1 2018-02-05 Martin Sebor <msebor@redhat.com>
3 PR tree-optimization/83369
4 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
7 2018-02-05 Martin Liska <mliska@suse.cz>
9 * doc/invoke.texi: Cherry-pick upstream r323995.
11 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
13 * ira.c (ira_init_register_move_cost): Adjust comment.
15 2018-02-05 Martin Liska <mliska@suse.cz>
18 * doc/gcov.texi: Fix typo in documentation.
20 2018-02-05 Martin Liska <mliska@suse.cz>
23 * doc/gcov.texi: Document necessity of --dynamic-list-data when
24 using dlopen functionality.
26 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
28 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
29 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
30 _mm_maskz_range_ss, _mm_mask_range_round_ss,
31 _mm_maskz_range_round_ss): New intrinsics.
32 (__builtin_ia32_rangesd128_round)
33 (__builtin_ia32_rangess128_round): Remove.
34 (__builtin_ia32_rangesd128_mask_round,
35 __builtin_ia32_rangess128_mask_round): New builtins.
36 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
37 __builtin_ia32_rangess128_round): Remove.
38 (__builtin_ia32_rangesd128_mask_round,
39 __builtin_ia32_rangess128_mask_round): New builtins.
40 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
41 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
42 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
43 "<round_saeonly_constraint>")): Changed to ...
44 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
45 "<round_saeonly_scalar_constraint>")): ... this.
46 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
47 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
48 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
49 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
50 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
52 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
54 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
56 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
57 Remove all values except native, 8540 and 8548.
59 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
61 * config/i386/i386.c (ix86_output_function_return): Pass
62 INVALID_REGNUM, instead of -1, as invalid register number to
63 indirect_thunk_name and output_indirect_thunk.
65 2018-02-02 Julia Koval <julia.koval@intel.com>
67 * config.gcc: Add -march=icelake.
68 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
69 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
70 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
71 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
72 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
73 (processor_target_table): Add icelake.
74 (ix86_option_override_internal): Handle new PTAs.
75 (get_builtin_code_for_version): Handle icelake.
76 (M_INTEL_COREI7_ICELAKE): New.
77 (fold_builtin_cpu): Handle icelake.
78 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
79 * doc/invoke.texi: Add -march=icelake.
81 2018-02-02 Julia Koval <julia.koval@intel.com>
83 * config/i386/i386.c (ix86_option_override_internal): Change flags type
85 * wide-int-bitmask.h: New.
87 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
90 * config/i386/i386.md: Replace Pmode with word_mode in
91 builtin_setjmp_setup and builtin_longjmp to support x32.
93 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
97 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
99 (rs6000_supported_cpu_names): New static variable.
100 (linux_cpu_translation_table): Likewise.
101 (elf_platform) <cpu>: Define new static variable and use it.
102 Translate kernel AT_PLATFORM name to canonical name if needed.
103 Error if platform name is unknown.
105 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
108 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
110 2018-02-01 Jeff Law <law@redhat.com>
113 * config/i386/i386.c (release_scratch_register_on_entry): Add new
114 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
115 the scratch if RELEASE_VIA_POP is false.
116 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
117 If we have to save a temporary register, decrement SIZE appropriately.
118 Pass new arguments to release_scratch_register_on_entry.
119 (ix86_adjust_stack_and_probe): Likewise.
120 (ix86_emit_probe_stack_range): Pass new arguments to
121 release_scratch_register_on_entry.
123 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
125 PR rtl-optimization/84157
126 * combine.c (change_zero_ext): Use REG_P predicate in
127 front of HARD_REGISTER_P predicate.
129 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
131 * config/avr/avr.c (avr_option_override): Move disabling of
132 -fdelete-null-pointer-checks to...
133 * common/config/avr/avr-common.c (avr_option_optimization_table):
136 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
138 PR tree-optimization/81635
139 * tree-data-ref.c (split_constant_offset_1): For types that
140 wrap on overflow, try to use range info to prove that wrapping
143 2018-02-01 Renlin Li <renlin.li@arm.com>
146 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
148 (aarch64_register_move_cost): Likewise.
149 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
151 (REG_CLASS_NAMES): Likewise.
152 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
153 TAILCALL_ADDR_REGS. Remove IP registers.
154 * config/aarch64/aarch64.md (Ucs): Update register constraint.
156 2018-02-01 Richard Biener <rguenther@suse.de>
158 * domwalk.h (dom_walker::dom_walker): Add additional constructor
159 for specifying RPO order and allow NULL for that.
160 * domwalk.c (dom_walker::dom_walker): Likewise.
161 (dom_walker::walk): Handle NULL RPO order.
162 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
164 (rewrite_update_dom_walker): Likewise.
165 (mark_def_dom_walker): Likewise.
167 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
169 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
170 (aarch64_maybe_expand_sve_subreg_move): Declare.
171 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
172 * config/aarch64/predicates.md (aarch64_any_register_operand): New
174 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
175 that are semantically a reverse operation.
176 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
177 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
178 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
180 (aarch64_can_change_mode_class): For big-endian, forbid changes
181 between two SVE modes if they have different element sizes.
183 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
185 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
186 the TImode handling for big-endian targets.
188 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
190 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
191 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
193 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
194 Remove BSWAP handing for big-endian targets and use the form of
195 LD1RQ appropariate for the mode.
197 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
199 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
200 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
203 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
206 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
207 check for operands that need to go through aarch64_sve_reload_be.
209 2018-02-01 Jakub Jelinek <jakub@redhat.com>
211 PR tree-optimization/81661
212 PR tree-optimization/84117
213 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
214 * tree-eh.c: Include gimplify.h.
215 (find_trapping_overflow, replace_trapping_overflow,
216 rewrite_to_non_trapping_overflow): New functions.
217 * tree-vect-loop.c: Include tree-eh.h.
218 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
219 * tree-data-ref.c: Include tree-eh.h.
220 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
222 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
224 PR rtl-optimization/84123
225 * combine.c (change_zero_ext): Check if hard register satisfies
226 can_change_dest_mode before calling gen_lowpart_SUBREG.
228 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
231 * ira.c (ira_init_register_move_cost): Remove assert.
233 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
235 PR rtl-optimization/84071
236 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
237 * doc/tm.texi: Regenerate.
239 2018-01-31 Richard Biener <rguenther@suse.de>
241 PR tree-optimization/84132
242 * tree-data-ref.c (analyze_miv_subscript): Properly
243 check whether evolution_function_is_affine_multivariate_p
244 before calling gcd_of_steps_may_divide_p.
246 2018-01-31 Julia Koval <julia.koval@intel.com>
249 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
250 * config/i386/i386.md (rdpid_rex64) New.
251 (rdpid): Make 32bit only.
253 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
256 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
257 an IDENTIFIER_NODE for FUNCTION_TYPE's.
259 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
262 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
264 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
266 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
268 PR rtl-optimization/84071
269 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
270 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
272 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
274 * config/arc/arc.c (arc_handle_aux_attribute): New function.
275 (arc_attribute_table): Add 'aux' attribute.
276 (arc_in_small_data_p): Consider aux like variables.
277 (arc_is_aux_reg_p): New function.
278 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
279 (arc_get_aux_arg): New function.
280 (prepare_move_operands): Handle aux-register access.
281 (arc_handle_aux_attribute): New function.
282 * doc/extend.texi (ARC Variable attributes): Add subsection.
284 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
286 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
287 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
288 (arc_attribute_table): Add 'uncached' attribute.
289 (arc_print_operand): Print '.di' flag for uncached memory
291 (arc_in_small_data_p): Do not consider for small data the uncached
293 (arc_is_uncached_mem_p): New function.
294 * config/arc/predicates.md (compact_store_memory_operand): Check
295 for uncached memory accesses.
296 (nonvol_nonimm_operand): Likewise.
297 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
299 2018-01-31 Jakub Jelinek <jakub@redhat.com>
302 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
303 falign-loops=): Add Optimization flag.
305 2018-01-30 Jeff Law <law@redhat.com>
308 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
309 INT_REGISTERS_SAVED. Check it prior to calling
310 get_scratch_register_on_entry.
311 (ix86_adjust_stack_and_probe): Similarly.
312 (ix86_emit_probe_stack_range): Similarly.
313 (ix86_expand_prologue): Corresponding changes.
315 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
318 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
319 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
321 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
324 * lra-constraints.c (curr_insn_transform): Process AND in the
327 2018-01-30 Jakub Jelinek <jakub@redhat.com>
329 PR rtl-optimization/83986
330 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
331 dependence against last_pending_memory_flush in addition to
334 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
336 PR tree-optimization/81611
337 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
340 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
342 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
345 2018-01-30 Richard Biener <rguenther@suse.de>
346 Jakub Jelinek <jakub@redhat.com>
348 PR tree-optimization/84111
349 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
350 inner loops added during recursion, as they don't have up-to-date
353 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
356 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
357 (can_inline_edge_by_limits_p): ... here.
358 (can_early_inline_edge_p, check_callers,
359 update_caller_keys, update_callee_keys, recursive_inlining,
360 add_new_edges_to_heap, speculation_useful_p,
361 inline_small_functions,
362 inline_small_functions, flatten_function,
363 inline_to_all_callers_1): Update.
365 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
367 * profile-count.c (profile_count::combine_with_ipa_count): Handle
370 2018-01-30 Richard Biener <rguenther@suse.de>
372 PR tree-optimization/83008
373 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
374 invariant and constant vector uses in stmts when they need
377 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
380 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
381 * configure: Regenerate.
383 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
385 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
387 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
388 Use gen_rtx_REG rather than gen_lowpart.
390 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
392 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
393 rather than 0 when creating partial subregs.
395 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
397 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
400 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
403 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
404 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
405 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
406 flags. This restores the settings used before the 2017-07-24.
407 Turning off pre increment/decrement/modify allows IVOPTS to
408 optimize DF/SF loops where the index is an int.
410 2018-01-29 Richard Biener <rguenther@suse.de>
411 Kelvin Nilsen <kelvin@gcc.gnu.org>
414 * tree-vect-stmts.c (vectorizable_call): Don't call
415 targetm.vectorize_builtin_md_vectorized_function if callee is
418 2018-01-22 Carl Love <cel@us.ibm.com>
420 * doc/extend.tex: Fix typo in second arg in
421 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
423 2018-01-29 Richard Biener <rguenther@suse.de>
425 PR tree-optimization/84086
426 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
427 (flush_ssaname_freelist): When SSA names were released reset
430 2018-01-29 Richard Biener <rguenther@suse.de>
432 PR tree-optimization/84057
433 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
434 removed paths when removing edges.
436 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
438 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
439 -mfunction-return=@var{choice}.
441 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
444 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
446 (layout::print_source_line): Likewise.
447 (test_get_line_width_without_trailing_whitespace): Add test cases.
449 2018-01-27 Jakub Jelinek <jakub@redhat.com>
452 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
455 2018-01-26 Jim Wilson <jimw@sifive.com>
457 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
459 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
462 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
464 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
465 and CMP + SUB-immediate -> SUBS.
467 2018-01-26 Martin Sebor <msebor@redhat.com>
469 PR tree-optimization/83896
470 * tree-ssa-strlen.c (get_string_len): Rename...
471 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
472 Avoid assuming length is constant.
473 (handle_char_store): Use HOST_WIDE_INT for string length.
475 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
478 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
479 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
481 2018-01-26 Richard Biener <rguenther@suse.de>
483 PR rtl-optimization/84003
484 * dse.c (record_store): Only record redundant stores when
485 the earlier store aliases at least all accesses the later one does.
487 2018-01-26 Jakub Jelinek <jakub@redhat.com>
489 PR rtl-optimization/83985
490 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
491 REG_CFA_RESTORE insns.
492 (delete_unmarked_insns): Don't ignore separate shrink wrapping
493 REG_CFA_RESTORE insns here.
496 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
497 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
499 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
501 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
502 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
503 (arc_init): Likewise.
504 (arc_override_options): Likewise.
505 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
507 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
509 * config/arc/arc.h (TARGET_DBNZ): Define.
510 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
511 properly set the tune attribute.
512 (dbnz): Use TARGET_DBNZ guard.
513 * config/arc/arc.opt (mtune): Add core3 option.
515 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
517 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
518 recognize new pic like addresses.
519 (arc_delegitimize_address): Clean up.
521 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
523 * config/arc/arc-arches.def: Option mrf16 valid for all
525 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
526 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
527 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
528 * config/arc/arc-tables.opt: Regenerate.
529 * config/arc/arc.c (arc_conditional_register_usage): Handle
530 reduced register file case.
531 (arc_file_start): Set must have build attributes.
532 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
534 * config/arc/arc.opt (mrf16): Add new option.
535 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
536 * config/arc/genmultilib.awk: Handle new mrf16 option.
537 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
538 * config/arc/t-multilib: Regenerate.
539 * doc/invoke.texi (ARC Options): Document mrf16 option.
541 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
543 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
544 * config/arc/arc.c (arc_handle_secure_attribute): New function.
545 (arc_attribute_table): Add 'secure_call' attribute.
546 (arc_print_operand): Print secure call operand.
547 (arc_function_ok_for_sibcall): Don't optimize tail calls when
549 (arc_is_secure_call_p): New function. * config/arc/arc.md
550 (call_i): Add support for sjli instruction.
551 (call_value_i): Likewise.
552 * config/arc/constraints.md (Csc): New constraint.
554 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
555 John Eric Martin <John.Martin@emmicro-us.com>
557 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
558 * config/arc/arc.c (_arc_jli_section): New struct.
559 (arc_jli_section): New type.
560 (rc_jli_sections): New static variable.
561 (arc_handle_jli_attribute): New function.
562 (arc_attribute_table): Add jli_always and jli_fixed attribute.
563 (arc_file_end): New function.
564 (TARGET_ASM_FILE_END): Define.
565 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
566 (arc_add_jli_section): New function.
567 (jli_call_scan): Likewise.
568 (arc_reorg): Call jli_call_scan.
569 (arc_output_addsi): Remove 'S' from printing asm operand.
570 (arc_is_jli_call_p): New function.
571 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
573 (movhi_insn): Likewise.
574 (movsi_insn): Likewise.
575 (movsi_set_cc_insn): Likewise.
576 (loadqi_update): Likewise.
577 (load_zeroextendqisi_update): Likewise.
578 (load_signextendqisi_update): Likewise.
579 (loadhi_update): Likewise.
580 (load_zeroextendhisi_update): Likewise.
581 (load_signextendhisi_update): Likewise.
582 (loadsi_update): Likewise.
583 (loadsf_update): Likewise.
584 (movsicc_insn): Likewise.
585 (bset_insn): Likewise.
586 (bxor_insn): Likewise.
587 (bclr_insn): Likewise.
588 (bmsk_insn): Likewise.
589 (bicsi3_insn): Likewise.
590 (cmpsi_cc_c_insn): Likewise.
591 (movsi_ne): Likewise.
592 (movsi_cond_exec): Likewise.
593 (clrsbsi2): Likewise.
604 (call_i): Remove 'S' asm letter, add jli instruction.
605 (call_value_i): Likewise.
606 * config/arc/arc.op (mjli-always): New option.
607 * config/arc/constraints.md (Cji): New constraint.
608 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
610 (subsf3_fpx): Likewise.
611 (mulsf3_fpx): Likewise.
612 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
614 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
616 * doc/invoke.texi (ARC): Document mjli-always option.
618 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
620 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
621 and use incw and decw where possible
622 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
624 2018-01-26 Richard Biener <rguenther@suse.de>
626 PR tree-optimization/81082
627 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
628 association if it requires casting to unsigned.
629 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
630 from fold_plusminus_mult_expr to catch important cases late when
631 range info is available.
633 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
635 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
636 * configure.ac (hidden_linkonce): New test.
637 * configure: Regenerate.
638 * config.in: Regenerate.
640 2018-01-26 Julia Koval <julia.koval@intel.com>
642 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
643 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
644 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
645 _mm_mask_bitshuffle_epi64_mask): Fix type.
646 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
647 USI_FTYPE_V4DI_V4DI_USI): Remove.
648 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
649 __builtin_ia32_vpshufbitqmb256_mask,
650 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
651 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
652 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
654 2018-01-26 Alan Modra <amodra@gmail.com>
657 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
658 UNSPEC_VBPERMQ. Sort other unspecs.
660 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
662 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
664 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
667 * predict.c (drop_profile): Do not push/pop cfun; update also
669 (handle_missing_profiles): Fix logic looking for zero profiles.
671 2018-01-25 Jakub Jelinek <jakub@redhat.com>
674 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
675 on functions with #pragma omp declare simd or functions with simd
677 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
678 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
679 Remove trailing \n from warning_at calls.
681 2018-01-25 Tom de Vries <tom@codesourcery.com>
684 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
685 for neutered workers.
687 2018-01-24 Joseph Myers <joseph@codesourcery.com>
690 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
691 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
693 2018-01-24 Jeff Law <law@redhat.com>
696 * i386.c (get_probe_interval): Move to earlier point.
697 (ix86_compute_frame_layout): If -fstack-clash-protection and
698 the frame is larger than the probe interval, then use pushes
699 to save registers rather than reg->mem moves.
700 (ix86_expand_prologue): Remove conditional for int_registers_saved
703 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
706 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
707 min/max for never referenced object.
709 2018-01-24 Jakub Jelinek <jakub@redhat.com>
712 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
714 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
715 attributes from DECL_ATTRIBUTES (decl) without affecting
716 DECL_ATTRIBUTES (current_function_decl).
717 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
718 functions with non-NULL DECL_ABSTRACT_ORIGIN.
720 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
722 PR tree-optimization/83979
723 * fold-const.c (fold_comparison): Use constant_boolean_node
724 instead of boolean_{true,false}_node.
726 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
728 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
731 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
733 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
734 Simplify the clause that sets the length attribute.
735 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
736 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
737 clause that sets the length attribute.
738 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
740 2018-01-24 Tom de Vries <tom@codesourcery.com>
743 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
744 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
745 Add strict parameter.
746 (prevent_branch_around_nothing): Insert dummy insn between branch to
747 label and label with no ptx insn inbetween.
748 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
750 2018-01-24 Tom de Vries <tom@codesourcery.com>
753 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
754 for neutered threads in warp.
755 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
757 2018-01-24 Richard Biener <rguenther@suse.de>
759 PR tree-optimization/83176
760 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
763 2018-01-24 Richard Biener <rguenther@suse.de>
765 PR tree-optimization/82819
766 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
767 code generating pluses that are no-ops in the target precision.
769 2018-01-24 Richard Biener <rguenther@suse.de>
772 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
774 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
776 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
777 to merge probabilities.
778 * predict.c (probably_never_executed): Also mark as cold functions
779 with global 0 profile and guessed local profile.
780 * profile-count.c (profile_probability::combine_with_count): New
782 * profile-count.h (profile_probability::operator*,
783 profile_probability::operator*=, profile_probability::operator/,
784 profile_probability::operator/=): Reduce precision to adjusted
785 and set value to guessed on contradictory divisions.
786 (profile_probability::combine_with_freq): Remove.
787 (profile_probability::combine_wiht_count): Declare.
788 (profile_count::force_nonzero):: Set to adjusted.
789 (profile_count::probability_in):: Set quality to adjusted.
790 * tree-ssa-tail-merge.c (replace_block_by): Use
793 2018-01-23 Andrew Waterman <andrew@sifive.com>
794 Jim Wilson <jimw@sifive.com>
796 * config/riscv/riscv.c (riscv_stack_boundary): New.
797 (riscv_option_override): Set riscv_stack_boundary. Handle
798 riscv_preferred_stack_boundary_arg.
799 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
800 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
801 (STACK_BOUNDARY): Set to riscv_stack_boundary.
802 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
803 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
804 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
806 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
809 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
810 of struct ix86_frame.
811 (ix86_expand_epilogue): Likewise. Add a local variable for
812 the reg_save_offset field in struct ix86_frame.
814 2018-01-23 Bin Cheng <bin.cheng@arm.com>
816 PR tree-optimization/82604
817 * tree-loop-distribution.c (enum partition_kind): New enum item
818 PKIND_PARTIAL_MEMSET.
819 (partition_builtin_p): Support above new enum item.
820 (generate_code_for_partition): Ditto.
821 (compute_access_range): Differentiate cases that equality can be
822 proven at all loops, the innermost loops or no loops.
823 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
824 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
825 (finalize_partitions, distribute_loop): Don't fuse partition of
826 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
827 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
830 2018-01-23 Martin Liska <mliska@suse.cz>
832 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
833 order to ignore the predictor.
834 (PRED_POLYMORPHIC_CALL): Likewise.
835 (PRED_RECURSIVE_CALL): Likewise.
837 2018-01-23 Martin Liska <mliska@suse.cz>
839 * tree-profile.c (tree_profiling): Print function header to
840 aware reader which function we are working on.
841 * value-prof.c (gimple_find_values_to_profile): Do not print
842 not interesting value histograms.
844 2018-01-23 Martin Liska <mliska@suse.cz>
846 * profile-count.h (enum profile_quality): Add
847 profile_uninitialized as the first value. Do not number values
848 as they are zero based.
849 (profile_count::verify): Update sanity check.
850 (profile_probability::verify): Likewise.
852 2018-01-23 Nathan Sidwell <nathan@acm.org>
854 * doc/invoke.texi (ffor-scope): Deprecate.
856 2018-01-23 David Malcolm <dmalcolm@redhat.com>
858 PR tree-optimization/83510
859 * domwalk.c (set_all_edges_as_executable): New function.
