1 2018-02-07 Jakub Jelinek <jakub@redhat.com>
3 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
4 *DIV_EXPR and *MOD_EXPR.
6 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
9 * config/i386/i386.c (ix86_option_override_internal): Mask out
10 the CF_SET bit when checking -fcf-protection.
12 2018-02-07 Tom de Vries <tom@codesourcery.com>
15 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
18 2018-02-07 Richard Biener <rguenther@suse.de>
20 PR tree-optimization/84204
21 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
24 PR tree-optimization/84205
25 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
26 special-case isl_ast_op_zdiv_r.
28 PR tree-optimization/84223
29 * graphite-scop-detection.c (gather_bbs::before_dom_children):
30 Only add conditions from within the region.
31 (gather_bbs::after_dom_children): Adjust.
33 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
36 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
37 * config/avr/avr.md: Only post-reload split REG-REG moves if
38 either register is REGERAL_REG_P.
40 2018-02-07 Jakub Jelinek <jakub@redhat.com>
42 PR tree-optimization/84235
43 * tree-ssa-scopedtables.c
44 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
45 if the subtraction is performed in floating point type where NaNs are
46 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
47 build 1. Formatting fix.
49 2018-02-06 Jakub Jelinek <jakub@redhat.com>
52 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
53 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
54 and skip it regardless of bb boundaries. Use CALL_P macro,
55 don't test INSN_P (insn) together with CALL_P or JUMP_P check
56 unnecessarily, formatting fix.
58 2018-02-06 Michael Collison <michael.collison@arm.com>
60 * config/arm/thumb2.md:
61 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
62 (*thumb_mov_notscc): Ditto.
64 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
67 * config/rs6000/rs6000.md (su code attribute): Use "u" for
68 unsigned_fix, not "s".
70 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
72 * configure.ac (gcc_fn_eh_frame_ro): New function.
73 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
74 correct .eh_frame permissions.
75 * configure: Regenerate.
77 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
79 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
82 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
84 * config/rs6000/rs6000.c (rs6000_option_override_internal):
85 Display warning message for -mno-speculate-indirect-jumps.
87 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
89 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
91 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
93 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
95 PR tree-optimization/84225
96 * tree-eh.c (find_trapping_overflow): Only call
97 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
99 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
102 * config/i386/i386.c: Reimplement the check of possible options
103 -mibt/-mshstk conbination. Change error messages.
104 * doc/invoke.texi: Fix a typo: remove extra '='.
106 2018-02-06 Marek Polacek <polacek@redhat.com>
108 PR tree-optimization/84228
109 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
111 2018-02-06 Tamar Christina <tamar.christina@arm.com>
114 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
115 emitted arch directives.
116 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
117 __ARM_FEATURE_COPROC before changing architectures.
119 2018-02-06 Richard Biener <rguenther@suse.de>
121 * config/i386/i386.c (print_reg): Fix typo.
122 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
124 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
126 * configure: Regenerate.
128 2018-02-05 Martin Sebor <msebor@redhat.com>
130 PR tree-optimization/83369
131 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
134 2018-02-05 Martin Liska <mliska@suse.cz>
136 * doc/invoke.texi: Cherry-pick upstream r323995.
138 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
140 * ira.c (ira_init_register_move_cost): Adjust comment.
142 2018-02-05 Martin Liska <mliska@suse.cz>
144 PR gcov-profile/84137
145 * doc/gcov.texi: Fix typo in documentation.
147 2018-02-05 Martin Liska <mliska@suse.cz>
149 PR gcov-profile/83879
150 * doc/gcov.texi: Document necessity of --dynamic-list-data when
151 using dlopen functionality.
153 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
155 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
156 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
157 _mm_maskz_range_ss, _mm_mask_range_round_ss,
158 _mm_maskz_range_round_ss): New intrinsics.
159 (__builtin_ia32_rangesd128_round)
160 (__builtin_ia32_rangess128_round): Remove.
161 (__builtin_ia32_rangesd128_mask_round,
162 __builtin_ia32_rangess128_mask_round): New builtins.
163 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
164 __builtin_ia32_rangess128_round): Remove.
165 (__builtin_ia32_rangesd128_mask_round,
166 __builtin_ia32_rangess128_mask_round): New builtins.
167 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
168 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
169 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
170 "<round_saeonly_constraint>")): Changed to ...
171 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
172 "<round_saeonly_scalar_constraint>")): ... this.
173 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
174 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
175 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
176 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
177 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
179 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
181 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
183 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
184 Remove all values except native, 8540 and 8548.
186 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
188 * config/i386/i386.c (ix86_output_function_return): Pass
189 INVALID_REGNUM, instead of -1, as invalid register number to
190 indirect_thunk_name and output_indirect_thunk.
192 2018-02-02 Julia Koval <julia.koval@intel.com>
194 * config.gcc: Add -march=icelake.
195 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
196 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
197 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
198 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
199 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
200 (processor_target_table): Add icelake.
201 (ix86_option_override_internal): Handle new PTAs.
202 (get_builtin_code_for_version): Handle icelake.
203 (M_INTEL_COREI7_ICELAKE): New.
204 (fold_builtin_cpu): Handle icelake.
205 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
206 * doc/invoke.texi: Add -march=icelake.
208 2018-02-02 Julia Koval <julia.koval@intel.com>
210 * config/i386/i386.c (ix86_option_override_internal): Change flags type
212 * wide-int-bitmask.h: New.
214 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
217 * config/i386/i386.md: Replace Pmode with word_mode in
218 builtin_setjmp_setup and builtin_longjmp to support x32.
220 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
224 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
226 (rs6000_supported_cpu_names): New static variable.
227 (linux_cpu_translation_table): Likewise.
228 (elf_platform) <cpu>: Define new static variable and use it.
229 Translate kernel AT_PLATFORM name to canonical name if needed.
230 Error if platform name is unknown.
232 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
235 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
237 2018-02-01 Jeff Law <law@redhat.com>
240 * config/i386/i386.c (release_scratch_register_on_entry): Add new
241 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
242 the scratch if RELEASE_VIA_POP is false.
243 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
244 If we have to save a temporary register, decrement SIZE appropriately.
245 Pass new arguments to release_scratch_register_on_entry.
246 (ix86_adjust_stack_and_probe): Likewise.
247 (ix86_emit_probe_stack_range): Pass new arguments to
248 release_scratch_register_on_entry.
250 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
252 PR rtl-optimization/84157
253 * combine.c (change_zero_ext): Use REG_P predicate in
254 front of HARD_REGISTER_P predicate.
256 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
258 * config/avr/avr.c (avr_option_override): Move disabling of
259 -fdelete-null-pointer-checks to...
260 * common/config/avr/avr-common.c (avr_option_optimization_table):
263 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
265 PR tree-optimization/81635
266 * tree-data-ref.c (split_constant_offset_1): For types that
267 wrap on overflow, try to use range info to prove that wrapping
270 2018-02-01 Renlin Li <renlin.li@arm.com>
273 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
275 (aarch64_register_move_cost): Likewise.
276 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
278 (REG_CLASS_NAMES): Likewise.
279 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
280 TAILCALL_ADDR_REGS. Remove IP registers.
281 * config/aarch64/aarch64.md (Ucs): Update register constraint.
283 2018-02-01 Richard Biener <rguenther@suse.de>
285 * domwalk.h (dom_walker::dom_walker): Add additional constructor
286 for specifying RPO order and allow NULL for that.
287 * domwalk.c (dom_walker::dom_walker): Likewise.
288 (dom_walker::walk): Handle NULL RPO order.
289 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
291 (rewrite_update_dom_walker): Likewise.
292 (mark_def_dom_walker): Likewise.
294 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
296 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
297 (aarch64_maybe_expand_sve_subreg_move): Declare.
298 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
299 * config/aarch64/predicates.md (aarch64_any_register_operand): New
301 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
302 that are semantically a reverse operation.
303 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
304 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
305 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
307 (aarch64_can_change_mode_class): For big-endian, forbid changes
308 between two SVE modes if they have different element sizes.
310 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
312 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
313 the TImode handling for big-endian targets.
315 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
317 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
318 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
320 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
321 Remove BSWAP handing for big-endian targets and use the form of
322 LD1RQ appropariate for the mode.
324 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
326 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
327 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
330 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
333 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
334 check for operands that need to go through aarch64_sve_reload_be.
336 2018-02-01 Jakub Jelinek <jakub@redhat.com>
338 PR tree-optimization/81661
339 PR tree-optimization/84117
340 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
341 * tree-eh.c: Include gimplify.h.
342 (find_trapping_overflow, replace_trapping_overflow,
343 rewrite_to_non_trapping_overflow): New functions.
344 * tree-vect-loop.c: Include tree-eh.h.
345 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
346 * tree-data-ref.c: Include tree-eh.h.
347 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
349 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
351 PR rtl-optimization/84123
352 * combine.c (change_zero_ext): Check if hard register satisfies
353 can_change_dest_mode before calling gen_lowpart_SUBREG.
355 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
358 * ira.c (ira_init_register_move_cost): Remove assert.
360 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
362 PR rtl-optimization/84071
363 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
364 * doc/tm.texi: Regenerate.
366 2018-01-31 Richard Biener <rguenther@suse.de>
368 PR tree-optimization/84132
369 * tree-data-ref.c (analyze_miv_subscript): Properly
370 check whether evolution_function_is_affine_multivariate_p
371 before calling gcd_of_steps_may_divide_p.
373 2018-01-31 Julia Koval <julia.koval@intel.com>
376 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
377 * config/i386/i386.md (rdpid_rex64) New.
378 (rdpid): Make 32bit only.
380 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
383 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
384 an IDENTIFIER_NODE for FUNCTION_TYPE's.
386 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
389 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
391 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
393 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
395 PR rtl-optimization/84071
396 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
397 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
399 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
401 * config/arc/arc.c (arc_handle_aux_attribute): New function.
402 (arc_attribute_table): Add 'aux' attribute.
403 (arc_in_small_data_p): Consider aux like variables.
404 (arc_is_aux_reg_p): New function.
405 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
406 (arc_get_aux_arg): New function.
407 (prepare_move_operands): Handle aux-register access.
408 (arc_handle_aux_attribute): New function.
409 * doc/extend.texi (ARC Variable attributes): Add subsection.
411 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
413 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
414 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
415 (arc_attribute_table): Add 'uncached' attribute.
416 (arc_print_operand): Print '.di' flag for uncached memory
418 (arc_in_small_data_p): Do not consider for small data the uncached
420 (arc_is_uncached_mem_p): New function.
421 * config/arc/predicates.md (compact_store_memory_operand): Check
422 for uncached memory accesses.
423 (nonvol_nonimm_operand): Likewise.
424 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
426 2018-01-31 Jakub Jelinek <jakub@redhat.com>
429 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
430 falign-loops=): Add Optimization flag.
432 2018-01-30 Jeff Law <law@redhat.com>
435 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
436 INT_REGISTERS_SAVED. Check it prior to calling
437 get_scratch_register_on_entry.
438 (ix86_adjust_stack_and_probe): Similarly.
439 (ix86_emit_probe_stack_range): Similarly.
440 (ix86_expand_prologue): Corresponding changes.
442 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
445 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
446 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
448 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
451 * lra-constraints.c (curr_insn_transform): Process AND in the
454 2018-01-30 Jakub Jelinek <jakub@redhat.com>
456 PR rtl-optimization/83986
457 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
458 dependence against last_pending_memory_flush in addition to
461 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
463 PR tree-optimization/81611
464 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
467 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
469 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
472 2018-01-30 Richard Biener <rguenther@suse.de>
473 Jakub Jelinek <jakub@redhat.com>
475 PR tree-optimization/84111
476 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
477 inner loops added during recursion, as they don't have up-to-date
480 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
483 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
484 (can_inline_edge_by_limits_p): ... here.
485 (can_early_inline_edge_p, check_callers,
486 update_caller_keys, update_callee_keys, recursive_inlining,
487 add_new_edges_to_heap, speculation_useful_p,
488 inline_small_functions,
489 inline_small_functions, flatten_function,
490 inline_to_all_callers_1): Update.
492 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
494 * profile-count.c (profile_count::combine_with_ipa_count): Handle
497 2018-01-30 Richard Biener <rguenther@suse.de>
499 PR tree-optimization/83008
500 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
501 invariant and constant vector uses in stmts when they need
504 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
507 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
508 * configure: Regenerate.
510 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
512 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
514 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
515 Use gen_rtx_REG rather than gen_lowpart.
517 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
519 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
520 rather than 0 when creating partial subregs.
522 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
524 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
527 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
530 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
531 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
532 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
533 flags. This restores the settings used before the 2017-07-24.
534 Turning off pre increment/decrement/modify allows IVOPTS to
535 optimize DF/SF loops where the index is an int.
537 2018-01-29 Richard Biener <rguenther@suse.de>
538 Kelvin Nilsen <kelvin@gcc.gnu.org>
541 * tree-vect-stmts.c (vectorizable_call): Don't call
542 targetm.vectorize_builtin_md_vectorized_function if callee is
545 2018-01-22 Carl Love <cel@us.ibm.com>
547 * doc/extend.tex: Fix typo in second arg in
548 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
550 2018-01-29 Richard Biener <rguenther@suse.de>
552 PR tree-optimization/84086
553 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
554 (flush_ssaname_freelist): When SSA names were released reset
557 2018-01-29 Richard Biener <rguenther@suse.de>
559 PR tree-optimization/84057
560 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
561 removed paths when removing edges.
563 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
565 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
566 -mfunction-return=@var{choice}.
568 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
571 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
573 (layout::print_source_line): Likewise.
574 (test_get_line_width_without_trailing_whitespace): Add test cases.
576 2018-01-27 Jakub Jelinek <jakub@redhat.com>
579 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
582 2018-01-26 Jim Wilson <jimw@sifive.com>
584 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
586 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
589 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
591 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
592 and CMP + SUB-immediate -> SUBS.
594 2018-01-26 Martin Sebor <msebor@redhat.com>
596 PR tree-optimization/83896
597 * tree-ssa-strlen.c (get_string_len): Rename...
598 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
599 Avoid assuming length is constant.
600 (handle_char_store): Use HOST_WIDE_INT for string length.
602 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
605 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
606 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
608 2018-01-26 Richard Biener <rguenther@suse.de>
610 PR rtl-optimization/84003
611 * dse.c (record_store): Only record redundant stores when
612 the earlier store aliases at least all accesses the later one does.
614 2018-01-26 Jakub Jelinek <jakub@redhat.com>
616 PR rtl-optimization/83985
617 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
618 REG_CFA_RESTORE insns.
619 (delete_unmarked_insns): Don't ignore separate shrink wrapping
620 REG_CFA_RESTORE insns here.
623 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
624 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
626 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
628 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
629 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
630 (arc_init): Likewise.
631 (arc_override_options): Likewise.
632 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
634 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
636 * config/arc/arc.h (TARGET_DBNZ): Define.
637 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
638 properly set the tune attribute.
639 (dbnz): Use TARGET_DBNZ guard.
640 * config/arc/arc.opt (mtune): Add core3 option.
642 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
644 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
645 recognize new pic like addresses.
646 (arc_delegitimize_address): Clean up.
648 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
650 * config/arc/arc-arches.def: Option mrf16 valid for all
652 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
653 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
654 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
655 * config/arc/arc-tables.opt: Regenerate.
656 * config/arc/arc.c (arc_conditional_register_usage): Handle
657 reduced register file case.
658 (arc_file_start): Set must have build attributes.
659 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
661 * config/arc/arc.opt (mrf16): Add new option.
662 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
663 * config/arc/genmultilib.awk: Handle new mrf16 option.
664 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
665 * config/arc/t-multilib: Regenerate.
666 * doc/invoke.texi (ARC Options): Document mrf16 option.
668 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
670 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
671 * config/arc/arc.c (arc_handle_secure_attribute): New function.
672 (arc_attribute_table): Add 'secure_call' attribute.
673 (arc_print_operand): Print secure call operand.
674 (arc_function_ok_for_sibcall): Don't optimize tail calls when
676 (arc_is_secure_call_p): New function. * config/arc/arc.md
677 (call_i): Add support for sjli instruction.
678 (call_value_i): Likewise.
679 * config/arc/constraints.md (Csc): New constraint.
681 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
682 John Eric Martin <John.Martin@emmicro-us.com>
684 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
685 * config/arc/arc.c (_arc_jli_section): New struct.
686 (arc_jli_section): New type.
687 (rc_jli_sections): New static variable.
688 (arc_handle_jli_attribute): New function.
689 (arc_attribute_table): Add jli_always and jli_fixed attribute.
690 (arc_file_end): New function.
691 (TARGET_ASM_FILE_END): Define.
692 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
693 (arc_add_jli_section): New function.
694 (jli_call_scan): Likewise.
695 (arc_reorg): Call jli_call_scan.
696 (arc_output_addsi): Remove 'S' from printing asm operand.
697 (arc_is_jli_call_p): New function.
698 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
700 (movhi_insn): Likewise.
701 (movsi_insn): Likewise.
702 (movsi_set_cc_insn): Likewise.
703 (loadqi_update): Likewise.
704 (load_zeroextendqisi_update): Likewise.
705 (load_signextendqisi_update): Likewise.
706 (loadhi_update): Likewise.
707 (load_zeroextendhisi_update): Likewise.
708 (load_signextendhisi_update): Likewise.
709 (loadsi_update): Likewise.
710 (loadsf_update): Likewise.
711 (movsicc_insn): Likewise.
712 (bset_insn): Likewise.
713 (bxor_insn): Likewise.
714 (bclr_insn): Likewise.
715 (bmsk_insn): Likewise.
716 (bicsi3_insn): Likewise.
717 (cmpsi_cc_c_insn): Likewise.
718 (movsi_ne): Likewise.
719 (movsi_cond_exec): Likewise.
720 (clrsbsi2): Likewise.
731 (call_i): Remove 'S' asm letter, add jli instruction.
732 (call_value_i): Likewise.
733 * config/arc/arc.op (mjli-always): New option.
734 * config/arc/constraints.md (Cji): New constraint.
735 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
737 (subsf3_fpx): Likewise.
738 (mulsf3_fpx): Likewise.
739 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
741 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
743 * doc/invoke.texi (ARC): Document mjli-always option.
745 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
747 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
748 and use incw and decw where possible
749 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
751 2018-01-26 Richard Biener <rguenther@suse.de>
753 PR tree-optimization/81082
754 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
755 association if it requires casting to unsigned.
756 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
757 from fold_plusminus_mult_expr to catch important cases late when
758 range info is available.
760 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
762 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
763 * configure.ac (hidden_linkonce): New test.
764 * configure: Regenerate.
765 * config.in: Regenerate.
767 2018-01-26 Julia Koval <julia.koval@intel.com>
769 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
770 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
771 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
772 _mm_mask_bitshuffle_epi64_mask): Fix type.
773 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
774 USI_FTYPE_V4DI_V4DI_USI): Remove.
775 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
776 __builtin_ia32_vpshufbitqmb256_mask,
777 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
778 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
779 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
781 2018-01-26 Alan Modra <amodra@gmail.com>
784 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
785 UNSPEC_VBPERMQ. Sort other unspecs.
787 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
789 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
791 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
794 * predict.c (drop_profile): Do not push/pop cfun; update also
796 (handle_missing_profiles): Fix logic looking for zero profiles.
798 2018-01-25 Jakub Jelinek <jakub@redhat.com>
801 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
802 on functions with #pragma omp declare simd or functions with simd
804 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
805 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
806 Remove trailing \n from warning_at calls.
808 2018-01-25 Tom de Vries <tom@codesourcery.com>
811 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
812 for neutered workers.
814 2018-01-24 Joseph Myers <joseph@codesourcery.com>
817 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
818 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
820 2018-01-24 Jeff Law <law@redhat.com>
823 * i386.c (get_probe_interval): Move to earlier point.
824 (ix86_compute_frame_layout): If -fstack-clash-protection and
825 the frame is larger than the probe interval, then use pushes
826 to save registers rather than reg->mem moves.
827 (ix86_expand_prologue): Remove conditional for int_registers_saved
830 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
833 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
834 min/max for never referenced object.
836 2018-01-24 Jakub Jelinek <jakub@redhat.com>
839 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
841 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
842 attributes from DECL_ATTRIBUTES (decl) without affecting
843 DECL_ATTRIBUTES (current_function_decl).
844 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
845 functions with non-NULL DECL_ABSTRACT_ORIGIN.
847 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
849 PR tree-optimization/83979
850 * fold-const.c (fold_comparison): Use constant_boolean_node
851 instead of boolean_{true,false}_node.
853 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
855 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
858 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
860 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
861 Simplify the clause that sets the length attribute.
862 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
863 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
864 clause that sets the length attribute.
865 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
867 2018-01-24 Tom de Vries <tom@codesourcery.com>
870 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
871 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
872 Add strict parameter.
873 (prevent_branch_around_nothing): Insert dummy insn between branch to
874 label and label with no ptx insn inbetween.
875 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
877 2018-01-24 Tom de Vries <tom@codesourcery.com>
880 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
881 for neutered threads in warp.
882 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
884 2018-01-24 Richard Biener <rguenther@suse.de>
886 PR tree-optimization/83176
887 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
890 2018-01-24 Richard Biener <rguenther@suse.de>
892 PR tree-optimization/82819
893 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
894 code generating pluses that are no-ops in the target precision.
896 2018-01-24 Richard Biener <rguenther@suse.de>
899 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
901 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
903 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
904 to merge probabilities.
905 * predict.c (probably_never_executed): Also mark as cold functions
906 with global 0 profile and guessed local profile.
907 * profile-count.c (profile_probability::combine_with_count): New
909 * profile-count.h (profile_probability::operator*,
910 profile_probability::operator*=, profile_probability::operator/,
911 profile_probability::operator/=): Reduce precision to adjusted
912 and set value to guessed on contradictory divisions.
913 (profile_probability::combine_with_freq): Remove.
