 (vect_grouped_load_supported): Likewise.
6163 (vect_shift_permute_load_chain): Likewise.
6164 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6165 (vect_transform_slp_perm_load): Likewise.
6166 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6167 (vectorizable_bswap): Likewise.
6168 (vect_gen_perm_mask_checked): Likewise.
6169 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6170 implementations of variable permutation vectors into account
6171 when deciding which selector to use.
6172 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6173 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6174 with a false third argument.
6175 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6176 to test whether the constant selector is valid and can_vec_perm_var_p
6177 to test whether a variable selector is valid.
6179 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6181 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6182 * optabs-query.c (can_vec_perm_p): Likewise.
6183 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6184 instead of vec_perm_indices.
6185 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6186 (vect_gen_perm_mask_checked): Likewise,
6187 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6188 (vect_gen_perm_mask_checked): Likewise,
6190 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6192 * optabs-query.h (qimode_for_vec_perm): Declare.
6193 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6194 (qimode_for_vec_perm): ...this new function.
6195 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6197 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6199 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6200 does not have a conditional at the top.
6202 2018-01-02 Richard Biener <rguenther@suse.de>
6204 * ipa-inline.c (big_speedup_p): Fix expression.
6206 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6209 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6212 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6216 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6217 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6218 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6219 cond_taken_branch_cost 3->4.
6221 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6223 PR tree-optimization/83581
6224 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6225 TODO_cleanup_cfg if any changes have been made.
6228 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6229 convert_modes if target mode has the right side, but different mode
6233 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6234 last argument when extracting from CONCAT. If either from_real or
6235 from_imag is NULL, use expansion through memory. If result is not
6236 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6237 the parts directly to inner mode, if even that fails, use expansion
6241 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6242 check for bswap in mode rather than HImode and use that in expand_unop
6245 Copyright (C) 2018 Free Software Foundation, Inc.
6247 Copying and distribution of this file, with or without modification,
6248 are permitted in any medium without royalty provided the copyright
6249 notice and this notice are preserved.