860 (dom_walker::dom_walker): Convert bool param
861 "skip_unreachable_blocks" to enum reachability. Move setup of
862 edge flags to set_all_edges_as_executable and only do it when
863 reachability is REACHABLE_BLOCKS.
864 * domwalk.h (enum dom_walker::reachability): New enum.
865 (dom_walker::dom_walker): Convert bool param
866 "skip_unreachable_blocks" to enum reachability.
867 (set_all_edges_as_executable): New decl.
868 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
869 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
871 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
872 but converting true to REACHABLE_BLOCKS.
873 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
875 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
876 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
877 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
879 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
880 if check_all_array_refs will be called.
882 2018-01-23 David Malcolm <dmalcolm@redhat.com>
884 * tree.c (selftest::test_location_wrappers): Add more test
887 2018-01-23 David Malcolm <dmalcolm@redhat.com>
889 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
890 (selftest::test_bit_in_range): Likewise.
892 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
895 * doc/sourcebuild.texi (vect_float): Say that the selector
896 only describes the situation when -funsafe-math-optimizations is on.
897 (vect_float_strict): Document.
899 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
901 PR tree-optimization/83965
902 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
903 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
904 instead of checking only for a reduction.
905 (vect_recog_widen_sum_pattern): Likewise.
907 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
909 * predict.c (probably_never_executed): Only use precise profile info.
910 (compute_function_frequency): Skip after inlining hack since we now
911 have quality checking.
913 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
915 * profile-count.h (profile_probability::very_unlikely,
916 profile_probability::unlikely, profile_probability::even): Set
917 precision to guessed.
919 2018-01-23 Richard Biener <rguenther@suse.de>
921 PR tree-optimization/83963
922 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
923 Properly terminate dominator walk when crossing the exit edge not
924 when visiting its source block.
926 2018-01-23 Jakub Jelinek <jakub@redhat.com>
929 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
930 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
932 2018-01-22 Jakub Jelinek <jakub@redhat.com>
934 PR tree-optimization/83957
935 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
936 semicolon after for body surrounded by braces.
938 PR tree-optimization/83081
939 * profile-count.h (profile_probability::split): New method.
940 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
941 Use profile_probability::split.
942 (do_compare_rtx_and_jump): Fix adjustment of probabilities
943 when splitting a single conditional jump into 2.
945 2018-01-22 David Malcolm <dmalcolm@redhat.com>
947 PR tree-optimization/69452
948 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
951 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
953 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
954 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
955 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
957 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
959 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
960 * config/rl78/rl78.md: New define_expand "movdi"
961 * config/rl78/rl78.c: New function definition rl78_split_movdi
963 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
966 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
968 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
969 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
970 128-bit to produce an UNSPEC move to get the double word with the
971 signbit and then a shift directly to do signbit.
972 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
973 implementation with a new version that just does either a direct
974 move or a regular move. Move memory interface to separate insns.
975 Move insns so they are next to the expander.
976 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
977 with signbit move. Split big and little endian case.
978 (signbit<mode>2_dm_mem_le): Likewise.
979 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
980 (signbit<mode>2_dm2): Likewise.
982 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
984 * config/rl78/rl78.md: New define_expand "anddi3".
986 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
988 * config/rl78/rl78.md: New define_expand "umindi3".
990 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
992 * config/rl78/rl78.md: New define_expand "smindi3".
994 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
996 * config/rl78/rl78.md: New define_expand "smaxdi3".
998 2018-01-22 Carl Love <cel@us.ibm.com>
1000 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1001 LVX_V1TI): Add macro expansion.
1002 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1003 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1004 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1005 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1006 Change check to determine if the instruction is a byte reversing
1007 entry. Fix typo in comment.
1008 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1009 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1010 Add def_builtin calls for new builtins.
1011 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1012 Add define_insn expansion.
1014 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1016 * config/rl78/rl78.md: New define_expand "umaxdi3".
1018 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1020 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1021 for non-QImode registers
1023 2018-01-22 Richard Biener <rguenther@suse.de>
1025 PR tree-optimization/83963
1026 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1027 including the loop exit block.
1028 (scop_detection::merge_sese): Likewise.
1029 (scop_detection::add_scop): Do it here instead.
1031 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1033 * doc/sourcebuild.texi (arm_softfloat): Document.
1035 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1038 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1039 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1040 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1042 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1043 David Edelsohn <dje.gcc@gmail.com>
1046 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1047 Change "crset eq" to "crset 2".
1048 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1049 (*call_indirect_aix<mode>_nospec): Likewise.
1050 (*call_value_indirect_aix<mode>_nospec): Likewise.
1051 (*call_indirect_elfv2<mode>_nospec): Likewise.
1052 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1053 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1054 change assembly output from . to $.
1055 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1056 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1057 (*tablejump<mode>_internal1_nospec): Likewise.
1059 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1062 * config/sh/sh_optimize_sett_clrt.cc:
1063 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1065 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1067 PR tree-optimization/83940
1068 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1069 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1070 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1071 use it to pass back the definition type.
1072 (vect_check_store_rhs): Likewise rhs_dt_out.
1073 (vect_build_gather_load_calls): Add a mask_dt argument and use
1074 it instead of a call to vect_is_simple_use.
1075 (vectorizable_store): Update calls to vect_check_load_store_mask
1076 and vect_check_store_rhs. Use the dt returned by the latter instead
1077 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1078 instead of calls to vect_is_simple_use. Pass the scalar rather
1079 than the vector operand to vect_is_simple_use when handling
1080 second and subsequent copies of an rhs value.
1081 (vectorizable_load): Update calls to vect_check_load_store_mask
1082 and vect_build_gather_load_calls. Use the cached mask_dt and
1083 gs_info.offset_dt instead of calls to vect_is_simple_use.
1085 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1088 * tree-emutls.c: Include gimplify.h.
1089 (lower_emutls_2): New function.
1090 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1091 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1092 it before further processing.
1095 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1096 UINTVAL (trueop1) instead of INTVAL (op1).
1098 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1102 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1103 INCOMING_FRAME_SP_OFFSET if not defined.
1104 (scan_trace): Add ENTRY argument. If true and
1105 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1106 emit a note to adjust the CFA offset.
1107 (create_cfi_notes): Adjust scan_trace callers.
1108 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1109 INCOMING_FRAME_SP_OFFSET in the CIE.
1110 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1111 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1113 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1114 * doc/tm.texi: Regenerated.
1116 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1118 PR rtl-optimization/83147
1119 * lra-constraints.c (remove_inheritance_pseudos): Use
1120 lra_substitute_pseudo_within_insn.
1122 2018-01-19 Tom de Vries <tom@codesourcery.com>
1123 Cesar Philippidis <cesar@codesourcery.com>
1126 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1128 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1131 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1132 spaces for function labels.
1134 2018-01-19 Martin Liska <mliska@suse.cz>
1136 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1137 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1138 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1139 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1140 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1141 (PRED_CONST_RETURN): Change from 69 to 65.
1142 (PRED_NULL_RETURN): Change from 91 to 71.
1143 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1144 (PRED_LOOP_GUARD): Change from 66 to 73.
1146 2018-01-19 Martin Liska <mliska@suse.cz>
1148 * predict.c (predict_insn_def): Add new assert.
1149 (struct branch_predictor): Change type to signed integer.
1150 (test_prediction_value_range): Amend test to cover
1152 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1153 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1154 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1155 (PRED_LOOP_IV_COMPARE): Likewise.
1156 * predict.h (PROB_UNINITIALIZED): Define new constant.
1158 2018-01-19 Martin Liska <mliska@suse.cz>
1160 * predict.c (dump_prediction): Add new format for
1161 analyze_brprob.py script which is enabled with -details
1163 * profile-count.h (precise_p): New function.
1165 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1167 PR tree-optimization/83922
1168 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1169 there are no statements that need masking.
1170 (vect_active_double_reduction_p): New function.
1171 (vect_analyze_loop_operations): Use it when handling phis that
1172 are not in the loop header.
1174 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1176 PR tree-optimization/83914
1177 * tree-vect-loop.c (vectorizable_induction): Don't convert
1178 init_expr or apply the peeling adjustment for inductions
1179 that are nested within the vectorized loop.
1181 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1183 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1186 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1190 * function.h (gimplify_parameters): Add gimple_seq * argument.
1191 * function.c: Include gimple.h and options.h.
1192 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1193 for the added local temporaries if needed.
1194 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1195 if there are any parameter cleanups, wrap whole body into a
1196 try/finally with the cleanups.
1198 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1201 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1202 Use GET_MODE_CLASS for scalar floating point.
1204 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1208 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1209 Fix call of call_cgraph_insertion_hooks.
1211 2018-01-18 Martin Sebor <msebor@redhat.com>
1213 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1215 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1218 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1221 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1224 * common.opt: (-ffile-prefix-map): New option.
1225 * opts.c (common_handle_option): Defer it.
1226 * opts-global.c (handle_common_deferred_options): Handle it.
1227 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1228 * file-prefix-map.h: New file.
1229 (remap_debug_filename, add_debug_prefix_map): ...here.
1230 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1231 * final.c (debug_prefix_map, add_debug_prefix_map
1232 remap_debug_filename): Move to...
1233 * file-prefix-map.c: New file.
1234 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1235 generalize, get rid of alloca(), use strrchr() instead of strchr().
1236 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1237 Implement in terms of add_prefix_map().
1238 (remap_macro_filename, remap_debug_filename): Implement in term of
1240 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1241 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1242 * dbxout.c: Include file-prefix-map.h.
1243 * varasm.c: Likewise.
1244 * vmsdbgout.c: Likewise.
1245 * xcoffout.c: Likewise.
1246 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1247 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1248 * doc/invoke.texi (-ffile-prefix-map): Document.
1249 (-fdebug-prefix-map): Update description.
1251 2018-01-18 Martin Liska <mliska@suse.cz>
1253 * config/i386/i386.c (indirect_thunk_name): Document that also
1255 (output_indirect_thunk): Document why both instructions
1256 (pause and lfence) are generated.
1258 2018-01-18 Richard Biener <rguenther@suse.de>
1260 PR tree-optimization/83887
1261 * graphite-scop-detection.c
1262 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1263 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1264 (scop_detection::merge_sese): Re-implement with a flood-fill
1265 algorithm that properly finds a SESE region if it exists.
1267 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1270 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1271 pointer_diff optimizations use view_convert instead of convert.
1273 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1275 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1276 Generate different code for -mno-speculate-indirect-jumps.
1277 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1278 (*call_indirect_aix<mode>): Disable for
1279 -mno-speculate-indirect-jumps.
1280 (*call_indirect_aix<mode>_nospec): New define_insn.
1281 (*call_value_indirect_aix<mode>): Disable for
1282 -mno-speculate-indirect-jumps.
1283 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1284 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1285 -mno-speculate-indirect-jumps.
1286 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1288 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1290 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1291 long double type, set the flags for noting the default long double
1292 type, even if we don't pass or return a long double type.
1294 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1297 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1300 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1302 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1303 support for merge[hl].
1304 (fold_mergehl_helper): New helper function.
1305 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1306 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1307 (altivec_vmrglw_direct): Add xxmrglw insn.
1309 2018-01-17 Andrew Waterman <andrew@sifive.com>
1311 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1312 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1314 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1317 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1318 call the lto_location_cache before reading the
1319 DECL_SOURCE_LOCATION of the types.
1321 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1322 Richard Sandiford <richard.sandiford@linaro.org>
1324 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1325 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1326 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1327 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1328 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1330 * config/aarch64/constraints.md (aarch64_movti_operand):
1332 * config/aarch64/predicates.md (Uti): Add new constraint.
1334 2018-01-17 Carl Love <cel@us.ibm.com>
1335 * config/rs6000/vsx.md (define_expand xl_len_r,
1336 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1337 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1339 (define_expand, define_insn): Move the shift left from the
1340 define_insn to the define_expand for lxvl and stxvl instructions.
1341 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1342 and XL_LEN_R definitions to PURE.
1344 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1346 * config/i386/i386.c (indirect_thunk_name): Declare regno
1347 as unsigned int. Compare regno with INVALID_REGNUM.
1348 (output_indirect_thunk): Ditto.
1349 (output_indirect_thunk_function): Ditto.
1350 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1351 in the call to output_indirect_thunk_function.
1353 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1356 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1357 rather than the size of inner_type to determine the stack slot size
1358 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1360 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1363 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1366 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1368 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1369 endian Linux systems to optionally enable multilibs for selecting
1370 the long double type if the user configured an explicit type.
1371 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1372 have no long double multilibs if not defined.
1373 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1374 warn if the user used -mabi={ieee,ibm}longdouble and we built
1375 multilibs for long double.
1376 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1377 appropriate multilib option.
1378 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1380 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1381 for building long double multilibs.
1382 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1384 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1386 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1389 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1391 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1394 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1397 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1400 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1402 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1403 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1405 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1407 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1408 different rtl trees depending on TARGET_64BIT.
1409 (rs6000_gen_lvx): Likewise.
1411 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1413 * config/visium/visium.md (nop): Tweak comment.
1414 (hazard_nop): Likewise.
1416 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1418 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1419 -mspeculate-indirect-jumps.
1420 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1421 for -mno-speculate-indirect-jumps.
1422 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1423 (*call_value_indirect_elfv2<mode>): Disable for
1424 -mno-speculate-indirect-jumps.
1425 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1426 (indirect_jump): Emit different RTL for
1427 -mno-speculate-indirect-jumps.
1428 (*indirect_jump<mode>): Disable for
1429 -mno-speculate-indirect-jumps.
1430 (*indirect_jump<mode>_nospec): New define_insn.
1431 (tablejump): Emit different RTL for
1432 -mno-speculate-indirect-jumps.
1433 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1434 (tablejumpsi_nospec): New define_expand.
1435 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1436 (tablejumpdi_nospec): New define_expand.
1437 (*tablejump<mode>_internal1): Disable for
1438 -mno-speculate-indirect-jumps.
1439 (*tablejump<mode>_internal1_nospec): New define_insn.
1440 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1443 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1445 * caller-save.c (insert_save): Drop unnecessary parameter. All
1448 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1449 Richard Biener <rguenth@suse.de>
1452 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1453 return early, inline manually is_gimple_sizepos. Make sure if we
1454 call gimplify_expr we don't end up with a gimple constant.
1455 * tree.c (variably_modified_type_p): Don't return true for
1456 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1457 * gimplify.h (is_gimple_sizepos): Remove.
1459 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1461 PR tree-optimization/83857
1462 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1463 vectorizable_live_operation for pure SLP statements.
1464 (vectorizable_live_operation): Handle PHIs.
1466 2018-01-16 Richard Biener <rguenther@suse.de>
1468 PR tree-optimization/83867
1469 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1470 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1472 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1475 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1476 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1477 If off is not INTEGER_CST, issue a may not be aligned warning
1478 rather than isn't aligned. Use isn%'t rather than isn't.
1479 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1481 <case MULT_EXPR>: Improve the case when bottom and one of the
1482 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1483 operand, in that case check if the other operand is multiple of
1484 bottom divided by the INTEGER_CST operand.
1486 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1489 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1490 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1491 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1492 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1493 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1494 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1495 (pa_function_arg_size): New function.
1497 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1499 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1500 in a separate statement.
1502 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1504 PR tree-optimization/83847
1505 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1506 group gathers and scatters.
1508 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1510 PR rtl-optimization/86620
1511 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1513 PR rtl-optimization/83213
1514 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1515 to last if both are JUMP_INSNs.
1517 PR tree-optimization/83843
1518 * gimple-ssa-store-merging.c
1519 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1520 store_immediate_info for bswap/nop orig_stores.
1522 2018-01-15 Andrew Waterman <andrew@sifive.com>
1524 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1526 <UDIV>: Increase cost if !TARGET_DIV.
1528 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1530 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1531 (define_attr "cr_logical_3op"): New.
1532 (cceq_ior_compare): Adjust.
1533 (cceq_ior_compare_complement): Adjust.
1534 (*cceq_rev_compare): Adjust.
1535 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1536 (is_cracked_insn): Adjust.
1537 (insn_must_be_first_in_group): Adjust.
1538 * config/rs6000/40x.md: Adjust.
1539 * config/rs6000/440.md: Adjust.
1540 * config/rs6000/476.md: Adjust.
1541 * config/rs6000/601.md: Adjust.
1542 * config/rs6000/603.md: Adjust.
1543 * config/rs6000/6xx.md: Adjust.
1544 * config/rs6000/7450.md: Adjust.
1545 * config/rs6000/7xx.md: Adjust.
1546 * config/rs6000/8540.md: Adjust.
1547 * config/rs6000/cell.md: Adjust.
1548 * config/rs6000/e300c2c3.md: Adjust.
1549 * config/rs6000/e500mc.md: Adjust.
1550 * config/rs6000/e500mc64.md: Adjust.
1551 * config/rs6000/e5500.md: Adjust.
1552 * config/rs6000/e6500.md: Adjust.
1553 * config/rs6000/mpc.md: Adjust.
1554 * config/rs6000/power4.md: Adjust.
1555 * config/rs6000/power5.md: Adjust.
1556 * config/rs6000/power6.md: Adjust.
1557 * config/rs6000/power7.md: Adjust.
1558 * config/rs6000/power8.md: Adjust.
1559 * config/rs6000/power9.md: Adjust.
1560 * config/rs6000/rs64.md: Adjust.
1561 * config/rs6000/titan.md: Adjust.
1563 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1565 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1566 ix86_indirect_branch_register logic.
1568 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1570 * config/i386/constraints.md (Bs): Update
1571 ix86_indirect_branch_register check. Don't check
1572 ix86_indirect_branch_register with GOT_memory_operand.
1574 * config/i386/predicates.md (GOT_memory_operand): Don't check
1575 ix86_indirect_branch_register here.
1576 (GOT32_symbol_operand): Likewise.
1578 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1580 * config/i386/predicates.md (constant_call_address_operand):
1581 Rewrite ix86_indirect_branch_register logic.
1582 (sibcall_insn_operand): Likewise.
1584 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1586 * config/i386/constraints.md (Bs): Replace
1587 ix86_indirect_branch_thunk_register with
1588 ix86_indirect_branch_register.
1590 * config/i386/i386.md (indirect_jump): Likewise.
1591 (tablejump): Likewise.
1592 (*sibcall_memory): Likewise.
1593 (*sibcall_value_memory): Likewise.
1594 Peepholes of indirect call and jump via memory: Likewise.
1595 * config/i386/i386.opt: Likewise.
1596 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1597 (GOT_memory_operand): Likewise.
1598 (call_insn_operand): Likewise.
1599 (sibcall_insn_operand): Likewise.
1600 (GOT32_symbol_operand): Likewise.
1602 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1605 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1606 type rather than type addr's type points to.
1607 (expand_omp_atomic_mutex): Likewise.
1608 (expand_omp_atomic): Likewise.
1610 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1613 * config/i386/i386.c (output_indirect_thunk_function): Use
1614 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1615 for __x86_return_thunk.
1617 2018-01-15 Richard Biener <rguenther@suse.de>
1620 * expmed.c (extract_bit_field_1): Fix typo.
1622 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1625 * config/arm/iterators.md (VF): New mode iterator.
1626 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1627 Remove integer-related logic from pattern.
1628 (neon_vabd<mode>_3): Likewise.
1630 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1633 * common.opt (fstrict-overflow): No longer an alias.
1634 (fwrapv-pointer): New option.
1635 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1636 also for pointer types based on flag_wrapv_pointer.
1637 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1638 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1639 opts->x_flag_wrapv got set.
1640 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1641 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1642 POINTER_TYPE_OVERFLOW_UNDEFINED.
1643 * match.pd: Likewise in address comparison pattern.
1644 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1646 2018-01-15 Richard Biener <rguenther@suse.de>
1649 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1650 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1651 Reset type names to their identifier if their TYPE_DECL doesn't
1652 have linkage (and thus is used for ODR and devirt).
1653 (save_debug_info_for_decl): Remove.
1654 (save_debug_info_for_type): Likewise.
1655 (add_tree_to_fld_list): Adjust.
1656 * tree-pretty-print.c (dump_generic_node): Make dumping of
1657 type names more robust.
1659 2018-01-15 Richard Biener <rguenther@suse.de>
1661 * BASE-VER: Bump to 8.0.1.
1663 2018-01-14 Martin Sebor <msebor@redhat.com>
1666 * builtins.c (check_access): Avoid warning when the no-warning bit
1669 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1671 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1672 * ira-color (allocno_hard_regs_compare): Likewise.
1674 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1677 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1678 Use .pushsection/.popsection.
1680 2018-01-14 Martin Sebor <msebor@redhat.com>
1683 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1685 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1687 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1688 entry from extra_headers.
1689 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1690 extra_headers, make the list bitwise identical to the i?86-*-* one.
1692 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1694 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1695 -mcmodel=large with -mindirect-branch=thunk,
1696 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1697 -mfunction-return=thunk-extern.
1698 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1699 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1700 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1702 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1704 * config/i386/i386.c (print_reg): Print the name of the full
1705 integer register without '%'.
1706 (ix86_print_operand): Handle 'V'.
1707 * doc/extend.texi: Document 'V' modifier.
1709 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1711 * config/i386/constraints.md (Bs): Disallow memory operand for
1712 -mindirect-branch-register.
1714 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1715 (GOT_memory_operand): Likewise.
1716 (call_insn_operand): Likewise.
1717 (sibcall_insn_operand): Likewise.
1718 (GOT32_symbol_operand): Likewise.
1719 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1720 for -mindirect-branch-register.
1721 (tablejump): Likewise.
1722 (*sibcall_memory): Likewise.