914 (profile_probability::combine_wiht_count): Declare.
915 (profile_count::force_nonzero):: Set to adjusted.
916 (profile_count::probability_in):: Set quality to adjusted.
917 * tree-ssa-tail-merge.c (replace_block_by): Use
920 2018-01-23 Andrew Waterman <andrew@sifive.com>
921 Jim Wilson <jimw@sifive.com>
923 * config/riscv/riscv.c (riscv_stack_boundary): New.
924 (riscv_option_override): Set riscv_stack_boundary. Handle
925 riscv_preferred_stack_boundary_arg.
926 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
927 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
928 (STACK_BOUNDARY): Set to riscv_stack_boundary.
929 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
930 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
931 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
933 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
936 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
937 of struct ix86_frame.
938 (ix86_expand_epilogue): Likewise. Add a local variable for
939 the reg_save_offset field in struct ix86_frame.
941 2018-01-23 Bin Cheng <bin.cheng@arm.com>
943 PR tree-optimization/82604
944 * tree-loop-distribution.c (enum partition_kind): New enum item
945 PKIND_PARTIAL_MEMSET.
946 (partition_builtin_p): Support above new enum item.
947 (generate_code_for_partition): Ditto.
948 (compute_access_range): Differentiate cases that equality can be
949 proven at all loops, the innermost loops or no loops.
950 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
951 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
952 (finalize_partitions, distribute_loop): Don't fuse partition of
953 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
954 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
957 2018-01-23 Martin Liska <mliska@suse.cz>
959 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
960 order to ignore the predictor.
961 (PRED_POLYMORPHIC_CALL): Likewise.
962 (PRED_RECURSIVE_CALL): Likewise.
964 2018-01-23 Martin Liska <mliska@suse.cz>
966 * tree-profile.c (tree_profiling): Print function header to
967 aware reader which function we are working on.
968 * value-prof.c (gimple_find_values_to_profile): Do not print
969 not interesting value histograms.
971 2018-01-23 Martin Liska <mliska@suse.cz>
973 * profile-count.h (enum profile_quality): Add
974 profile_uninitialized as the first value. Do not number values
975 as they are zero based.
976 (profile_count::verify): Update sanity check.
977 (profile_probability::verify): Likewise.
979 2018-01-23 Nathan Sidwell <nathan@acm.org>
981 * doc/invoke.texi (ffor-scope): Deprecate.
983 2018-01-23 David Malcolm <dmalcolm@redhat.com>
985 PR tree-optimization/83510
986 * domwalk.c (set_all_edges_as_executable): New function.
987 (dom_walker::dom_walker): Convert bool param
988 "skip_unreachable_blocks" to enum reachability. Move setup of
989 edge flags to set_all_edges_as_executable and only do it when
990 reachability is REACHABLE_BLOCKS.
991 * domwalk.h (enum dom_walker::reachability): New enum.
992 (dom_walker::dom_walker): Convert bool param
993 "skip_unreachable_blocks" to enum reachability.
994 (set_all_edges_as_executable): New decl.
995 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
996 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
998 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
999 but converting true to REACHABLE_BLOCKS.
1000 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1002 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1003 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1004 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1006 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1007 if check_all_array_refs will be called.
1009 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1011 * tree.c (selftest::test_location_wrappers): Add more test
1014 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1016 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1017 (selftest::test_bit_in_range): Likewise.
1019 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1022 * doc/sourcebuild.texi (vect_float): Say that the selector
1023 only describes the situation when -funsafe-math-optimizations is on.
1024 (vect_float_strict): Document.
1026 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1028 PR tree-optimization/83965
1029 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1030 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1031 instead of checking only for a reduction.
1032 (vect_recog_widen_sum_pattern): Likewise.
1034 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1036 * predict.c (probably_never_executed): Only use precise profile info.
1037 (compute_function_frequency): Skip after inlining hack since we now
1038 have quality checking.
1040 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1042 * profile-count.h (profile_probability::very_unlikely,
1043 profile_probability::unlikely, profile_probability::even): Set
1044 precision to guessed.
1046 2018-01-23 Richard Biener <rguenther@suse.de>
1048 PR tree-optimization/83963
1049 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1050 Properly terminate dominator walk when crossing the exit edge not
1051 when visiting its source block.
1053 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1056 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1057 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1059 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1061 PR tree-optimization/83957
1062 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1063 semicolon after for body surrounded by braces.
1065 PR tree-optimization/83081
1066 * profile-count.h (profile_probability::split): New method.
1067 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1068 Use profile_probability::split.
1069 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1070 when splitting a single conditional jump into 2.
1072 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1074 PR tree-optimization/69452
1075 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1078 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1080 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1081 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1082 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1084 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1086 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1087 * config/rl78/rl78.md: New define_expand "movdi"
1088 * config/rl78/rl78.c: New function definition rl78_split_movdi
1090 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1093 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1095 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1096 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1097 128-bit to produce an UNSPEC move to get the double word with the
1098 signbit and then a shift directly to do signbit.
1099 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1100 implementation with a new version that just does either a direct
1101 move or a regular move. Move memory interface to separate insns.
1102 Move insns so they are next to the expander.
1103 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1104 with signbit move. Split big and little endian case.
1105 (signbit<mode>2_dm_mem_le): Likewise.
1106 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1107 (signbit<mode>2_dm2): Likewise.
1109 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1111 * config/rl78/rl78.md: New define_expand "anddi3".
1113 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1115 * config/rl78/rl78.md: New define_expand "umindi3".
1117 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1119 * config/rl78/rl78.md: New define_expand "smindi3".
1121 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1123 * config/rl78/rl78.md: New define_expand "smaxdi3".
1125 2018-01-22 Carl Love <cel@us.ibm.com>
1127 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1128 LVX_V1TI): Add macro expansion.
1129 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1130 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1131 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1132 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1133 Change check to determine if the instruction is a byte reversing
1134 entry. Fix typo in comment.
1135 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1136 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1137 Add def_builtin calls for new builtins.
1138 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1139 Add define_insn expansion.
1141 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1143 * config/rl78/rl78.md: New define_expand "umaxdi3".
1145 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1147 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1148 for non-QImode registers
1150 2018-01-22 Richard Biener <rguenther@suse.de>
1152 PR tree-optimization/83963
1153 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1154 including the loop exit block.
1155 (scop_detection::merge_sese): Likewise.
1156 (scop_detection::add_scop): Do it here instead.
1158 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1160 * doc/sourcebuild.texi (arm_softfloat): Document.
1162 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1165 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1166 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1167 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1169 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1170 David Edelsohn <dje.gcc@gmail.com>
1173 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1174 Change "crset eq" to "crset 2".
1175 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1176 (*call_indirect_aix<mode>_nospec): Likewise.
1177 (*call_value_indirect_aix<mode>_nospec): Likewise.
1178 (*call_indirect_elfv2<mode>_nospec): Likewise.
1179 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1180 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1181 change assembly output from . to $.
1182 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1183 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1184 (*tablejump<mode>_internal1_nospec): Likewise.
1186 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1189 * config/sh/sh_optimize_sett_clrt.cc:
1190 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1192 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1194 PR tree-optimization/83940
1195 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1196 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1197 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1198 use it to pass back the definition type.
1199 (vect_check_store_rhs): Likewise rhs_dt_out.
1200 (vect_build_gather_load_calls): Add a mask_dt argument and use
1201 it instead of a call to vect_is_simple_use.
1202 (vectorizable_store): Update calls to vect_check_load_store_mask
1203 and vect_check_store_rhs. Use the dt returned by the latter instead
1204 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1205 instead of calls to vect_is_simple_use. Pass the scalar rather
1206 than the vector operand to vect_is_simple_use when handling
1207 second and subsequent copies of an rhs value.
1208 (vectorizable_load): Update calls to vect_check_load_store_mask
1209 and vect_build_gather_load_calls. Use the cached mask_dt and
1210 gs_info.offset_dt instead of calls to vect_is_simple_use.
1212 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1215 * tree-emutls.c: Include gimplify.h.
1216 (lower_emutls_2): New function.
1217 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1218 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1219 it before further processing.
1222 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1223 UINTVAL (trueop1) instead of INTVAL (op1).
1225 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1229 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1230 INCOMING_FRAME_SP_OFFSET if not defined.
1231 (scan_trace): Add ENTRY argument. If true and
1232 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1233 emit a note to adjust the CFA offset.
1234 (create_cfi_notes): Adjust scan_trace callers.
1235 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1236 INCOMING_FRAME_SP_OFFSET in the CIE.
1237 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1238 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1240 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1241 * doc/tm.texi: Regenerated.
1243 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1245 PR rtl-optimization/83147
1246 * lra-constraints.c (remove_inheritance_pseudos): Use
1247 lra_substitute_pseudo_within_insn.
1249 2018-01-19 Tom de Vries <tom@codesourcery.com>
1250 Cesar Philippidis <cesar@codesourcery.com>
1253 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1255 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1258 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1259 spaces for function labels.
1261 2018-01-19 Martin Liska <mliska@suse.cz>
1263 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1264 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1265 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1266 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1267 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1268 (PRED_CONST_RETURN): Change from 69 to 65.
1269 (PRED_NULL_RETURN): Change from 91 to 71.
1270 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1271 (PRED_LOOP_GUARD): Change from 66 to 73.
1273 2018-01-19 Martin Liska <mliska@suse.cz>
1275 * predict.c (predict_insn_def): Add new assert.
1276 (struct branch_predictor): Change type to signed integer.
1277 (test_prediction_value_range): Amend test to cover
1279 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1280 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1281 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1282 (PRED_LOOP_IV_COMPARE): Likewise.
1283 * predict.h (PROB_UNINITIALIZED): Define new constant.
1285 2018-01-19 Martin Liska <mliska@suse.cz>
1287 * predict.c (dump_prediction): Add new format for
1288 analyze_brprob.py script which is enabled with -details
1290 * profile-count.h (precise_p): New function.
1292 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1294 PR tree-optimization/83922
1295 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1296 there are no statements that need masking.
1297 (vect_active_double_reduction_p): New function.
1298 (vect_analyze_loop_operations): Use it when handling phis that
1299 are not in the loop header.
1301 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1303 PR tree-optimization/83914
1304 * tree-vect-loop.c (vectorizable_induction): Don't convert
1305 init_expr or apply the peeling adjustment for inductions
1306 that are nested within the vectorized loop.
1308 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1310 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1313 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1317 * function.h (gimplify_parameters): Add gimple_seq * argument.
1318 * function.c: Include gimple.h and options.h.
1319 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1320 for the added local temporaries if needed.
1321 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1322 if there are any parameter cleanups, wrap whole body into a
1323 try/finally with the cleanups.
1325 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1328 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1329 Use GET_MODE_CLASS for scalar floating point.
1331 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1335 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1336 Fix call of call_cgraph_insertion_hooks.
1338 2018-01-18 Martin Sebor <msebor@redhat.com>
1340 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1342 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1345 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1348 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1351 * common.opt: (-ffile-prefix-map): New option.
1352 * opts.c (common_handle_option): Defer it.
1353 * opts-global.c (handle_common_deferred_options): Handle it.
1354 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1355 * file-prefix-map.h: New file.
1356 (remap_debug_filename, add_debug_prefix_map): ...here.
1357 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1358 * final.c (debug_prefix_map, add_debug_prefix_map
1359 remap_debug_filename): Move to...
1360 * file-prefix-map.c: New file.
1361 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1362 generalize, get rid of alloca(), use strrchr() instead of strchr().
1363 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1364 Implement in terms of add_prefix_map().
1365 (remap_macro_filename, remap_debug_filename): Implement in term of
1367 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1368 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1369 * dbxout.c: Include file-prefix-map.h.
1370 * varasm.c: Likewise.
1371 * vmsdbgout.c: Likewise.
1372 * xcoffout.c: Likewise.
1373 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1374 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1375 * doc/invoke.texi (-ffile-prefix-map): Document.
1376 (-fdebug-prefix-map): Update description.
1378 2018-01-18 Martin Liska <mliska@suse.cz>
1380 * config/i386/i386.c (indirect_thunk_name): Document that also
1382 (output_indirect_thunk): Document why both instructions
1383 (pause and lfence) are generated.
1385 2018-01-18 Richard Biener <rguenther@suse.de>
1387 PR tree-optimization/83887
1388 * graphite-scop-detection.c
1389 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1390 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1391 (scop_detection::merge_sese): Re-implement with a flood-fill
1392 algorithm that properly finds a SESE region if it exists.
1394 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1397 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1398 pointer_diff optimizations use view_convert instead of convert.
1400 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1402 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1403 Generate different code for -mno-speculate-indirect-jumps.
1404 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1405 (*call_indirect_aix<mode>): Disable for
1406 -mno-speculate-indirect-jumps.
1407 (*call_indirect_aix<mode>_nospec): New define_insn.
1408 (*call_value_indirect_aix<mode>): Disable for
1409 -mno-speculate-indirect-jumps.
1410 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1411 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1412 -mno-speculate-indirect-jumps.
1413 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1415 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1417 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1418 long double type, set the flags for noting the default long double
1419 type, even if we don't pass or return a long double type.
1421 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1424 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1427 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1429 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1430 support for merge[hl].
1431 (fold_mergehl_helper): New helper function.
1432 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1433 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1434 (altivec_vmrglw_direct): Add xxmrglw insn.
1436 2018-01-17 Andrew Waterman <andrew@sifive.com>
1438 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1439 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1441 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1444 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1445 call the lto_location_cache before reading the
1446 DECL_SOURCE_LOCATION of the types.
1448 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1449 Richard Sandiford <richard.sandiford@linaro.org>
1451 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1452 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1453 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1454 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1455 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1457 * config/aarch64/constraints.md (aarch64_movti_operand):
1459 * config/aarch64/predicates.md (Uti): Add new constraint.
1461 2018-01-17 Carl Love <cel@us.ibm.com>
1462 * config/rs6000/vsx.md (define_expand xl_len_r,
1463 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1464 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1466 (define_expand, define_insn): Move the shift left from the
1467 define_insn to the define_expand for lxvl and stxvl instructions.
1468 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1469 and XL_LEN_R definitions to PURE.
1471 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1473 * config/i386/i386.c (indirect_thunk_name): Declare regno
1474 as unsigned int. Compare regno with INVALID_REGNUM.
1475 (output_indirect_thunk): Ditto.
1476 (output_indirect_thunk_function): Ditto.
1477 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1478 in the call to output_indirect_thunk_function.
1480 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1483 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1484 rather than the size of inner_type to determine the stack slot size
1485 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1487 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1490 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1493 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1495 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1496 endian Linux systems to optionally enable multilibs for selecting
1497 the long double type if the user configured an explicit type.
1498 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1499 have no long double multilibs if not defined.
1500 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1501 warn if the user used -mabi={ieee,ibm}longdouble and we built
1502 multilibs for long double.
1503 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1504 appropriate multilib option.
1505 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1507 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1508 for building long double multilibs.
1509 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1511 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1513 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1516 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1518 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1521 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1524 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1527 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1529 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1530 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1532 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1534 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1535 different rtl trees depending on TARGET_64BIT.
1536 (rs6000_gen_lvx): Likewise.
1538 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1540 * config/visium/visium.md (nop): Tweak comment.
1541 (hazard_nop): Likewise.
1543 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1545 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1546 -mspeculate-indirect-jumps.
1547 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1548 for -mno-speculate-indirect-jumps.
1549 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1550 (*call_value_indirect_elfv2<mode>): Disable for
1551 -mno-speculate-indirect-jumps.
1552 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1553 (indirect_jump): Emit different RTL for
1554 -mno-speculate-indirect-jumps.
1555 (*indirect_jump<mode>): Disable for
1556 -mno-speculate-indirect-jumps.
1557 (*indirect_jump<mode>_nospec): New define_insn.
1558 (tablejump): Emit different RTL for
1559 -mno-speculate-indirect-jumps.
1560 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1561 (tablejumpsi_nospec): New define_expand.
1562 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1563 (tablejumpdi_nospec): New define_expand.
1564 (*tablejump<mode>_internal1): Disable for
1565 -mno-speculate-indirect-jumps.
1566 (*tablejump<mode>_internal1_nospec): New define_insn.
1567 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1570 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1572 * caller-save.c (insert_save): Drop unnecessary parameter. All
1575 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1576 Richard Biener <rguenth@suse.de>
1579 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1580 return early, inline manually is_gimple_sizepos. Make sure if we
1581 call gimplify_expr we don't end up with a gimple constant.
1582 * tree.c (variably_modified_type_p): Don't return true for
1583 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1584 * gimplify.h (is_gimple_sizepos): Remove.
1586 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1588 PR tree-optimization/83857
1589 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1590 vectorizable_live_operation for pure SLP statements.
1591 (vectorizable_live_operation): Handle PHIs.
1593 2018-01-16 Richard Biener <rguenther@suse.de>
1595 PR tree-optimization/83867
1596 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1597 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1599 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1602 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1603 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1604 If off is not INTEGER_CST, issue a may not be aligned warning
1605 rather than isn't aligned. Use isn%'t rather than isn't.
1606 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1608 <case MULT_EXPR>: Improve the case when bottom and one of the
1609 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1610 operand, in that case check if the other operand is multiple of
1611 bottom divided by the INTEGER_CST operand.
1613 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1616 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1617 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1618 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1619 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1620 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1621 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1622 (pa_function_arg_size): New function.
1624 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1626 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1627 in a separate statement.
1629 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1631 PR tree-optimization/83847
1632 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1633 group gathers and scatters.
1635 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1637 PR rtl-optimization/86620
1638 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1640 PR rtl-optimization/83213
1641 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1642 to last if both are JUMP_INSNs.
1644 PR tree-optimization/83843
1645 * gimple-ssa-store-merging.c
1646 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1647 store_immediate_info for bswap/nop orig_stores.
1649 2018-01-15 Andrew Waterman <andrew@sifive.com>
1651 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1653 <UDIV>: Increase cost if !TARGET_DIV.
1655 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1657 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1658 (define_attr "cr_logical_3op"): New.
1659 (cceq_ior_compare): Adjust.
1660 (cceq_ior_compare_complement): Adjust.
1661 (*cceq_rev_compare): Adjust.
1662 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1663 (is_cracked_insn): Adjust.
1664 (insn_must_be_first_in_group): Adjust.
1665 * config/rs6000/40x.md: Adjust.
1666 * config/rs6000/440.md: Adjust.
1667 * config/rs6000/476.md: Adjust.
1668 * config/rs6000/601.md: Adjust.
1669 * config/rs6000/603.md: Adjust.
1670 * config/rs6000/6xx.md: Adjust.
1671 * config/rs6000/7450.md: Adjust.
1672 * config/rs6000/7xx.md: Adjust.
1673 * config/rs6000/8540.md: Adjust.
1674 * config/rs6000/cell.md: Adjust.
1675 * config/rs6000/e300c2c3.md: Adjust.
1676 * config/rs6000/e500mc.md: Adjust.
1677 * config/rs6000/e500mc64.md: Adjust.
1678 * config/rs6000/e5500.md: Adjust.
1679 * config/rs6000/e6500.md: Adjust.
1680 * config/rs6000/mpc.md: Adjust.
1681 * config/rs6000/power4.md: Adjust.
1682 * config/rs6000/power5.md: Adjust.
1683 * config/rs6000/power6.md: Adjust.
1684 * config/rs6000/power7.md: Adjust.
1685 * config/rs6000/power8.md: Adjust.
1686 * config/rs6000/power9.md: Adjust.
1687 * config/rs6000/rs64.md: Adjust.
1688 * config/rs6000/titan.md: Adjust.
1690 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1692 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1693 ix86_indirect_branch_register logic.
1695 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1697 * config/i386/constraints.md (Bs): Update
1698 ix86_indirect_branch_register check. Don't check
1699 ix86_indirect_branch_register with GOT_memory_operand.
1701 * config/i386/predicates.md (GOT_memory_operand): Don't check
1702 ix86_indirect_branch_register here.
1703 (GOT32_symbol_operand): Likewise.
1705 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1707 * config/i386/predicates.md (constant_call_address_operand):
1708 Rewrite ix86_indirect_branch_register logic.
1709 (sibcall_insn_operand): Likewise.
1711 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1713 * config/i386/constraints.md (Bs): Replace
1714 ix86_indirect_branch_thunk_register with
1715 ix86_indirect_branch_register.
1717 * config/i386/i386.md (indirect_jump): Likewise.
1718 (tablejump): Likewise.
1719 (*sibcall_memory): Likewise.
1720 (*sibcall_value_memory): Likewise.
1721 Peepholes of indirect call and jump via memory: Likewise.
1722 * config/i386/i386.opt: Likewise.
1723 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1724 (GOT_memory_operand): Likewise.
1725 (call_insn_operand): Likewise.
1726 (sibcall_insn_operand): Likewise.
1727 (GOT32_symbol_operand): Likewise.
1729 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1732 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1733 type rather than type addr's type points to.
1734 (expand_omp_atomic_mutex): Likewise.
1735 (expand_omp_atomic): Likewise.
1737 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1740 * config/i386/i386.c (output_indirect_thunk_function): Use
1741 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1742 for __x86_return_thunk.
1744 2018-01-15 Richard Biener <rguenther@suse.de>
1747 * expmed.c (extract_bit_field_1): Fix typo.
1749 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1752 * config/arm/iterators.md (VF): New mode iterator.
1753 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1754 Remove integer-related logic from pattern.
1755 (neon_vabd<mode>_3): Likewise.
1757 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1760 * common.opt (fstrict-overflow): No longer an alias.
1761 (fwrapv-pointer): New option.
1762 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1763 also for pointer types based on flag_wrapv_pointer.
1764 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1765 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1766 opts->x_flag_wrapv got set.
1767 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1768 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1769 POINTER_TYPE_OVERFLOW_UNDEFINED.
1770 * match.pd: Likewise in address comparison pattern.