1723 (*sibcall_value_memory): Likewise.
1724 Disallow peepholes of indirect call and jump via memory for
1725 -mindirect-branch-register.
1726 (*call_pop): Replace m with Bw.
1727 (*call_value_pop): Likewise.
1728 (*sibcall_pop_memory): Replace m with Bs.
1729 * config/i386/i386.opt (mindirect-branch-register): New option.
1730 * doc/invoke.texi: Document -mindirect-branch-register option.
1732 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1734 * config/i386/i386-protos.h (ix86_output_function_return): New.
1735 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1736 set function_return_type.
1737 (indirect_thunk_name): Add ret_p to indicate thunk for function
1739 (output_indirect_thunk_function): Pass false to
1740 indirect_thunk_name.
1741 (ix86_output_indirect_branch_via_reg): Likewise.
1742 (ix86_output_indirect_branch_via_push): Likewise.
1743 (output_indirect_thunk_function): Create alias for function
1744 return thunk if regno < 0.
1745 (ix86_output_function_return): New function.
1746 (ix86_handle_fndecl_attribute): Handle function_return.
1747 (ix86_attribute_table): Add function_return.
1748 * config/i386/i386.h (machine_function): Add
1749 function_return_type.
1750 * config/i386/i386.md (simple_return_internal): Use
1751 ix86_output_function_return.
1752 (simple_return_internal_long): Likewise.
1753 * config/i386/i386.opt (mfunction-return=): New option.
1754 (indirect_branch): Mention -mfunction-return=.
1755 * doc/extend.texi: Document function_return function attribute.
1756 * doc/invoke.texi: Document -mfunction-return= option.
1758 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1760 * config/i386/i386-opts.h (indirect_branch): New.
1761 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1762 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1763 with local indirect jump when converting indirect call and jump.
1764 (ix86_set_indirect_branch_type): New.
1765 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1766 (indirectlabelno): New.
1767 (indirect_thunk_needed): Likewise.
1768 (indirect_thunk_bnd_needed): Likewise.
1769 (indirect_thunks_used): Likewise.
1770 (indirect_thunks_bnd_used): Likewise.
1771 (INDIRECT_LABEL): Likewise.
1772 (indirect_thunk_name): Likewise.
1773 (output_indirect_thunk): Likewise.
1774 (output_indirect_thunk_function): Likewise.
1775 (ix86_output_indirect_branch_via_reg): Likewise.
1776 (ix86_output_indirect_branch_via_push): Likewise.
1777 (ix86_output_indirect_branch): Likewise.
1778 (ix86_output_indirect_jmp): Likewise.
1779 (ix86_code_end): Call output_indirect_thunk_function if needed.
1780 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1782 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1783 (ix86_attribute_table): Add indirect_branch.
1784 * config/i386/i386.h (machine_function): Add indirect_branch_type
1785 and has_local_indirect_jump.
1786 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1788 (tablejump): Likewise.
1789 (*indirect_jump): Use ix86_output_indirect_jmp.
1790 (*tablejump_1): Likewise.
1791 (simple_return_indirect_internal): Likewise.
1792 * config/i386/i386.opt (mindirect-branch=): New option.
1793 (indirect_branch): New.
1796 (thunk-inline): Likewise.
1797 (thunk-extern): Likewise.
1798 * doc/extend.texi: Document indirect_branch function attribute.
1799 * doc/invoke.texi: Document -mindirect-branch= option.
1801 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1804 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1806 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1808 * ipa-inline.c (want_inline_small_function_p): Return false if
1809 inlining has already failed with CIF_FINAL_ERROR.
1810 (update_caller_keys): Call want_inline_small_function_p before
1812 (update_callee_keys): Likewise.
1814 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1816 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1818 (rs6000_quadword_masked_address_p): Likewise.
1819 (quad_aligned_load_p): Likewise.
1820 (quad_aligned_store_p): Likewise.
1821 (const_load_sequence_p): Add comment to describe the outer-most loop.
1822 (mimic_memory_attributes_and_flags): New function.
1823 (rs6000_gen_stvx): Likewise.
1824 (replace_swapped_aligned_store): Likewise.
1825 (rs6000_gen_lvx): Likewise.
1826 (replace_swapped_aligned_load): Likewise.
1827 (replace_swapped_load_constant): Capitalize argument name in
1828 comment describing this function.
1829 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1830 and stores that access quad-word aligned addresses and replace
1831 with stvx or lvx instructions when appropriate.
1832 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1833 New function prototype.
1834 (rs6000_quadword_masked_address_p): Likewise.
1835 (rs6000_gen_lvx): Likewise.
1836 (rs6000_gen_stvx): Likewise.
1837 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1838 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1839 when memory address is aligned.
1840 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1841 this split to select lvx instruction when memory address is aligned.
1842 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1843 instruction when memory address is aligned.
1844 (*vsx_le_perm_load_v16qi): Likewise.
1845 (four unnamed splitters): Modify to select the stvx instruction
1846 when memory is aligned.
1848 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1850 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1851 which appears in the queue multiple times.
1853 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1854 Alan Hayward <alan.hayward@arm.com>
1855 David Sherwood <david.sherwood@arm.com>
1857 * tree-vectorizer.h (vec_lower_bound): New structure.
1858 (_loop_vec_info): Add check_nonzero and lower_bounds.
1859 (LOOP_VINFO_CHECK_NONZERO): New macro.
1860 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1861 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1862 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1863 fields. Make seg_len the distance travelled, not including the
1865 (dr_direction_indicator): Declare.
1866 (dr_zero_step_indicator): Likewise.
1867 (dr_known_forward_stride_p): Likewise.
1868 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1870 (runtime_alias_check_p): Allow runtime alias checks with
1872 (operator ==): Compare access_size and align.
1873 (prune_runtime_alias_test_list): Rework for new distinction between
1874 the access_size and seg_len.
1875 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1877 (get_segment_min_max): New function.
1878 (create_intersect_range_checks): Use it.
1879 (dr_step_indicator): New function.
1880 (dr_direction_indicator): Likewise.
1881 (dr_zero_step_indicator): Likewise.
1882 (dr_known_forward_stride_p): Likewise.
1883 * tree-loop-distribution.c (data_ref_segment_size): Return
1884 DR_STEP * (niters - 1).
1885 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1887 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1888 (vect_preserves_scalar_order_p): New function, split out from...
1889 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1890 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1891 (vect_vfa_access_size): New function.
1892 (vect_vfa_align): Likewise.
1893 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1894 (dump_lower_bound): New function.
1895 (vect_check_lower_bound): Likewise.
1896 (vect_small_gap_p): Likewise.
1897 (vectorizable_with_step_bound_p): Likewise.
1898 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1899 depencies if the vectorization factor is 1. Convert the checks
1900 for nonzero steps into checks on the bounds of DR_STEP. Try using
1901 a bunds check for variable steps if the minimum required step is
1902 relatively small. Update calls to the dr_with_seg_len
1903 constructor and to vect_compile_time_alias.
1904 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1906 (vect_loop_versioning): Call it.
1907 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1909 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1911 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1912 Alan Hayward <alan.hayward@arm.com>
1913 David Sherwood <david.sherwood@arm.com>
1915 * doc/sourcebuild.texi (vect_scatter_store): Document.
1916 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1918 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1920 * genopinit.c (main): Add supports_vec_scatter_store and
1921 supports_vec_scatter_store_cached to target_optabs.
1922 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1923 IFN_MASK_SCATTER_STORE.
1924 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1926 * internal-fn.h (internal_store_fn_p): Declare.
1927 (internal_fn_stored_value_index): Likewise.
1928 * internal-fn.c (scatter_store_direct): New macro.
1929 (expand_scatter_store_optab_fn): New function.
1930 (direct_scatter_store_optab_supported_p): New macro.
1931 (internal_store_fn_p): New function.
1932 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1933 IFN_MASK_SCATTER_STORE.
1934 (internal_fn_mask_index): Likewise.
1935 (internal_fn_stored_value_index): New function.
1936 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1938 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1939 * optabs-query.c (supports_vec_scatter_store_p): New function.
1940 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1941 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1942 true for scatter stores.
1943 (vect_gather_scatter_fn_p): Handle scatter stores too.
1944 (vect_check_gather_scatter): Consider using scatter stores if
1945 supports_vec_scatter_store_p.
1946 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1948 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1949 internal_fn_stored_value_index.
1950 (check_load_store_masking): Handle scatter stores too.
1951 (vect_get_store_rhs): Make public.
1952 (vectorizable_call): Use internal_store_fn_p.
1953 (vectorizable_store): Handle scatter store internal functions.
1954 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1955 when deciding whether the end of the group has been reached.
1956 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1957 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1958 (mask_scatter_store<mode>): New insns.
1960 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1961 Alan Hayward <alan.hayward@arm.com>
1962 David Sherwood <david.sherwood@arm.com>
1964 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1965 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1966 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1968 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1969 Use vect_truncate_gather_scatter_offset if we can't treat the
1970 operation as a normal gather load or scatter store.
1971 (get_group_load_store_type): Take the gather_scatter_info
1972 as argument. Try using a gather load or scatter store for
1973 single-element groups.
1974 (get_load_store_type): Update calls to get_group_load_store_type
1975 and vect_use_strided_gather_scatters_p.
1977 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1978 Alan Hayward <alan.hayward@arm.com>
1979 David Sherwood <david.sherwood@arm.com>
1981 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1982 optional tree argument.
1983 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1985 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1986 but continue to use the current value as a fallback.
1987 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1988 to compare the updates.
1989 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1990 (get_load_store_type): Use it when handling a strided access.
1991 (vect_get_strided_load_store_ops): New function.
1992 (vect_get_data_ptr_increment): Likewise.
1993 (vectorizable_load): Handle strided gather loads. Always pass
1994 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1996 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1997 Alan Hayward <alan.hayward@arm.com>
1998 David Sherwood <david.sherwood@arm.com>
2000 * doc/md.texi (gather_load@var{m}): Document.
2001 (mask_gather_load@var{m}): Likewise.
2002 * genopinit.c (main): Add supports_vec_gather_load and
2003 supports_vec_gather_load_cached to target_optabs.
2004 * optabs-tree.c (init_tree_optimization_optabs): Use
2005 ggc_cleared_alloc to allocate target_optabs.
2006 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2007 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2009 * internal-fn.h (internal_load_fn_p): Declare.
2010 (internal_gather_scatter_fn_p): Likewise.
2011 (internal_fn_mask_index): Likewise.
2012 (internal_gather_scatter_fn_supported_p): Likewise.
2013 * internal-fn.c (gather_load_direct): New macro.
2014 (expand_gather_load_optab_fn): New function.
2015 (direct_gather_load_optab_supported_p): New macro.
2016 (direct_internal_fn_optab): New function.
2017 (internal_load_fn_p): Likewise.
2018 (internal_gather_scatter_fn_p): Likewise.
2019 (internal_fn_mask_index): Likewise.
2020 (internal_gather_scatter_fn_supported_p): Likewise.
2021 * optabs-query.c (supports_at_least_one_mode_p): New function.
2022 (supports_vec_gather_load_p): Likewise.
2023 * optabs-query.h (supports_vec_gather_load_p): Declare.
2024 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2025 and memory_type field.
2026 (NUM_PATTERNS): Bump to 15.
2027 * tree-vect-data-refs.c: Include internal-fn.h.
2028 (vect_gather_scatter_fn_p): New function.
2029 (vect_describe_gather_scatter_call): Likewise.
2030 (vect_check_gather_scatter): Try using internal functions for
2031 gather loads. Recognize existing calls to a gather load function.
2032 (vect_analyze_data_refs): Consider using gather loads if
2033 supports_vec_gather_load_p.
2034 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2035 (vect_get_gather_scatter_offset_type): Likewise.
2036 (vect_convert_mask_for_vectype): Likewise.
2037 (vect_add_conversion_to_patterm): Likewise.
2038 (vect_try_gather_scatter_pattern): Likewise.
2039 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2040 (vect_vect_recog_func_ptrs): Add it.
2041 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2042 internal_fn_mask_index and internal_gather_scatter_fn_p.
2043 (check_load_store_masking): Take the gather_scatter_info as an
2044 argument and handle gather loads.
2045 (vect_get_gather_scatter_ops): New function.
2046 (vectorizable_call): Check internal_load_fn_p.
2047 (vectorizable_load): Likewise. Handle gather load internal
2049 (vectorizable_store): Update call to check_load_store_masking.
2050 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2051 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2052 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2053 (aarch64_gather_scale_operand_d): New predicates.
2054 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2055 (mask_gather_load<mode>): New insns.
2057 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2058 Alan Hayward <alan.hayward@arm.com>
2059 David Sherwood <david.sherwood@arm.com>
2061 * optabs.def (fold_left_plus_optab): New optab.
2062 * doc/md.texi (fold_left_plus_@var{m}): Document.
2063 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2064 * internal-fn.c (fold_left_direct): Define.
2065 (expand_fold_left_optab_fn): Likewise.
2066 (direct_fold_left_optab_supported_p): Likewise.
2067 * fold-const-call.c (fold_const_fold_left): New function.
2068 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2069 * tree-parloops.c (valid_reduction_p): New function.
2070 (gather_scalar_reductions): Use it.
2071 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2072 (vect_finish_replace_stmt): Declare.
2073 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2074 (needs_fold_left_reduction_p): New function, split out from...
2075 (vect_is_simple_reduction): ...here. Accept reductions that
2076 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2077 (vect_force_simple_reduction): Also store the reduction type in
2078 the assignment's STMT_VINFO_REDUC_TYPE.
2079 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2080 (merge_with_identity): New function.
2081 (vect_expand_fold_left): Likewise.
2082 (vectorize_fold_left_reduction): Likewise.
2083 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2084 scalar phi in place for it. Check for target support and reject
2085 cases that would reassociate the operation. Defer the transform
2086 phase to vectorize_fold_left_reduction.
2087 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2088 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2089 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2091 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2093 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2094 call to ifc_temp_var.
2096 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2097 Alan Hayward <alan.hayward@arm.com>
2098 David Sherwood <david.sherwood@arm.com>
2100 * target.def (legitimize_address_displacement): Take the original
2101 offset as a poly_int.
2102 * targhooks.h (default_legitimize_address_displacement): Update
2104 * targhooks.c (default_legitimize_address_displacement): Likewise.
2105 * doc/tm.texi: Regenerate.
2106 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2107 as an argument, moving assert of ad->disp == ad->disp_term to...
2108 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2109 Try calling targetm.legitimize_address_displacement before expanding
2110 the address rather than afterwards, and adjust for the new interface.
2111 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2112 Match the new hook interface. Handle SVE addresses.
2113 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2116 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2118 * Makefile.in (OBJS): Add early-remat.o.
2119 * target.def (select_early_remat_modes): New hook.
2120 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2121 * doc/tm.texi: Regenerate.
2122 * targhooks.h (default_select_early_remat_modes): Declare.
2123 * targhooks.c (default_select_early_remat_modes): New function.
2124 * timevar.def (TV_EARLY_REMAT): New timevar.
2125 * passes.def (pass_early_remat): New pass.
2126 * tree-pass.h (make_pass_early_remat): Declare.
2127 * early-remat.c: New file.
2128 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2130 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2132 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2133 Alan Hayward <alan.hayward@arm.com>
2134 David Sherwood <david.sherwood@arm.com>
2136 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2137 vfm1 with a bound_epilog parameter.
2138 (vect_do_peeling): Update calls accordingly, and move the prologue
2139 call earlier in the function. Treat the base bound_epilog as 0 for
2140 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2141 this base when peeling for gaps.
2142 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2143 with fully-masked loops.
2144 (vect_estimate_min_profitable_iters): Handle the single peeled
2145 iteration in that case.
2147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2148 Alan Hayward <alan.hayward@arm.com>
2149 David Sherwood <david.sherwood@arm.com>
2151 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2152 single-element interleaving even if the size is not a power of 2.
2153 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2154 accesses for single-element interleaving if the group size is
2157 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2158 Alan Hayward <alan.hayward@arm.com>
2159 David Sherwood <david.sherwood@arm.com>
2161 * doc/md.texi (fold_extract_last_@var{m}): Document.
2162 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2163 * optabs.def (fold_extract_last_optab): New optab.
2164 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2165 * internal-fn.c (fold_extract_direct): New macro.
2166 (expand_fold_extract_optab_fn): Likewise.
2167 (direct_fold_extract_optab_supported_p): Likewise.
2168 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2169 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2170 EXTRACT_LAST_REDUCTION.
2171 (get_initial_def_for_reduction): Do not create an initial vector
2172 for EXTRACT_LAST_REDUCTION reductions.
2173 (vectorizable_reduction): Leave the scalar phi in place for
2174 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2175 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2176 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2177 transform phase to vectorizable_condition.
2178 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2180 (vect_finish_stmt_generation): ...here.
2181 (vect_finish_replace_stmt): New function.
2182 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2183 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2185 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2187 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2188 Alan Hayward <alan.hayward@arm.com>
2189 David Sherwood <david.sherwood@arm.com>
2191 * doc/md.texi (extract_last_@var{m}): Document.
2192 * optabs.def (extract_last_optab): New optab.
2193 * internal-fn.def (EXTRACT_LAST): New internal function.
2194 * internal-fn.c (cond_unary_direct): New macro.
2195 (expand_cond_unary_optab_fn): Likewise.
2196 (direct_cond_unary_optab_supported_p): Likewise.
2197 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2198 loops using EXTRACT_LAST.
2199 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2200 (extract_last_<mode>): ...this optab.
2201 (vec_extract<mode><Vel>): Update accordingly.
2203 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2204 Alan Hayward <alan.hayward@arm.com>
2205 David Sherwood <david.sherwood@arm.com>
2207 * target.def (empty_mask_is_expensive): New hook.
2208 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2209 * doc/tm.texi: Regenerate.
2210 * targhooks.h (default_empty_mask_is_expensive): Declare.
2211 * targhooks.c (default_empty_mask_is_expensive): New function.
2212 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2213 if the target says that empty masks are expensive.
2214 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2216 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2219 Alan Hayward <alan.hayward@arm.com>
2220 David Sherwood <david.sherwood@arm.com>
2222 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2223 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2224 (vect_use_loop_mask_for_alignment_p): New function.
2225 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2226 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2227 niters_skip argument. Make sure that the first niters_skip elements
2228 of the first iteration are inactive.
2229 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2230 Update call to vect_set_loop_masks_directly.
2231 (get_misalign_in_elems): New function, split out from...
2232 (vect_gen_prolog_loop_niters): ...here.
2233 (vect_update_init_of_dr): Take a code argument that specifies whether
2234 the adjustment should be added or subtracted.
2235 (vect_update_init_of_drs): Likewise.
2236 (vect_prepare_for_masked_peels): New function.
2237 (vect_do_peeling): Skip prologue peeling if we're using a mask
2238 instead. Update call to vect_update_inits_of_drs.
2239 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2241 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2242 alignment. Do not include the number of peeled iterations in
2243 the minimum threshold in that case.
2244 (vectorizable_induction): Adjust the start value down by
2245 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2246 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2247 Take the number of skipped iterations into account when calculating
2249 * tree-vect-stmts.c (vect_gen_while_not): New function.
2251 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2252 Alan Hayward <alan.hayward@arm.com>
2253 David Sherwood <david.sherwood@arm.com>
2255 * doc/sourcebuild.texi (vect_fully_masked): Document.
2256 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2258 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2260 (vect_analyze_loop_2): ...here. Don't check the vectorization
2261 factor against the number of loop iterations if the loop is
2264 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2265 Alan Hayward <alan.hayward@arm.com>
2266 David Sherwood <david.sherwood@arm.com>
2268 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2269 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2270 (dump_groups): Update accordingly.
2271 (iv_use::mem_type): New member variable.
2272 (address_p): New function.
2273 (record_use): Add a mem_type argument and initialize the new
2275 (record_group_use): Add a mem_type argument. Use address_p.
2276 Remove obsolete null checks of base_object. Update call to record_use.
2277 (find_interesting_uses_op): Update call to record_group_use.
2278 (find_interesting_uses_cond): Likewise.
2279 (find_interesting_uses_address): Likewise.
2280 (get_mem_type_for_internal_fn): New function.
2281 (find_address_like_use): Likewise.
2282 (find_interesting_uses_stmt): Try find_address_like_use before
2283 calling find_interesting_uses_op.
2284 (addr_offset_valid_p): Use the iv mem_type field as the type
2285 of the addressed memory.
2286 (add_autoinc_candidates): Likewise.
2287 (get_address_cost): Likewise.
2288 (split_small_address_groups_p): Use address_p.
2289 (split_address_groups): Likewise.
2290 (add_iv_candidate_for_use): Likewise.
2291 (autoinc_possible_for_pair): Likewise.
2292 (rewrite_groups): Likewise.
2293 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2294 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2295 (get_alias_ptr_type_for_ptr_address): New function.
2296 (rewrite_use_address): Rewrite address uses in calls that were
2297 identified by find_address_like_use.
2299 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2300 Alan Hayward <alan.hayward@arm.com>
2301 David Sherwood <david.sherwood@arm.com>
2303 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2305 * gimple-expr.h (is_gimple_addressable: Likewise.
2306 * gimple-expr.c (is_gimple_address): Likewise.
2307 * internal-fn.c (expand_call_mem_ref): New function.
2308 (expand_mask_load_optab_fn): Use it.
2309 (expand_mask_store_optab_fn): Likewise.
2311 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2312 Alan Hayward <alan.hayward@arm.com>
2313 David Sherwood <david.sherwood@arm.com>
2315 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2316 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2317 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2318 (cond_umax@var{mode}): Document.
2319 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2320 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2321 (cond_umin_optab, cond_umax_optab): New optabs.
2322 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2323 (COND_IOR, COND_XOR): New internal functions.
2324 * internal-fn.h (get_conditional_internal_fn): Declare.
2325 * internal-fn.c (cond_binary_direct): New macro.
2326 (expand_cond_binary_optab_fn): Likewise.
2327 (direct_cond_binary_optab_supported_p): Likewise.
2328 (get_conditional_internal_fn): New function.
2329 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2330 Cope with reduction statements that are vectorized as calls rather
2332 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2333 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2334 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2335 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2336 (UNSPEC_COND_EOR): New unspecs.
2337 (optab): Add mappings for them.
2338 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2339 (sve_int_op, sve_fp_op): New int attributes.
2341 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2342 Alan Hayward <alan.hayward@arm.com>
2343 David Sherwood <david.sherwood@arm.com>
2345 * optabs.def (while_ult_optab): New optab.
2346 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2347 * internal-fn.def (WHILE_ULT): New internal function.
2348 * internal-fn.h (direct_internal_fn_supported_p): New override
2349 that takes two types as argument.
2350 * internal-fn.c (while_direct): New macro.
2351 (expand_while_optab_fn): New function.
2352 (convert_optab_supported_p): Likewise.
2353 (direct_while_optab_supported_p): New macro.
2354 * wide-int.h (wi::udiv_ceil): New function.
2355 * tree-vectorizer.h (rgroup_masks): New structure.
2356 (vec_loop_masks): New typedef.
2357 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2359 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2360 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2361 (vect_max_vf): New function.
2362 (slpeel_make_loop_iterate_ntimes): Delete.
2363 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2364 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2365 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2366 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2367 internal-fn.h, stor-layout.h and optabs-query.h.
2368 (vect_set_loop_mask): New function.
2369 (add_preheader_seq): Likewise.
2370 (add_header_seq): Likewise.
2371 (interleave_supported_p): Likewise.
2372 (vect_maybe_permute_loop_masks): Likewise.
2373 (vect_set_loop_masks_directly): Likewise.
2374 (vect_set_loop_condition_masked): Likewise.
2375 (vect_set_loop_condition_unmasked): New function, split out from
2376 slpeel_make_loop_iterate_ntimes.
2377 (slpeel_make_loop_iterate_ntimes): Rename to..
2378 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2379 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2380 (vect_do_peeling): Update call accordingly.
2381 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2383 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2384 mask_compare_type, can_fully_mask_p and fully_masked_p.
2385 (release_vec_loop_masks): New function.
2386 (_loop_vec_info): Use it to free the loop masks.
2387 (can_produce_all_loop_masks_p): New function.
2388 (vect_get_max_nscalars_per_iter): Likewise.
2389 (vect_verify_full_masking): Likewise.
2390 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2391 retries, and free the mask rgroups before retrying. Check loop-wide
2392 reasons for disallowing fully-masked loops. Make the final decision
2393 about whether use a fully-masked loop or not.
2394 (vect_estimate_min_profitable_iters): Do not assume that peeling
2395 for the number of iterations will be needed for fully-masked loops.
2396 (vectorizable_reduction): Disable fully-masked loops.
2397 (vectorizable_live_operation): Likewise.
2398 (vect_halve_mask_nunits): New function.
2399 (vect_double_mask_nunits): Likewise.
2400 (vect_record_loop_mask): Likewise.
2401 (vect_get_loop_mask): Likewise.
2402 (vect_transform_loop): Handle the case in which the final loop
2403 iteration might handle a partial vector. Call vect_set_loop_condition
2404 instead of slpeel_make_loop_iterate_ntimes.
2405 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2406 (check_load_store_masking): New function.
2407 (prepare_load_store_mask): Likewise.
2408 (vectorizable_store): Handle fully-masked loops.
2409 (vectorizable_load): Likewise.
2410 (supportable_widening_operation): Use vect_halve_mask_nunits for
2412 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2413 (vect_gen_while): New function.
2414 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2415 (aarch64_uqdec<mode>): New insn.
2417 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2418 Alan Hayward <alan.hayward@arm.com>
2419 David Sherwood <david.sherwood@arm.com>
2421 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2422 (reduc_xor_scal_optab): New optabs.
2423 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2424 (reduc_xor_scal_@var{m}): Document.
2425 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2426 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2428 * fold-const-call.c (fold_const_call): Handle them.
2429 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2430 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2431 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2432 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2433 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2434 (UNSPEC_XORV): New unspecs.
2435 (optab): Add entries for them.
2436 (BITWISEV): New int iterator.
2437 (bit_reduc_op): New int attributes.
2439 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2440 Alan Hayward <alan.hayward@arm.com>
2441 David Sherwood <david.sherwood@arm.com>
2443 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2444 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2445 * optabs.def (vec_shl_insert_optab): New optab.
2446 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2447 (duplicate_and_interleave): Likewise.
2448 * tree-vect-loop.c: Include internal-fn.h.
2449 (neutral_op_for_slp_reduction): New function, split out from
2450 get_initial_defs_for_reduction.
2451 (get_initial_def_for_reduction): Handle option 2 for variable-length
2452 vectors by loading the neutral value into a vector and then shifting
2453 the initial value into element 0.
2454 (get_initial_defs_for_reduction): Replace the code argument with
2455 the neutral value calculated by neutral_op_for_slp_reduction.
2456 Use gimple_build_vector for constant-length vectors.
2457 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2458 but the first group_size elements have a neutral value.
2459 Use duplicate_and_interleave otherwise.
2460 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2461 Update call to get_initial_defs_for_reduction. Handle SLP
2462 reductions for variable-length vectors by creating one vector
2463 result for each scalar result, with the elements associated
2464 with other scalar results stubbed out with the neutral value.
2465 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2466 Require IFN_VEC_SHL_INSERT for double reductions on
2467 variable-length vectors, or SLP reductions that have
2468 a neutral value. Require can_duplicate_and_interleave_p
2469 support for variable-length unchained SLP reductions if there
2470 is no neutral value, such as for MIN/MAX reductions. Also require
2471 the number of vector elements to be a multiple of the number of
2472 SLP statements when doing variable-length unchained SLP reductions.
2473 Update call to vect_create_epilog_for_reduction.
2474 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2475 and remove initial values.
2476 (duplicate_and_interleave): Make public.
2477 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2478 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2480 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2481 Alan Hayward <alan.hayward@arm.com>
2482 David Sherwood <david.sherwood@arm.com>
2484 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2485 (can_duplicate_and_interleave_p): New function.
2486 (vect_get_and_check_slp_defs): Take the vector of statements
2487 rather than just the current one. Remove excess parentheses.
2488 Restriction rejectinon of vect_constant_def and vect_external_def
2489 for variable-length vectors to boolean types, or types for which
2490 can_duplicate_and_interleave_p is false.
2491 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2492 (duplicate_and_interleave): New function.
2493 (vect_get_constant_vectors): Use gimple_build_vector for
2494 constant-length vectors and suitable variable-length constant
2495 vectors. Use duplicate_and_interleave for other variable-length
2496 vectors. Don't defer the update when inserting new statements.
2498 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2499 Alan Hayward <alan.hayward@arm.com>
2500 David Sherwood <david.sherwood@arm.com>
2502 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2503 min_profitable_iters doesn't go negative.
2505 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2506 Alan Hayward <alan.hayward@arm.com>
2507 David Sherwood <david.sherwood@arm.com>
2509 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2510 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2511 * optabs.def (vec_mask_load_lanes_optab): New optab.
2512 (vec_mask_store_lanes_optab): Likewise.
2513 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2514 (MASK_STORE_LANES): Likewise.
2515 * internal-fn.c (mask_load_lanes_direct): New macro.
2516 (mask_store_lanes_direct): Likewise.
2517 (expand_mask_load_optab_fn): Handle masked operations.
2518 (expand_mask_load_lanes_optab_fn): New macro.
2519 (expand_mask_store_optab_fn): Handle masked operations.
2520 (expand_mask_store_lanes_optab_fn): New macro.
2521 (direct_mask_load_lanes_optab_supported_p): Likewise.
2522 (direct_mask_store_lanes_optab_supported_p): Likewise.
2523 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2525 (vect_load_lanes_supported): Likewise.
2526 * tree-vect-data-refs.c (strip_conversion): New function.
2527 (can_group_stmts_p): Likewise.
2528 (vect_analyze_data_ref_accesses): Use it instead of checking
2529 for a pair of assignments.
2530 (vect_store_lanes_supported): Take a masked_p parameter.
2531 (vect_load_lanes_supported): Likewise.
2532 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2533 vect_store_lanes_supported and vect_load_lanes_supported.
2534 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2535 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2536 parameter. Don't allow gaps for masked accesses.
2537 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2538 and vect_load_lanes_supported.
2539 (get_load_store_type): Take a masked_p parameter and update
2540 call to get_group_load_store_type.
2541 (vectorizable_store): Update call to get_load_store_type.
2542 Handle IFN_MASK_STORE_LANES.
2543 (vectorizable_load): Update call to get_load_store_type.
2544 Handle IFN_MASK_LOAD_LANES.
2546 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2547 Alan Hayward <alan.hayward@arm.com>
2548 David Sherwood <david.sherwood@arm.com>
2550 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2552 * config/aarch64/aarch64-protos.h
2553 (aarch64_sve_struct_memory_operand_p): Declare.
2554 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2555 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2556 (VPRED, vpred): Handle SVE structure modes.
2557 * config/aarch64/constraints.md (Utx): New constraint.
2558 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2559 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2560 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2561 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2562 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2563 structure modes. Split into pieces after RA.
2564 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2565 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2567 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2568 SVE structure modes.
2569 (aarch64_classify_address): Likewise.
2570 (sizetochar): Move earlier in file.
2571 (aarch64_print_operand): Handle SVE register lists.
2572 (aarch64_array_mode): New function.
2573 (aarch64_sve_struct_memory_operand_p): Likewise.
2574 (TARGET_ARRAY_MODE): Redefine.
2576 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2577 Alan Hayward <alan.hayward@arm.com>
2578 David Sherwood <david.sherwood@arm.com>
2580 * target.def (array_mode): New target hook.
2581 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2582 * doc/tm.texi: Regenerate.
2583 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2584 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2585 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2587 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2590 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2591 Alan Hayward <alan.hayward@arm.com>
2592 David Sherwood <david.sherwood@arm.com>
2594 * fold-const.c (fold_binary_loc): Check the argument types
2595 rather than the result type when testing for a vector operation.
2597 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2599 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2600 * doc/tm.texi: Regenerate.
2602 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2603 Alan Hayward <alan.hayward@arm.com>
2604 David Sherwood <david.sherwood@arm.com>
2606 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2607 (sve): Document new AArch64 extension.
2608 * doc/md.texi (w): Extend the description of the AArch64
2609 constraint to include SVE vectors.
2610 (Upl, Upa): Document new AArch64 predicate constraints.
2611 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2613 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2614 (msve-vector-bits=): New option.
2615 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2616 SVE when these are disabled.
2617 (sve): New extension.
2618 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2619 modes. Adjust their number of units based on aarch64_sve_vg.
2620 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2621 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2622 aarch64_addr_query_type.
2623 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2624 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2625 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2626 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2627 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2628 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2629 (aarch64_simd_imm_zero_p): Delete.
2630 (aarch64_check_zero_based_sve_index_immediate): Declare.
2631 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2632 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2633 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2634 (aarch64_sve_float_mul_immediate_p): Likewise.
2635 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2637 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2638 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2639 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2640 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2641 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2642 (aarch64_regmode_natural_size): Likewise.
2643 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2644 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2646 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2647 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2648 for VG and the SVE predicate registers.
2649 (V_ALIASES): Add a "z"-prefixed alias.
2650 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2651 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2652 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2653 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2654 (REG_CLASS_NAMES): Add entries for them.
2655 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2656 and the predicate registers.
2657 (aarch64_sve_vg): Declare.
2658 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2659 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2660 (REGMODE_NATURAL_SIZE): Define.
2661 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2663 * config/aarch64/aarch64.c: Include cfgrtl.h.
2664 (simd_immediate_info): Add a constructor for series vectors,
2665 and an associated step field.
2666 (aarch64_sve_vg): New variable.
2667 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2668 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2669 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2670 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2671 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2672 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2673 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2674 (aarch64_get_mask_mode): New functions.
2675 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2676 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2677 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2678 predicate modes and predicate registers. Explicitly restrict
2679 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2680 to store a vector mode if it is recognized by
2681 aarch64_classify_vector_mode.
2682 (aarch64_regmode_natural_size): New function.
2683 (aarch64_hard_regno_caller_save_mode): Return the original mode
2685 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2686 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2687 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2688 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2690 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2691 does not overlap dest if the function is frame-related. Handle
2693 (aarch64_split_add_offset): New function.
2694 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2695 them aarch64_add_offset.
2696 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2697 and update call to aarch64_sub_sp.
2698 (aarch64_add_cfa_expression): New function.
2699 (aarch64_expand_prologue): Pass extra temporary registers to the
2700 functions above. Handle the case in which we need to emit new
2701 DW_CFA_expressions for registers that were originally saved
2702 relative to the stack pointer, but now have to be expressed
2703 relative to the frame pointer.
2704 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2706 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2707 IP0 and IP1 values for SVE frames.
2708 (aarch64_expand_vec_series): New function.
2709 (aarch64_expand_sve_widened_duplicate): Likewise.
2710 (aarch64_expand_sve_const_vector): Likewise.
2711 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2712 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2713 into the register, rather than emitting a SET directly.
2714 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2715 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2716 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2717 (offset_9bit_signed_scaled_p): New functions.
2718 (aarch64_replicate_bitmask_imm): New function.
2719 (aarch64_bitmask_imm): Use it.
2720 (aarch64_cannot_force_const_mem): Reject expressions involving
2721 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2722 (aarch64_classify_index): Handle SVE indices, by requiring
2723 a plain register index with a scale that matches the element size.
2724 (aarch64_classify_address): Handle SVE addresses. Assert that
2725 the mode of the address is VOIDmode or an integer mode.
2726 Update call to aarch64_classify_symbol.
2727 (aarch64_classify_symbolic_expression): Update call to
2728 aarch64_classify_symbol.
2729 (aarch64_const_vec_all_in_range_p): New function.
2730 (aarch64_print_vector_float_operand): Likewise.
2731 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2732 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2733 and the FP immediates 1.0 and 0.5.
2734 (aarch64_print_address_internal): Handle SVE addresses.
2735 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2736 (aarch64_regno_regclass): Handle predicate registers.
2737 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2739 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2740 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2741 (aarch64_convert_sve_vector_bits): New function.
2742 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2743 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2745 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2746 Handle SVE vector and predicate modes. Accept VL-based constants
2747 that need only one temporary register, and VL offsets that require
2748 no temporary registers.
2749 (aarch64_conditional_register_usage): Mark the predicate registers
2750 as fixed if SVE isn't available.
2751 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2752 Return true for SVE vector and predicate modes.
2753 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2754 rather than an unsigned int. Handle SVE modes.
2755 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2757 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2759 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2760 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2761 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2762 (aarch64_sve_float_mul_immediate_p): New functions.
2763 (aarch64_sve_valid_immediate): New function.
2764 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2765 Explicitly reject structure modes. Check for INDEX constants.
2766 Handle PTRUE and PFALSE constants.
2767 (aarch64_check_zero_based_sve_index_immediate): New function.
2768 (aarch64_simd_imm_zero_p): Delete.
2769 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2770 vector modes. Accept constants in the range of CNT[BHWD].
2771 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2772 ask for an Advanced SIMD mode.
2773 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2774 (aarch64_simd_vector_alignment): Handle SVE predicates.
2775 (aarch64_vectorize_preferred_vector_alignment): New function.
2776 (aarch64_simd_vector_alignment_reachable): Use it instead of
2778 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2779 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2781 (MAX_VECT_LEN): Delete.
2782 (expand_vec_perm_d): Add a vec_flags field.
2783 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2784 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2785 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2787 (aarch64_evpc_rev): Rename to...
2788 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2789 (aarch64_evpc_rev_global): New function.
2790 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2791 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2793 (aarch64_evpc_sve_tbl): New function.
2794 (aarch64_expand_vec_perm_const_1): Update after rename of
2795 aarch64_evpc_rev. Handle SVE permutes too, trying
2796 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2797 than aarch64_evpc_tbl.
2798 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2799 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2800 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2801 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2802 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2803 (aarch64_expand_sve_vcond): New functions.
2804 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2805 of aarch64_vector_mode_p.
2806 (aarch64_dwarf_poly_indeterminate_value): New function.
2807 (aarch64_compute_pressure_classes): Likewise.
2808 (aarch64_can_change_mode_class): Likewise.
2809 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2810 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2811 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2812 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2813 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2814 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2815 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2816 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2818 (Dn, Dl, Dr): Accept const as well as const_vector.
2819 (Dz): Likewise. Compare against CONST0_RTX.
2820 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2821 of "vector" where appropriate.
2822 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2823 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2824 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2825 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2826 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2827 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2828 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2829 (v_int_equiv): Extend to SVE modes.
2830 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2832 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2833 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2834 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2835 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2836 (SVE_COND_FP_CMP): New int iterators.
2837 (perm_hilo): Handle the new unpack unspecs.
2838 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2840 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2841 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2842 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2843 (aarch64_equality_operator, aarch64_constant_vector_operand)
2844 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2845 (aarch64_sve_nonimmediate_operand): Likewise.
2846 (aarch64_sve_general_operand): Likewise.
2847 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2848 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2849 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2850 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2851 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2852 (aarch64_sve_float_arith_immediate): Likewise.
2853 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2854 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2855 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2856 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2857 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2858 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2859 (aarch64_sve_float_arith_operand): Likewise.
2860 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2861 (aarch64_sve_float_mul_operand): Likewise.
2862 (aarch64_sve_vec_perm_operand): Likewise.
2863 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2864 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2865 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2866 as well as const_vector.
2867 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2868 in file. Use CONST0_RTX and CONSTM1_RTX.
2869 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2870 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2871 Use aarch64_simd_imm_zero.
2872 * config/aarch64/aarch64-sve.md: New file.
2873 * config/aarch64/aarch64.md: Include it.
2874 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2875 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2876 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2877 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2878 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2879 (sve): New attribute.
2880 (enabled): Disable instructions with the sve attribute unless
2882 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2883 aarch64_expand_mov_immediate.
2884 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2885 CNT[BHSD] immediates.
2886 (movti): Split CONST_POLY_INT moves into two halves.
2887 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2888 Split additions that need a temporary here if the destination
2889 is the stack pointer.
2890 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2891 (*add<mode>3_poly_1): New instruction.
2892 (set_clobber_cc): New expander.
2894 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2896 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2897 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2898 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2899 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2900 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2901 Change innermode from fixed_mode_size to machine_mode.
2902 (simplify_subreg): Update call accordingly. Handle a constant-sized
2903 subreg of a variable-length CONST_VECTOR.
2905 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2906 Alan Hayward <alan.hayward@arm.com>
2907 David Sherwood <david.sherwood@arm.com>
2909 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2910 (add_offset_to_base): New function, split out from...
2911 (create_mem_ref): ...here. When handling a scale other than 1,
2912 check first whether the address is valid without the offset.
2913 Add it into the base if so, leaving the index and scale as-is.
2915 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2918 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2919 fold_for_warn before checking if arg2 is INTEGER_CST.
2921 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2923 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2924 (store_multiple_operation): Delete.
2925 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2926 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2927 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2928 guarded by TARGET_STRING.
2929 (rs6000_output_load_multiple): Delete.
2930 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2931 OPTION_MASK_STRING / TARGET_STRING handling.
2932 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2933 (const rs6000_opt_masks) <"string">: Change mask to 0.
2934 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2935 (MASK_STRING): Delete.
2936 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2938 (load_multiple): Delete.
2945 (store_multiple): Delete.
2952 (movmemsi_8reg): Delete.
2953 (corresponding unnamed define_insn): Delete.
2954 (movmemsi_6reg): Delete.
2955 (corresponding unnamed define_insn): Delete.
2956 (movmemsi_4reg): Delete.
2957 (corresponding unnamed define_insn): Delete.
2958 (movmemsi_2reg): Delete.
2959 (corresponding unnamed define_insn): Delete.
2960 (movmemsi_1reg): Delete.
2961 (corresponding unnamed define_insn): Delete.
2962 * config/rs6000/rs6000.opt (mno-string): New.
2963 (mstring): Replace by deprecation warning stub.
2964 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2966 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2968 * regrename.c (regrename_do_replace): If replacing the same
2969 reg multiple times, try to reuse last created gen_raw_REG.
2972 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2973 main to workaround a bug in GDB.
2975 2018-01-12 Tom de Vries <tom@codesourcery.com>
2978 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2980 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2982 PR rtl-optimization/80481
2983 * ira-color.c (get_cap_member): New function.
2984 (allocnos_conflict_by_live_ranges_p): Use it.
2985 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2986 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2988 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2991 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2992 (*saddl_se_1): Ditto.
2994 (*ssubl_se_1): Ditto.
2996 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2998 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2999 rather than wi::to_widest for DR_INITs.
3000 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3001 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3002 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3004 (vect_analyze_group_access_1): Note that here.
3006 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3008 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3009 polynomial type sizes.
3011 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3013 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3014 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3015 (gimple_add_tmp_var): Likewise.