1771 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1773 2018-01-15 Richard Biener <rguenther@suse.de>
1776 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1777 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1778 Reset type names to their identifier if their TYPE_DECL doesn't
1779 have linkage (and thus is used for ODR and devirt).
1780 (save_debug_info_for_decl): Remove.
1781 (save_debug_info_for_type): Likewise.
1782 (add_tree_to_fld_list): Adjust.
1783 * tree-pretty-print.c (dump_generic_node): Make dumping of
1784 type names more robust.
1786 2018-01-15 Richard Biener <rguenther@suse.de>
1788 * BASE-VER: Bump to 8.0.1.
1790 2018-01-14 Martin Sebor <msebor@redhat.com>
1793 * builtins.c (check_access): Avoid warning when the no-warning bit
1796 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1798 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1799 * ira-color (allocno_hard_regs_compare): Likewise.
1801 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1804 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1805 Use .pushsection/.popsection.
1807 2018-01-14 Martin Sebor <msebor@redhat.com>
1810 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1812 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1814 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1815 entry from extra_headers.
1816 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1817 extra_headers, make the list bitwise identical to the i?86-*-* one.
1819 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1821 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1822 -mcmodel=large with -mindirect-branch=thunk,
1823 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1824 -mfunction-return=thunk-extern.
1825 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1826 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1827 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1829 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1831 * config/i386/i386.c (print_reg): Print the name of the full
1832 integer register without '%'.
1833 (ix86_print_operand): Handle 'V'.
1834 * doc/extend.texi: Document 'V' modifier.
1836 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1838 * config/i386/constraints.md (Bs): Disallow memory operand for
1839 -mindirect-branch-register.
1841 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1842 (GOT_memory_operand): Likewise.
1843 (call_insn_operand): Likewise.
1844 (sibcall_insn_operand): Likewise.
1845 (GOT32_symbol_operand): Likewise.
1846 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1847 for -mindirect-branch-register.
1848 (tablejump): Likewise.
1849 (*sibcall_memory): Likewise.
1850 (*sibcall_value_memory): Likewise.
1851 Disallow peepholes of indirect call and jump via memory for
1852 -mindirect-branch-register.
1853 (*call_pop): Replace m with Bw.
1854 (*call_value_pop): Likewise.
1855 (*sibcall_pop_memory): Replace m with Bs.
1856 * config/i386/i386.opt (mindirect-branch-register): New option.
1857 * doc/invoke.texi: Document -mindirect-branch-register option.
1859 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1861 * config/i386/i386-protos.h (ix86_output_function_return): New.
1862 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1863 set function_return_type.
1864 (indirect_thunk_name): Add ret_p to indicate thunk for function
1866 (output_indirect_thunk_function): Pass false to
1867 indirect_thunk_name.
1868 (ix86_output_indirect_branch_via_reg): Likewise.
1869 (ix86_output_indirect_branch_via_push): Likewise.
1870 (output_indirect_thunk_function): Create alias for function
1871 return thunk if regno < 0.
1872 (ix86_output_function_return): New function.
1873 (ix86_handle_fndecl_attribute): Handle function_return.
1874 (ix86_attribute_table): Add function_return.
1875 * config/i386/i386.h (machine_function): Add
1876 function_return_type.
1877 * config/i386/i386.md (simple_return_internal): Use
1878 ix86_output_function_return.
1879 (simple_return_internal_long): Likewise.
1880 * config/i386/i386.opt (mfunction-return=): New option.
1881 (indirect_branch): Mention -mfunction-return=.
1882 * doc/extend.texi: Document function_return function attribute.
1883 * doc/invoke.texi: Document -mfunction-return= option.
1885 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1887 * config/i386/i386-opts.h (indirect_branch): New.
1888 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1889 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1890 with local indirect jump when converting indirect call and jump.
1891 (ix86_set_indirect_branch_type): New.
1892 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1893 (indirectlabelno): New.
1894 (indirect_thunk_needed): Likewise.
1895 (indirect_thunk_bnd_needed): Likewise.
1896 (indirect_thunks_used): Likewise.
1897 (indirect_thunks_bnd_used): Likewise.
1898 (INDIRECT_LABEL): Likewise.
1899 (indirect_thunk_name): Likewise.
1900 (output_indirect_thunk): Likewise.
1901 (output_indirect_thunk_function): Likewise.
1902 (ix86_output_indirect_branch_via_reg): Likewise.
1903 (ix86_output_indirect_branch_via_push): Likewise.
1904 (ix86_output_indirect_branch): Likewise.
1905 (ix86_output_indirect_jmp): Likewise.
1906 (ix86_code_end): Call output_indirect_thunk_function if needed.
1907 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1909 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1910 (ix86_attribute_table): Add indirect_branch.
1911 * config/i386/i386.h (machine_function): Add indirect_branch_type
1912 and has_local_indirect_jump.
1913 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1915 (tablejump): Likewise.
1916 (*indirect_jump): Use ix86_output_indirect_jmp.
1917 (*tablejump_1): Likewise.
1918 (simple_return_indirect_internal): Likewise.
1919 * config/i386/i386.opt (mindirect-branch=): New option.
1920 (indirect_branch): New.
1923 (thunk-inline): Likewise.
1924 (thunk-extern): Likewise.
1925 * doc/extend.texi: Document indirect_branch function attribute.
1926 * doc/invoke.texi: Document -mindirect-branch= option.
1928 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1931 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1933 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1935 * ipa-inline.c (want_inline_small_function_p): Return false if
1936 inlining has already failed with CIF_FINAL_ERROR.
1937 (update_caller_keys): Call want_inline_small_function_p before
1939 (update_callee_keys): Likewise.
1941 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1943 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1945 (rs6000_quadword_masked_address_p): Likewise.
1946 (quad_aligned_load_p): Likewise.
1947 (quad_aligned_store_p): Likewise.
1948 (const_load_sequence_p): Add comment to describe the outer-most loop.
1949 (mimic_memory_attributes_and_flags): New function.
1950 (rs6000_gen_stvx): Likewise.
1951 (replace_swapped_aligned_store): Likewise.
1952 (rs6000_gen_lvx): Likewise.
1953 (replace_swapped_aligned_load): Likewise.
1954 (replace_swapped_load_constant): Capitalize argument name in
1955 comment describing this function.
1956 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1957 and stores that access quad-word aligned addresses and replace
1958 with stvx or lvx instructions when appropriate.
1959 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1960 New function prototype.
1961 (rs6000_quadword_masked_address_p): Likewise.
1962 (rs6000_gen_lvx): Likewise.
1963 (rs6000_gen_stvx): Likewise.
1964 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1965 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1966 when memory address is aligned.
1967 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1968 this split to select lvx instruction when memory address is aligned.
1969 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1970 instruction when memory address is aligned.
1971 (*vsx_le_perm_load_v16qi): Likewise.
1972 (four unnamed splitters): Modify to select the stvx instruction
1973 when memory is aligned.
1975 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1977 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1978 which appears in the queue multiple times.
1980 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1981 Alan Hayward <alan.hayward@arm.com>
1982 David Sherwood <david.sherwood@arm.com>
1984 * tree-vectorizer.h (vec_lower_bound): New structure.
1985 (_loop_vec_info): Add check_nonzero and lower_bounds.
1986 (LOOP_VINFO_CHECK_NONZERO): New macro.
1987 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1988 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1989 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1990 fields. Make seg_len the distance travelled, not including the
1992 (dr_direction_indicator): Declare.
1993 (dr_zero_step_indicator): Likewise.
1994 (dr_known_forward_stride_p): Likewise.
1995 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1997 (runtime_alias_check_p): Allow runtime alias checks with
1999 (operator ==): Compare access_size and align.
2000 (prune_runtime_alias_test_list): Rework for new distinction between
2001 the access_size and seg_len.
2002 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2004 (get_segment_min_max): New function.
2005 (create_intersect_range_checks): Use it.
2006 (dr_step_indicator): New function.
2007 (dr_direction_indicator): Likewise.
2008 (dr_zero_step_indicator): Likewise.
2009 (dr_known_forward_stride_p): Likewise.
2010 * tree-loop-distribution.c (data_ref_segment_size): Return
2011 DR_STEP * (niters - 1).
2012 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2014 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2015 (vect_preserves_scalar_order_p): New function, split out from...
2016 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2017 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2018 (vect_vfa_access_size): New function.
2019 (vect_vfa_align): Likewise.
2020 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2021 (dump_lower_bound): New function.
2022 (vect_check_lower_bound): Likewise.
2023 (vect_small_gap_p): Likewise.
2024 (vectorizable_with_step_bound_p): Likewise.
2025 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2026 depencies if the vectorization factor is 1. Convert the checks
2027 for nonzero steps into checks on the bounds of DR_STEP. Try using
2028 a bunds check for variable steps if the minimum required step is
2029 relatively small. Update calls to the dr_with_seg_len
2030 constructor and to vect_compile_time_alias.
2031 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2033 (vect_loop_versioning): Call it.
2034 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2036 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2038 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2039 Alan Hayward <alan.hayward@arm.com>
2040 David Sherwood <david.sherwood@arm.com>
2042 * doc/sourcebuild.texi (vect_scatter_store): Document.
2043 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2045 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2047 * genopinit.c (main): Add supports_vec_scatter_store and
2048 supports_vec_scatter_store_cached to target_optabs.
2049 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2050 IFN_MASK_SCATTER_STORE.
2051 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2053 * internal-fn.h (internal_store_fn_p): Declare.
2054 (internal_fn_stored_value_index): Likewise.
2055 * internal-fn.c (scatter_store_direct): New macro.
2056 (expand_scatter_store_optab_fn): New function.
2057 (direct_scatter_store_optab_supported_p): New macro.
2058 (internal_store_fn_p): New function.
2059 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2060 IFN_MASK_SCATTER_STORE.
2061 (internal_fn_mask_index): Likewise.
2062 (internal_fn_stored_value_index): New function.
2063 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2065 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2066 * optabs-query.c (supports_vec_scatter_store_p): New function.
2067 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2068 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2069 true for scatter stores.
2070 (vect_gather_scatter_fn_p): Handle scatter stores too.
2071 (vect_check_gather_scatter): Consider using scatter stores if
2072 supports_vec_scatter_store_p.
2073 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2075 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2076 internal_fn_stored_value_index.
2077 (check_load_store_masking): Handle scatter stores too.
2078 (vect_get_store_rhs): Make public.
2079 (vectorizable_call): Use internal_store_fn_p.
2080 (vectorizable_store): Handle scatter store internal functions.
2081 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2082 when deciding whether the end of the group has been reached.
2083 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2084 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2085 (mask_scatter_store<mode>): New insns.
2087 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2088 Alan Hayward <alan.hayward@arm.com>
2089 David Sherwood <david.sherwood@arm.com>
2091 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2092 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2093 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2095 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2096 Use vect_truncate_gather_scatter_offset if we can't treat the
2097 operation as a normal gather load or scatter store.
2098 (get_group_load_store_type): Take the gather_scatter_info
2099 as argument. Try using a gather load or scatter store for
2100 single-element groups.
2101 (get_load_store_type): Update calls to get_group_load_store_type
2102 and vect_use_strided_gather_scatters_p.
2104 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2105 Alan Hayward <alan.hayward@arm.com>
2106 David Sherwood <david.sherwood@arm.com>
2108 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2109 optional tree argument.
2110 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2112 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2113 but continue to use the current value as a fallback.
2114 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2115 to compare the updates.
2116 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2117 (get_load_store_type): Use it when handling a strided access.
2118 (vect_get_strided_load_store_ops): New function.
2119 (vect_get_data_ptr_increment): Likewise.
2120 (vectorizable_load): Handle strided gather loads. Always pass
2121 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2123 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2124 Alan Hayward <alan.hayward@arm.com>
2125 David Sherwood <david.sherwood@arm.com>
2127 * doc/md.texi (gather_load@var{m}): Document.
2128 (mask_gather_load@var{m}): Likewise.
2129 * genopinit.c (main): Add supports_vec_gather_load and
2130 supports_vec_gather_load_cached to target_optabs.
2131 * optabs-tree.c (init_tree_optimization_optabs): Use
2132 ggc_cleared_alloc to allocate target_optabs.
2133 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2134 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2136 * internal-fn.h (internal_load_fn_p): Declare.
2137 (internal_gather_scatter_fn_p): Likewise.
2138 (internal_fn_mask_index): Likewise.
2139 (internal_gather_scatter_fn_supported_p): Likewise.
2140 * internal-fn.c (gather_load_direct): New macro.
2141 (expand_gather_load_optab_fn): New function.
2142 (direct_gather_load_optab_supported_p): New macro.
2143 (direct_internal_fn_optab): New function.
2144 (internal_load_fn_p): Likewise.
2145 (internal_gather_scatter_fn_p): Likewise.
2146 (internal_fn_mask_index): Likewise.
2147 (internal_gather_scatter_fn_supported_p): Likewise.
2148 * optabs-query.c (supports_at_least_one_mode_p): New function.
2149 (supports_vec_gather_load_p): Likewise.
2150 * optabs-query.h (supports_vec_gather_load_p): Declare.
2151 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2152 and memory_type field.
2153 (NUM_PATTERNS): Bump to 15.
2154 * tree-vect-data-refs.c: Include internal-fn.h.
2155 (vect_gather_scatter_fn_p): New function.
2156 (vect_describe_gather_scatter_call): Likewise.
2157 (vect_check_gather_scatter): Try using internal functions for
2158 gather loads. Recognize existing calls to a gather load function.
2159 (vect_analyze_data_refs): Consider using gather loads if
2160 supports_vec_gather_load_p.
2161 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2162 (vect_get_gather_scatter_offset_type): Likewise.
2163 (vect_convert_mask_for_vectype): Likewise.
2164 (vect_add_conversion_to_patterm): Likewise.
2165 (vect_try_gather_scatter_pattern): Likewise.
2166 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2167 (vect_vect_recog_func_ptrs): Add it.
2168 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2169 internal_fn_mask_index and internal_gather_scatter_fn_p.
2170 (check_load_store_masking): Take the gather_scatter_info as an
2171 argument and handle gather loads.
2172 (vect_get_gather_scatter_ops): New function.
2173 (vectorizable_call): Check internal_load_fn_p.
2174 (vectorizable_load): Likewise. Handle gather load internal
2176 (vectorizable_store): Update call to check_load_store_masking.
2177 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2178 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2179 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2180 (aarch64_gather_scale_operand_d): New predicates.
2181 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2182 (mask_gather_load<mode>): New insns.
2184 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2185 Alan Hayward <alan.hayward@arm.com>
2186 David Sherwood <david.sherwood@arm.com>
2188 * optabs.def (fold_left_plus_optab): New optab.
2189 * doc/md.texi (fold_left_plus_@var{m}): Document.
2190 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2191 * internal-fn.c (fold_left_direct): Define.
2192 (expand_fold_left_optab_fn): Likewise.
2193 (direct_fold_left_optab_supported_p): Likewise.
2194 * fold-const-call.c (fold_const_fold_left): New function.
2195 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2196 * tree-parloops.c (valid_reduction_p): New function.
2197 (gather_scalar_reductions): Use it.
2198 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2199 (vect_finish_replace_stmt): Declare.
2200 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2201 (needs_fold_left_reduction_p): New function, split out from...
2202 (vect_is_simple_reduction): ...here. Accept reductions that
2203 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2204 (vect_force_simple_reduction): Also store the reduction type in
2205 the assignment's STMT_VINFO_REDUC_TYPE.
2206 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2207 (merge_with_identity): New function.
2208 (vect_expand_fold_left): Likewise.
2209 (vectorize_fold_left_reduction): Likewise.
2210 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2211 scalar phi in place for it. Check for target support and reject
2212 cases that would reassociate the operation. Defer the transform
2213 phase to vectorize_fold_left_reduction.
2214 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2215 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2216 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2220 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2221 call to ifc_temp_var.
2223 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2224 Alan Hayward <alan.hayward@arm.com>
2225 David Sherwood <david.sherwood@arm.com>
2227 * target.def (legitimize_address_displacement): Take the original
2228 offset as a poly_int.
2229 * targhooks.h (default_legitimize_address_displacement): Update
2231 * targhooks.c (default_legitimize_address_displacement): Likewise.
2232 * doc/tm.texi: Regenerate.
2233 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2234 as an argument, moving assert of ad->disp == ad->disp_term to...
2235 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2236 Try calling targetm.legitimize_address_displacement before expanding
2237 the address rather than afterwards, and adjust for the new interface.
2238 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2239 Match the new hook interface. Handle SVE addresses.
2240 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2243 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2245 * Makefile.in (OBJS): Add early-remat.o.
2246 * target.def (select_early_remat_modes): New hook.
2247 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2248 * doc/tm.texi: Regenerate.
2249 * targhooks.h (default_select_early_remat_modes): Declare.
2250 * targhooks.c (default_select_early_remat_modes): New function.
2251 * timevar.def (TV_EARLY_REMAT): New timevar.
2252 * passes.def (pass_early_remat): New pass.
2253 * tree-pass.h (make_pass_early_remat): Declare.
2254 * early-remat.c: New file.
2255 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2257 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2259 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2260 Alan Hayward <alan.hayward@arm.com>
2261 David Sherwood <david.sherwood@arm.com>
2263 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2264 vfm1 with a bound_epilog parameter.
2265 (vect_do_peeling): Update calls accordingly, and move the prologue
2266 call earlier in the function. Treat the base bound_epilog as 0 for
2267 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2268 this base when peeling for gaps.
2269 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2270 with fully-masked loops.
2271 (vect_estimate_min_profitable_iters): Handle the single peeled
2272 iteration in that case.
2274 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2275 Alan Hayward <alan.hayward@arm.com>
2276 David Sherwood <david.sherwood@arm.com>
2278 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2279 single-element interleaving even if the size is not a power of 2.
2280 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2281 accesses for single-element interleaving if the group size is
2284 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2285 Alan Hayward <alan.hayward@arm.com>
2286 David Sherwood <david.sherwood@arm.com>
2288 * doc/md.texi (fold_extract_last_@var{m}): Document.
2289 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2290 * optabs.def (fold_extract_last_optab): New optab.
2291 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2292 * internal-fn.c (fold_extract_direct): New macro.
2293 (expand_fold_extract_optab_fn): Likewise.
2294 (direct_fold_extract_optab_supported_p): Likewise.
2295 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2296 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2297 EXTRACT_LAST_REDUCTION.
2298 (get_initial_def_for_reduction): Do not create an initial vector
2299 for EXTRACT_LAST_REDUCTION reductions.
2300 (vectorizable_reduction): Leave the scalar phi in place for
2301 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2302 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2303 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2304 transform phase to vectorizable_condition.
2305 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2307 (vect_finish_stmt_generation): ...here.
2308 (vect_finish_replace_stmt): New function.
2309 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2310 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2312 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2314 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2315 Alan Hayward <alan.hayward@arm.com>
2316 David Sherwood <david.sherwood@arm.com>
2318 * doc/md.texi (extract_last_@var{m}): Document.
2319 * optabs.def (extract_last_optab): New optab.
2320 * internal-fn.def (EXTRACT_LAST): New internal function.
2321 * internal-fn.c (cond_unary_direct): New macro.
2322 (expand_cond_unary_optab_fn): Likewise.
2323 (direct_cond_unary_optab_supported_p): Likewise.
2324 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2325 loops using EXTRACT_LAST.
2326 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2327 (extract_last_<mode>): ...this optab.
2328 (vec_extract<mode><Vel>): Update accordingly.
2330 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2331 Alan Hayward <alan.hayward@arm.com>
2332 David Sherwood <david.sherwood@arm.com>
2334 * target.def (empty_mask_is_expensive): New hook.
2335 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2336 * doc/tm.texi: Regenerate.
2337 * targhooks.h (default_empty_mask_is_expensive): Declare.
2338 * targhooks.c (default_empty_mask_is_expensive): New function.
2339 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2340 if the target says that empty masks are expensive.
2341 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2343 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2345 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2346 Alan Hayward <alan.hayward@arm.com>
2347 David Sherwood <david.sherwood@arm.com>
2349 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2350 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2351 (vect_use_loop_mask_for_alignment_p): New function.
2352 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2353 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2354 niters_skip argument. Make sure that the first niters_skip elements
2355 of the first iteration are inactive.
2356 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2357 Update call to vect_set_loop_masks_directly.
2358 (get_misalign_in_elems): New function, split out from...
2359 (vect_gen_prolog_loop_niters): ...here.
2360 (vect_update_init_of_dr): Take a code argument that specifies whether
2361 the adjustment should be added or subtracted.
2362 (vect_update_init_of_drs): Likewise.
2363 (vect_prepare_for_masked_peels): New function.
2364 (vect_do_peeling): Skip prologue peeling if we're using a mask
2365 instead. Update call to vect_update_inits_of_drs.
2366 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2368 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2369 alignment. Do not include the number of peeled iterations in
2370 the minimum threshold in that case.
2371 (vectorizable_induction): Adjust the start value down by
2372 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2373 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2374 Take the number of skipped iterations into account when calculating
2376 * tree-vect-stmts.c (vect_gen_while_not): New function.
2378 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2379 Alan Hayward <alan.hayward@arm.com>
2380 David Sherwood <david.sherwood@arm.com>
2382 * doc/sourcebuild.texi (vect_fully_masked): Document.
2383 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2385 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2387 (vect_analyze_loop_2): ...here. Don't check the vectorization
2388 factor against the number of loop iterations if the loop is
2391 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2392 Alan Hayward <alan.hayward@arm.com>
2393 David Sherwood <david.sherwood@arm.com>
2395 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2396 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2397 (dump_groups): Update accordingly.
2398 (iv_use::mem_type): New member variable.
2399 (address_p): New function.
2400 (record_use): Add a mem_type argument and initialize the new
2402 (record_group_use): Add a mem_type argument. Use address_p.
2403 Remove obsolete null checks of base_object. Update call to record_use.
2404 (find_interesting_uses_op): Update call to record_group_use.
2405 (find_interesting_uses_cond): Likewise.
2406 (find_interesting_uses_address): Likewise.
2407 (get_mem_type_for_internal_fn): New function.
2408 (find_address_like_use): Likewise.
2409 (find_interesting_uses_stmt): Try find_address_like_use before
2410 calling find_interesting_uses_op.
2411 (addr_offset_valid_p): Use the iv mem_type field as the type
2412 of the addressed memory.
2413 (add_autoinc_candidates): Likewise.
2414 (get_address_cost): Likewise.
2415 (split_small_address_groups_p): Use address_p.
2416 (split_address_groups): Likewise.
2417 (add_iv_candidate_for_use): Likewise.
2418 (autoinc_possible_for_pair): Likewise.
2419 (rewrite_groups): Likewise.
2420 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2421 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2422 (get_alias_ptr_type_for_ptr_address): New function.
2423 (rewrite_use_address): Rewrite address uses in calls that were
2424 identified by find_address_like_use.
2426 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2427 Alan Hayward <alan.hayward@arm.com>
2428 David Sherwood <david.sherwood@arm.com>
2430 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2432 * gimple-expr.h (is_gimple_addressable: Likewise.
2433 * gimple-expr.c (is_gimple_address): Likewise.
2434 * internal-fn.c (expand_call_mem_ref): New function.
2435 (expand_mask_load_optab_fn): Use it.
2436 (expand_mask_store_optab_fn): Likewise.
2438 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2439 Alan Hayward <alan.hayward@arm.com>
2440 David Sherwood <david.sherwood@arm.com>
2442 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2443 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2444 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2445 (cond_umax@var{mode}): Document.
2446 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2447 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2448 (cond_umin_optab, cond_umax_optab): New optabs.
2449 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2450 (COND_IOR, COND_XOR): New internal functions.
2451 * internal-fn.h (get_conditional_internal_fn): Declare.
2452 * internal-fn.c (cond_binary_direct): New macro.
2453 (expand_cond_binary_optab_fn): Likewise.
2454 (direct_cond_binary_optab_supported_p): Likewise.
2455 (get_conditional_internal_fn): New function.
2456 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2457 Cope with reduction statements that are vectorized as calls rather
2459 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2460 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2461 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2462 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2463 (UNSPEC_COND_EOR): New unspecs.
2464 (optab): Add mappings for them.
2465 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2466 (sve_int_op, sve_fp_op): New int attributes.
2468 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2469 Alan Hayward <alan.hayward@arm.com>
2470 David Sherwood <david.sherwood@arm.com>
2472 * optabs.def (while_ult_optab): New optab.
2473 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2474 * internal-fn.def (WHILE_ULT): New internal function.
2475 * internal-fn.h (direct_internal_fn_supported_p): New override
2476 that takes two types as argument.
2477 * internal-fn.c (while_direct): New macro.
2478 (expand_while_optab_fn): New function.
2479 (convert_optab_supported_p): Likewise.
2480 (direct_while_optab_supported_p): New macro.
2481 * wide-int.h (wi::udiv_ceil): New function.
2482 * tree-vectorizer.h (rgroup_masks): New structure.
2483 (vec_loop_masks): New typedef.
2484 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2486 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2487 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2488 (vect_max_vf): New function.
2489 (slpeel_make_loop_iterate_ntimes): Delete.
2490 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2491 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2492 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2493 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2494 internal-fn.h, stor-layout.h and optabs-query.h.
2495 (vect_set_loop_mask): New function.
2496 (add_preheader_seq): Likewise.
2497 (add_header_seq): Likewise.
2498 (interleave_supported_p): Likewise.
2499 (vect_maybe_permute_loop_masks): Likewise.
2500 (vect_set_loop_masks_directly): Likewise.
2501 (vect_set_loop_condition_masked): Likewise.
2502 (vect_set_loop_condition_unmasked): New function, split out from
2503 slpeel_make_loop_iterate_ntimes.
2504 (slpeel_make_loop_iterate_ntimes): Rename to..
2505 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2506 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2507 (vect_do_peeling): Update call accordingly.
2508 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2510 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2511 mask_compare_type, can_fully_mask_p and fully_masked_p.
2512 (release_vec_loop_masks): New function.
2513 (_loop_vec_info): Use it to free the loop masks.
2514 (can_produce_all_loop_masks_p): New function.
2515 (vect_get_max_nscalars_per_iter): Likewise.
2516 (vect_verify_full_masking): Likewise.
2517 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2518 retries, and free the mask rgroups before retrying. Check loop-wide
2519 reasons for disallowing fully-masked loops. Make the final decision
2520 about whether use a fully-masked loop or not.
2521 (vect_estimate_min_profitable_iters): Do not assume that peeling
2522 for the number of iterations will be needed for fully-masked loops.
2523 (vectorizable_reduction): Disable fully-masked loops.
2524 (vectorizable_live_operation): Likewise.
2525 (vect_halve_mask_nunits): New function.
2526 (vect_double_mask_nunits): Likewise.
2527 (vect_record_loop_mask): Likewise.
2528 (vect_get_loop_mask): Likewise.
2529 (vect_transform_loop): Handle the case in which the final loop
2530 iteration might handle a partial vector. Call vect_set_loop_condition
2531 instead of slpeel_make_loop_iterate_ntimes.
2532 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2533 (check_load_store_masking): New function.
2534 (prepare_load_store_mask): Likewise.
2535 (vectorizable_store): Handle fully-masked loops.
2536 (vectorizable_load): Likewise.
2537 (supportable_widening_operation): Use vect_halve_mask_nunits for
2539 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2540 (vect_gen_while): New function.
2541 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2542 (aarch64_uqdec<mode>): New insn.
2544 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2545 Alan Hayward <alan.hayward@arm.com>
2546 David Sherwood <david.sherwood@arm.com>
2548 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2549 (reduc_xor_scal_optab): New optabs.
2550 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2551 (reduc_xor_scal_@var{m}): Document.
2552 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2553 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2555 * fold-const-call.c (fold_const_call): Handle them.
2556 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2557 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2558 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2559 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2560 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2561 (UNSPEC_XORV): New unspecs.
2562 (optab): Add entries for them.
2563 (BITWISEV): New int iterator.
2564 (bit_reduc_op): New int attributes.
2566 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2567 Alan Hayward <alan.hayward@arm.com>
2568 David Sherwood <david.sherwood@arm.com>
2570 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2571 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2572 * optabs.def (vec_shl_insert_optab): New optab.
2573 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2574 (duplicate_and_interleave): Likewise.
2575 * tree-vect-loop.c: Include internal-fn.h.
2576 (neutral_op_for_slp_reduction): New function, split out from
2577 get_initial_defs_for_reduction.
2578 (get_initial_def_for_reduction): Handle option 2 for variable-length
2579 vectors by loading the neutral value into a vector and then shifting
2580 the initial value into element 0.
2581 (get_initial_defs_for_reduction): Replace the code argument with
2582 the neutral value calculated by neutral_op_for_slp_reduction.
2583 Use gimple_build_vector for constant-length vectors.
2584 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2585 but the first group_size elements have a neutral value.
2586 Use duplicate_and_interleave otherwise.
2587 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2588 Update call to get_initial_defs_for_reduction. Handle SLP
2589 reductions for variable-length vectors by creating one vector
2590 result for each scalar result, with the elements associated
2591 with other scalar results stubbed out with the neutral value.
2592 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2593 Require IFN_VEC_SHL_INSERT for double reductions on
2594 variable-length vectors, or SLP reductions that have
2595 a neutral value. Require can_duplicate_and_interleave_p
2596 support for variable-length unchained SLP reductions if there
2597 is no neutral value, such as for MIN/MAX reductions. Also require
2598 the number of vector elements to be a multiple of the number of
2599 SLP statements when doing variable-length unchained SLP reductions.
2600 Update call to vect_create_epilog_for_reduction.
2601 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2602 and remove initial values.
2603 (duplicate_and_interleave): Make public.
2604 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2605 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2607 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2608 Alan Hayward <alan.hayward@arm.com>
2609 David Sherwood <david.sherwood@arm.com>
2611 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2612 (can_duplicate_and_interleave_p): New function.
2613 (vect_get_and_check_slp_defs): Take the vector of statements
2614 rather than just the current one. Remove excess parentheses.
2615 Restriction rejectinon of vect_constant_def and vect_external_def
2616 for variable-length vectors to boolean types, or types for which
2617 can_duplicate_and_interleave_p is false.
2618 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2619 (duplicate_and_interleave): New function.
2620 (vect_get_constant_vectors): Use gimple_build_vector for
2621 constant-length vectors and suitable variable-length constant
2622 vectors. Use duplicate_and_interleave for other variable-length
2623 vectors. Don't defer the update when inserting new statements.
2625 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2626 Alan Hayward <alan.hayward@arm.com>
2627 David Sherwood <david.sherwood@arm.com>
2629 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2630 min_profitable_iters doesn't go negative.
2632 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2633 Alan Hayward <alan.hayward@arm.com>
2634 David Sherwood <david.sherwood@arm.com>
2636 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2637 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2638 * optabs.def (vec_mask_load_lanes_optab): New optab.
2639 (vec_mask_store_lanes_optab): Likewise.
2640 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2641 (MASK_STORE_LANES): Likewise.
2642 * internal-fn.c (mask_load_lanes_direct): New macro.
2643 (mask_store_lanes_direct): Likewise.
2644 (expand_mask_load_optab_fn): Handle masked operations.
2645 (expand_mask_load_lanes_optab_fn): New macro.
2646 (expand_mask_store_optab_fn): Handle masked operations.
2647 (expand_mask_store_lanes_optab_fn): New macro.
2648 (direct_mask_load_lanes_optab_supported_p): Likewise.
2649 (direct_mask_store_lanes_optab_supported_p): Likewise.
2650 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2652 (vect_load_lanes_supported): Likewise.
2653 * tree-vect-data-refs.c (strip_conversion): New function.
2654 (can_group_stmts_p): Likewise.
2655 (vect_analyze_data_ref_accesses): Use it instead of checking
2656 for a pair of assignments.
2657 (vect_store_lanes_supported): Take a masked_p parameter.
2658 (vect_load_lanes_supported): Likewise.
2659 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2660 vect_store_lanes_supported and vect_load_lanes_supported.
2661 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2662 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2663 parameter. Don't allow gaps for masked accesses.
2664 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2665 and vect_load_lanes_supported.
2666 (get_load_store_type): Take a masked_p parameter and update
2667 call to get_group_load_store_type.
2668 (vectorizable_store): Update call to get_load_store_type.
2669 Handle IFN_MASK_STORE_LANES.
2670 (vectorizable_load): Update call to get_load_store_type.
2671 Handle IFN_MASK_LOAD_LANES.
2673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2674 Alan Hayward <alan.hayward@arm.com>
2675 David Sherwood <david.sherwood@arm.com>
2677 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2679 * config/aarch64/aarch64-protos.h
2680 (aarch64_sve_struct_memory_operand_p): Declare.
2681 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2682 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2683 (VPRED, vpred): Handle SVE structure modes.
2684 * config/aarch64/constraints.md (Utx): New constraint.
2685 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2686 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2687 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2688 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2689 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2690 structure modes. Split into pieces after RA.
2691 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2692 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2694 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2695 SVE structure modes.
2696 (aarch64_classify_address): Likewise.
2697 (sizetochar): Move earlier in file.
2698 (aarch64_print_operand): Handle SVE register lists.
2699 (aarch64_array_mode): New function.
2700 (aarch64_sve_struct_memory_operand_p): Likewise.
2701 (TARGET_ARRAY_MODE): Redefine.
2703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2704 Alan Hayward <alan.hayward@arm.com>
2705 David Sherwood <david.sherwood@arm.com>
2707 * target.def (array_mode): New target hook.
2708 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2709 * doc/tm.texi: Regenerate.
2710 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2711 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2712 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2714 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2718 Alan Hayward <alan.hayward@arm.com>
2719 David Sherwood <david.sherwood@arm.com>
2721 * fold-const.c (fold_binary_loc): Check the argument types
2722 rather than the result type when testing for a vector operation.
2724 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2726 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2727 * doc/tm.texi: Regenerate.
2729 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2730 Alan Hayward <alan.hayward@arm.com>
2731 David Sherwood <david.sherwood@arm.com>
2733 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2734 (sve): Document new AArch64 extension.
2735 * doc/md.texi (w): Extend the description of the AArch64
2736 constraint to include SVE vectors.
2737 (Upl, Upa): Document new AArch64 predicate constraints.
2738 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2740 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2741 (msve-vector-bits=): New option.
2742 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2743 SVE when these are disabled.
2744 (sve): New extension.
2745 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2746 modes. Adjust their number of units based on aarch64_sve_vg.
2747 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2748 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2749 aarch64_addr_query_type.
2750 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2751 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2752 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2753 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2754 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2755 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2756 (aarch64_simd_imm_zero_p): Delete.
2757 (aarch64_check_zero_based_sve_index_immediate): Declare.
2758 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2759 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2760 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2761 (aarch64_sve_float_mul_immediate_p): Likewise.
2762 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2764 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2765 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2766 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2767 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2768 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2769 (aarch64_regmode_natural_size): Likewise.
2770 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2771 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2773 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2774 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2775 for VG and the SVE predicate registers.
2776 (V_ALIASES): Add a "z"-prefixed alias.
2777 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2778 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2779 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2780 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2781 (REG_CLASS_NAMES): Add entries for them.
2782 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2783 and the predicate registers.
2784 (aarch64_sve_vg): Declare.
2785 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2786 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2787 (REGMODE_NATURAL_SIZE): Define.
2788 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2790 * config/aarch64/aarch64.c: Include cfgrtl.h.
2791 (simd_immediate_info): Add a constructor for series vectors,
2792 and an associated step field.
2793 (aarch64_sve_vg): New variable.
2794 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2795 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2796 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2797 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2798 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2799 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2800 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2801 (aarch64_get_mask_mode): New functions.
2802 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2803 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2804 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2805 predicate modes and predicate registers. Explicitly restrict
2806 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2807 to store a vector mode if it is recognized by
2808 aarch64_classify_vector_mode.
2809 (aarch64_regmode_natural_size): New function.
2810 (aarch64_hard_regno_caller_save_mode): Return the original mode
2812 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2813 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2814 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2815 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2817 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2818 does not overlap dest if the function is frame-related. Handle
2820 (aarch64_split_add_offset): New function.
2821 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2822 them aarch64_add_offset.
2823 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2824 and update call to aarch64_sub_sp.
2825 (aarch64_add_cfa_expression): New function.
2826 (aarch64_expand_prologue): Pass extra temporary registers to the
2827 functions above. Handle the case in which we need to emit new
2828 DW_CFA_expressions for registers that were originally saved
2829 relative to the stack pointer, but now have to be expressed
2830 relative to the frame pointer.
2831 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2833 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2834 IP0 and IP1 values for SVE frames.
2835 (aarch64_expand_vec_series): New function.
2836 (aarch64_expand_sve_widened_duplicate): Likewise.
2837 (aarch64_expand_sve_const_vector): Likewise.
2838 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2839 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2840 into the register, rather than emitting a SET directly.
2841 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2842 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2843 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2844 (offset_9bit_signed_scaled_p): New functions.
2845 (aarch64_replicate_bitmask_imm): New function.
2846 (aarch64_bitmask_imm): Use it.
2847 (aarch64_cannot_force_const_mem): Reject expressions involving
2848 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2849 (aarch64_classify_index): Handle SVE indices, by requiring
2850 a plain register index with a scale that matches the element size.
2851 (aarch64_classify_address): Handle SVE addresses. Assert that
2852 the mode of the address is VOIDmode or an integer mode.
2853 Update call to aarch64_classify_symbol.
2854 (aarch64_classify_symbolic_expression): Update call to
2855 aarch64_classify_symbol.
2856 (aarch64_const_vec_all_in_range_p): New function.
2857 (aarch64_print_vector_float_operand): Likewise.
2858 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2859 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2860 and the FP immediates 1.0 and 0.5.
2861 (aarch64_print_address_internal): Handle SVE addresses.
2862 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2863 (aarch64_regno_regclass): Handle predicate registers.
2864 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2866 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2867 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2868 (aarch64_convert_sve_vector_bits): New function.
2869 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2870 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2872 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2873 Handle SVE vector and predicate modes. Accept VL-based constants
2874 that need only one temporary register, and VL offsets that require
2875 no temporary registers.
2876 (aarch64_conditional_register_usage): Mark the predicate registers
2877 as fixed if SVE isn't available.
2878 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2879 Return true for SVE vector and predicate modes.
2880 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2881 rather than an unsigned int. Handle SVE modes.
2882 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2884 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2886 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2887 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2888 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2889 (aarch64_sve_float_mul_immediate_p): New functions.
2890 (aarch64_sve_valid_immediate): New function.
2891 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2892 Explicitly reject structure modes. Check for INDEX constants.
2893 Handle PTRUE and PFALSE constants.
2894 (aarch64_check_zero_based_sve_index_immediate): New function.
2895 (aarch64_simd_imm_zero_p): Delete.
2896 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2897 vector modes. Accept constants in the range of CNT[BHWD].
2898 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2899 ask for an Advanced SIMD mode.
2900 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2901 (aarch64_simd_vector_alignment): Handle SVE predicates.
2902 (aarch64_vectorize_preferred_vector_alignment): New function.
2903 (aarch64_simd_vector_alignment_reachable): Use it instead of
2905 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2906 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2908 (MAX_VECT_LEN): Delete.
2909 (expand_vec_perm_d): Add a vec_flags field.
2910 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2911 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2912 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2914 (aarch64_evpc_rev): Rename to...
2915 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2916 (aarch64_evpc_rev_global): New function.
2917 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2918 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2920 (aarch64_evpc_sve_tbl): New function.
2921 (aarch64_expand_vec_perm_const_1): Update after rename of
2922 aarch64_evpc_rev. Handle SVE permutes too, trying
2923 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2924 than aarch64_evpc_tbl.
2925 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2926 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2927 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2928 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2929 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2930 (aarch64_expand_sve_vcond): New functions.
2931 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2932 of aarch64_vector_mode_p.
2933 (aarch64_dwarf_poly_indeterminate_value): New function.
2934 (aarch64_compute_pressure_classes): Likewise.
2935 (aarch64_can_change_mode_class): Likewise.
2936 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2937 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2938 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2939 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2940 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2941 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2942 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2943 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2945 (Dn, Dl, Dr): Accept const as well as const_vector.
2946 (Dz): Likewise. Compare against CONST0_RTX.
2947 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2948 of "vector" where appropriate.
2949 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2950 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2951 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2952 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2953 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2954 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2955 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2956 (v_int_equiv): Extend to SVE modes.
2957 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2959 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2960 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2961 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2962 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2963 (SVE_COND_FP_CMP): New int iterators.
2964 (perm_hilo): Handle the new unpack unspecs.
2965 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2967 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2968 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2969 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2970 (aarch64_equality_operator, aarch64_constant_vector_operand)
2971 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2972 (aarch64_sve_nonimmediate_operand): Likewise.
2973 (aarch64_sve_general_operand): Likewise.
2974 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2975 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2976 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2977 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2978 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2979 (aarch64_sve_float_arith_immediate): Likewise.
2980 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2981 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2982 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2983 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2984 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2985 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2986 (aarch64_sve_float_arith_operand): Likewise.
2987 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2988 (aarch64_sve_float_mul_operand): Likewise.
2989 (aarch64_sve_vec_perm_operand): Likewise.
2990 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2991 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2992 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2993 as well as const_vector.
2994 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2995 in file. Use CONST0_RTX and CONSTM1_RTX.
2996 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2997 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2998 Use aarch64_simd_imm_zero.
2999 * config/aarch64/aarch64-sve.md: New file.
3000 * config/aarch64/aarch64.md: Include it.
3001 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3002 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3003 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3004 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3005 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3006 (sve): New attribute.
3007 (enabled): Disable instructions with the sve attribute unless
3009 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3010 aarch64_expand_mov_immediate.
3011 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3012 CNT[BHSD] immediates.
3013 (movti): Split CONST_POLY_INT moves into two halves.
3014 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3015 Split additions that need a temporary here if the destination
3016 is the stack pointer.
3017 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3018 (*add<mode>3_poly_1): New instruction.
3019 (set_clobber_cc): New expander.
3021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3023 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3024 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3025 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3026 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3027 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3028 Change innermode from fixed_mode_size to machine_mode.
3029 (simplify_subreg): Update call accordingly. Handle a constant-sized
3030 subreg of a variable-length CONST_VECTOR.
3032 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3033 Alan Hayward <alan.hayward@arm.com>
3034 David Sherwood <david.sherwood@arm.com>
3036 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3037 (add_offset_to_base): New function, split out from...
3038 (create_mem_ref): ...here. When handling a scale other than 1,
3039 check first whether the address is valid without the offset.
3040 Add it into the base if so, leaving the index and scale as-is.
3042 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3045 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3046 fold_for_warn before checking if arg2 is INTEGER_CST.
3048 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3050 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3051 (store_multiple_operation): Delete.
3052 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3053 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3054 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3055 guarded by TARGET_STRING.
3056 (rs6000_output_load_multiple): Delete.
3057 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3058 OPTION_MASK_STRING / TARGET_STRING handling.
3059 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3060 (const rs6000_opt_masks) <"string">: Change mask to 0.
3061 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3062 (MASK_STRING): Delete.
3063 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3065 (load_multiple): Delete.
3072 (store_multiple): Delete.
3079 (movmemsi_8reg): Delete.
3080 (corresponding unnamed define_insn): Delete.
3081 (movmemsi_6reg): Delete.
3082 (corresponding unnamed define_insn): Delete.
3083 (movmemsi_4reg): Delete.
3084 (corresponding unnamed define_insn): Delete.
3085 (movmemsi_2reg): Delete.
3086 (corresponding unnamed define_insn): Delete.
3087 (movmemsi_1reg): Delete.
3088 (corresponding unnamed define_insn): Delete.