3017 2018-01-12 Martin Liska <mliska@suse.cz>
3019 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3020 (gimple_alloc_sizes): Likewise.
3021 (dump_gimple_statistics): Use PRIu64 in printf format.
3022 * gimple.h: Change uint64_t to int.
3024 2018-01-12 Martin Liska <mliska@suse.cz>
3026 * tree-core.h: Use uint64_t instead of int.
3027 * tree.c (tree_node_counts): Likewise.
3028 (tree_node_sizes): Likewise.
3029 (dump_tree_statistics): Use PRIu64 in printf format.
3031 2018-01-12 Martin Liska <mliska@suse.cz>
3033 * Makefile.in: As qsort_chk is implemented in vec.c, add
3034 vec.o to linkage of gencfn-macros.
3035 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3036 passing the info to record_node_allocation_statistics.
3037 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3039 * ggc-common.c (struct ggc_usage): Add operator== and use
3040 it in operator< and compare function.
3041 * mem-stats.h (struct mem_usage): Likewise.
3042 * vec.c (struct vec_usage): Remove operator< and compare
3043 function. Can be simply inherited.
3045 2018-01-12 Martin Jambor <mjambor@suse.cz>
3048 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3049 * tree-ssa-math-opts.c: Include domwalk.h.
3050 (convert_mult_to_fma_1): New function.
3051 (fma_transformation_info): New type.
3052 (fma_deferring_state): Likewise.
3053 (cancel_fma_deferring): New function.
3054 (result_of_phi): Likewise.
3055 (last_fma_candidate_feeds_initial_phi): Likewise.
3056 (convert_mult_to_fma): Added deferring logic, split actual
3057 transformation to convert_mult_to_fma_1.
3058 (math_opts_dom_walker): New type.
3059 (math_opts_dom_walker::after_dom_children): New method, body moved
3060 here from pass_optimize_widening_mul::execute, added deferring logic
3062 (pass_optimize_widening_mul::execute): Moved most of code to
3063 math_opts_dom_walker::after_dom_children.
3064 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3065 * config/i386/i386.c (ix86_option_override_internal): Added
3066 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3068 2018-01-12 Richard Biener <rguenther@suse.de>
3071 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3072 inline instance vars.
3074 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3077 * config/rx/rx.c (rx_is_restricted_memory_address):
3080 2018-01-12 Richard Biener <rguenther@suse.de>
3082 PR tree-optimization/80846
3083 * target.def (split_reduction): New target hook.
3084 * targhooks.c (default_split_reduction): New function.
3085 * targhooks.h (default_split_reduction): Declare.
3086 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3087 target requests first reduce vectors by combining low and high
3089 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3090 (get_vectype_for_scalar_type_and_size): Export.
3091 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3092 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3093 * doc/tm.texi: Regenerate.
3094 * config/i386/i386.c (ix86_split_reduction): Implement
3095 TARGET_VECTORIZE_SPLIT_REDUCTION.
3097 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3100 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3101 in PIC mode except for TARGET_VXWORKS_RTP.
3102 * config/sparc/sparc.c: Include cfgrtl.h.
3103 (TARGET_INIT_PIC_REG): Define.
3104 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3105 (sparc_pic_register_p): New predicate.
3106 (sparc_legitimate_address_p): Use it.
3107 (sparc_legitimize_pic_address): Likewise.
3108 (sparc_delegitimize_address): Likewise.
3109 (sparc_mode_dependent_address_p): Likewise.
3110 (gen_load_pcrel_sym): Remove 4th parameter.
3111 (load_got_register): Adjust call to above. Remove obsolete stuff.
3112 (sparc_expand_prologue): Do not call load_got_register here.
3113 (sparc_flat_expand_prologue): Likewise.
3114 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3115 (sparc_use_pseudo_pic_reg): New function.
3116 (sparc_init_pic_reg): Likewise.
3117 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3118 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3120 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3122 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3123 Add item for branch_cost.
3125 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3127 PR rtl-optimization/83565
3128 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3129 not extend the result to a larger mode for rotate operations.
3130 (num_sign_bit_copies1): Likewise.
3132 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3135 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3137 Use values-Xc.o for -pedantic.
3138 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3140 2018-01-12 Martin Liska <mliska@suse.cz>
3143 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3145 (possible_polymorphic_call_targets): Use it.
3146 (ipa_devirt): Likewise.
3148 2018-01-12 Martin Liska <mliska@suse.cz>
3150 * profile-count.h (enum profile_quality): Use 0 as invalid
3151 enum value of profile_quality.
3153 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3155 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3156 -mext-string options.
3158 2018-01-12 Richard Biener <rguenther@suse.de>
3160 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3161 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3162 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3164 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3166 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3168 * configure.ac (--with-long-double-format): Add support for the
3169 configuration option to change the default long double format on
3171 * config.gcc (powerpc*-linux*-*): Likewise.
3172 * configure: Regenerate.
3173 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3174 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3175 used without modification.
3177 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3179 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3180 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3181 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3182 MISC_BUILTIN_SPEC_BARRIER.
3183 (rs6000_init_builtins): Likewise.
3184 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3186 (speculation_barrier): New define_insn.
3187 * doc/extend.texi: Document __builtin_speculation_barrier.
3189 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3192 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3193 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3194 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3196 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3197 integral modes instead of "ss" and "sd".
3198 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3199 vectors with 32-bit and 64-bit elements.
3200 (vecdupssescalarmodesuffix): New mode attribute.
3201 (vec_dup<mode>): Use it.
3203 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3206 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3207 frame if argument is passed on stack.
3209 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3212 * ree.c (combine_reaching_defs): Optimize also
3213 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3214 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3216 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3219 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3221 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3224 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3225 after they are computed.
3227 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3229 PR tree-optimization/83695
3230 * gimple-loop-linterchange.cc
3231 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3232 reset cached scev information after interchange.
3233 (pass_linterchange::execute): Remove call to scev_reset_htab.
3235 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3237 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3238 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3239 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3240 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3241 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3242 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3243 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3244 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3245 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3246 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3247 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3248 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3249 (V_lane_reg): Likewise.
3250 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3252 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3253 (vfmal_lane_low<mode>_intrinsic,
3254 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3255 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3256 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3257 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3258 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3259 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3261 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3263 * config/arm/arm-cpus.in (fp16fml): New feature.
3264 (ALL_SIMD): Add fp16fml.
3265 (armv8.2-a): Add fp16fml as an option.
3266 (armv8.3-a): Likewise.
3267 (armv8.4-a): Add fp16fml as part of fp16.
3268 * config/arm/arm.h (TARGET_FP16FML): Define.
3269 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3271 * config/arm/arm-modes.def (V2HF): Define.
3272 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3273 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3274 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3275 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3276 vfmsl_low, vfmsl_high): New set of builtins.
3277 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3278 (vfml_op): New code attribute.
3279 (VFMLHALVES): New int iterator.
3280 (VFML, VFMLSEL): New mode attributes.
3281 (V_reg): Define mapping for V2HF.
3282 (V_hi, V_lo): New mode attributes.
3283 (VF_constraint): Likewise.
3284 (vfml_half, vfml_half_selector): New int attributes.
3285 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3287 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3288 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3290 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3291 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3292 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3293 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3295 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3296 Document new effective target and option set.
3298 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3300 * config/arm/arm-cpus.in (armv8_4): New feature.
3301 (ARMv8_4a): New fgroup.
3302 (armv8.4-a): New arch.
3303 * config/arm/arm-tables.opt: Regenerate.
3304 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3305 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3306 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3307 Add matching rules for -march=armv8.4-a and extensions.
3308 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3310 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3313 * config/rx/rx.md (BW): New mode attribute.
3314 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3316 2018-01-11 Richard Biener <rguenther@suse.de>
3318 PR tree-optimization/83435
3319 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3320 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3321 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3323 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3324 Alan Hayward <alan.hayward@arm.com>
3325 David Sherwood <david.sherwood@arm.com>
3327 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3329 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3330 (aarch64_print_address_internal): Use it to check for a zero offset.
3332 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3333 Alan Hayward <alan.hayward@arm.com>
3334 David Sherwood <david.sherwood@arm.com>
3336 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3337 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3338 Return a poly_int64 rather than a HOST_WIDE_INT.
3339 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3340 rather than a HOST_WIDE_INT.
3341 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3342 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3343 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3344 final_offset from HOST_WIDE_INT to poly_int64.
3345 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3346 to_constant when getting the number of units in an Advanced SIMD
3348 (aarch64_builtin_vectorized_function): Check for a constant number
3350 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3352 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3353 attribute instead of GET_MODE_NUNITS.
3354 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3355 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3356 GET_MODE_SIZE for fixed-size registers.
3357 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3358 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3359 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3360 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3361 (aarch64_print_operand, aarch64_print_address_internal)
3362 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3363 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3364 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3365 Handle polynomial GET_MODE_SIZE.
3366 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3367 wider than SImode without modification.
3368 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3369 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3370 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3371 passing and returning SVE modes.
3372 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3373 rather than GEN_INT.
3374 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3375 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3376 (aarch64_allocate_and_probe_stack_space): Likewise.
3377 (aarch64_layout_frame): Cope with polynomial offsets.
3378 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3379 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3381 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3382 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3383 poly_int64 rather than a HOST_WIDE_INT.
3384 (aarch64_get_separate_components, aarch64_process_components)
3385 (aarch64_expand_prologue, aarch64_expand_epilogue)
3386 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3387 (aarch64_anchor_offset): New function, split out from...
3388 (aarch64_legitimize_address): ...here.
3389 (aarch64_builtin_vectorization_cost): Handle polynomial
3390 TYPE_VECTOR_SUBPARTS.
3391 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3393 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3394 number of elements from the PARALLEL rather than the mode.
3395 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3396 rather than GET_MODE_BITSIZE.
3397 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3398 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3399 (aarch64_expand_vec_perm_const_1): Handle polynomial
3400 d->perm.length () and d->perm elements.
3401 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3402 Apply to_constant to d->perm elements.
3403 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3404 polynomial CONST_VECTOR_NUNITS.
3405 (aarch64_move_pointer): Take amount as a poly_int64 rather
3407 (aarch64_progress_pointer): Avoid temporary variable.
3408 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3409 the mode attribute instead of GET_MODE.
3411 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3412 Alan Hayward <alan.hayward@arm.com>
3413 David Sherwood <david.sherwood@arm.com>
3415 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3416 x exists before using it.
3417 (aarch64_add_constant_internal): Rename to...
3418 (aarch64_add_offset_1): ...this. Replace regnum with separate
3419 src and dest rtxes. Handle the case in which they're different,
3420 including when the offset is zero. Replace scratchreg with an rtx.
3421 Use 2 additions if there is no spare register into which we can
3422 move a 16-bit constant.
3423 (aarch64_add_constant): Delete.
3424 (aarch64_add_offset): Replace reg with separate src and dest
3425 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3426 Use aarch64_add_offset_1.
3427 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3428 an rtx rather than an int. Take the delta as a poly_int64
3429 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3430 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3431 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3432 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3433 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3435 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3436 aarch64_add_constant.
3438 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3440 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3441 Use scalar_float_mode.
3443 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3445 * config/aarch64/aarch64-simd.md
3446 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3447 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3448 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3449 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3450 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3451 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3452 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3453 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3454 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3455 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3457 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3460 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3461 targ_options->x_arm_arch_string is non NULL.
3463 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3465 * config/aarch64/aarch64.h
3466 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3468 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3471 * expmed.c (emit_store_flag_force): Swap if const op0
3472 and change VOIDmode to mode of op0.
3474 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3476 PR rtl-optimization/83761
3477 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3478 than bytes to mode_for_size.
3480 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3483 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3484 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3487 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3490 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3491 when in layout mode.
3492 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3493 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3496 2018-01-10 Michael Collison <michael.collison@arm.com>
3498 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3499 * config/aarch64/aarch64-option-extension.def: Add
3500 AARCH64_OPT_EXTENSION of 'fp16fml'.
3501 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3502 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3503 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3504 * config/aarch64/constraints.md (Ui7): New constraint.
3505 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3506 (VFMLA_SEL_W): Ditto.
3509 (VFMLA16_LOW): New int iterator.
3510 (VFMLA16_HIGH): Ditto.
3511 (UNSPEC_FMLAL): New unspec.
3512 (UNSPEC_FMLSL): Ditto.
3513 (UNSPEC_FMLAL2): Ditto.
3514 (UNSPEC_FMLSL2): Ditto.
3515 (f16mac): New code attribute.
3516 * config/aarch64/aarch64-simd-builtins.def
3517 (aarch64_fmlal_lowv2sf): Ditto.
3518 (aarch64_fmlsl_lowv2sf): Ditto.
3519 (aarch64_fmlalq_lowv4sf): Ditto.
3520 (aarch64_fmlslq_lowv4sf): Ditto.
3521 (aarch64_fmlal_highv2sf): Ditto.
3522 (aarch64_fmlsl_highv2sf): Ditto.
3523 (aarch64_fmlalq_highv4sf): Ditto.
3524 (aarch64_fmlslq_highv4sf): Ditto.
3525 (aarch64_fmlal_lane_lowv2sf): Ditto.
3526 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3527 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3528 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3529 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3530 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3531 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3532 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3533 (aarch64_fmlal_lane_highv2sf): Ditto.
3534 (aarch64_fmlsl_lane_highv2sf): Ditto.
3535 (aarch64_fmlal_laneq_highv2sf): Ditto.
3536 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3537 (aarch64_fmlalq_lane_highv4sf): Ditto.
3538 (aarch64_fmlsl_lane_highv4sf): Ditto.
3539 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3540 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3541 * config/aarch64/aarch64-simd.md:
3542 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3543 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3544 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3545 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3546 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3547 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3548 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3549 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3550 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3551 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3552 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3553 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3554 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3555 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3556 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3557 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3558 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3559 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3560 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3561 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3562 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3563 (vfmlsl_low_u32): Ditto.
3564 (vfmlalq_low_u32): Ditto.
3565 (vfmlslq_low_u32): Ditto.
3566 (vfmlal_high_u32): Ditto.
3567 (vfmlsl_high_u32): Ditto.
3568 (vfmlalq_high_u32): Ditto.
3569 (vfmlslq_high_u32): Ditto.
3570 (vfmlal_lane_low_u32): Ditto.
3571 (vfmlsl_lane_low_u32): Ditto.
3572 (vfmlal_laneq_low_u32): Ditto.
3573 (vfmlsl_laneq_low_u32): Ditto.
3574 (vfmlalq_lane_low_u32): Ditto.
3575 (vfmlslq_lane_low_u32): Ditto.
3576 (vfmlalq_laneq_low_u32): Ditto.
3577 (vfmlslq_laneq_low_u32): Ditto.
3578 (vfmlal_lane_high_u32): Ditto.
3579 (vfmlsl_lane_high_u32): Ditto.
3580 (vfmlal_laneq_high_u32): Ditto.
3581 (vfmlsl_laneq_high_u32): Ditto.
3582 (vfmlalq_lane_high_u32): Ditto.
3583 (vfmlslq_lane_high_u32): Ditto.
3584 (vfmlalq_laneq_high_u32): Ditto.
3585 (vfmlslq_laneq_high_u32): Ditto.
3586 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3587 (AARCH64_FL_FOR_ARCH8_4): New.
3588 (AARCH64_ISA_F16FML): New ISA flag.
3589 (TARGET_F16FML): New feature flag for fp16fml.
3590 (doc/invoke.texi): Document new fp16fml option.
3592 2018-01-10 Michael Collison <michael.collison@arm.com>
3594 * config/aarch64/aarch64-builtins.c:
3595 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3596 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3597 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3598 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3599 (AARCH64_ISA_SHA3): New ISA flag.
3600 (TARGET_SHA3): New feature flag for sha3.
3601 * config/aarch64/iterators.md (sha512_op): New int attribute.
3602 (CRYPTO_SHA512): New int iterator.
3603 (UNSPEC_SHA512H): New unspec.
3604 (UNSPEC_SHA512H2): Ditto.
3605 (UNSPEC_SHA512SU0): Ditto.
3606 (UNSPEC_SHA512SU1): Ditto.
3607 * config/aarch64/aarch64-simd-builtins.def
3608 (aarch64_crypto_sha512hqv2di): New builtin.
3609 (aarch64_crypto_sha512h2qv2di): Ditto.
3610 (aarch64_crypto_sha512su0qv2di): Ditto.
3611 (aarch64_crypto_sha512su1qv2di): Ditto.
3612 (aarch64_eor3qv8hi): Ditto.
3613 (aarch64_rax1qv2di): Ditto.
3614 (aarch64_xarqv2di): Ditto.
3615 (aarch64_bcaxqv8hi): Ditto.
3616 * config/aarch64/aarch64-simd.md:
3617 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3618 (aarch64_crypto_sha512su0qv2di): Ditto.
3619 (aarch64_crypto_sha512su1qv2di): Ditto.
3620 (aarch64_eor3qv8hi): Ditto.
3621 (aarch64_rax1qv2di): Ditto.
3622 (aarch64_xarqv2di): Ditto.
3623 (aarch64_bcaxqv8hi): Ditto.
3624 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3625 (vsha512h2q_u64): Ditto.
3626 (vsha512su0q_u64): Ditto.
3627 (vsha512su1q_u64): Ditto.
3628 (veor3q_u16): Ditto.
3629 (vrax1q_u64): Ditto.
3631 (vbcaxq_u16): Ditto.
3632 * config/arm/types.md (crypto_sha512): New type attribute.
3633 (crypto_sha3): Ditto.
3634 (doc/invoke.texi): Document new sha3 option.
3636 2018-01-10 Michael Collison <michael.collison@arm.com>
3638 * config/aarch64/aarch64-builtins.c:
3639 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3640 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3641 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3642 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3643 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3644 (AARCH64_ISA_SM4): New ISA flag.
3645 (TARGET_SM4): New feature flag for sm4.
3646 * config/aarch64/aarch64-simd-builtins.def
3647 (aarch64_sm3ss1qv4si): Ditto.
3648 (aarch64_sm3tt1aq4si): Ditto.
3649 (aarch64_sm3tt1bq4si): Ditto.
3650 (aarch64_sm3tt2aq4si): Ditto.
3651 (aarch64_sm3tt2bq4si): Ditto.
3652 (aarch64_sm3partw1qv4si): Ditto.
3653 (aarch64_sm3partw2qv4si): Ditto.
3654 (aarch64_sm4eqv4si): Ditto.
3655 (aarch64_sm4ekeyqv4si): Ditto.
3656 * config/aarch64/aarch64-simd.md:
3657 (aarch64_sm3ss1qv4si): Ditto.
3658 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3659 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3660 (aarch64_sm4eqv4si): Ditto.
3661 (aarch64_sm4ekeyqv4si): Ditto.
3662 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3663 (sm3part_op): Ditto.
3664 (CRYPTO_SM3TT): Ditto.
3665 (CRYPTO_SM3PART): Ditto.
3666 (UNSPEC_SM3SS1): New unspec.
3667 (UNSPEC_SM3TT1A): Ditto.
3668 (UNSPEC_SM3TT1B): Ditto.
3669 (UNSPEC_SM3TT2A): Ditto.
3670 (UNSPEC_SM3TT2B): Ditto.
3671 (UNSPEC_SM3PARTW1): Ditto.
3672 (UNSPEC_SM3PARTW2): Ditto.
3673 (UNSPEC_SM4E): Ditto.
3674 (UNSPEC_SM4EKEY): Ditto.
3675 * config/aarch64/constraints.md (Ui2): New constraint.
3676 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3677 * config/arm/types.md (crypto_sm3): New type attribute.
3678 (crypto_sm4): Ditto.
3679 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3680 (vsm3tt1aq_u32): Ditto.
3681 (vsm3tt1bq_u32): Ditto.
3682 (vsm3tt2aq_u32): Ditto.
3683 (vsm3tt2bq_u32): Ditto.
3684 (vsm3partw1q_u32): Ditto.
3685 (vsm3partw2q_u32): Ditto.
3686 (vsm4eq_u32): Ditto.
3687 (vsm4ekeyq_u32): Ditto.
3688 (doc/invoke.texi): Document new sm4 option.
3690 2018-01-10 Michael Collison <michael.collison@arm.com>
3692 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3693 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3694 (AARCH64_FL_FOR_ARCH8_4): New.
3695 (AARCH64_FL_V8_4): New flag.
3696 (doc/invoke.texi): Document new armv8.4-a option.
3698 2018-01-10 Michael Collison <michael.collison@arm.com>
3700 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3701 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3702 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3703 * config/aarch64/aarch64-option-extension.def: Add
3704 AARCH64_OPT_EXTENSION of 'sha2'.
3705 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3706 (crypto): Disable sha2 and aes if crypto disabled.
3707 (crypto): Enable aes and sha2 if enabled.
3708 (simd): Disable sha2 and aes if simd disabled.
3709 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3711 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3712 (TARGET_SHA2): New feature flag for sha2.
3713 (TARGET_AES): New feature flag for aes.
3714 * config/aarch64/aarch64-simd.md:
3715 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3716 conditional on TARGET_AES.
3717 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3718 (aarch64_crypto_sha1hsi): Make pattern conditional
3720 (aarch64_crypto_sha1hv4si): Ditto.
3721 (aarch64_be_crypto_sha1hv4si): Ditto.
3722 (aarch64_crypto_sha1su1v4si): Ditto.
3723 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3724 (aarch64_crypto_sha1su0v4si): Ditto.
3725 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3726 (aarch64_crypto_sha256su0v4si): Ditto.
3727 (aarch64_crypto_sha256su1v4si): Ditto.