3089 * config/rs6000/rs6000.opt (mno-string): New.
3090 (mstring): Replace by deprecation warning stub.
3091 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3093 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3095 * regrename.c (regrename_do_replace): If replacing the same
3096 reg multiple times, try to reuse last created gen_raw_REG.
3099 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3100 main to workaround a bug in GDB.
3102 2018-01-12 Tom de Vries <tom@codesourcery.com>
3105 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3107 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3109 PR rtl-optimization/80481
3110 * ira-color.c (get_cap_member): New function.
3111 (allocnos_conflict_by_live_ranges_p): Use it.
3112 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3113 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3115 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3118 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3119 (*saddl_se_1): Ditto.
3121 (*ssubl_se_1): Ditto.
3123 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3125 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3126 rather than wi::to_widest for DR_INITs.
3127 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3128 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3129 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3131 (vect_analyze_group_access_1): Note that here.
3133 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3135 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3136 polynomial type sizes.
3138 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3140 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3141 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3142 (gimple_add_tmp_var): Likewise.
3144 2018-01-12 Martin Liska <mliska@suse.cz>
3146 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3147 (gimple_alloc_sizes): Likewise.
3148 (dump_gimple_statistics): Use PRIu64 in printf format.
3149 * gimple.h: Change uint64_t to int.
3151 2018-01-12 Martin Liska <mliska@suse.cz>
3153 * tree-core.h: Use uint64_t instead of int.
3154 * tree.c (tree_node_counts): Likewise.
3155 (tree_node_sizes): Likewise.
3156 (dump_tree_statistics): Use PRIu64 in printf format.
3158 2018-01-12 Martin Liska <mliska@suse.cz>
3160 * Makefile.in: As qsort_chk is implemented in vec.c, add
3161 vec.o to linkage of gencfn-macros.
3162 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3163 passing the info to record_node_allocation_statistics.
3164 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3166 * ggc-common.c (struct ggc_usage): Add operator== and use
3167 it in operator< and compare function.
3168 * mem-stats.h (struct mem_usage): Likewise.
3169 * vec.c (struct vec_usage): Remove operator< and compare
3170 function. Can be simply inherited.
3172 2018-01-12 Martin Jambor <mjambor@suse.cz>
3175 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3176 * tree-ssa-math-opts.c: Include domwalk.h.
3177 (convert_mult_to_fma_1): New function.
3178 (fma_transformation_info): New type.
3179 (fma_deferring_state): Likewise.
3180 (cancel_fma_deferring): New function.
3181 (result_of_phi): Likewise.
3182 (last_fma_candidate_feeds_initial_phi): Likewise.
3183 (convert_mult_to_fma): Added deferring logic, split actual
3184 transformation to convert_mult_to_fma_1.
3185 (math_opts_dom_walker): New type.
3186 (math_opts_dom_walker::after_dom_children): New method, body moved
3187 here from pass_optimize_widening_mul::execute, added deferring logic
3189 (pass_optimize_widening_mul::execute): Moved most of code to
3190 math_opts_dom_walker::after_dom_children.
3191 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3192 * config/i386/i386.c (ix86_option_override_internal): Added
3193 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3195 2018-01-12 Richard Biener <rguenther@suse.de>
3198 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3199 inline instance vars.
3201 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3204 * config/rx/rx.c (rx_is_restricted_memory_address):
3207 2018-01-12 Richard Biener <rguenther@suse.de>
3209 PR tree-optimization/80846
3210 * target.def (split_reduction): New target hook.
3211 * targhooks.c (default_split_reduction): New function.
3212 * targhooks.h (default_split_reduction): Declare.
3213 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3214 target requests first reduce vectors by combining low and high
3216 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3217 (get_vectype_for_scalar_type_and_size): Export.
3218 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3219 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3220 * doc/tm.texi: Regenerate.
3221 * config/i386/i386.c (ix86_split_reduction): Implement
3222 TARGET_VECTORIZE_SPLIT_REDUCTION.
3224 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3227 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3228 in PIC mode except for TARGET_VXWORKS_RTP.
3229 * config/sparc/sparc.c: Include cfgrtl.h.
3230 (TARGET_INIT_PIC_REG): Define.
3231 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3232 (sparc_pic_register_p): New predicate.
3233 (sparc_legitimate_address_p): Use it.
3234 (sparc_legitimize_pic_address): Likewise.
3235 (sparc_delegitimize_address): Likewise.
3236 (sparc_mode_dependent_address_p): Likewise.
3237 (gen_load_pcrel_sym): Remove 4th parameter.
3238 (load_got_register): Adjust call to above. Remove obsolete stuff.
3239 (sparc_expand_prologue): Do not call load_got_register here.
3240 (sparc_flat_expand_prologue): Likewise.
3241 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3242 (sparc_use_pseudo_pic_reg): New function.
3243 (sparc_init_pic_reg): Likewise.
3244 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3245 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3247 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3249 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3250 Add item for branch_cost.
3252 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3254 PR rtl-optimization/83565
3255 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3256 not extend the result to a larger mode for rotate operations.
3257 (num_sign_bit_copies1): Likewise.
3259 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3262 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3264 Use values-Xc.o for -pedantic.
3265 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3267 2018-01-12 Martin Liska <mliska@suse.cz>
3270 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3272 (possible_polymorphic_call_targets): Use it.
3273 (ipa_devirt): Likewise.
3275 2018-01-12 Martin Liska <mliska@suse.cz>
3277 * profile-count.h (enum profile_quality): Use 0 as invalid
3278 enum value of profile_quality.
3280 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3282 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3283 -mext-string options.
3285 2018-01-12 Richard Biener <rguenther@suse.de>
3287 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3288 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3289 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3291 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3293 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3295 * configure.ac (--with-long-double-format): Add support for the
3296 configuration option to change the default long double format on
3298 * config.gcc (powerpc*-linux*-*): Likewise.
3299 * configure: Regenerate.
3300 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3301 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3302 used without modification.
3304 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3306 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3307 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3308 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3309 MISC_BUILTIN_SPEC_BARRIER.
3310 (rs6000_init_builtins): Likewise.
3311 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3313 (speculation_barrier): New define_insn.
3314 * doc/extend.texi: Document __builtin_speculation_barrier.
3316 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3319 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3320 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3321 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3323 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3324 integral modes instead of "ss" and "sd".
3325 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3326 vectors with 32-bit and 64-bit elements.
3327 (vecdupssescalarmodesuffix): New mode attribute.
3328 (vec_dup<mode>): Use it.
3330 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3333 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3334 frame if argument is passed on stack.
3336 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3339 * ree.c (combine_reaching_defs): Optimize also
3340 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3341 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3343 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3346 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3348 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3351 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3352 after they are computed.
3354 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3356 PR tree-optimization/83695
3357 * gimple-loop-linterchange.cc
3358 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3359 reset cached scev information after interchange.
3360 (pass_linterchange::execute): Remove call to scev_reset_htab.
3362 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3364 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3365 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3366 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3367 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3368 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3369 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3370 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3371 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3372 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3373 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3374 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3375 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3376 (V_lane_reg): Likewise.
3377 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3379 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3380 (vfmal_lane_low<mode>_intrinsic,
3381 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3382 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3383 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3384 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3385 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3386 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3388 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3390 * config/arm/arm-cpus.in (fp16fml): New feature.
3391 (ALL_SIMD): Add fp16fml.
3392 (armv8.2-a): Add fp16fml as an option.
3393 (armv8.3-a): Likewise.
3394 (armv8.4-a): Add fp16fml as part of fp16.
3395 * config/arm/arm.h (TARGET_FP16FML): Define.
3396 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3398 * config/arm/arm-modes.def (V2HF): Define.
3399 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3400 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3401 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3402 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3403 vfmsl_low, vfmsl_high): New set of builtins.
3404 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3405 (vfml_op): New code attribute.
3406 (VFMLHALVES): New int iterator.
3407 (VFML, VFMLSEL): New mode attributes.
3408 (V_reg): Define mapping for V2HF.
3409 (V_hi, V_lo): New mode attributes.
3410 (VF_constraint): Likewise.
3411 (vfml_half, vfml_half_selector): New int attributes.
3412 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3414 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3415 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3417 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3418 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3419 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3420 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3422 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3423 Document new effective target and option set.
3425 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3427 * config/arm/arm-cpus.in (armv8_4): New feature.
3428 (ARMv8_4a): New fgroup.
3429 (armv8.4-a): New arch.
3430 * config/arm/arm-tables.opt: Regenerate.
3431 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3432 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3433 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3434 Add matching rules for -march=armv8.4-a and extensions.
3435 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3437 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3440 * config/rx/rx.md (BW): New mode attribute.
3441 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3443 2018-01-11 Richard Biener <rguenther@suse.de>
3445 PR tree-optimization/83435
3446 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3447 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3448 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3450 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3451 Alan Hayward <alan.hayward@arm.com>
3452 David Sherwood <david.sherwood@arm.com>
3454 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3456 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3457 (aarch64_print_address_internal): Use it to check for a zero offset.
3459 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3460 Alan Hayward <alan.hayward@arm.com>
3461 David Sherwood <david.sherwood@arm.com>
3463 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3464 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3465 Return a poly_int64 rather than a HOST_WIDE_INT.
3466 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3467 rather than a HOST_WIDE_INT.
3468 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3469 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3470 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3471 final_offset from HOST_WIDE_INT to poly_int64.
3472 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3473 to_constant when getting the number of units in an Advanced SIMD
3475 (aarch64_builtin_vectorized_function): Check for a constant number
3477 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3479 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3480 attribute instead of GET_MODE_NUNITS.
3481 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3482 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3483 GET_MODE_SIZE for fixed-size registers.
3484 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3485 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3486 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3487 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3488 (aarch64_print_operand, aarch64_print_address_internal)
3489 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3490 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3491 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3492 Handle polynomial GET_MODE_SIZE.
3493 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3494 wider than SImode without modification.
3495 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3496 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3497 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3498 passing and returning SVE modes.
3499 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3500 rather than GEN_INT.
3501 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3502 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3503 (aarch64_allocate_and_probe_stack_space): Likewise.
3504 (aarch64_layout_frame): Cope with polynomial offsets.
3505 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3506 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3508 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3509 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3510 poly_int64 rather than a HOST_WIDE_INT.
3511 (aarch64_get_separate_components, aarch64_process_components)
3512 (aarch64_expand_prologue, aarch64_expand_epilogue)
3513 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3514 (aarch64_anchor_offset): New function, split out from...
3515 (aarch64_legitimize_address): ...here.
3516 (aarch64_builtin_vectorization_cost): Handle polynomial
3517 TYPE_VECTOR_SUBPARTS.
3518 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3520 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3521 number of elements from the PARALLEL rather than the mode.
3522 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3523 rather than GET_MODE_BITSIZE.
3524 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3525 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3526 (aarch64_expand_vec_perm_const_1): Handle polynomial
3527 d->perm.length () and d->perm elements.
3528 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3529 Apply to_constant to d->perm elements.
3530 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3531 polynomial CONST_VECTOR_NUNITS.
3532 (aarch64_move_pointer): Take amount as a poly_int64 rather
3534 (aarch64_progress_pointer): Avoid temporary variable.
3535 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3536 the mode attribute instead of GET_MODE.
3538 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3539 Alan Hayward <alan.hayward@arm.com>
3540 David Sherwood <david.sherwood@arm.com>
3542 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3543 x exists before using it.
3544 (aarch64_add_constant_internal): Rename to...
3545 (aarch64_add_offset_1): ...this. Replace regnum with separate
3546 src and dest rtxes. Handle the case in which they're different,
3547 including when the offset is zero. Replace scratchreg with an rtx.
3548 Use 2 additions if there is no spare register into which we can
3549 move a 16-bit constant.
3550 (aarch64_add_constant): Delete.
3551 (aarch64_add_offset): Replace reg with separate src and dest
3552 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3553 Use aarch64_add_offset_1.
3554 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3555 an rtx rather than an int. Take the delta as a poly_int64
3556 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3557 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3558 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3559 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3560 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3562 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3563 aarch64_add_constant.
3565 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3567 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3568 Use scalar_float_mode.
3570 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3572 * config/aarch64/aarch64-simd.md
3573 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3574 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3575 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3576 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3577 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3578 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3579 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3580 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3581 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3582 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3584 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3587 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3588 targ_options->x_arm_arch_string is non NULL.
3590 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3592 * config/aarch64/aarch64.h
3593 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3595 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3598 * expmed.c (emit_store_flag_force): Swap if const op0
3599 and change VOIDmode to mode of op0.
3601 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3603 PR rtl-optimization/83761
3604 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3605 than bytes to mode_for_size.
3607 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3610 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3611 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3614 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3617 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3618 when in layout mode.
3619 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3620 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3623 2018-01-10 Michael Collison <michael.collison@arm.com>
3625 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3626 * config/aarch64/aarch64-option-extension.def: Add
3627 AARCH64_OPT_EXTENSION of 'fp16fml'.
3628 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3629 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3630 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3631 * config/aarch64/constraints.md (Ui7): New constraint.
3632 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3633 (VFMLA_SEL_W): Ditto.
3636 (VFMLA16_LOW): New int iterator.
3637 (VFMLA16_HIGH): Ditto.
3638 (UNSPEC_FMLAL): New unspec.
3639 (UNSPEC_FMLSL): Ditto.
3640 (UNSPEC_FMLAL2): Ditto.
3641 (UNSPEC_FMLSL2): Ditto.
3642 (f16mac): New code attribute.
3643 * config/aarch64/aarch64-simd-builtins.def
3644 (aarch64_fmlal_lowv2sf): Ditto.
3645 (aarch64_fmlsl_lowv2sf): Ditto.
3646 (aarch64_fmlalq_lowv4sf): Ditto.
3647 (aarch64_fmlslq_lowv4sf): Ditto.
3648 (aarch64_fmlal_highv2sf): Ditto.
3649 (aarch64_fmlsl_highv2sf): Ditto.
3650 (aarch64_fmlalq_highv4sf): Ditto.
3651 (aarch64_fmlslq_highv4sf): Ditto.
3652 (aarch64_fmlal_lane_lowv2sf): Ditto.
3653 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3654 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3655 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3656 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3657 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3658 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3659 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3660 (aarch64_fmlal_lane_highv2sf): Ditto.
3661 (aarch64_fmlsl_lane_highv2sf): Ditto.
3662 (aarch64_fmlal_laneq_highv2sf): Ditto.
3663 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3664 (aarch64_fmlalq_lane_highv4sf): Ditto.
3665 (aarch64_fmlsl_lane_highv4sf): Ditto.
3666 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3667 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3668 * config/aarch64/aarch64-simd.md:
3669 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3670 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3671 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3672 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3673 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3674 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3675 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3676 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3677 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3678 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3679 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3680 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3681 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3682 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3683 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3684 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3685 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3686 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3687 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3688 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3689 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3690 (vfmlsl_low_u32): Ditto.
3691 (vfmlalq_low_u32): Ditto.
3692 (vfmlslq_low_u32): Ditto.
3693 (vfmlal_high_u32): Ditto.
3694 (vfmlsl_high_u32): Ditto.
3695 (vfmlalq_high_u32): Ditto.
3696 (vfmlslq_high_u32): Ditto.
3697 (vfmlal_lane_low_u32): Ditto.
3698 (vfmlsl_lane_low_u32): Ditto.
3699 (vfmlal_laneq_low_u32): Ditto.
3700 (vfmlsl_laneq_low_u32): Ditto.
3701 (vfmlalq_lane_low_u32): Ditto.
3702 (vfmlslq_lane_low_u32): Ditto.
3703 (vfmlalq_laneq_low_u32): Ditto.
3704 (vfmlslq_laneq_low_u32): Ditto.
3705 (vfmlal_lane_high_u32): Ditto.
3706 (vfmlsl_lane_high_u32): Ditto.
3707 (vfmlal_laneq_high_u32): Ditto.
3708 (vfmlsl_laneq_high_u32): Ditto.
3709 (vfmlalq_lane_high_u32): Ditto.
3710 (vfmlslq_lane_high_u32): Ditto.
3711 (vfmlalq_laneq_high_u32): Ditto.
3712 (vfmlslq_laneq_high_u32): Ditto.
3713 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3714 (AARCH64_FL_FOR_ARCH8_4): New.
3715 (AARCH64_ISA_F16FML): New ISA flag.
3716 (TARGET_F16FML): New feature flag for fp16fml.
3717 (doc/invoke.texi): Document new fp16fml option.
3719 2018-01-10 Michael Collison <michael.collison@arm.com>
3721 * config/aarch64/aarch64-builtins.c:
3722 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3723 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3724 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3725 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3726 (AARCH64_ISA_SHA3): New ISA flag.
3727 (TARGET_SHA3): New feature flag for sha3.
3728 * config/aarch64/iterators.md (sha512_op): New int attribute.
3729 (CRYPTO_SHA512): New int iterator.
3730 (UNSPEC_SHA512H): New unspec.
3731 (UNSPEC_SHA512H2): Ditto.
3732 (UNSPEC_SHA512SU0): Ditto.
3733 (UNSPEC_SHA512SU1): Ditto.
3734 * config/aarch64/aarch64-simd-builtins.def
3735 (aarch64_crypto_sha512hqv2di): New builtin.
3736 (aarch64_crypto_sha512h2qv2di): Ditto.
3737 (aarch64_crypto_sha512su0qv2di): Ditto.
3738 (aarch64_crypto_sha512su1qv2di): Ditto.
3739 (aarch64_eor3qv8hi): Ditto.
3740 (aarch64_rax1qv2di): Ditto.
3741 (aarch64_xarqv2di): Ditto.
3742 (aarch64_bcaxqv8hi): Ditto.
3743 * config/aarch64/aarch64-simd.md:
3744 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3745 (aarch64_crypto_sha512su0qv2di): Ditto.
3746 (aarch64_crypto_sha512su1qv2di): Ditto.
3747 (aarch64_eor3qv8hi): Ditto.
3748 (aarch64_rax1qv2di): Ditto.
3749 (aarch64_xarqv2di): Ditto.
3750 (aarch64_bcaxqv8hi): Ditto.
3751 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3752 (vsha512h2q_u64): Ditto.
3753 (vsha512su0q_u64): Ditto.
3754 (vsha512su1q_u64): Ditto.
3755 (veor3q_u16): Ditto.
3756 (vrax1q_u64): Ditto.
3758 (vbcaxq_u16): Ditto.
3759 * config/arm/types.md (crypto_sha512): New type attribute.
3760 (crypto_sha3): Ditto.
3761 (doc/invoke.texi): Document new sha3 option.
3763 2018-01-10 Michael Collison <michael.collison@arm.com>
3765 * config/aarch64/aarch64-builtins.c:
3766 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3767 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3768 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3769 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3770 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3771 (AARCH64_ISA_SM4): New ISA flag.
3772 (TARGET_SM4): New feature flag for sm4.
3773 * config/aarch64/aarch64-simd-builtins.def
3774 (aarch64_sm3ss1qv4si): Ditto.
3775 (aarch64_sm3tt1aq4si): Ditto.
3776 (aarch64_sm3tt1bq4si): Ditto.
3777 (aarch64_sm3tt2aq4si): Ditto.
3778 (aarch64_sm3tt2bq4si): Ditto.
3779 (aarch64_sm3partw1qv4si): Ditto.
3780 (aarch64_sm3partw2qv4si): Ditto.
3781 (aarch64_sm4eqv4si): Ditto.
3782 (aarch64_sm4ekeyqv4si): Ditto.
3783 * config/aarch64/aarch64-simd.md:
3784 (aarch64_sm3ss1qv4si): Ditto.
3785 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3786 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3787 (aarch64_sm4eqv4si): Ditto.
3788 (aarch64_sm4ekeyqv4si): Ditto.
3789 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3790 (sm3part_op): Ditto.
3791 (CRYPTO_SM3TT): Ditto.
3792 (CRYPTO_SM3PART): Ditto.
3793 (UNSPEC_SM3SS1): New unspec.
3794 (UNSPEC_SM3TT1A): Ditto.
3795 (UNSPEC_SM3TT1B): Ditto.
3796 (UNSPEC_SM3TT2A): Ditto.
3797 (UNSPEC_SM3TT2B): Ditto.
3798 (UNSPEC_SM3PARTW1): Ditto.
3799 (UNSPEC_SM3PARTW2): Ditto.
3800 (UNSPEC_SM4E): Ditto.
3801 (UNSPEC_SM4EKEY): Ditto.
3802 * config/aarch64/constraints.md (Ui2): New constraint.
3803 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3804 * config/arm/types.md (crypto_sm3): New type attribute.
3805 (crypto_sm4): Ditto.
3806 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3807 (vsm3tt1aq_u32): Ditto.
3808 (vsm3tt1bq_u32): Ditto.
3809 (vsm3tt2aq_u32): Ditto.
3810 (vsm3tt2bq_u32): Ditto.
3811 (vsm3partw1q_u32): Ditto.
3812 (vsm3partw2q_u32): Ditto.
3813 (vsm4eq_u32): Ditto.
3814 (vsm4ekeyq_u32): Ditto.
3815 (doc/invoke.texi): Document new sm4 option.
3817 2018-01-10 Michael Collison <michael.collison@arm.com>
3819 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3820 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3821 (AARCH64_FL_FOR_ARCH8_4): New.
3822 (AARCH64_FL_V8_4): New flag.
3823 (doc/invoke.texi): Document new armv8.4-a option.
3825 2018-01-10 Michael Collison <michael.collison@arm.com>
3827 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3828 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3829 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3830 * config/aarch64/aarch64-option-extension.def: Add
3831 AARCH64_OPT_EXTENSION of 'sha2'.
3832 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3833 (crypto): Disable sha2 and aes if crypto disabled.
3834 (crypto): Enable aes and sha2 if enabled.
3835 (simd): Disable sha2 and aes if simd disabled.
3836 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3838 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3839 (TARGET_SHA2): New feature flag for sha2.
3840 (TARGET_AES): New feature flag for aes.