3728 (doc/invoke.texi): Document new aes and sha2 options.
3730 2018-01-10 Martin Sebor <msebor@redhat.com>
3732 PR tree-optimization/83781
3733 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3736 2018-01-11 Martin Sebor <msebor@gmail.com>
3737 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3739 PR tree-optimization/83501
3740 PR tree-optimization/81703
3742 * tree-ssa-strlen.c (get_string_cst): Rename...
3743 (get_string_len): ...to this. Handle global constants.
3744 (handle_char_store): Adjust.
3746 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3747 Jim Wilson <jimw@sifive.com>
3749 * config/riscv/riscv-protos.h (riscv_output_return): New.
3750 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3751 (riscv_attribute_table, riscv_output_return),
3752 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3753 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3754 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3756 (riscv_expand_prologue): Add early return for naked function.
3757 (riscv_expand_epilogue): Likewise.
3758 (riscv_function_ok_for_sibcall): Return false for naked function.
3759 (riscv_set_current_function): New.
3760 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3761 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3762 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3763 * doc/extend.texi (RISC-V Function Attributes): New.
3765 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3767 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3768 check for 128-bit long double before checking TCmode.
3769 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3770 128-bit long doubles before checking TFmode or TCmode.
3771 (FLOAT128_IBM_P): Likewise.
3773 2018-01-10 Martin Sebor <msebor@redhat.com>
3775 PR tree-optimization/83671
3776 * builtins.c (c_strlen): Unconditionally return zero for the empty
3778 Use -Warray-bounds for warnings.
3779 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3780 for non-constant array indices with COMPONENT_REF, arrays of
3781 arrays, and pointers to arrays.
3782 (gimple_fold_builtin_strlen): Determine and set length range for
3783 non-constant character arrays.
3785 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3788 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3791 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3793 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3795 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3798 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3799 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3800 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3801 indexed_or_indirect_operand predicate.
3802 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3803 (*vsx_le_perm_load_v8hi): Likewise.
3804 (*vsx_le_perm_load_v16qi): Likewise.
3805 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3806 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3807 (*vsx_le_perm_store_v8hi): Likewise.
3808 (*vsx_le_perm_store_v16qi): Likewise.
3809 (eight unnamed splitters): Likewise.
3811 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3813 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3814 * config/rs6000/emmintrin.h: Likewise.
3815 * config/rs6000/mmintrin.h: Likewise.
3816 * config/rs6000/xmmintrin.h: Likewise.
3818 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3821 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3823 * tree.c (tree_nop_conversion): Return true for location wrapper
3825 (maybe_wrap_with_location): New function.
3826 (selftest::check_strip_nops): New function.
3827 (selftest::test_location_wrappers): New function.
3828 (selftest::tree_c_tests): Call it.
3829 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3830 (maybe_wrap_with_location): New decl.
3831 (EXPR_LOCATION_WRAPPER_P): New macro.
3832 (location_wrapper_p): New inline function.
3833 (tree_strip_any_location_wrapper): New inline function.
3835 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3838 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3839 stack_realign_offset for the largest alignment of stack slot
3841 (ix86_find_max_used_stack_alignment): New function.
3842 (ix86_finalize_stack_frame_flags): Use it. Set
3843 max_used_stack_alignment if we don't realign stack.
3844 * config/i386/i386.h (machine_function): Add
3845 max_used_stack_alignment.
3847 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3849 * config/arm/arm.opt (-mbranch-cost): New option.
3850 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3853 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3856 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3857 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3859 2018-01-10 Richard Biener <rguenther@suse.de>
3862 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3863 early out so it also covers the case where we have a non-NULL
3866 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3868 PR tree-optimization/83753
3869 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3870 for non-strided grouped accesses if the number of elements is 1.
3872 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3875 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3876 * i386.h (TARGET_USE_GATHER): Define.
3877 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3879 2018-01-10 Martin Liska <mliska@suse.cz>
3882 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3883 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3885 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3886 CLEANUP_NO_PARTITIONING is not set.
3888 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3890 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3891 for vectors, as a partial revert of r254296.
3892 * rtl.h (const_vec_p): Delete.
3893 (const_vec_duplicate_p): Don't test for vector CONSTs.
3894 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3895 * expmed.c (make_tree): Likewise.
3898 * common.md (E, F): Use CONSTANT_P instead of checking for
3900 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3901 checking for CONST_VECTOR.
3903 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3906 * predict.c (force_edge_cold): Handle in more sane way edges
3909 2018-01-09 Carl Love <cel@us.ibm.com>
3911 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3913 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3914 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3915 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3916 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3917 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3918 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3919 * config/rs6000/rs6000-protos.h: Add extern defition for
3920 rs6000_generate_float2_double_code.
3921 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3923 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3924 (float2_v2df): Add define_expand.
3926 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3929 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3930 op_mode in the force_to_mode call.
3932 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3934 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3935 instead of checking each element individually.
3936 (aarch64_evpc_uzp): Likewise.
3937 (aarch64_evpc_zip): Likewise.
3938 (aarch64_evpc_ext): Likewise.
3939 (aarch64_evpc_rev): Likewise.
3940 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3941 instead of checking each element individually. Return true without
3943 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3944 whether all selected elements come from the same input, instead of
3945 checking each element individually. Remove calls to gen_rtx_REG,
3946 start_sequence and end_sequence and instead assert that no rtl is
3949 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3951 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3952 order of HIGH and CONST checks.
3954 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3956 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3957 if the destination isn't an SSA_NAME.
3959 2018-01-09 Richard Biener <rguenther@suse.de>
3961 PR tree-optimization/83668
3962 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3964 (canonicalize_loop_form): ... here, renamed from ...
3965 (canonicalize_loop_closed_ssa_form): ... this and amended to
3966 swap successor edges for loop exit blocks to make us use
3967 the RPO order we need for initial schedule generation.
3969 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3971 PR tree-optimization/64811
3972 * match.pd: When optimizing comparisons with Inf, avoid
3973 introducing or losing exceptions from comparisons with NaN.
3975 2018-01-09 Martin Liska <mliska@suse.cz>
3978 * asan.c (shadow_mem_size): Add gcc_assert.
3980 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3982 Don't save registers in main().
3985 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3986 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3987 * config/avr/avr.c (avr_set_current_function): Don't error if
3988 naked, OS_task or OS_main are specified at the same time.
3989 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3991 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3993 * common/config/avr/avr-common.c (avr_option_optimization_table):
3994 Switch on -mmain-is-OS_task for optimizing compilations.
3996 2018-01-09 Richard Biener <rguenther@suse.de>
3998 PR tree-optimization/83572
3999 * graphite.c: Include cfganal.h.
4000 (graphite_transform_loops): Connect infinite loops to exit
4001 and remove fake edges at the end.
4003 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4005 * ipa-inline.c (edge_badness): Revert accidental checkin.
4007 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4010 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4011 symbols; not inline clones.
4013 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4016 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4017 hard registers. Formatting fixes.
4019 PR preprocessor/83722
4020 * gcc.c (try_generate_repro): Pass
4021 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4022 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4025 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4026 Kito Cheng <kito.cheng@gmail.com>
4028 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4029 (riscv_leaf_function_p): Delete.
4030 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4032 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4034 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4036 (do_ifelse): New function.
4037 (do_isel): New function.
4038 (do_sub3): New function.
4039 (do_add3): New function.
4040 (do_load_mask_compare): New function.
4041 (do_overlap_load_compare): New function.
4042 (expand_compare_loop): New function.
4043 (expand_block_compare): Call expand_compare_loop() when appropriate.
4044 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4046 (-mblock-compare-inline-loop-limit): New option.
4048 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4051 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4052 Reverse order of second and third operands in first alternative.
4053 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4054 of first and second elements in UNSPEC_VPERMR vector.
4055 (altivec_expand_vec_perm_le): Likewise.
4057 2018-01-08 Jeff Law <law@redhat.com>
4059 PR rtl-optimizatin/81308
4060 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4061 (process_switch): If group_case_labels makes a change, then set
4063 (pass_convert_switch::execute): If a switch is converted, then
4064 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4066 PR rtl-optimization/81308
4067 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4070 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4072 PR target/83663 - Revert r255946
4073 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4074 generation for cases where splatting a value is not useful.
4075 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4076 across a vec_duplicate and a paradoxical subreg forming a vector
4077 mode to a vec_concat.
4079 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4081 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4082 -march=armv8.3-a variants.
4083 * config/arm/t-multilib: Likewise.
4084 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4086 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4088 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4090 (cceq_ior_compare_complement): Give it a name so I can use it, and
4091 change boolean_or_operator predicate to boolean_operator so it can
4092 be used to generate a crand.
4093 (eqne): New code iterator.
4094 (bd/bd_neg): New code_attrs.
4095 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4096 a single define_insn.
4097 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4098 decrement (bdnzt/bdnzf/bdzt/bdzf).
4099 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4100 with the new names of the branch decrement patterns, and added the
4101 names of the branch decrement conditional patterns.
4103 2018-01-08 Richard Biener <rguenther@suse.de>
4105 PR tree-optimization/83563
4106 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4109 2018-01-08 Richard Biener <rguenther@suse.de>
4112 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4114 2018-01-08 Richard Biener <rguenther@suse.de>
4116 PR tree-optimization/83685
4117 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4118 references to abnormals.
4120 2018-01-08 Richard Biener <rguenther@suse.de>
4123 * dwarf2out.c (output_indirect_strings): Handle empty
4124 skeleton_debug_str_hash.
4125 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4127 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4129 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4130 (emit_store_direct): Likewise.
4131 (arc_trampoline_adjust_address): Likewise.
4132 (arc_asm_trampoline_template): New function.
4133 (arc_initialize_trampoline): Use asm_trampoline_template.
4134 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4135 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4136 * config/arc/arc.md (flush_icache): Delete pattern.
4138 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4140 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4141 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4144 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4147 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4148 by not USED_FOR_TARGET.
4149 (make_pass_resolve_sw_modes): Likewise.
4151 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4153 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4156 2018-01-08 Richard Biener <rguenther@suse.de>
4159 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4161 2018-01-08 Richard Biener <rguenther@suse.de>
4164 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4166 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4169 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4170 basic blocks with a small number of successors.
4171 (convert_control_dep_chain_into_preds): Improve handling of
4173 (dump_predicates): Split apart into...
4174 (dump_pred_chain): ...here...
4175 (dump_pred_info): ...and here.
4176 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4177 (can_chain_union_be_invalidated_p): Improve check for invalidation
4179 (uninit_uses_cannot_happen): Avoid unnecessary if
4180 convert_control_dep_chain_into_preds yielded nothing.
4182 2018-01-06 Martin Sebor <msebor@redhat.com>
4184 PR tree-optimization/83640
4185 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4186 subtracting negative offset from size.
4187 (builtin_access::overlap): Adjust offset bounds of the access to fall
4188 within the size of the object if possible.
4190 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4192 PR rtl-optimization/83699
4193 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4194 extract_bit_field_as_subreg to cases in which the extracted
4195 value is also a vector.
4197 * lra-constraints.c (process_alt_operands): Test for the equivalence
4198 substitutions when detecting a possible reload cycle.
4200 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4203 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4204 by default if flag_selective_schedling{,2}. Formatting fixes.
4206 PR rtl-optimization/83682
4207 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4208 if it has non-VECTOR_MODE element mode.
4209 (vec_duplicate_p): Likewise.
4212 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4213 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4215 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4218 * config/i386/i386-builtin.def
4219 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4220 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4221 Require also OPTION_MASK_ISA_AVX512F in addition to
4222 OPTION_MASK_ISA_GFNI.
4223 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4224 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4225 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4226 to OPTION_MASK_ISA_GFNI.
4227 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4228 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4229 OPTION_MASK_ISA_AVX512BW.
4230 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4231 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4232 addition to OPTION_MASK_ISA_GFNI.
4233 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4234 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4235 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4236 to OPTION_MASK_ISA_GFNI.
4237 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4238 a requirement for all ISAs rather than any of them with a few
4240 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4242 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4243 bitmasks to be enabled with 3 exceptions, instead of requiring any
4244 enabled ISA with lots of exceptions.
4245 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4246 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4247 Change avx512bw in isa attribute to avx512f.
4248 * config/i386/sgxintrin.h: Add license boilerplate.
4249 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4250 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4251 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4252 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4254 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4255 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4256 temporarily sse2 rather than sse if not enabled already.
4259 * config/i386/sse.md (VI248_VLBW): Rename to ...
4260 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4261 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4262 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4263 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4264 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4265 mode iterator instead of VI248_VLBW.
4267 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4269 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4270 (record_modified): Skip clobbers; add debug output.
4271 (param_change_prob): Use sreal frequencies.
4273 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4275 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4276 punt for user-aligned variables.
4278 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4280 * tree-chrec.c (chrec_contains_symbols): Return true for
4283 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4286 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4287 of (x|y) == x for BICS pattern.
4289 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4291 PR tree-optimization/83605
4292 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4293 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4296 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4298 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4299 * config/epiphany/rtems.h: New file.
4301 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4302 Uros Bizjak <ubizjak@gmail.com>
4305 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4306 QIreg_operand instead of register_operand predicate.
4307 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4308 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4309 comments instead of -fmitigate[-_]rop.
4311 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4314 * cgraphunit.c (symbol_table::compile): Switch to text_section
4315 before calling assembly_start debug hook.
4316 * run-rtl-passes.c (run_rtl_passes): Likewise.
4319 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4321 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4322 range_int_cst_p rather than !symbolic_range_p before calling
4323 extract_range_from_multiplicative_op_1.
4325 2018-01-04 Jeff Law <law@redhat.com>
4327 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4328 redundant test in assertion.
4330 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4332 * doc/rtl.texi: Document machine_mode wrapper classes.
4334 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4336 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4339 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4341 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4342 the VEC_PERM_EXPR fold to fail.
4344 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4347 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4348 to switched_sections.
4350 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4353 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4356 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4359 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4360 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4362 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4365 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4366 is BLKmode and bitpos not zero or mode change is needed.
4368 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4371 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4374 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4377 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4378 instead of MULT rtx. Update all corresponding splitters.
4380 (*ssub<modesuffix>): Ditto.
4382 (*cmp_sadd_di): Update split patterns.
4383 (*cmp_sadd_si): Ditto.
4384 (*cmp_sadd_sidi): Ditto.
4385 (*cmp_ssub_di): Ditto.
4386 (*cmp_ssub_si): Ditto.
4387 (*cmp_ssub_sidi): Ditto.
4388 * config/alpha/predicates.md (const23_operand): New predicate.
4389 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4390 Look for ASHIFT, not MULT inner operand.
4391 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4393 2018-01-04 Martin Liska <mliska@suse.cz>
4395 PR gcov-profile/83669
4396 * gcov.c (output_intermediate_file): Add version to intermediate
4398 * doc/gcov.texi: Document new field 'version' in intermediate
4399 file format. Fix location of '-k' option of gcov command.
4401 2018-01-04 Martin Liska <mliska@suse.cz>
4404 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4406 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4408 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4410 2018-01-03 Martin Sebor <msebor@redhat.com>
4412 PR tree-optimization/83655
4413 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4414 checking calls with invalid arguments.
4416 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4418 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4419 (vectorizable_mask_load_store): Delete.
4420 (vectorizable_call): Return false for masked loads and stores.
4421 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4422 instead of gimple_assign_rhs1.
4423 (vectorizable_load): Handle IFN_MASK_LOAD.
4424 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4428 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4430 (vectorizable_mask_load_store): ...here.
4431 (vectorizable_load): ...and here.
4433 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4435 * tree-vect-stmts.c (vect_build_all_ones_mask)
4436 (vect_build_zero_merge_argument): New functions, split out from...
4437 (vectorizable_load): ...here.
4439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4441 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4443 (vectorizable_mask_load_store): ...here.
4444 (vectorizable_store): ...and here.
4446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4448 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4450 (vectorizable_mask_load_store): ...here.
4452 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4454 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4455 (vect_model_store_cost): Take a vec_load_store_type instead of a
4457 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4458 (vect_model_store_cost): Take a vec_load_store_type instead of a
4460 (vectorizable_mask_load_store): Update accordingly.
4461 (vectorizable_store): Likewise.
4462 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4466 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4467 IFN_MASK_LOAD calls here rather than...
4468 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4471 Alan Hayward <alan.hayward@arm.com>
4472 David Sherwood <david.sherwood@arm.com>
4474 * expmed.c (extract_bit_field_1): For vector extracts,
4475 fall back to extract_bit_field_as_subreg if vec_extract
4478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4479 Alan Hayward <alan.hayward@arm.com>
4480 David Sherwood <david.sherwood@arm.com>
4482 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4483 they are variable or constant sized.
4484 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4485 slots for constant-sized data.
4487 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4488 Alan Hayward <alan.hayward@arm.com>
4489 David Sherwood <david.sherwood@arm.com>
4491 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4492 handling COND_EXPRs with boolean comparisons, try to find a better
4493 basis for the mask type than the boolean itself.
4495 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4497 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4498 is calculated and how it can be overridden.
4499 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4500 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4502 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4505 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4506 Alan Hayward <alan.hayward@arm.com>
4507 David Sherwood <david.sherwood@arm.com>
4509 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4510 Remove the mode argument.
4511 (aarch64_simd_valid_immediate): Remove the mode and inverse
4513 * config/aarch64/iterators.md (bitsize): New iterator.
4514 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4515 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4516 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4517 aarch64_simd_valid_immediate.
4518 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4519 (aarch64_reg_or_bic_imm): Likewise.
4520 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4521 with an insn_type enum and msl with a modifier_type enum.
4522 Replace element_width with a scalar_mode. Change the shift
4523 to unsigned int. Add constructors for scalar_float_mode and
4524 scalar_int_mode elements.
4525 (aarch64_vect_float_const_representable_p): Delete.
4526 (aarch64_can_const_movi_rtx_p)
4527 (aarch64_simd_scalar_immediate_valid_for_move)
4528 (aarch64_simd_make_constant): Update call to
4529 aarch64_simd_valid_immediate.
4530 (aarch64_advsimd_valid_immediate_hs): New function.
4531 (aarch64_advsimd_valid_immediate): Likewise.
4532 (aarch64_simd_valid_immediate): Remove mode and inverse
4533 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4534 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4535 and aarch64_float_const_representable_p on the result.
4536 (aarch64_output_simd_mov_immediate): Remove mode argument.
4537 Update call to aarch64_simd_valid_immediate and use of
4538 simd_immediate_info.
4539 (aarch64_output_scalar_simd_mov_immediate): Update call
4542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4543 Alan Hayward <alan.hayward@arm.com>
4544 David Sherwood <david.sherwood@arm.com>
4546 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4547 (mode_nunits): Likewise CONST_MODE_NUNITS.
4548 * machmode.def (ADJUST_NUNITS): Document.
4549 * genmodes.c (mode_data::need_nunits_adj): New field.
4550 (blank_mode): Update accordingly.
4551 (adj_nunits): New variable.
4552 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4554 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4555 listed in adj_nunits.
4556 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4557 listed in adj_nunits. Don't emit case statements for such modes.
4558 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4559 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4560 nothing if adj_nunits is nonnull.
4561 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4562 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4563 (emit_mode_fbit): Update use of print_maybe_const_decl.
4564 (emit_move_size): Likewise. Treat the array as non-const
4566 (emit_mode_adjustments): Handle adj_nunits.
4568 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4570 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4571 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4572 (VECTOR_MODES): Use it.
4573 (make_vector_modes): Take the prefix as an argument.
4575 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4576 Alan Hayward <alan.hayward@arm.com>
4577 David Sherwood <david.sherwood@arm.com>
4579 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4580 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4581 for MODE_VECTOR_BOOL.
4582 * machmode.def (VECTOR_BOOL_MODE): Document.
4583 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4584 (make_vector_bool_mode): New function.
4585 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4587 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4588 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4590 * stor-layout.c (int_mode_for_mode): Likewise.
4591 * tree.c (build_vector_type_for_mode): Likewise.
4592 * varasm.c (output_constant_pool_2): Likewise.
4593 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4594 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4595 for MODE_VECTOR_BOOL.
4596 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4597 of mode class checks.
4598 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4599 instead of a list of mode class checks.
4600 (expand_vector_scalar_condition): Likewise.
4601 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4604 Alan Hayward <alan.hayward@arm.com>
4605 David Sherwood <david.sherwood@arm.com>
4607 * machmode.h (mode_size): Change from unsigned short to
4609 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4610 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4611 or if measurement_type is not polynomial.
4612 (fixed_size_mode::includes_p): Check for constant-sized modes.
4613 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4614 return a poly_uint16 rather than an unsigned short.
4615 (emit_mode_size): Change the type of mode_size from unsigned short
4616 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4617 (emit_mode_adjustments): Cope with polynomial vector sizes.
4618 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4620 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4622 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4623 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4624 * caller-save.c (setup_save_areas): Likewise.
4625 (replace_reg_with_saved_mem): Likewise.
4626 * calls.c (emit_library_call_value_1): Likewise.
4627 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4628 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4629 (gen_lowpart_for_combine): Likewise.
4630 * convert.c (convert_to_integer_1): Likewise.
4631 * cse.c (equiv_constant, cse_insn): Likewise.
4632 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4633 (cselib_subst_to_values): Likewise.
4634 * dce.c (word_dce_process_block): Likewise.
4635 * df-problems.c (df_word_lr_mark_ref): Likewise.
4636 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4637 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4638 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4639 (rtl_for_decl_location): Likewise.
4640 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4641 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4642 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4643 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4644 (expand_expr_real_1): Likewise.
4645 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4646 (pad_below): Likewise.
4647 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4648 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4649 * ira.c (get_subreg_tracking_sizes): Likewise.
4650 * ira-build.c (ira_create_allocno_objects): Likewise.
4651 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4652 (ira_sort_regnos_for_alter_reg): Likewise.
4653 * ira-costs.c (record_operand_costs): Likewise.
4654 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4655 (resolve_simple_move): Likewise.
4656 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4657 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4658 (lra_constraints): Likewise.
4659 (CONST_POOL_OK_P): Reject variable-sized modes.
4660 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4661 (add_pseudo_to_slot, lra_spill): Likewise.
4662 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4663 * optabs-query.c (get_best_extraction_insn): Likewise.
4664 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4665 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4666 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4667 * recog.c (offsettable_address_addr_space_p): Likewise.
4668 * regcprop.c (maybe_mode_change): Likewise.
4669 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4670 * regrename.c (build_def_use): Likewise.
4671 * regstat.c (dump_reg_info): Likewise.
4672 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4673 (find_reloads, find_reloads_subreg_address): Likewise.
4674 * reload1.c (eliminate_regs_1): Likewise.
4675 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4676 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4677 (simplify_binary_operation_1, simplify_subreg): Likewise.
4678 * targhooks.c (default_function_arg_padding): Likewise.
4679 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4680 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4681 (verify_gimple_assign_ternary): Likewise.
4682 * tree-inline.c (estimate_move_cost): Likewise.
4683 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4684 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4685 (get_address_cost_ainc): Likewise.
4686 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4687 (vect_supportable_dr_alignment): Likewise.
4688 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4689 (vectorizable_reduction): Likewise.
4690 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4691 (vectorizable_operation, vectorizable_load): Likewise.
4692 * tree.c (build_same_sized_truth_vector_type): Likewise.
4693 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4694 * var-tracking.c (emit_note_insn_var_location): Likewise.
4695 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4696 (ADDR_VEC_ALIGN): Likewise.
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 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4704 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4705 or if measurement_type is polynomial.
4706 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4707 * combine.c (make_extraction): Likewise.
4708 * dse.c (find_shift_sequence): Likewise.
4709 * dwarf2out.c (mem_loc_descriptor): Likewise.
4710 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4711 (extract_bit_field, extract_low_bits): Likewise.
4712 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4713 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4714 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4715 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4716 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4717 * reload.c (find_reloads): Likewise.
4718 * reload1.c (alter_reg): Likewise.
4719 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4720 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4721 * tree-if-conv.c (predicate_mem_writes): Likewise.
4722 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4723 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4724 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4725 * valtrack.c (dead_debug_insert_temp): Likewise.
4726 * varasm.c (mergeable_constant_section): Likewise.
4727 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4730 Alan Hayward <alan.hayward@arm.com>
4731 David Sherwood <david.sherwood@arm.com>
4733 * expr.c (expand_assignment): Cope with polynomial mode sizes
4734 when assigning to a CONCAT.
4736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4737 Alan Hayward <alan.hayward@arm.com>
4738 David Sherwood <david.sherwood@arm.com>
4740 * machmode.h (mode_precision): Change from unsigned short to
4742 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4744 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4745 or if measurement_type is not polynomial.
4746 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4747 in which the mode is already known to be a scalar_int_mode.
4748 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4749 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4751 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4752 for GET_MODE_PRECISION.
4753 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4754 for GET_MODE_PRECISION.
4755 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4757 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4758 (expand_field_assignment, make_extraction): Likewise.
4759 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4760 (get_last_value): Likewise.
4761 * convert.c (convert_to_integer_1): Likewise.
4762 * cse.c (cse_insn): Likewise.
4763 * expr.c (expand_expr_real_1): Likewise.
4764 * lra-constraints.c (simplify_operand_subreg): Likewise.
4765 * optabs-query.c (can_atomic_load_p): Likewise.
4766 * optabs.c (expand_atomic_load): Likewise.
4767 (expand_atomic_store): Likewise.
4768 * ree.c (combine_reaching_defs): Likewise.
4769 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4770 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4771 * tree.h (type_has_mode_precision_p): Likewise.
4772 * ubsan.c (instrument_si_overflow): Likewise.
4774 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4775 Alan Hayward <alan.hayward@arm.com>
4776 David Sherwood <david.sherwood@arm.com>
4778 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4779 polynomial numbers of units.
4780 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4781 (valid_vector_subparts_p): New function.
4782 (build_vector_type): Remove temporary shim and take the number
4783 of units as a poly_uint64 rather than an int.
4784 (build_opaque_vector_type): Take the number of units as a
4785 poly_uint64 rather than an int.
4786 * tree.c (build_vector_from_ctor): Handle polynomial
4787 TYPE_VECTOR_SUBPARTS.
4788 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4789 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4790 (build_vector_from_val): If the number of units is variable,
4791 use build_vec_duplicate_cst for constant operands and
4792 VEC_DUPLICATE_EXPR otherwise.
4793 (make_vector_type): Remove temporary is_constant ().
4794 (build_vector_type, build_opaque_vector_type): Take the number of
4795 units as a poly_uint64 rather than an int.
4796 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4798 * cfgexpand.c (expand_debug_expr): Likewise.
4799 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4800 (store_constructor, expand_expr_real_1): Likewise.
4801 (const_scalar_mask_from_tree): Likewise.
4802 * fold-const-call.c (fold_const_reduction): Likewise.
4803 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4804 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4805 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4806 (fold_relational_const): Likewise.
4807 (native_interpret_vector): Likewise. Change the size from an
4808 int to an unsigned int.
4809 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4810 TYPE_VECTOR_SUBPARTS.
4811 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4812 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4813 duplicating a non-constant operand into a variable-length vector.
4814 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4815 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4816 * ipa-icf.c (sem_variable::equals): Likewise.
4817 * match.pd: Likewise.
4818 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4819 * print-tree.c (print_node): Likewise.
4820 * stor-layout.c (layout_type): Likewise.
4821 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4822 * tree-cfg.c (verify_gimple_comparison): Likewise.
4823 (verify_gimple_assign_binary): Likewise.
4824 (verify_gimple_assign_ternary): Likewise.
4825 (verify_gimple_assign_single): Likewise.
4826 * tree-pretty-print.c (dump_generic_node): Likewise.
4827 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4828 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4829 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4830 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4831 (vect_shift_permute_load_chain): Likewise.
4832 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4833 (expand_vector_condition, optimize_vector_constructor): Likewise.
4834 (lower_vec_perm, get_compute_type): Likewise.
4835 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4836 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4837 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4838 (vect_recog_mask_conversion_pattern): Likewise.
4839 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4840 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4841 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4842 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4843 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4844 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4845 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4846 (supportable_widening_operation): Likewise.
4847 (supportable_narrowing_operation): Likewise.
4848 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4850 * varasm.c (output_constant): Likewise.
4852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4853 Alan Hayward <alan.hayward@arm.com>
4854 David Sherwood <david.sherwood@arm.com>
4856 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4857 so that both the length == 3 and length != 3 cases set up their
4858 own permute vectors. Add comments explaining why we know the
4859 number of elements is constant.
4860 (vect_permute_load_chain): Likewise.
4862 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4863 Alan Hayward <alan.hayward@arm.com>
4864 David Sherwood <david.sherwood@arm.com>
4866 * machmode.h (mode_nunits): Change from unsigned char to
4868 (ONLY_FIXED_SIZE_MODES): New macro.
4869 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4870 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4871 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4873 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4874 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4875 or if measurement_type is not polynomial.
4876 * genmodes.c (ZERO_COEFFS): New macro.
4877 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4879 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4880 Use ZERO_COEFFS when emitting initializers.
4881 * data-streamer.h (bp_pack_poly_value): New function.
4882 (bp_unpack_poly_value): Likewise.
4883 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4884 for GET_MODE_NUNITS.
4885 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4886 for GET_MODE_NUNITS.
4887 * tree.c (make_vector_type): Remove temporary shim and make
4888 the real function take the number of units as a poly_uint64
4890 (build_vector_type_for_mode): Handle polynomial nunits.
4891 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4892 * emit-rtl.c (const_vec_series_p_1): Likewise.
4893 (gen_rtx_CONST_VECTOR): Likewise.
4894 * fold-const.c (test_vec_duplicate_folding): Likewise.
4895 * genrecog.c (validate_pattern): Likewise.
4896 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4897 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4898 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4899 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4900 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4901 * rtlanal.c (subreg_get_info): Likewise.
4902 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4903 (vect_grouped_load_supported): Likewise.
4904 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4905 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4906 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4907 (simplify_const_unary_operation, simplify_binary_operation_1)
4908 (simplify_const_binary_operation, simplify_ternary_operation)
4909 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4910 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4911 instead of CONST_VECTOR_NUNITS.
4912 * varasm.c (output_constant_pool_2): Likewise.
4913 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4914 explicit-encoded elements in the XVEC for variable-length vectors.
4916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4918 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4921 Alan Hayward <alan.hayward@arm.com>
4922 David Sherwood <david.sherwood@arm.com>
4924 * coretypes.h (fixed_size_mode): Declare.
4925 (fixed_size_mode_pod): New typedef.
4926 * builtins.h (target_builtins::x_apply_args_mode)
4927 (target_builtins::x_apply_result_mode): Change type to
4928 fixed_size_mode_pod.
4929 * builtins.c (apply_args_size, apply_result_size, result_vector)
4930 (expand_builtin_apply_args_1, expand_builtin_apply)
4931 (expand_builtin_return): Update accordingly.
4933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4935 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4936 * cselib.c (cselib_hash_rtx): Likewise.
4937 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4938 CONST_VECTOR encoding.
4940 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4941 Jeff Law <law@redhat.com>
4944 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4945 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4946 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4947 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4950 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4951 explicitly probe *sp in a noreturn function if there were any callee
4952 register saves or frame pointer is needed.
4954 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4957 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4958 BLKmode for ternary, binary or unary expressions.
4961 * var-tracking.c (delete_vta_debug_insn): New inline function.
4962 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4963 insns from get_insns () to NULL instead of each bb separately.
4964 Use delete_vta_debug_insn. No longer static.
4965 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4966 delete_vta_debug_insns callers.
4967 * rtl.h (delete_vta_debug_insns): Declare.
4968 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4969 instead of variable_tracking_main.
4971 2018-01-03 Martin Sebor <msebor@redhat.com>
4973 PR tree-optimization/83603
4974 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4975 arguments past the endof the argument list in functions declared
4976 without a prototype.
4977 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4978 Avoid checking when arguments are null.
4980 2018-01-03 Martin Sebor <msebor@redhat.com>
4983 * doc/extend.texi (attribute const): Fix a typo.
4984 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4985 issuing -Wsuggest-attribute for void functions.
4987 2018-01-03 Martin Sebor <msebor@redhat.com>
4989 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4990 offset_int::from instead of wide_int::to_shwi.
4991 (maybe_diag_overlap): Remove assertion.
4992 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4993 * gimple-ssa-sprintf.c (format_directive): Same.
4994 (parse_directive): Same.
4995 (sprintf_dom_walker::compute_format_length): Same.
4996 (try_substitute_return_value): Same.
4998 2018-01-03 Jeff Law <law@redhat.com>
5001 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5002 non-constant residual for zero at runtime and avoid probing in
5003 that case. Reorganize code for trailing problem to mirror handling
5006 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5008 PR tree-optimization/83501
5009 * tree-ssa-strlen.c (get_string_cst): New.
5010 (handle_char_store): Call get_string_cst.
5012 2018-01-03 Martin Liska <mliska@suse.cz>
5014 PR tree-optimization/83593
5015 * tree-ssa-strlen.c: Include tree-cfg.h.
5016 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5017 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5018 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5020 (strlen_dom_walker::before_dom_children): Call
5021 gimple_purge_dead_eh_edges. Dump tranformation with details
5023 (strlen_dom_walker::before_dom_children): Update call by adding
5024 new argument cleanup_eh.
5025 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5027 2018-01-03 Martin Liska <mliska@suse.cz>
5030 * cif-code.def (VARIADIC_THUNK): New enum value.
5031 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5034 2018-01-03 Jan Beulich <jbeulich@suse.com>
5036 * sse.md (mov<mode>_internal): Tighten condition for when to use
5037 vmovdqu<ssescalarsize> for TI and OI modes.
5039 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5041 Update copyright years.
5043 2018-01-03 Martin Liska <mliska@suse.cz>
5046 * ipa-visibility.c (function_and_variable_visibility): Skip
5047 functions with noipa attribure.
5049 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5051 * gcc.c (process_command): Update copyright notice dates.
5052 * gcov-dump.c (print_version): Ditto.
5053 * gcov.c (print_version): Ditto.
5054 * gcov-tool.c (print_version): Ditto.
5055 * gengtype.c (create_file): Ditto.
5056 * doc/cpp.texi: Bump @copying's copyright year.
5057 * doc/cppinternals.texi: Ditto.
5058 * doc/gcc.texi: Ditto.
5059 * doc/gccint.texi: Ditto.
5060 * doc/gcov.texi: Ditto.
5061 * doc/install.texi: Ditto.
5062 * doc/invoke.texi: Ditto.
5064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5066 * vector-builder.h (vector_builder::m_full_nelts): Change from
5067 unsigned int to poly_uint64.
5068 (vector_builder::full_nelts): Update prototype accordingly.
5069 (vector_builder::new_vector): Likewise.
5070 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5071 (vector_builder::operator ==): Likewise.
5072 (vector_builder::finalize): Likewise.
5073 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5074 Take the number of elements as a poly_uint64 rather than an
5076 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5077 from unsigned int to poly_uint64.
5078 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5079 (vec_perm_indices::new_vector): Likewise.
5080 (vec_perm_indices::length): Likewise.
5081 (vec_perm_indices::nelts_per_input): Likewise.
5082 (vec_perm_indices::input_nelts): Likewise.
5083 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5084 number of elements per input as a poly_uint64 rather than an
5085 unsigned int. Use the original encoding for variable-length
5086 vectors, rather than clamping each individual element.
5087 For the second and subsequent elements in each pattern,
5088 clamp the step and base before clamping their sum.
5089 (vec_perm_indices::series_p): Handle polynomial element counts.
5090 (vec_perm_indices::all_in_range_p): Likewise.
5091 (vec_perm_indices_to_tree): Likewise.
5092 (vec_perm_indices_to_rtx): Likewise.
5093 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5094 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5095 (tree_vector_builder::new_binary_operation): Handle polynomial
5096 element counts. Return false if we need to know the number
5097 of elements at compile time.
5098 * fold-const.c (fold_vec_perm): Punt if the number of elements
5099 isn't known at compile time.
5101 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5103 * vec-perm-indices.h (vec_perm_builder): Change element type
5104 from HOST_WIDE_INT to poly_int64.
5105 (vec_perm_indices::element_type): Update accordingly.
5106 (vec_perm_indices::clamp): Handle polynomial element_types.
5107 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5108 (vec_perm_indices::all_in_range_p): Likewise.
5109 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5111 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5112 polynomial vec_perm_indices element types.
5113 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5114 * fold-const.c (fold_vec_perm): Likewise.
5115 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5116 * tree-vect-generic.c (lower_vec_perm): Likewise.
5117 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5118 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5119 element type to HOST_WIDE_INT.
5121 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5122 Alan Hayward <alan.hayward@arm.com>
5123 David Sherwood <david.sherwood@arm.com>
5125 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5126 rather than an int. Use plus_constant.
5127 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5128 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5131 Alan Hayward <alan.hayward@arm.com>
5132 David Sherwood <david.sherwood@arm.com>
5134 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5135 a HOST_WIDE_INT to a poly_int64.
5137 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5138 Alan Hayward <alan.hayward@arm.com>
5139 David Sherwood <david.sherwood@arm.com>
5141 * calls.c (load_register_parameters): Cope with polynomial
5142 mode sizes. Require a constant size for BLKmode parameters
5143 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5144 forces a parameter to be padded at the lsb end in order to
5145 fill a complete number of words, require the parameter size
5146 to be ordered wrt UNITS_PER_WORD.
5148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5149 Alan Hayward <alan.hayward@arm.com>
5150 David Sherwood <david.sherwood@arm.com>
5152 * reload1.c (spill_stack_slot_width): Change element type
5153 from unsigned int to poly_uint64_pod.
5154 (alter_reg): Treat mode sizes as polynomial.
5156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5157 Alan Hayward <alan.hayward@arm.com>
5158 David Sherwood <david.sherwood@arm.com>
5160 * reload.c (complex_word_subreg_p): New function.
5161 (reload_inner_reg_of_subreg, push_reload): Use it.
5163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5164 Alan Hayward <alan.hayward@arm.com>
5165 David Sherwood <david.sherwood@arm.com>
5167 * lra-constraints.c (process_alt_operands): Reject matched
5168 operands whose sizes aren't ordered.
5169 (match_reload): Refer to this check here.
5171 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5172 Alan Hayward <alan.hayward@arm.com>
5173 David Sherwood <david.sherwood@arm.com>
5175 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5176 that the mode size is in the set {1, 2, 4, 8, 16}.
5178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5179 Alan Hayward <alan.hayward@arm.com>
5180 David Sherwood <david.sherwood@arm.com>
5182 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5183 Use plus_constant instead of gen_rtx_PLUS.
5185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5186 Alan Hayward <alan.hayward@arm.com>
5187 David Sherwood <david.sherwood@arm.com>
5189 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5190 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5191 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5192 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5193 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5194 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5195 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5196 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5197 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5198 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5200 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5201 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5202 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5203 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5204 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5205 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5206 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5207 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5208 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5209 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5211 * expr.c (emit_move_resolve_push): Treat the input and result
5212 of PUSH_ROUNDING as a poly_int64.
5213 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5214 (emit_push_insn): Likewise.
5215 * lra-eliminations.c (mark_not_eliminable): Likewise.
5216 * recog.c (push_operand): Likewise.
5217 * reload1.c (elimination_effects): Likewise.
5218 * rtlanal.c (nonzero_bits1): Likewise.
5219 * calls.c (store_one_arg): Likewise. Require the padding to be
5220 known at compile time.
5222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5223 Alan Hayward <alan.hayward@arm.com>
5224 David Sherwood <david.sherwood@arm.com>
5226 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5227 Use plus_constant instead of gen_rtx_PLUS.
5229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5230 Alan Hayward <alan.hayward@arm.com>
5231 David Sherwood <david.sherwood@arm.com>
5233 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5236 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5237 Alan Hayward <alan.hayward@arm.com>
5238 David Sherwood <david.sherwood@arm.com>
5240 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5241 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5242 via stack temporaries. Treat the mode size as polynomial too.
5244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5245 Alan Hayward <alan.hayward@arm.com>
5246 David Sherwood <david.sherwood@arm.com>
5248 * expr.c (expand_expr_real_2): When handling conversions involving
5249 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5250 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5251 as a poly_uint64 too.
5253 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5254 Alan Hayward <alan.hayward@arm.com>
5255 David Sherwood <david.sherwood@arm.com>
5257 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5260 Alan Hayward <alan.hayward@arm.com>
5261 David Sherwood <david.sherwood@arm.com>
5263 * combine.c (can_change_dest_mode): Handle polynomial
5264 REGMODE_NATURAL_SIZE.
5265 * expmed.c (store_bit_field_1): Likewise.
5266 * expr.c (store_constructor): Likewise.
5267 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5268 and polynomial REGMODE_NATURAL_SIZE.
5269 (gen_lowpart_common): Likewise.
5270 * reginfo.c (record_subregs_of_mode): Likewise.
5271 * rtlanal.c (read_modify_subreg_p): Likewise.
5273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5274 Alan Hayward <alan.hayward@arm.com>
5275 David Sherwood <david.sherwood@arm.com>
5277 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5278 numbers of elements.
5280 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5281 Alan Hayward <alan.hayward@arm.com>
5282 David Sherwood <david.sherwood@arm.com>
5284 * match.pd: Cope with polynomial numbers of vector elements.
5286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5287 Alan Hayward <alan.hayward@arm.com>
5288 David Sherwood <david.sherwood@arm.com>
5290 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5291 in a POINTER_PLUS_EXPR.
5293 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5294 Alan Hayward <alan.hayward@arm.com>
5295 David Sherwood <david.sherwood@arm.com>
5297 * omp-simd-clone.c (simd_clone_subparts): New function.
5298 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5299 (ipa_simd_modify_function_body): Likewise.
5301 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5302 Alan Hayward <alan.hayward@arm.com>
5303 David Sherwood <david.sherwood@arm.com>
5305 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5306 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5307 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5308 (expand_vector_condition, vector_element): Likewise.
5309 (subparts_gt): New function.
5310 (get_compute_type): Use subparts_gt.
5311 (count_type_subparts): Delete.
5312 (expand_vector_operations_1): Use subparts_gt instead of
5313 count_type_subparts.
5315 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5316 Alan Hayward <alan.hayward@arm.com>
5317 David Sherwood <david.sherwood@arm.com>
5319 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5320 (vect_compile_time_alias): ...this new function. Do the calculation
5321 on poly_ints rather than trees.
5322 (vect_prune_runtime_alias_test_list): Update call accordingly.
5324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5325 Alan Hayward <alan.hayward@arm.com>
5326 David Sherwood <david.sherwood@arm.com>
5328 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5330 (vect_schedule_slp_instance): Likewise.
5332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5333 Alan Hayward <alan.hayward@arm.com>
5334 David Sherwood <david.sherwood@arm.com>
5336 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5337 constant and extern definitions for variable-length vectors.