3841 * config/aarch64/aarch64-simd.md:
3842 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3843 conditional on TARGET_AES.
3844 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3845 (aarch64_crypto_sha1hsi): Make pattern conditional
3847 (aarch64_crypto_sha1hv4si): Ditto.
3848 (aarch64_be_crypto_sha1hv4si): Ditto.
3849 (aarch64_crypto_sha1su1v4si): Ditto.
3850 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3851 (aarch64_crypto_sha1su0v4si): Ditto.
3852 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3853 (aarch64_crypto_sha256su0v4si): Ditto.
3854 (aarch64_crypto_sha256su1v4si): Ditto.
3855 (doc/invoke.texi): Document new aes and sha2 options.
3857 2018-01-10 Martin Sebor <msebor@redhat.com>
3859 PR tree-optimization/83781
3860 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3863 2018-01-11 Martin Sebor <msebor@gmail.com>
3864 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3866 PR tree-optimization/83501
3867 PR tree-optimization/81703
3869 * tree-ssa-strlen.c (get_string_cst): Rename...
3870 (get_string_len): ...to this. Handle global constants.
3871 (handle_char_store): Adjust.
3873 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3874 Jim Wilson <jimw@sifive.com>
3876 * config/riscv/riscv-protos.h (riscv_output_return): New.
3877 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3878 (riscv_attribute_table, riscv_output_return),
3879 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3880 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3881 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3883 (riscv_expand_prologue): Add early return for naked function.
3884 (riscv_expand_epilogue): Likewise.
3885 (riscv_function_ok_for_sibcall): Return false for naked function.
3886 (riscv_set_current_function): New.
3887 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3888 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3889 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3890 * doc/extend.texi (RISC-V Function Attributes): New.
3892 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3894 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3895 check for 128-bit long double before checking TCmode.
3896 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3897 128-bit long doubles before checking TFmode or TCmode.
3898 (FLOAT128_IBM_P): Likewise.
3900 2018-01-10 Martin Sebor <msebor@redhat.com>
3902 PR tree-optimization/83671
3903 * builtins.c (c_strlen): Unconditionally return zero for the empty
3905 Use -Warray-bounds for warnings.
3906 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3907 for non-constant array indices with COMPONENT_REF, arrays of
3908 arrays, and pointers to arrays.
3909 (gimple_fold_builtin_strlen): Determine and set length range for
3910 non-constant character arrays.
3912 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3915 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3918 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3920 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3922 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3925 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3926 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3927 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3928 indexed_or_indirect_operand predicate.
3929 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3930 (*vsx_le_perm_load_v8hi): Likewise.
3931 (*vsx_le_perm_load_v16qi): Likewise.
3932 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3933 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3934 (*vsx_le_perm_store_v8hi): Likewise.
3935 (*vsx_le_perm_store_v16qi): Likewise.
3936 (eight unnamed splitters): Likewise.
3938 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3940 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3941 * config/rs6000/emmintrin.h: Likewise.
3942 * config/rs6000/mmintrin.h: Likewise.
3943 * config/rs6000/xmmintrin.h: Likewise.
3945 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3948 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3950 * tree.c (tree_nop_conversion): Return true for location wrapper
3952 (maybe_wrap_with_location): New function.
3953 (selftest::check_strip_nops): New function.
3954 (selftest::test_location_wrappers): New function.
3955 (selftest::tree_c_tests): Call it.
3956 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3957 (maybe_wrap_with_location): New decl.
3958 (EXPR_LOCATION_WRAPPER_P): New macro.
3959 (location_wrapper_p): New inline function.
3960 (tree_strip_any_location_wrapper): New inline function.
3962 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3965 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3966 stack_realign_offset for the largest alignment of stack slot
3968 (ix86_find_max_used_stack_alignment): New function.
3969 (ix86_finalize_stack_frame_flags): Use it. Set
3970 max_used_stack_alignment if we don't realign stack.
3971 * config/i386/i386.h (machine_function): Add
3972 max_used_stack_alignment.
3974 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3976 * config/arm/arm.opt (-mbranch-cost): New option.
3977 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3980 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3983 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3984 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3986 2018-01-10 Richard Biener <rguenther@suse.de>
3989 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3990 early out so it also covers the case where we have a non-NULL
3993 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3995 PR tree-optimization/83753
3996 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3997 for non-strided grouped accesses if the number of elements is 1.
3999 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4002 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4003 * i386.h (TARGET_USE_GATHER): Define.
4004 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4006 2018-01-10 Martin Liska <mliska@suse.cz>
4009 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4010 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4012 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4013 CLEANUP_NO_PARTITIONING is not set.
4015 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4017 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4018 for vectors, as a partial revert of r254296.
4019 * rtl.h (const_vec_p): Delete.
4020 (const_vec_duplicate_p): Don't test for vector CONSTs.
4021 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4022 * expmed.c (make_tree): Likewise.
4025 * common.md (E, F): Use CONSTANT_P instead of checking for
4027 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4028 checking for CONST_VECTOR.
4030 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4033 * predict.c (force_edge_cold): Handle in more sane way edges
4036 2018-01-09 Carl Love <cel@us.ibm.com>
4038 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4040 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4041 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4042 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4043 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4044 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4045 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4046 * config/rs6000/rs6000-protos.h: Add extern defition for
4047 rs6000_generate_float2_double_code.
4048 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4050 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4051 (float2_v2df): Add define_expand.
4053 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4056 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4057 op_mode in the force_to_mode call.
4059 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4061 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4062 instead of checking each element individually.
4063 (aarch64_evpc_uzp): Likewise.
4064 (aarch64_evpc_zip): Likewise.
4065 (aarch64_evpc_ext): Likewise.
4066 (aarch64_evpc_rev): Likewise.
4067 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4068 instead of checking each element individually. Return true without
4070 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4071 whether all selected elements come from the same input, instead of
4072 checking each element individually. Remove calls to gen_rtx_REG,
4073 start_sequence and end_sequence and instead assert that no rtl is
4076 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4078 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4079 order of HIGH and CONST checks.
4081 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4083 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4084 if the destination isn't an SSA_NAME.
4086 2018-01-09 Richard Biener <rguenther@suse.de>
4088 PR tree-optimization/83668
4089 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4091 (canonicalize_loop_form): ... here, renamed from ...
4092 (canonicalize_loop_closed_ssa_form): ... this and amended to
4093 swap successor edges for loop exit blocks to make us use
4094 the RPO order we need for initial schedule generation.
4096 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4098 PR tree-optimization/64811
4099 * match.pd: When optimizing comparisons with Inf, avoid
4100 introducing or losing exceptions from comparisons with NaN.
4102 2018-01-09 Martin Liska <mliska@suse.cz>
4105 * asan.c (shadow_mem_size): Add gcc_assert.
4107 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4109 Don't save registers in main().
4112 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4113 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4114 * config/avr/avr.c (avr_set_current_function): Don't error if
4115 naked, OS_task or OS_main are specified at the same time.
4116 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4118 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4120 * common/config/avr/avr-common.c (avr_option_optimization_table):
4121 Switch on -mmain-is-OS_task for optimizing compilations.
4123 2018-01-09 Richard Biener <rguenther@suse.de>
4125 PR tree-optimization/83572
4126 * graphite.c: Include cfganal.h.
4127 (graphite_transform_loops): Connect infinite loops to exit
4128 and remove fake edges at the end.
4130 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4132 * ipa-inline.c (edge_badness): Revert accidental checkin.
4134 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4137 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4138 symbols; not inline clones.
4140 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4143 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4144 hard registers. Formatting fixes.
4146 PR preprocessor/83722
4147 * gcc.c (try_generate_repro): Pass
4148 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4149 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4152 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4153 Kito Cheng <kito.cheng@gmail.com>
4155 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4156 (riscv_leaf_function_p): Delete.
4157 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4159 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4161 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4163 (do_ifelse): New function.
4164 (do_isel): New function.
4165 (do_sub3): New function.
4166 (do_add3): New function.
4167 (do_load_mask_compare): New function.
4168 (do_overlap_load_compare): New function.
4169 (expand_compare_loop): New function.
4170 (expand_block_compare): Call expand_compare_loop() when appropriate.
4171 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4173 (-mblock-compare-inline-loop-limit): New option.
4175 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4178 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4179 Reverse order of second and third operands in first alternative.
4180 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4181 of first and second elements in UNSPEC_VPERMR vector.
4182 (altivec_expand_vec_perm_le): Likewise.
4184 2018-01-08 Jeff Law <law@redhat.com>
4186 PR rtl-optimizatin/81308
4187 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4188 (process_switch): If group_case_labels makes a change, then set
4190 (pass_convert_switch::execute): If a switch is converted, then
4191 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4193 PR rtl-optimization/81308
4194 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4197 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4199 PR target/83663 - Revert r255946
4200 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4201 generation for cases where splatting a value is not useful.
4202 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4203 across a vec_duplicate and a paradoxical subreg forming a vector
4204 mode to a vec_concat.
4206 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4208 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4209 -march=armv8.3-a variants.
4210 * config/arm/t-multilib: Likewise.
4211 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4213 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4215 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4217 (cceq_ior_compare_complement): Give it a name so I can use it, and
4218 change boolean_or_operator predicate to boolean_operator so it can
4219 be used to generate a crand.
4220 (eqne): New code iterator.
4221 (bd/bd_neg): New code_attrs.
4222 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4223 a single define_insn.
4224 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4225 decrement (bdnzt/bdnzf/bdzt/bdzf).
4226 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4227 with the new names of the branch decrement patterns, and added the
4228 names of the branch decrement conditional patterns.
4230 2018-01-08 Richard Biener <rguenther@suse.de>
4232 PR tree-optimization/83563
4233 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4236 2018-01-08 Richard Biener <rguenther@suse.de>
4239 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4241 2018-01-08 Richard Biener <rguenther@suse.de>
4243 PR tree-optimization/83685
4244 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4245 references to abnormals.
4247 2018-01-08 Richard Biener <rguenther@suse.de>
4250 * dwarf2out.c (output_indirect_strings): Handle empty
4251 skeleton_debug_str_hash.
4252 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4254 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4256 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4257 (emit_store_direct): Likewise.
4258 (arc_trampoline_adjust_address): Likewise.
4259 (arc_asm_trampoline_template): New function.
4260 (arc_initialize_trampoline): Use asm_trampoline_template.
4261 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4262 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4263 * config/arc/arc.md (flush_icache): Delete pattern.
4265 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4267 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4268 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4271 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4274 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4275 by not USED_FOR_TARGET.
4276 (make_pass_resolve_sw_modes): Likewise.
4278 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4280 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4283 2018-01-08 Richard Biener <rguenther@suse.de>
4286 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4288 2018-01-08 Richard Biener <rguenther@suse.de>
4291 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4293 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4296 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4297 basic blocks with a small number of successors.
4298 (convert_control_dep_chain_into_preds): Improve handling of
4300 (dump_predicates): Split apart into...
4301 (dump_pred_chain): ...here...
4302 (dump_pred_info): ...and here.
4303 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4304 (can_chain_union_be_invalidated_p): Improve check for invalidation
4306 (uninit_uses_cannot_happen): Avoid unnecessary if
4307 convert_control_dep_chain_into_preds yielded nothing.
4309 2018-01-06 Martin Sebor <msebor@redhat.com>
4311 PR tree-optimization/83640
4312 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4313 subtracting negative offset from size.
4314 (builtin_access::overlap): Adjust offset bounds of the access to fall
4315 within the size of the object if possible.
4317 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4319 PR rtl-optimization/83699
4320 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4321 extract_bit_field_as_subreg to cases in which the extracted
4322 value is also a vector.
4324 * lra-constraints.c (process_alt_operands): Test for the equivalence
4325 substitutions when detecting a possible reload cycle.
4327 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4330 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4331 by default if flag_selective_schedling{,2}. Formatting fixes.
4333 PR rtl-optimization/83682
4334 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4335 if it has non-VECTOR_MODE element mode.
4336 (vec_duplicate_p): Likewise.
4339 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4340 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4342 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4345 * config/i386/i386-builtin.def
4346 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4347 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4348 Require also OPTION_MASK_ISA_AVX512F in addition to
4349 OPTION_MASK_ISA_GFNI.
4350 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4351 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4352 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4353 to OPTION_MASK_ISA_GFNI.
4354 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4355 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4356 OPTION_MASK_ISA_AVX512BW.
4357 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4358 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4359 addition to OPTION_MASK_ISA_GFNI.
4360 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4361 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4362 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4363 to OPTION_MASK_ISA_GFNI.
4364 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4365 a requirement for all ISAs rather than any of them with a few
4367 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4369 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4370 bitmasks to be enabled with 3 exceptions, instead of requiring any
4371 enabled ISA with lots of exceptions.
4372 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4373 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4374 Change avx512bw in isa attribute to avx512f.
4375 * config/i386/sgxintrin.h: Add license boilerplate.
4376 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4377 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4378 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4379 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4381 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4382 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4383 temporarily sse2 rather than sse if not enabled already.
4386 * config/i386/sse.md (VI248_VLBW): Rename to ...
4387 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4388 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4389 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4390 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4391 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4392 mode iterator instead of VI248_VLBW.
4394 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4396 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4397 (record_modified): Skip clobbers; add debug output.
4398 (param_change_prob): Use sreal frequencies.
4400 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4402 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4403 punt for user-aligned variables.
4405 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4407 * tree-chrec.c (chrec_contains_symbols): Return true for
4410 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4413 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4414 of (x|y) == x for BICS pattern.
4416 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4418 PR tree-optimization/83605
4419 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4420 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4423 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4425 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4426 * config/epiphany/rtems.h: New file.
4428 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4429 Uros Bizjak <ubizjak@gmail.com>
4432 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4433 QIreg_operand instead of register_operand predicate.
4434 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4435 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4436 comments instead of -fmitigate[-_]rop.
4438 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4441 * cgraphunit.c (symbol_table::compile): Switch to text_section
4442 before calling assembly_start debug hook.
4443 * run-rtl-passes.c (run_rtl_passes): Likewise.
4446 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4448 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4449 range_int_cst_p rather than !symbolic_range_p before calling
4450 extract_range_from_multiplicative_op_1.
4452 2018-01-04 Jeff Law <law@redhat.com>
4454 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4455 redundant test in assertion.
4457 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4459 * doc/rtl.texi: Document machine_mode wrapper classes.
4461 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4463 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4466 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4468 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4469 the VEC_PERM_EXPR fold to fail.
4471 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4474 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4475 to switched_sections.
4477 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4480 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4483 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4486 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4487 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4489 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4492 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4493 is BLKmode and bitpos not zero or mode change is needed.
4495 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4498 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4501 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4504 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4505 instead of MULT rtx. Update all corresponding splitters.
4507 (*ssub<modesuffix>): Ditto.
4509 (*cmp_sadd_di): Update split patterns.
4510 (*cmp_sadd_si): Ditto.
4511 (*cmp_sadd_sidi): Ditto.
4512 (*cmp_ssub_di): Ditto.
4513 (*cmp_ssub_si): Ditto.
4514 (*cmp_ssub_sidi): Ditto.
4515 * config/alpha/predicates.md (const23_operand): New predicate.
4516 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4517 Look for ASHIFT, not MULT inner operand.
4518 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4520 2018-01-04 Martin Liska <mliska@suse.cz>
4522 PR gcov-profile/83669
4523 * gcov.c (output_intermediate_file): Add version to intermediate
4525 * doc/gcov.texi: Document new field 'version' in intermediate
4526 file format. Fix location of '-k' option of gcov command.
4528 2018-01-04 Martin Liska <mliska@suse.cz>
4531 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4533 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4535 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4537 2018-01-03 Martin Sebor <msebor@redhat.com>
4539 PR tree-optimization/83655
4540 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4541 checking calls with invalid arguments.
4543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4545 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4546 (vectorizable_mask_load_store): Delete.
4547 (vectorizable_call): Return false for masked loads and stores.
4548 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4549 instead of gimple_assign_rhs1.
4550 (vectorizable_load): Handle IFN_MASK_LOAD.
4551 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4553 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4555 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4557 (vectorizable_mask_load_store): ...here.
4558 (vectorizable_load): ...and here.
4560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4562 * tree-vect-stmts.c (vect_build_all_ones_mask)
4563 (vect_build_zero_merge_argument): New functions, split out from...
4564 (vectorizable_load): ...here.
4566 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4568 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4570 (vectorizable_mask_load_store): ...here.
4571 (vectorizable_store): ...and here.
4573 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4575 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4577 (vectorizable_mask_load_store): ...here.
4579 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4581 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4582 (vect_model_store_cost): Take a vec_load_store_type instead of a
4584 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4585 (vect_model_store_cost): Take a vec_load_store_type instead of a
4587 (vectorizable_mask_load_store): Update accordingly.
4588 (vectorizable_store): Likewise.
4589 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4591 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4593 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4594 IFN_MASK_LOAD calls here rather than...
4595 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4598 Alan Hayward <alan.hayward@arm.com>
4599 David Sherwood <david.sherwood@arm.com>
4601 * expmed.c (extract_bit_field_1): For vector extracts,
4602 fall back to extract_bit_field_as_subreg if vec_extract
4605 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4606 Alan Hayward <alan.hayward@arm.com>
4607 David Sherwood <david.sherwood@arm.com>
4609 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4610 they are variable or constant sized.
4611 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4612 slots for constant-sized data.
4614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4615 Alan Hayward <alan.hayward@arm.com>
4616 David Sherwood <david.sherwood@arm.com>
4618 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4619 handling COND_EXPRs with boolean comparisons, try to find a better
4620 basis for the mask type than the boolean itself.
4622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4624 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4625 is calculated and how it can be overridden.
4626 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4627 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4629 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4632 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4633 Alan Hayward <alan.hayward@arm.com>
4634 David Sherwood <david.sherwood@arm.com>
4636 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4637 Remove the mode argument.
4638 (aarch64_simd_valid_immediate): Remove the mode and inverse
4640 * config/aarch64/iterators.md (bitsize): New iterator.
4641 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4642 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4643 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4644 aarch64_simd_valid_immediate.
4645 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4646 (aarch64_reg_or_bic_imm): Likewise.
4647 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4648 with an insn_type enum and msl with a modifier_type enum.
4649 Replace element_width with a scalar_mode. Change the shift
4650 to unsigned int. Add constructors for scalar_float_mode and
4651 scalar_int_mode elements.
4652 (aarch64_vect_float_const_representable_p): Delete.
4653 (aarch64_can_const_movi_rtx_p)
4654 (aarch64_simd_scalar_immediate_valid_for_move)
4655 (aarch64_simd_make_constant): Update call to
4656 aarch64_simd_valid_immediate.
4657 (aarch64_advsimd_valid_immediate_hs): New function.
4658 (aarch64_advsimd_valid_immediate): Likewise.
4659 (aarch64_simd_valid_immediate): Remove mode and inverse
4660 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4661 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4662 and aarch64_float_const_representable_p on the result.
4663 (aarch64_output_simd_mov_immediate): Remove mode argument.
4664 Update call to aarch64_simd_valid_immediate and use of
4665 simd_immediate_info.
4666 (aarch64_output_scalar_simd_mov_immediate): Update call
4669 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4670 Alan Hayward <alan.hayward@arm.com>
4671 David Sherwood <david.sherwood@arm.com>
4673 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4674 (mode_nunits): Likewise CONST_MODE_NUNITS.
4675 * machmode.def (ADJUST_NUNITS): Document.
4676 * genmodes.c (mode_data::need_nunits_adj): New field.
4677 (blank_mode): Update accordingly.
4678 (adj_nunits): New variable.
4679 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4681 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4682 listed in adj_nunits.
4683 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4684 listed in adj_nunits. Don't emit case statements for such modes.
4685 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4686 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4687 nothing if adj_nunits is nonnull.
4688 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4689 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4690 (emit_mode_fbit): Update use of print_maybe_const_decl.
4691 (emit_move_size): Likewise. Treat the array as non-const
4693 (emit_mode_adjustments): Handle adj_nunits.
4695 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4697 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4698 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4699 (VECTOR_MODES): Use it.
4700 (make_vector_modes): Take the prefix as an argument.
4702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4703 Alan Hayward <alan.hayward@arm.com>
4704 David Sherwood <david.sherwood@arm.com>
4706 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4707 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4708 for MODE_VECTOR_BOOL.
4709 * machmode.def (VECTOR_BOOL_MODE): Document.
4710 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4711 (make_vector_bool_mode): New function.
4712 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4714 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4715 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4717 * stor-layout.c (int_mode_for_mode): Likewise.
4718 * tree.c (build_vector_type_for_mode): Likewise.
4719 * varasm.c (output_constant_pool_2): Likewise.
4720 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4721 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4722 for MODE_VECTOR_BOOL.
4723 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4724 of mode class checks.
4725 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4726 instead of a list of mode class checks.
4727 (expand_vector_scalar_condition): Likewise.
4728 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4730 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4731 Alan Hayward <alan.hayward@arm.com>
4732 David Sherwood <david.sherwood@arm.com>
4734 * machmode.h (mode_size): Change from unsigned short to
4736 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4737 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4738 or if measurement_type is not polynomial.
4739 (fixed_size_mode::includes_p): Check for constant-sized modes.
4740 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4741 return a poly_uint16 rather than an unsigned short.
4742 (emit_mode_size): Change the type of mode_size from unsigned short
4743 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4744 (emit_mode_adjustments): Cope with polynomial vector sizes.
4745 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4747 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4749 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4750 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4751 * caller-save.c (setup_save_areas): Likewise.
4752 (replace_reg_with_saved_mem): Likewise.
4753 * calls.c (emit_library_call_value_1): Likewise.
4754 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4755 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4756 (gen_lowpart_for_combine): Likewise.
4757 * convert.c (convert_to_integer_1): Likewise.
4758 * cse.c (equiv_constant, cse_insn): Likewise.
4759 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4760 (cselib_subst_to_values): Likewise.
4761 * dce.c (word_dce_process_block): Likewise.
4762 * df-problems.c (df_word_lr_mark_ref): Likewise.
4763 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4764 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4765 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4766 (rtl_for_decl_location): Likewise.
4767 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4768 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4769 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4770 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4771 (expand_expr_real_1): Likewise.
4772 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4773 (pad_below): Likewise.
4774 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4775 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4776 * ira.c (get_subreg_tracking_sizes): Likewise.
4777 * ira-build.c (ira_create_allocno_objects): Likewise.
4778 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4779 (ira_sort_regnos_for_alter_reg): Likewise.
4780 * ira-costs.c (record_operand_costs): Likewise.
4781 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4782 (resolve_simple_move): Likewise.
4783 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4784 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4785 (lra_constraints): Likewise.
4786 (CONST_POOL_OK_P): Reject variable-sized modes.
4787 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4788 (add_pseudo_to_slot, lra_spill): Likewise.
4789 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4790 * optabs-query.c (get_best_extraction_insn): Likewise.
4791 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4792 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4793 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4794 * recog.c (offsettable_address_addr_space_p): Likewise.
4795 * regcprop.c (maybe_mode_change): Likewise.
4796 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4797 * regrename.c (build_def_use): Likewise.
4798 * regstat.c (dump_reg_info): Likewise.
4799 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4800 (find_reloads, find_reloads_subreg_address): Likewise.
4801 * reload1.c (eliminate_regs_1): Likewise.
4802 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4803 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4804 (simplify_binary_operation_1, simplify_subreg): Likewise.
4805 * targhooks.c (default_function_arg_padding): Likewise.
4806 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4807 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4808 (verify_gimple_assign_ternary): Likewise.
4809 * tree-inline.c (estimate_move_cost): Likewise.
4810 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4811 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4812 (get_address_cost_ainc): Likewise.
4813 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4814 (vect_supportable_dr_alignment): Likewise.
4815 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4816 (vectorizable_reduction): Likewise.
4817 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4818 (vectorizable_operation, vectorizable_load): Likewise.
4819 * tree.c (build_same_sized_truth_vector_type): Likewise.
4820 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4821 * var-tracking.c (emit_note_insn_var_location): Likewise.
4822 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4823 (ADDR_VEC_ALIGN): Likewise.
4825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4826 Alan Hayward <alan.hayward@arm.com>
4827 David Sherwood <david.sherwood@arm.com>
4829 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4831 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4832 or if measurement_type is polynomial.
4833 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4834 * combine.c (make_extraction): Likewise.
4835 * dse.c (find_shift_sequence): Likewise.
4836 * dwarf2out.c (mem_loc_descriptor): Likewise.
4837 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4838 (extract_bit_field, extract_low_bits): Likewise.
4839 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4840 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4841 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4842 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4843 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4844 * reload.c (find_reloads): Likewise.
4845 * reload1.c (alter_reg): Likewise.
4846 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4847 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4848 * tree-if-conv.c (predicate_mem_writes): Likewise.
4849 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4850 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4851 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4852 * valtrack.c (dead_debug_insert_temp): Likewise.
4853 * varasm.c (mergeable_constant_section): Likewise.
4854 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4856 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4857 Alan Hayward <alan.hayward@arm.com>
4858 David Sherwood <david.sherwood@arm.com>
4860 * expr.c (expand_assignment): Cope with polynomial mode sizes
4861 when assigning to a CONCAT.
4863 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4864 Alan Hayward <alan.hayward@arm.com>
4865 David Sherwood <david.sherwood@arm.com>
4867 * machmode.h (mode_precision): Change from unsigned short to
4869 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4871 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4872 or if measurement_type is not polynomial.
4873 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4874 in which the mode is already known to be a scalar_int_mode.
4875 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4876 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4878 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4879 for GET_MODE_PRECISION.
4880 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4881 for GET_MODE_PRECISION.
4882 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4884 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4885 (expand_field_assignment, make_extraction): Likewise.
4886 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4887 (get_last_value): Likewise.
4888 * convert.c (convert_to_integer_1): Likewise.
4889 * cse.c (cse_insn): Likewise.
4890 * expr.c (expand_expr_real_1): Likewise.
4891 * lra-constraints.c (simplify_operand_subreg): Likewise.
4892 * optabs-query.c (can_atomic_load_p): Likewise.
4893 * optabs.c (expand_atomic_load): Likewise.
4894 (expand_atomic_store): Likewise.
4895 * ree.c (combine_reaching_defs): Likewise.
4896 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4897 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4898 * tree.h (type_has_mode_precision_p): Likewise.
4899 * ubsan.c (instrument_si_overflow): Likewise.
4901 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4902 Alan Hayward <alan.hayward@arm.com>
4903 David Sherwood <david.sherwood@arm.com>
4905 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4906 polynomial numbers of units.
4907 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4908 (valid_vector_subparts_p): New function.
4909 (build_vector_type): Remove temporary shim and take the number
4910 of units as a poly_uint64 rather than an int.
4911 (build_opaque_vector_type): Take the number of units as a
4912 poly_uint64 rather than an int.
4913 * tree.c (build_vector_from_ctor): Handle polynomial
4914 TYPE_VECTOR_SUBPARTS.
4915 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4916 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4917 (build_vector_from_val): If the number of units is variable,
4918 use build_vec_duplicate_cst for constant operands and
4919 VEC_DUPLICATE_EXPR otherwise.
4920 (make_vector_type): Remove temporary is_constant ().
4921 (build_vector_type, build_opaque_vector_type): Take the number of
4922 units as a poly_uint64 rather than an int.
4923 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4925 * cfgexpand.c (expand_debug_expr): Likewise.
4926 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4927 (store_constructor, expand_expr_real_1): Likewise.
4928 (const_scalar_mask_from_tree): Likewise.
4929 * fold-const-call.c (fold_const_reduction): Likewise.
4930 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4931 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4932 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4933 (fold_relational_const): Likewise.
4934 (native_interpret_vector): Likewise. Change the size from an
4935 int to an unsigned int.
4936 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4937 TYPE_VECTOR_SUBPARTS.
4938 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4939 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4940 duplicating a non-constant operand into a variable-length vector.
4941 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4942 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4943 * ipa-icf.c (sem_variable::equals): Likewise.
4944 * match.pd: Likewise.
4945 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4946 * print-tree.c (print_node): Likewise.
4947 * stor-layout.c (layout_type): Likewise.
4948 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4949 * tree-cfg.c (verify_gimple_comparison): Likewise.
4950 (verify_gimple_assign_binary): Likewise.
4951 (verify_gimple_assign_ternary): Likewise.
4952 (verify_gimple_assign_single): Likewise.
4953 * tree-pretty-print.c (dump_generic_node): Likewise.
4954 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4955 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4956 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4957 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4958 (vect_shift_permute_load_chain): Likewise.
4959 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4960 (expand_vector_condition, optimize_vector_constructor): Likewise.
4961 (lower_vec_perm, get_compute_type): Likewise.
4962 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4963 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4964 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4965 (vect_recog_mask_conversion_pattern): Likewise.
4966 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4967 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4968 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4969 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4970 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4971 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4972 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4973 (supportable_widening_operation): Likewise.
4974 (supportable_narrowing_operation): Likewise.
4975 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4977 * varasm.c (output_constant): Likewise.
4979 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4980 Alan Hayward <alan.hayward@arm.com>
4981 David Sherwood <david.sherwood@arm.com>
4983 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4984 so that both the length == 3 and length != 3 cases set up their
4985 own permute vectors. Add comments explaining why we know the
4986 number of elements is constant.
4987 (vect_permute_load_chain): Likewise.
4989 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4990 Alan Hayward <alan.hayward@arm.com>
4991 David Sherwood <david.sherwood@arm.com>
4993 * machmode.h (mode_nunits): Change from unsigned char to
4995 (ONLY_FIXED_SIZE_MODES): New macro.
4996 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4997 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4998 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5000 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5001 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5002 or if measurement_type is not polynomial.
5003 * genmodes.c (ZERO_COEFFS): New macro.
5004 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5006 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5007 Use ZERO_COEFFS when emitting initializers.
5008 * data-streamer.h (bp_pack_poly_value): New function.
5009 (bp_unpack_poly_value): Likewise.
5010 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5011 for GET_MODE_NUNITS.
5012 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5013 for GET_MODE_NUNITS.
5014 * tree.c (make_vector_type): Remove temporary shim and make
5015 the real function take the number of units as a poly_uint64
5017 (build_vector_type_for_mode): Handle polynomial nunits.
5018 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5019 * emit-rtl.c (const_vec_series_p_1): Likewise.
5020 (gen_rtx_CONST_VECTOR): Likewise.
5021 * fold-const.c (test_vec_duplicate_folding): Likewise.
5022 * genrecog.c (validate_pattern): Likewise.
5023 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5024 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5025 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5026 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5027 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5028 * rtlanal.c (subreg_get_info): Likewise.
5029 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5030 (vect_grouped_load_supported): Likewise.
5031 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5032 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5033 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5034 (simplify_const_unary_operation, simplify_binary_operation_1)
5035 (simplify_const_binary_operation, simplify_ternary_operation)
5036 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5037 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5038 instead of CONST_VECTOR_NUNITS.
5039 * varasm.c (output_constant_pool_2): Likewise.
5040 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5041 explicit-encoded elements in the XVEC for variable-length vectors.
5043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5045 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5048 Alan Hayward <alan.hayward@arm.com>
5049 David Sherwood <david.sherwood@arm.com>
5051 * coretypes.h (fixed_size_mode): Declare.
5052 (fixed_size_mode_pod): New typedef.
5053 * builtins.h (target_builtins::x_apply_args_mode)
5054 (target_builtins::x_apply_result_mode): Change type to
5055 fixed_size_mode_pod.
5056 * builtins.c (apply_args_size, apply_result_size, result_vector)
5057 (expand_builtin_apply_args_1, expand_builtin_apply)
5058 (expand_builtin_return): Update accordingly.
5060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5062 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5063 * cselib.c (cselib_hash_rtx): Likewise.
5064 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5065 CONST_VECTOR encoding.
5067 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5068 Jeff Law <law@redhat.com>
5071 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5072 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5073 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5074 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5077 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5078 explicitly probe *sp in a noreturn function if there were any callee
5079 register saves or frame pointer is needed.
5081 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5084 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5085 BLKmode for ternary, binary or unary expressions.
5088 * var-tracking.c (delete_vta_debug_insn): New inline function.
5089 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5090 insns from get_insns () to NULL instead of each bb separately.
5091 Use delete_vta_debug_insn. No longer static.
5092 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5093 delete_vta_debug_insns callers.
5094 * rtl.h (delete_vta_debug_insns): Declare.
5095 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5096 instead of variable_tracking_main.
5098 2018-01-03 Martin Sebor <msebor@redhat.com>
5100 PR tree-optimization/83603
5101 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5102 arguments past the endof the argument list in functions declared
5103 without a prototype.
5104 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5105 Avoid checking when arguments are null.
5107 2018-01-03 Martin Sebor <msebor@redhat.com>
5110 * doc/extend.texi (attribute const): Fix a typo.
5111 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5112 issuing -Wsuggest-attribute for void functions.
5114 2018-01-03 Martin Sebor <msebor@redhat.com>
5116 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5117 offset_int::from instead of wide_int::to_shwi.
5118 (maybe_diag_overlap): Remove assertion.
5119 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5120 * gimple-ssa-sprintf.c (format_directive): Same.
5121 (parse_directive): Same.
5122 (sprintf_dom_walker::compute_format_length): Same.
5123 (try_substitute_return_value): Same.
5125 2018-01-03 Jeff Law <law@redhat.com>
5128 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5129 non-constant residual for zero at runtime and avoid probing in
5130 that case. Reorganize code for trailing problem to mirror handling
5133 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5135 PR tree-optimization/83501
5136 * tree-ssa-strlen.c (get_string_cst): New.
5137 (handle_char_store): Call get_string_cst.
5139 2018-01-03 Martin Liska <mliska@suse.cz>
5141 PR tree-optimization/83593
5142 * tree-ssa-strlen.c: Include tree-cfg.h.
5143 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5144 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5145 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5147 (strlen_dom_walker::before_dom_children): Call
5148 gimple_purge_dead_eh_edges. Dump tranformation with details
5150 (strlen_dom_walker::before_dom_children): Update call by adding
5151 new argument cleanup_eh.
5152 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5154 2018-01-03 Martin Liska <mliska@suse.cz>
5157 * cif-code.def (VARIADIC_THUNK): New enum value.
5158 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5161 2018-01-03 Jan Beulich <jbeulich@suse.com>
5163 * sse.md (mov<mode>_internal): Tighten condition for when to use
5164 vmovdqu<ssescalarsize> for TI and OI modes.
5166 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5168 Update copyright years.
5170 2018-01-03 Martin Liska <mliska@suse.cz>
5173 * ipa-visibility.c (function_and_variable_visibility): Skip
5174 functions with noipa attribure.
5176 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5178 * gcc.c (process_command): Update copyright notice dates.
5179 * gcov-dump.c (print_version): Ditto.
5180 * gcov.c (print_version): Ditto.
5181 * gcov-tool.c (print_version): Ditto.
5182 * gengtype.c (create_file): Ditto.
5183 * doc/cpp.texi: Bump @copying's copyright year.
5184 * doc/cppinternals.texi: Ditto.
5185 * doc/gcc.texi: Ditto.
5186 * doc/gccint.texi: Ditto.
5187 * doc/gcov.texi: Ditto.
5188 * doc/install.texi: Ditto.
5189 * doc/invoke.texi: Ditto.
5191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5193 * vector-builder.h (vector_builder::m_full_nelts): Change from
5194 unsigned int to poly_uint64.
5195 (vector_builder::full_nelts): Update prototype accordingly.
5196 (vector_builder::new_vector): Likewise.
5197 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5198 (vector_builder::operator ==): Likewise.
5199 (vector_builder::finalize): Likewise.
5200 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5201 Take the number of elements as a poly_uint64 rather than an
5203 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5204 from unsigned int to poly_uint64.
5205 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5206 (vec_perm_indices::new_vector): Likewise.
5207 (vec_perm_indices::length): Likewise.
5208 (vec_perm_indices::nelts_per_input): Likewise.
5209 (vec_perm_indices::input_nelts): Likewise.
5210 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5211 number of elements per input as a poly_uint64 rather than an
5212 unsigned int. Use the original encoding for variable-length
5213 vectors, rather than clamping each individual element.
5214 For the second and subsequent elements in each pattern,
5215 clamp the step and base before clamping their sum.
5216 (vec_perm_indices::series_p): Handle polynomial element counts.
5217 (vec_perm_indices::all_in_range_p): Likewise.
5218 (vec_perm_indices_to_tree): Likewise.
5219 (vec_perm_indices_to_rtx): Likewise.
5220 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5221 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5222 (tree_vector_builder::new_binary_operation): Handle polynomial
5223 element counts. Return false if we need to know the number
5224 of elements at compile time.
5225 * fold-const.c (fold_vec_perm): Punt if the number of elements
5226 isn't known at compile time.
5228 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5230 * vec-perm-indices.h (vec_perm_builder): Change element type
5231 from HOST_WIDE_INT to poly_int64.
5232 (vec_perm_indices::element_type): Update accordingly.
5233 (vec_perm_indices::clamp): Handle polynomial element_types.
5234 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5235 (vec_perm_indices::all_in_range_p): Likewise.
5236 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5238 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5239 polynomial vec_perm_indices element types.
5240 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5241 * fold-const.c (fold_vec_perm): Likewise.
5242 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5243 * tree-vect-generic.c (lower_vec_perm): Likewise.
5244 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5245 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5246 element type to HOST_WIDE_INT.
5248 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5249 Alan Hayward <alan.hayward@arm.com>
5250 David Sherwood <david.sherwood@arm.com>
5252 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5253 rather than an int. Use plus_constant.
5254 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5255 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5258 Alan Hayward <alan.hayward@arm.com>
5259 David Sherwood <david.sherwood@arm.com>
5261 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5262 a HOST_WIDE_INT to a poly_int64.
5264 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5265 Alan Hayward <alan.hayward@arm.com>
5266 David Sherwood <david.sherwood@arm.com>
5268 * calls.c (load_register_parameters): Cope with polynomial
5269 mode sizes. Require a constant size for BLKmode parameters
5270 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5271 forces a parameter to be padded at the lsb end in order to
5272 fill a complete number of words, require the parameter size
5273 to be ordered wrt UNITS_PER_WORD.
5275 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5276 Alan Hayward <alan.hayward@arm.com>
5277 David Sherwood <david.sherwood@arm.com>
5279 * reload1.c (spill_stack_slot_width): Change element type
5280 from unsigned int to poly_uint64_pod.
5281 (alter_reg): Treat mode sizes as polynomial.
5283 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5284 Alan Hayward <alan.hayward@arm.com>
5285 David Sherwood <david.sherwood@arm.com>
5287 * reload.c (complex_word_subreg_p): New function.
5288 (reload_inner_reg_of_subreg, push_reload): Use it.
5290 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5291 Alan Hayward <alan.hayward@arm.com>
5292 David Sherwood <david.sherwood@arm.com>
5294 * lra-constraints.c (process_alt_operands): Reject matched
5295 operands whose sizes aren't ordered.
5296 (match_reload): Refer to this check here.
5298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5299 Alan Hayward <alan.hayward@arm.com>
5300 David Sherwood <david.sherwood@arm.com>
5302 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5303 that the mode size is in the set {1, 2, 4, 8, 16}.
5305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5306 Alan Hayward <alan.hayward@arm.com>
5307 David Sherwood <david.sherwood@arm.com>
5309 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5310 Use plus_constant instead of gen_rtx_PLUS.
5312 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5313 Alan Hayward <alan.hayward@arm.com>
5314 David Sherwood <david.sherwood@arm.com>
5316 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5317 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5318 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5319 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5320 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5321 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5322 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5323 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5324 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5325 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5327 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5328 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5329 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5330 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5331 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5332 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5333 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5334 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5335 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5336 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5338 * expr.c (emit_move_resolve_push): Treat the input and result
5339 of PUSH_ROUNDING as a poly_int64.
5340 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5341 (emit_push_insn): Likewise.
5342 * lra-eliminations.c (mark_not_eliminable): Likewise.
5343 * recog.c (push_operand): Likewise.
5344 * reload1.c (elimination_effects): Likewise.
5345 * rtlanal.c (nonzero_bits1): Likewise.
5346 * calls.c (store_one_arg): Likewise. Require the padding to be
5347 known at compile time.
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 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5354 Use plus_constant instead of gen_rtx_PLUS.
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 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5363 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5364 Alan Hayward <alan.hayward@arm.com>
5365 David Sherwood <david.sherwood@arm.com>
5367 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5368 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5369 via stack temporaries. Treat the mode size as polynomial too.
5371 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5372 Alan Hayward <alan.hayward@arm.com>
5373 David Sherwood <david.sherwood@arm.com>
5375 * expr.c (expand_expr_real_2): When handling conversions involving
5376 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5377 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5378 as a poly_uint64 too.
5380 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5381 Alan Hayward <alan.hayward@arm.com>
5382 David Sherwood <david.sherwood@arm.com>
5384 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5386 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5387 Alan Hayward <alan.hayward@arm.com>
5388 David Sherwood <david.sherwood@arm.com>
5390 * combine.c (can_change_dest_mode): Handle polynomial
5391 REGMODE_NATURAL_SIZE.
5392 * expmed.c (store_bit_field_1): Likewise.
5393 * expr.c (store_constructor): Likewise.
5394 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5395 and polynomial REGMODE_NATURAL_SIZE.
5396 (gen_lowpart_common): Likewise.
5397 * reginfo.c (record_subregs_of_mode): Likewise.
5398 * rtlanal.c (read_modify_subreg_p): Likewise.
5400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5401 Alan Hayward <alan.hayward@arm.com>
5402 David Sherwood <david.sherwood@arm.com>
5404 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5405 numbers of elements.
5407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5408 Alan Hayward <alan.hayward@arm.com>
5409 David Sherwood <david.sherwood@arm.com>
5411 * match.pd: Cope with polynomial numbers of vector elements.
5413 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5414 Alan Hayward <alan.hayward@arm.com>
5415 David Sherwood <david.sherwood@arm.com>
5417 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5418 in a POINTER_PLUS_EXPR.
5420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5421 Alan Hayward <alan.hayward@arm.com>
5422 David Sherwood <david.sherwood@arm.com>
5424 * omp-simd-clone.c (simd_clone_subparts): New function.
5425 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5426 (ipa_simd_modify_function_body): Likewise.
5428 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5429 Alan Hayward <alan.hayward@arm.com>
5430 David Sherwood <david.sherwood@arm.com>
5432 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5433 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5434 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5435 (expand_vector_condition, vector_element): Likewise.
5436 (subparts_gt): New function.
5437 (get_compute_type): Use subparts_gt.
5438 (count_type_subparts): Delete.
5439 (expand_vector_operations_1): Use subparts_gt instead of
5440 count_type_subparts.
5442 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5443 Alan Hayward <alan.hayward@arm.com>
5444 David Sherwood <david.sherwood@arm.com>
5446 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5447 (vect_compile_time_alias): ...this new function. Do the calculation
5448 on poly_ints rather than trees.
5449 (vect_prune_runtime_alias_test_list): Update call accordingly.
5451 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5452 Alan Hayward <alan.hayward@arm.com>
5453 David Sherwood <david.sherwood@arm.com>
5455 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5457 (vect_schedule_slp_instance): Likewise.
5459 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5460 Alan Hayward <alan.hayward@arm.com>
5461 David Sherwood <david.sherwood@arm.com>
5463 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5464 constant and extern definitions for variable-length vectors.
5465 (vect_get_constant_vectors): Note that the number of units
5466 is known to be constant.