5338 (vect_get_constant_vectors): Note that the number of units
5339 is known to be constant.
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 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5346 of units as polynomial. Choose between WIDE and NARROW based
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 * tree-vect-stmts.c (simd_clone_subparts): New function.
5354 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5361 vectors as polynomial. Use build_index_vector for
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 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5369 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5370 for variable-length vectors.
5371 (vectorizable_mask_load_store): Treat the number of units as
5372 polynomial, asserting that it is constant if the condition has
5373 already been enforced.
5374 (vectorizable_store, vectorizable_load): Likewise.
5376 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5377 Alan Hayward <alan.hayward@arm.com>
5378 David Sherwood <david.sherwood@arm.com>
5380 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5381 of units as polynomial. Punt if we can't tell at compile time
5382 which vector contains the final result.
5384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5385 Alan Hayward <alan.hayward@arm.com>
5386 David Sherwood <david.sherwood@arm.com>
5388 * tree-vect-loop.c (vectorizable_induction): Treat the number
5389 of units as polynomial. Punt on SLP inductions. Use an integer
5390 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5391 cast of such a series for variable-length floating-point
5394 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5395 Alan Hayward <alan.hayward@arm.com>
5396 David Sherwood <david.sherwood@arm.com>
5398 * tree.h (build_index_vector): Declare.
5399 * tree.c (build_index_vector): New function.
5400 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5401 of units as polynomial, forcibly converting it to a constant if
5402 vectorizable_reduction has already enforced the condition.
5403 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5404 to create a {1,2,3,...} vector.
5405 (vectorizable_reduction): Treat the number of units as polynomial.
5406 Choose vectype_in based on the largest scalar element size rather
5407 than the smallest number of units. Enforce the restrictions
5410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5411 Alan Hayward <alan.hayward@arm.com>
5412 David Sherwood <david.sherwood@arm.com>
5414 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5415 number of units as polynomial.
5417 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5418 Alan Hayward <alan.hayward@arm.com>
5419 David Sherwood <david.sherwood@arm.com>
5421 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5422 * target.def (autovectorize_vector_sizes): Return the vector sizes
5423 by pointer, using vector_sizes rather than a bitmask.
5424 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5425 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5426 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5428 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5429 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5430 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5431 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5432 * omp-general.c (omp_max_vf): Likewise.
5433 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5434 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5435 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5436 * tree-vect-slp.c (vect_slp_bb): Likewise.
5437 * doc/tm.texi: Regenerate.
5438 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5440 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5441 the vector size as a poly_uint64 rather than an unsigned int.
5442 (current_vector_size): Change from an unsigned int to a poly_uint64.
5443 (get_vectype_for_scalar_type): Update accordingly.
5444 * tree.h (build_truth_vector_type): Take the size and number of
5445 units as a poly_uint64 rather than an unsigned int.
5446 (build_vector_type): Add a temporary overload that takes
5447 the number of units as a poly_uint64 rather than an unsigned int.
5448 * tree.c (make_vector_type): Likewise.
5449 (build_truth_vector_type): Take the number of units as a poly_uint64
5450 rather than an unsigned int.
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 * target.def (get_mask_mode): Take the number of units and length
5457 as poly_uint64s rather than unsigned ints.
5458 * targhooks.h (default_get_mask_mode): Update accordingly.
5459 * targhooks.c (default_get_mask_mode): Likewise.
5460 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5461 * doc/tm.texi: Regenerate.
5463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5464 Alan Hayward <alan.hayward@arm.com>
5465 David Sherwood <david.sherwood@arm.com>
5467 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5468 * omp-general.c (omp_max_vf): Likewise.
5469 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5470 (expand_omp_simd): Handle polynomial safelen.
5471 * omp-low.c (omplow_simd_context): Add a default constructor.
5472 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5473 (lower_rec_simd_input_clauses): Update accordingly.
5474 (lower_rec_input_clauses): Likewise.
5476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5477 Alan Hayward <alan.hayward@arm.com>
5478 David Sherwood <david.sherwood@arm.com>
5480 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5481 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5482 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5483 (vect_analyze_slp_cost): Likewise.
5484 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5485 (vect_model_load_cost): Likewise.
5487 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5488 Alan Hayward <alan.hayward@arm.com>
5489 David Sherwood <david.sherwood@arm.com>
5491 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5492 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5493 from an unsigned int * to a poly_uint64_pod *.
5494 (calculate_unrolling_factor): New function.
5495 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5497 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5498 Alan Hayward <alan.hayward@arm.com>
5499 David Sherwood <david.sherwood@arm.com>
5501 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5502 from an unsigned int to a poly_uint64.
5503 (_loop_vec_info::slp_unrolling_factor): Likewise.
5504 (_loop_vec_info::vectorization_factor): Change from an int
5506 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5507 (vect_get_num_vectors): New function.
5508 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5509 (vect_get_num_copies): Use vect_get_num_vectors.
5510 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5511 to an unsigned int *.
5512 (vect_analyze_data_refs): Change min_vf from an int * to a
5514 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5515 than an unsigned HOST_WIDE_INT.
5516 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5517 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5518 to an unsigned int *.
5519 (vect_analyze_data_ref_dependences): Likewise.
5520 (vect_compute_data_ref_alignment): Handle polynomial vf.
5521 (vect_enhance_data_refs_alignment): Likewise.
5522 (vect_prune_runtime_alias_test_list): Likewise.
5523 (vect_shift_permute_load_chain): Likewise.
5524 (vect_supportable_dr_alignment): Likewise.
5525 (dependence_distance_ge_vf): Take the vectorization factor as a
5526 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5527 (vect_analyze_data_refs): Change min_vf from an int * to a
5529 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5530 vfm1 as a poly_uint64 rather than an int. Make the same change
5531 for the returned bound_scalar.
5532 (vect_gen_vector_loop_niters): Handle polynomial vf.
5533 (vect_do_peeling): Likewise. Update call to
5534 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5535 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5537 * tree-vect-loop.c (vect_determine_vectorization_factor)
5538 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5539 (vect_get_known_peeling_cost): Likewise.
5540 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5541 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5542 (vect_transform_loop): Likewise. Use the lowest possible VF when
5543 updating the upper bounds of the loop.
5544 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5546 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5547 polynomial unroll factors.
5548 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5549 (vect_make_slp_decision): Likewise.
5550 (vect_supported_load_permutation_p): Likewise, and polynomial
5552 (vect_analyze_slp_cost): Handle polynomial vf.
5553 (vect_slp_analyze_node_operations): Likewise.
5554 (vect_slp_analyze_bb_1): Likewise.
5555 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5556 than an unsigned HOST_WIDE_INT.
5557 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5558 (vectorizable_load): Handle polynomial vf.
5559 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5561 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5563 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5564 Alan Hayward <alan.hayward@arm.com>
5565 David Sherwood <david.sherwood@arm.com>
5567 * match.pd: Handle bit operations involving three constants
5568 and try to fold one pair.
5570 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5572 * tree-vect-loop-manip.c: Include gimple-fold.h.
5573 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5574 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5575 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5576 Add a path that uses a step of VF instead of 1, but disable it
5578 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5579 and niters_no_overflow parameters. Update calls to
5580 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5581 Create a new SSA name if the latter choses to use a ste other
5582 than zero, and return it via niters_vector_mult_vf_var.
5583 * tree-vect-loop.c (vect_transform_loop): Update calls to
5584 vect_do_peeling, vect_gen_vector_loop_niters and
5585 slpeel_make_loop_iterate_ntimes.
5586 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5587 (vect_gen_vector_loop_niters): Update declarations after above changes.
5589 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5591 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5592 128-bit round to integer instructions.
5593 (ceil<mode>2): Likewise.
5594 (btrunc<mode>2): Likewise.
5595 (round<mode>2): Likewise.
5597 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5599 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5600 unaligned VSX load/store on P8/P9.
5601 (expand_block_clear): Allow the use of unaligned VSX
5602 load/store on P8/P9.
5604 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5606 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5608 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5609 swap associated with both a load and a store.
5611 2018-01-02 Andrew Waterman <andrew@sifive.com>
5613 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5614 * config/riscv/riscv.md (clear_cache): Use it.
5616 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5618 * web.c: Remove out-of-date comment.
5620 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5622 * expr.c (fixup_args_size_notes): Check that any existing
5623 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5624 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5625 (emit_single_push_insn): ...here.
5627 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5629 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5630 (const_vector_encoded_nelts): New function.
5631 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5632 (const_vector_int_elt, const_vector_elt): Declare.
5633 * emit-rtl.c (const_vector_int_elt_1): New function.
5634 (const_vector_elt): Likewise.
5635 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5636 of CONST_VECTOR_ELT.
5638 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5640 * expr.c: Include rtx-vector-builder.h.
5641 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5642 directly on the tree encoding.
5643 (const_vector_from_tree): Likewise.
5644 * optabs.c: Include rtx-vector-builder.h.
5645 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5646 sequence of "u" values.
5647 * vec-perm-indices.c: Include rtx-vector-builder.h.
5648 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5649 directly on the vec_perm_indices encoding.
5651 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5653 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5654 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5655 * rtx-vector-builder.h: New file.
5656 * rtx-vector-builder.c: Likewise.
5657 * rtl.h (rtx_def::u2): Add a const_vector field.
5658 (CONST_VECTOR_NPATTERNS): New macro.
5659 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5660 (CONST_VECTOR_DUPLICATE_P): Likewise.
5661 (CONST_VECTOR_STEPPED_P): Likewise.
5662 (CONST_VECTOR_ENCODED_ELT): Likewise.
5663 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5664 (unwrap_const_vec_duplicate): Likewise.
5665 (const_vec_series_p): Check for a non-duplicated vector encoding.
5666 Say that the function only returns true for integer vectors.
5667 * emit-rtl.c: Include rtx-vector-builder.h.
5668 (gen_const_vec_duplicate_1): Delete.
5669 (gen_const_vector): Call gen_const_vec_duplicate instead of
5670 gen_const_vec_duplicate_1.
5671 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5672 (gen_const_vec_duplicate): Use rtx_vector_builder.
5673 (gen_const_vec_series): Likewise.
5674 (gen_rtx_CONST_VECTOR): Likewise.
5675 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5676 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5677 Build a new vector rather than modifying a CONST_VECTOR in-place.
5678 (handle_special_swappables): Update call accordingly.
5679 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5680 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5681 Build a new vector rather than modifying a CONST_VECTOR in-place.
5682 (handle_special_swappables): Update call accordingly.
5684 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5686 * simplify-rtx.c (simplify_const_binary_operation): Use
5687 CONST_VECTOR_ELT instead of XVECEXP.
5689 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5691 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5692 the selector elements to be different from the data elements
5693 if the selector is a VECTOR_CST.
5694 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5695 ssizetype for the selector.
5697 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5699 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5700 before testing each element individually.
5701 * tree-vect-generic.c (lower_vec_perm): Likewise.
5703 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5705 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5706 * selftest-run-tests.c (selftest::run_tests): Call it.
5707 * vector-builder.h (vector_builder::operator ==): New function.
5708 (vector_builder::operator !=): Likewise.
5709 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5710 (vec_perm_indices::all_from_input_p): New function.
5711 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5712 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5713 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5714 instead of reading the VECTOR_CST directly. Detect whether both
5715 vector inputs are the same before constructing the vec_perm_indices,
5716 and update the number of inputs argument accordingly. Use the
5717 utility functions added above. Only construct sel2 if we need to.
5719 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5721 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5722 the broadcast of the low byte.
5723 (expand_mult_highpart): Use an explicit encoding for the permutes.
5724 * optabs-query.c (can_mult_highpart_p): Likewise.
5725 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5726 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5727 (vectorizable_bswap): Likewise.
5728 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5729 explicit encoding for the power-of-2 permutes.
5730 (vect_permute_store_chain): Likewise.
5731 (vect_grouped_load_supported): Likewise.
5732 (vect_permute_load_chain): Likewise.
5734 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5736 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5737 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5738 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5739 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5740 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5741 (vect_gen_perm_mask_any): Likewise.
5743 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5745 * int-vector-builder.h: New file.
5746 * vec-perm-indices.h: Include int-vector-builder.h.
5747 (vec_perm_indices): Redefine as an int_vector_builder.
5748 (auto_vec_perm_indices): Delete.
5749 (vec_perm_builder): Redefine as a stand-alone class.
5750 (vec_perm_indices::vec_perm_indices): New function.
5751 (vec_perm_indices::clamp): Likewise.
5752 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5753 (vec_perm_indices::new_vector): New function.
5754 (vec_perm_indices::new_expanded_vector): Update for new
5755 vec_perm_indices class.
5756 (vec_perm_indices::rotate_inputs): New function.
5757 (vec_perm_indices::all_in_range_p): Operate directly on the
5758 encoded form, without computing elided elements.
5759 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5760 encoding. Update for new vec_perm_indices class.
5761 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5762 the given vec_perm_builder.
5763 (expand_vec_perm_var): Update vec_perm_builder constructor.
5764 (expand_mult_highpart): Use vec_perm_builder instead of
5765 auto_vec_perm_indices.
5766 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5767 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5768 or double series encoding as appropriate.
5769 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5770 vec_perm_indices instead of auto_vec_perm_indices.
5771 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5772 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5773 (vect_permute_store_chain): Likewise.
5774 (vect_grouped_load_supported): Likewise.
5775 (vect_permute_load_chain): Likewise.
5776 (vect_shift_permute_load_chain): Likewise.
5777 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5778 (vect_transform_slp_perm_load): Likewise.
5779 (vect_schedule_slp_instance): Likewise.
5780 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5781 (vectorizable_mask_load_store): Likewise.
5782 (vectorizable_bswap): Likewise.
5783 (vectorizable_store): Likewise.
5784 (vectorizable_load): Likewise.
5785 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5786 vec_perm_indices instead of auto_vec_perm_indices. Use
5787 tree_to_vec_perm_builder to read the vector from a tree.
5788 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5789 vec_perm_builder instead of a vec_perm_indices.
5790 (have_whole_vector_shift): Use vec_perm_builder and
5791 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5792 truncation to calc_vec_perm_mask_for_shift.
5793 (vect_create_epilog_for_reduction): Likewise.
5794 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5795 from auto_vec_perm_indices to vec_perm_indices.
5796 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5797 instead of changing individual elements.
5798 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5799 the vector in d.perm.
5800 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5801 from auto_vec_perm_indices to vec_perm_indices.
5802 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5803 instead of changing individual elements.
5804 (arm_vectorize_vec_perm_const): Use new_vector to install
5805 the vector in d.perm.
5806 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5807 Update vec_perm_builder constructor.
5808 (rs6000_expand_interleave): Likewise.
5809 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5810 (rs6000_expand_interleave): Likewise.
5812 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5814 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5815 to qimode could truncate the indices.
5816 * optabs.c (expand_vec_perm_var): Likewise.
5818 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5820 * Makefile.in (OBJS): Add vec-perm-indices.o.
5821 * vec-perm-indices.h: New file.
5822 * vec-perm-indices.c: Likewise.
5823 * target.h (vec_perm_indices): Replace with a forward class
5825 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5826 * optabs.h: Include vec-perm-indices.h.
5827 (expand_vec_perm): Delete.
5828 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5829 (expand_vec_perm_const): Declare.
5830 * target.def (vec_perm_const_ok): Replace with...
5831 (vec_perm_const): ...this new hook.
5832 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5833 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5834 * doc/tm.texi: Regenerate.
5835 * optabs.def (vec_perm_const): Delete.
5836 * doc/md.texi (vec_perm_const): Likewise.
5837 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5838 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5839 expand_vec_perm for constant permutation vectors. Assert that
5840 the mode of variable permutation vectors is the integer equivalent
5841 of the mode that is being permuted.
5842 * optabs-query.h (selector_fits_mode_p): Declare.
5843 * optabs-query.c: Include vec-perm-indices.h.
5844 (selector_fits_mode_p): New function.
5845 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5846 is defined, instead of checking whether the vec_perm_const_optab
5847 exists. Use targetm.vectorize.vec_perm_const instead of
5848 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5849 fit in the vector mode before using a variable permute.
5850 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5851 vec_perm_indices instead of an rtx.
5852 (expand_vec_perm): Replace with...
5853 (expand_vec_perm_const): ...this new function. Take the selector
5854 as a vec_perm_indices rather than an rtx. Also take the mode of
5855 the selector. Update call to shift_amt_for_vec_perm_mask.
5856 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5857 Use vec_perm_indices::new_expanded_vector to expand the original
5858 selector into bytes. Check whether the indices fit in the vector
5859 mode before using a variable permute.
5860 (expand_vec_perm_var): Make global.
5861 (expand_mult_highpart): Use expand_vec_perm_const.
5862 * fold-const.c: Includes vec-perm-indices.h.
5863 * tree-ssa-forwprop.c: Likewise.
5864 * tree-vect-data-refs.c: Likewise.
5865 * tree-vect-generic.c: Likewise.
5866 * tree-vect-loop.c: Likewise.
5867 * tree-vect-slp.c: Likewise.
5868 * tree-vect-stmts.c: Likewise.
5869 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5871 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5872 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5873 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5874 (aarch64_vectorize_vec_perm_const): ...this new function.
5875 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5876 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5877 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5878 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5879 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5880 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5881 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5883 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5884 check for NEON modes.
5885 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5886 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5887 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5888 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5890 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5891 the old VEC_PERM_CONST conditions.
5892 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5893 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5894 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5895 (ia64_vectorize_vec_perm_const_ok): Merge into...
5896 (ia64_vectorize_vec_perm_const): ...this new function.
5897 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5898 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5899 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5900 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5901 * config/mips/mips.c (mips_expand_vec_perm_const)
5902 (mips_vectorize_vec_perm_const_ok): Merge into...
5903 (mips_vectorize_vec_perm_const): ...this new function.
5904 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5905 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5906 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5907 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5908 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5909 (rs6000_expand_vec_perm_const): Delete.
5910 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5912 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5913 (altivec_expand_vec_perm_const_le): Take each operand individually.
5914 Operate on constant selectors rather than rtxes.
5915 (altivec_expand_vec_perm_const): Likewise. Update call to
5916 altivec_expand_vec_perm_const_le.
5917 (rs6000_expand_vec_perm_const): Delete.
5918 (rs6000_vectorize_vec_perm_const_ok): Delete.
5919 (rs6000_vectorize_vec_perm_const): New function.
5920 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5921 an element count and rtx array.
5922 (rs6000_expand_extract_even): Update call accordingly.
5923 (rs6000_expand_interleave): Likewise.
5924 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5925 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5926 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5927 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5928 (rs6000_expand_vec_perm_const): Delete.
5929 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5930 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5931 (altivec_expand_vec_perm_const_le): Take each operand individually.
5932 Operate on constant selectors rather than rtxes.
5933 (altivec_expand_vec_perm_const): Likewise. Update call to
5934 altivec_expand_vec_perm_const_le.
5935 (rs6000_expand_vec_perm_const): Delete.
5936 (rs6000_vectorize_vec_perm_const_ok): Delete.
5937 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5938 reference to the SPE evmerge intructions.
5939 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5940 an element count and rtx array.
5941 (rs6000_expand_extract_even): Update call accordingly.
5942 (rs6000_expand_interleave): Likewise.
5943 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5944 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5946 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5948 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5950 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5951 vector mode and that that mode matches the mode of the data
5953 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5954 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5955 directly using expand_vec_perm_1 when forcing selectors into
5957 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5959 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5961 * optabs-query.h (can_vec_perm_p): Delete.
5962 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5963 * optabs-query.c (can_vec_perm_p): Split into...
5964 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5965 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5966 particular selector is valid.
5967 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5968 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5969 (vect_grouped_load_supported): Likewise.
5970 (vect_shift_permute_load_chain): Likewise.
5971 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5972 (vect_transform_slp_perm_load): Likewise.
5973 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5974 (vectorizable_bswap): Likewise.
5975 (vect_gen_perm_mask_checked): Likewise.
5976 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5977 implementations of variable permutation vectors into account
5978 when deciding which selector to use.
5979 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5980 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5981 with a false third argument.
5982 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5983 to test whether the constant selector is valid and can_vec_perm_var_p
5984 to test whether a variable selector is valid.
5986 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5988 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5989 * optabs-query.c (can_vec_perm_p): Likewise.
5990 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5991 instead of vec_perm_indices.
5992 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5993 (vect_gen_perm_mask_checked): Likewise,
5994 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5995 (vect_gen_perm_mask_checked): Likewise,
5997 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5999 * optabs-query.h (qimode_for_vec_perm): Declare.
6000 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6001 (qimode_for_vec_perm): ...this new function.
6002 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6004 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6006 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6007 does not have a conditional at the top.
6009 2018-01-02 Richard Biener <rguenther@suse.de>
6011 * ipa-inline.c (big_speedup_p): Fix expression.
6013 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6016 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6019 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6023 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6024 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6025 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6026 cond_taken_branch_cost 3->4.
6028 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6030 PR tree-optimization/83581
6031 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6032 TODO_cleanup_cfg if any changes have been made.
6035 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6036 convert_modes if target mode has the right side, but different mode
6040 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6041 last argument when extracting from CONCAT. If either from_real or
6042 from_imag is NULL, use expansion through memory. If result is not
6043 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6044 the parts directly to inner mode, if even that fails, use expansion
6048 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6049 check for bswap in mode rather than HImode and use that in expand_unop
6052 Copyright (C) 2018 Free Software Foundation, Inc.
6054 Copying and distribution of this file, with or without modification,
6055 are permitted in any medium without royalty provided the copyright
6056 notice and this notice are preserved.