5468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5469 Alan Hayward <alan.hayward@arm.com>
5470 David Sherwood <david.sherwood@arm.com>
5472 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5473 of units as polynomial. Choose between WIDE and NARROW based
5476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5477 Alan Hayward <alan.hayward@arm.com>
5478 David Sherwood <david.sherwood@arm.com>
5480 * tree-vect-stmts.c (simd_clone_subparts): New function.
5481 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5484 Alan Hayward <alan.hayward@arm.com>
5485 David Sherwood <david.sherwood@arm.com>
5487 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5488 vectors as polynomial. Use build_index_vector for
5491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5492 Alan Hayward <alan.hayward@arm.com>
5493 David Sherwood <david.sherwood@arm.com>
5495 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5496 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5497 for variable-length vectors.
5498 (vectorizable_mask_load_store): Treat the number of units as
5499 polynomial, asserting that it is constant if the condition has
5500 already been enforced.
5501 (vectorizable_store, vectorizable_load): Likewise.
5503 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5504 Alan Hayward <alan.hayward@arm.com>
5505 David Sherwood <david.sherwood@arm.com>
5507 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5508 of units as polynomial. Punt if we can't tell at compile time
5509 which vector contains the final result.
5511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5512 Alan Hayward <alan.hayward@arm.com>
5513 David Sherwood <david.sherwood@arm.com>
5515 * tree-vect-loop.c (vectorizable_induction): Treat the number
5516 of units as polynomial. Punt on SLP inductions. Use an integer
5517 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5518 cast of such a series for variable-length floating-point
5521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5522 Alan Hayward <alan.hayward@arm.com>
5523 David Sherwood <david.sherwood@arm.com>
5525 * tree.h (build_index_vector): Declare.
5526 * tree.c (build_index_vector): New function.
5527 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5528 of units as polynomial, forcibly converting it to a constant if
5529 vectorizable_reduction has already enforced the condition.
5530 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5531 to create a {1,2,3,...} vector.
5532 (vectorizable_reduction): Treat the number of units as polynomial.
5533 Choose vectype_in based on the largest scalar element size rather
5534 than the smallest number of units. Enforce the restrictions
5537 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5538 Alan Hayward <alan.hayward@arm.com>
5539 David Sherwood <david.sherwood@arm.com>
5541 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5542 number of units as polynomial.
5544 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5545 Alan Hayward <alan.hayward@arm.com>
5546 David Sherwood <david.sherwood@arm.com>
5548 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5549 * target.def (autovectorize_vector_sizes): Return the vector sizes
5550 by pointer, using vector_sizes rather than a bitmask.
5551 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5552 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5553 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5555 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5556 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5557 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5558 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5559 * omp-general.c (omp_max_vf): Likewise.
5560 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5561 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5562 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5563 * tree-vect-slp.c (vect_slp_bb): Likewise.
5564 * doc/tm.texi: Regenerate.
5565 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5567 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5568 the vector size as a poly_uint64 rather than an unsigned int.
5569 (current_vector_size): Change from an unsigned int to a poly_uint64.
5570 (get_vectype_for_scalar_type): Update accordingly.
5571 * tree.h (build_truth_vector_type): Take the size and number of
5572 units as a poly_uint64 rather than an unsigned int.
5573 (build_vector_type): Add a temporary overload that takes
5574 the number of units as a poly_uint64 rather than an unsigned int.
5575 * tree.c (make_vector_type): Likewise.
5576 (build_truth_vector_type): Take the number of units as a poly_uint64
5577 rather than an unsigned int.
5579 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5580 Alan Hayward <alan.hayward@arm.com>
5581 David Sherwood <david.sherwood@arm.com>
5583 * target.def (get_mask_mode): Take the number of units and length
5584 as poly_uint64s rather than unsigned ints.
5585 * targhooks.h (default_get_mask_mode): Update accordingly.
5586 * targhooks.c (default_get_mask_mode): Likewise.
5587 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5588 * doc/tm.texi: Regenerate.
5590 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5591 Alan Hayward <alan.hayward@arm.com>
5592 David Sherwood <david.sherwood@arm.com>
5594 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5595 * omp-general.c (omp_max_vf): Likewise.
5596 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5597 (expand_omp_simd): Handle polynomial safelen.
5598 * omp-low.c (omplow_simd_context): Add a default constructor.
5599 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5600 (lower_rec_simd_input_clauses): Update accordingly.
5601 (lower_rec_input_clauses): Likewise.
5603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5604 Alan Hayward <alan.hayward@arm.com>
5605 David Sherwood <david.sherwood@arm.com>
5607 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5608 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5609 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5610 (vect_analyze_slp_cost): Likewise.
5611 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5612 (vect_model_load_cost): Likewise.
5614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5615 Alan Hayward <alan.hayward@arm.com>
5616 David Sherwood <david.sherwood@arm.com>
5618 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5619 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5620 from an unsigned int * to a poly_uint64_pod *.
5621 (calculate_unrolling_factor): New function.
5622 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5624 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5625 Alan Hayward <alan.hayward@arm.com>
5626 David Sherwood <david.sherwood@arm.com>
5628 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5629 from an unsigned int to a poly_uint64.
5630 (_loop_vec_info::slp_unrolling_factor): Likewise.
5631 (_loop_vec_info::vectorization_factor): Change from an int
5633 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5634 (vect_get_num_vectors): New function.
5635 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5636 (vect_get_num_copies): Use vect_get_num_vectors.
5637 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5638 to an unsigned int *.
5639 (vect_analyze_data_refs): Change min_vf from an int * to a
5641 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5642 than an unsigned HOST_WIDE_INT.
5643 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5644 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5645 to an unsigned int *.
5646 (vect_analyze_data_ref_dependences): Likewise.
5647 (vect_compute_data_ref_alignment): Handle polynomial vf.
5648 (vect_enhance_data_refs_alignment): Likewise.
5649 (vect_prune_runtime_alias_test_list): Likewise.
5650 (vect_shift_permute_load_chain): Likewise.
5651 (vect_supportable_dr_alignment): Likewise.
5652 (dependence_distance_ge_vf): Take the vectorization factor as a
5653 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5654 (vect_analyze_data_refs): Change min_vf from an int * to a
5656 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5657 vfm1 as a poly_uint64 rather than an int. Make the same change
5658 for the returned bound_scalar.
5659 (vect_gen_vector_loop_niters): Handle polynomial vf.
5660 (vect_do_peeling): Likewise. Update call to
5661 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5662 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5664 * tree-vect-loop.c (vect_determine_vectorization_factor)
5665 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5666 (vect_get_known_peeling_cost): Likewise.
5667 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5668 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5669 (vect_transform_loop): Likewise. Use the lowest possible VF when
5670 updating the upper bounds of the loop.
5671 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5673 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5674 polynomial unroll factors.
5675 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5676 (vect_make_slp_decision): Likewise.
5677 (vect_supported_load_permutation_p): Likewise, and polynomial
5679 (vect_analyze_slp_cost): Handle polynomial vf.
5680 (vect_slp_analyze_node_operations): Likewise.
5681 (vect_slp_analyze_bb_1): Likewise.
5682 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5683 than an unsigned HOST_WIDE_INT.
5684 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5685 (vectorizable_load): Handle polynomial vf.
5686 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5688 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5691 Alan Hayward <alan.hayward@arm.com>
5692 David Sherwood <david.sherwood@arm.com>
5694 * match.pd: Handle bit operations involving three constants
5695 and try to fold one pair.
5697 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5699 * tree-vect-loop-manip.c: Include gimple-fold.h.
5700 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5701 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5702 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5703 Add a path that uses a step of VF instead of 1, but disable it
5705 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5706 and niters_no_overflow parameters. Update calls to
5707 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5708 Create a new SSA name if the latter choses to use a ste other
5709 than zero, and return it via niters_vector_mult_vf_var.
5710 * tree-vect-loop.c (vect_transform_loop): Update calls to
5711 vect_do_peeling, vect_gen_vector_loop_niters and
5712 slpeel_make_loop_iterate_ntimes.
5713 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5714 (vect_gen_vector_loop_niters): Update declarations after above changes.
5716 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5718 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5719 128-bit round to integer instructions.
5720 (ceil<mode>2): Likewise.
5721 (btrunc<mode>2): Likewise.
5722 (round<mode>2): Likewise.
5724 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5726 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5727 unaligned VSX load/store on P8/P9.
5728 (expand_block_clear): Allow the use of unaligned VSX
5729 load/store on P8/P9.
5731 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5733 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5735 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5736 swap associated with both a load and a store.
5738 2018-01-02 Andrew Waterman <andrew@sifive.com>
5740 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5741 * config/riscv/riscv.md (clear_cache): Use it.
5743 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5745 * web.c: Remove out-of-date comment.
5747 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5749 * expr.c (fixup_args_size_notes): Check that any existing
5750 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5751 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5752 (emit_single_push_insn): ...here.
5754 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5756 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5757 (const_vector_encoded_nelts): New function.
5758 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5759 (const_vector_int_elt, const_vector_elt): Declare.
5760 * emit-rtl.c (const_vector_int_elt_1): New function.
5761 (const_vector_elt): Likewise.
5762 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5763 of CONST_VECTOR_ELT.
5765 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5767 * expr.c: Include rtx-vector-builder.h.
5768 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5769 directly on the tree encoding.
5770 (const_vector_from_tree): Likewise.
5771 * optabs.c: Include rtx-vector-builder.h.
5772 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5773 sequence of "u" values.
5774 * vec-perm-indices.c: Include rtx-vector-builder.h.
5775 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5776 directly on the vec_perm_indices encoding.
5778 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5780 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5781 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5782 * rtx-vector-builder.h: New file.
5783 * rtx-vector-builder.c: Likewise.
5784 * rtl.h (rtx_def::u2): Add a const_vector field.
5785 (CONST_VECTOR_NPATTERNS): New macro.
5786 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5787 (CONST_VECTOR_DUPLICATE_P): Likewise.
5788 (CONST_VECTOR_STEPPED_P): Likewise.
5789 (CONST_VECTOR_ENCODED_ELT): Likewise.
5790 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5791 (unwrap_const_vec_duplicate): Likewise.
5792 (const_vec_series_p): Check for a non-duplicated vector encoding.
5793 Say that the function only returns true for integer vectors.
5794 * emit-rtl.c: Include rtx-vector-builder.h.
5795 (gen_const_vec_duplicate_1): Delete.
5796 (gen_const_vector): Call gen_const_vec_duplicate instead of
5797 gen_const_vec_duplicate_1.
5798 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5799 (gen_const_vec_duplicate): Use rtx_vector_builder.
5800 (gen_const_vec_series): Likewise.
5801 (gen_rtx_CONST_VECTOR): Likewise.
5802 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5803 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5804 Build a new vector rather than modifying a CONST_VECTOR in-place.
5805 (handle_special_swappables): Update call accordingly.
5806 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5807 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5808 Build a new vector rather than modifying a CONST_VECTOR in-place.
5809 (handle_special_swappables): Update call accordingly.
5811 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5813 * simplify-rtx.c (simplify_const_binary_operation): Use
5814 CONST_VECTOR_ELT instead of XVECEXP.
5816 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5818 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5819 the selector elements to be different from the data elements
5820 if the selector is a VECTOR_CST.
5821 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5822 ssizetype for the selector.
5824 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5826 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5827 before testing each element individually.
5828 * tree-vect-generic.c (lower_vec_perm): Likewise.
5830 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5832 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5833 * selftest-run-tests.c (selftest::run_tests): Call it.
5834 * vector-builder.h (vector_builder::operator ==): New function.
5835 (vector_builder::operator !=): Likewise.
5836 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5837 (vec_perm_indices::all_from_input_p): New function.
5838 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5839 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5840 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5841 instead of reading the VECTOR_CST directly. Detect whether both
5842 vector inputs are the same before constructing the vec_perm_indices,
5843 and update the number of inputs argument accordingly. Use the
5844 utility functions added above. Only construct sel2 if we need to.
5846 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5848 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5849 the broadcast of the low byte.
5850 (expand_mult_highpart): Use an explicit encoding for the permutes.
5851 * optabs-query.c (can_mult_highpart_p): Likewise.
5852 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5853 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5854 (vectorizable_bswap): Likewise.
5855 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5856 explicit encoding for the power-of-2 permutes.
5857 (vect_permute_store_chain): Likewise.
5858 (vect_grouped_load_supported): Likewise.
5859 (vect_permute_load_chain): Likewise.
5861 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5863 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5864 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5865 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5866 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5867 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5868 (vect_gen_perm_mask_any): Likewise.
5870 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5872 * int-vector-builder.h: New file.
5873 * vec-perm-indices.h: Include int-vector-builder.h.
5874 (vec_perm_indices): Redefine as an int_vector_builder.
5875 (auto_vec_perm_indices): Delete.
5876 (vec_perm_builder): Redefine as a stand-alone class.
5877 (vec_perm_indices::vec_perm_indices): New function.
5878 (vec_perm_indices::clamp): Likewise.
5879 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5880 (vec_perm_indices::new_vector): New function.
5881 (vec_perm_indices::new_expanded_vector): Update for new
5882 vec_perm_indices class.
5883 (vec_perm_indices::rotate_inputs): New function.
5884 (vec_perm_indices::all_in_range_p): Operate directly on the
5885 encoded form, without computing elided elements.
5886 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5887 encoding. Update for new vec_perm_indices class.
5888 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5889 the given vec_perm_builder.
5890 (expand_vec_perm_var): Update vec_perm_builder constructor.
5891 (expand_mult_highpart): Use vec_perm_builder instead of
5892 auto_vec_perm_indices.
5893 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5894 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5895 or double series encoding as appropriate.
5896 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5897 vec_perm_indices instead of auto_vec_perm_indices.
5898 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5899 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5900 (vect_permute_store_chain): Likewise.
5901 (vect_grouped_load_supported): Likewise.
5902 (vect_permute_load_chain): Likewise.
5903 (vect_shift_permute_load_chain): Likewise.
5904 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5905 (vect_transform_slp_perm_load): Likewise.
5906 (vect_schedule_slp_instance): Likewise.
5907 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5908 (vectorizable_mask_load_store): Likewise.
5909 (vectorizable_bswap): Likewise.
5910 (vectorizable_store): Likewise.
5911 (vectorizable_load): Likewise.
5912 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5913 vec_perm_indices instead of auto_vec_perm_indices. Use
5914 tree_to_vec_perm_builder to read the vector from a tree.
5915 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5916 vec_perm_builder instead of a vec_perm_indices.
5917 (have_whole_vector_shift): Use vec_perm_builder and
5918 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5919 truncation to calc_vec_perm_mask_for_shift.
5920 (vect_create_epilog_for_reduction): Likewise.
5921 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5922 from auto_vec_perm_indices to vec_perm_indices.
5923 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5924 instead of changing individual elements.
5925 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5926 the vector in d.perm.
5927 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5928 from auto_vec_perm_indices to vec_perm_indices.
5929 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5930 instead of changing individual elements.
5931 (arm_vectorize_vec_perm_const): Use new_vector to install
5932 the vector in d.perm.
5933 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5934 Update vec_perm_builder constructor.
5935 (rs6000_expand_interleave): Likewise.
5936 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5937 (rs6000_expand_interleave): Likewise.
5939 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5941 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5942 to qimode could truncate the indices.
5943 * optabs.c (expand_vec_perm_var): Likewise.
5945 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5947 * Makefile.in (OBJS): Add vec-perm-indices.o.
5948 * vec-perm-indices.h: New file.
5949 * vec-perm-indices.c: Likewise.
5950 * target.h (vec_perm_indices): Replace with a forward class
5952 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5953 * optabs.h: Include vec-perm-indices.h.
5954 (expand_vec_perm): Delete.
5955 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5956 (expand_vec_perm_const): Declare.
5957 * target.def (vec_perm_const_ok): Replace with...
5958 (vec_perm_const): ...this new hook.
5959 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5960 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5961 * doc/tm.texi: Regenerate.
5962 * optabs.def (vec_perm_const): Delete.
5963 * doc/md.texi (vec_perm_const): Likewise.
5964 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5965 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5966 expand_vec_perm for constant permutation vectors. Assert that
5967 the mode of variable permutation vectors is the integer equivalent
5968 of the mode that is being permuted.
5969 * optabs-query.h (selector_fits_mode_p): Declare.
5970 * optabs-query.c: Include vec-perm-indices.h.
5971 (selector_fits_mode_p): New function.
5972 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5973 is defined, instead of checking whether the vec_perm_const_optab
5974 exists. Use targetm.vectorize.vec_perm_const instead of
5975 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5976 fit in the vector mode before using a variable permute.
5977 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5978 vec_perm_indices instead of an rtx.
5979 (expand_vec_perm): Replace with...
5980 (expand_vec_perm_const): ...this new function. Take the selector
5981 as a vec_perm_indices rather than an rtx. Also take the mode of
5982 the selector. Update call to shift_amt_for_vec_perm_mask.
5983 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5984 Use vec_perm_indices::new_expanded_vector to expand the original
5985 selector into bytes. Check whether the indices fit in the vector
5986 mode before using a variable permute.
5987 (expand_vec_perm_var): Make global.
5988 (expand_mult_highpart): Use expand_vec_perm_const.
5989 * fold-const.c: Includes vec-perm-indices.h.
5990 * tree-ssa-forwprop.c: Likewise.
5991 * tree-vect-data-refs.c: Likewise.
5992 * tree-vect-generic.c: Likewise.
5993 * tree-vect-loop.c: Likewise.
5994 * tree-vect-slp.c: Likewise.
5995 * tree-vect-stmts.c: Likewise.
5996 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5998 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5999 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6000 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6001 (aarch64_vectorize_vec_perm_const): ...this new function.
6002 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6003 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6004 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6005 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6006 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6007 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6008 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6010 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6011 check for NEON modes.
6012 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6013 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6014 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6015 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6017 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6018 the old VEC_PERM_CONST conditions.
6019 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6020 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6021 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6022 (ia64_vectorize_vec_perm_const_ok): Merge into...
6023 (ia64_vectorize_vec_perm_const): ...this new function.
6024 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6025 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6026 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6027 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6028 * config/mips/mips.c (mips_expand_vec_perm_const)
6029 (mips_vectorize_vec_perm_const_ok): Merge into...
6030 (mips_vectorize_vec_perm_const): ...this new function.
6031 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6032 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6033 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6034 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6035 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6036 (rs6000_expand_vec_perm_const): Delete.
6037 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6039 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6040 (altivec_expand_vec_perm_const_le): Take each operand individually.
6041 Operate on constant selectors rather than rtxes.
6042 (altivec_expand_vec_perm_const): Likewise. Update call to
6043 altivec_expand_vec_perm_const_le.
6044 (rs6000_expand_vec_perm_const): Delete.
6045 (rs6000_vectorize_vec_perm_const_ok): Delete.
6046 (rs6000_vectorize_vec_perm_const): New function.
6047 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6048 an element count and rtx array.
6049 (rs6000_expand_extract_even): Update call accordingly.
6050 (rs6000_expand_interleave): Likewise.
6051 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6052 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6053 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6054 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6055 (rs6000_expand_vec_perm_const): Delete.
6056 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6057 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6058 (altivec_expand_vec_perm_const_le): Take each operand individually.
6059 Operate on constant selectors rather than rtxes.
6060 (altivec_expand_vec_perm_const): Likewise. Update call to
6061 altivec_expand_vec_perm_const_le.
6062 (rs6000_expand_vec_perm_const): Delete.
6063 (rs6000_vectorize_vec_perm_const_ok): Delete.
6064 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6065 reference to the SPE evmerge intructions.
6066 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6067 an element count and rtx array.
6068 (rs6000_expand_extract_even): Update call accordingly.
6069 (rs6000_expand_interleave): Likewise.
6070 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6071 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6073 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6075 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6077 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6078 vector mode and that that mode matches the mode of the data
6080 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6081 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6082 directly using expand_vec_perm_1 when forcing selectors into
6084 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6086 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6088 * optabs-query.h (can_vec_perm_p): Delete.
6089 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6090 * optabs-query.c (can_vec_perm_p): Split into...
6091 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6092 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6093 particular selector is valid.
6094 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6095 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6096 (vect_grouped_load_supported): Likewise.
6097 (vect_shift_permute_load_chain): Likewise.
6098 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6099 (vect_transform_slp_perm_load): Likewise.
6100 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6101 (vectorizable_bswap): Likewise.
6102 (vect_gen_perm_mask_checked): Likewise.
6103 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6104 implementations of variable permutation vectors into account
6105 when deciding which selector to use.
6106 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6107 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6108 with a false third argument.
6109 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6110 to test whether the constant selector is valid and can_vec_perm_var_p
6111 to test whether a variable selector is valid.
6113 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6115 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6116 * optabs-query.c (can_vec_perm_p): Likewise.
6117 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6118 instead of vec_perm_indices.
6119 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6120 (vect_gen_perm_mask_checked): Likewise,
6121 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6122 (vect_gen_perm_mask_checked): Likewise,
6124 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6126 * optabs-query.h (qimode_for_vec_perm): Declare.
6127 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6128 (qimode_for_vec_perm): ...this new function.
6129 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6131 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6133 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6134 does not have a conditional at the top.
6136 2018-01-02 Richard Biener <rguenther@suse.de>
6138 * ipa-inline.c (big_speedup_p): Fix expression.
6140 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6143 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6146 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6150 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6151 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6152 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6153 cond_taken_branch_cost 3->4.
6155 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6157 PR tree-optimization/83581
6158 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6159 TODO_cleanup_cfg if any changes have been made.
6162 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6163 convert_modes if target mode has the right side, but different mode
6167 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6168 last argument when extracting from CONCAT. If either from_real or
6169 from_imag is NULL, use expansion through memory. If result is not
6170 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6171 the parts directly to inner mode, if even that fails, use expansion
6175 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6176 check for bswap in mode rather than HImode and use that in expand_unop
6179 Copyright (C) 2018 Free Software Foundation, Inc.
6181 Copying and distribution of this file, with or without modification,
6182 are permitted in any medium without royalty provided the copyright
6183 notice and this notice are preserved.