1 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
4 * config/i386/i386.c (ix86_option_override_internal): Mask out
5 the CF_SET bit when checking -fcf-protection.
7 2018-02-07 Tom de Vries <tom@codesourcery.com>
10 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
13 2018-02-07 Richard Biener <rguenther@suse.de>
15 PR tree-optimization/84204
16 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
19 PR tree-optimization/84205
20 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
21 special-case isl_ast_op_zdiv_r.
23 PR tree-optimization/84223
24 * graphite-scop-detection.c (gather_bbs::before_dom_children):
25 Only add conditions from within the region.
26 (gather_bbs::after_dom_children): Adjust.
28 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
31 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
32 * config/avr/avr.md: Only post-reload split REG-REG moves if
33 either register is REGERAL_REG_P.
35 2018-02-07 Jakub Jelinek <jakub@redhat.com>
37 PR tree-optimization/84235
38 * tree-ssa-scopedtables.c
39 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
40 if the subtraction is performed in floating point type where NaNs are
41 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
42 build 1. Formatting fix.
44 2018-02-06 Jakub Jelinek <jakub@redhat.com>
47 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
48 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
49 and skip it regardless of bb boundaries. Use CALL_P macro,
50 don't test INSN_P (insn) together with CALL_P or JUMP_P check
51 unnecessarily, formatting fix.
53 2018-02-06 Michael Collison <michael.collison@arm.com>
55 * config/arm/thumb2.md:
56 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
57 (*thumb_mov_notscc): Ditto.
59 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
62 * config/rs6000/rs6000.md (su code attribute): Use "u" for
63 unsigned_fix, not "s".
65 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
67 * configure.ac (gcc_fn_eh_frame_ro): New function.
68 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
69 correct .eh_frame permissions.
70 * configure: Regenerate.
72 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
74 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
77 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
79 * config/rs6000/rs6000.c (rs6000_option_override_internal):
80 Display warning message for -mno-speculate-indirect-jumps.
82 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
84 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
86 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
88 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
90 PR tree-optimization/84225
91 * tree-eh.c (find_trapping_overflow): Only call
92 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
94 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
97 * config/i386/i386.c: Reimplement the check of possible options
98 -mibt/-mshstk conbination. Change error messages.
99 * doc/invoke.texi: Fix a typo: remove extra '='.
101 2018-02-06 Marek Polacek <polacek@redhat.com>
103 PR tree-optimization/84228
104 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
106 2018-02-06 Tamar Christina <tamar.christina@arm.com>
109 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
110 emitted arch directives.
111 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
112 __ARM_FEATURE_COPROC before changing architectures.
114 2018-02-06 Richard Biener <rguenther@suse.de>
116 * config/i386/i386.c (print_reg): Fix typo.
117 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
119 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
121 * configure: Regenerate.
123 2018-02-05 Martin Sebor <msebor@redhat.com>
125 PR tree-optimization/83369
126 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
129 2018-02-05 Martin Liska <mliska@suse.cz>
131 * doc/invoke.texi: Cherry-pick upstream r323995.
133 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
135 * ira.c (ira_init_register_move_cost): Adjust comment.
137 2018-02-05 Martin Liska <mliska@suse.cz>
139 PR gcov-profile/84137
140 * doc/gcov.texi: Fix typo in documentation.
142 2018-02-05 Martin Liska <mliska@suse.cz>
144 PR gcov-profile/83879
145 * doc/gcov.texi: Document necessity of --dynamic-list-data when
146 using dlopen functionality.
148 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
150 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
151 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
152 _mm_maskz_range_ss, _mm_mask_range_round_ss,
153 _mm_maskz_range_round_ss): New intrinsics.
154 (__builtin_ia32_rangesd128_round)
155 (__builtin_ia32_rangess128_round): Remove.
156 (__builtin_ia32_rangesd128_mask_round,
157 __builtin_ia32_rangess128_mask_round): New builtins.
158 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
159 __builtin_ia32_rangess128_round): Remove.
160 (__builtin_ia32_rangesd128_mask_round,
161 __builtin_ia32_rangess128_mask_round): New builtins.
162 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
163 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
164 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
165 "<round_saeonly_constraint>")): Changed to ...
166 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
167 "<round_saeonly_scalar_constraint>")): ... this.
168 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
169 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
170 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
171 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
172 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
174 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
176 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
178 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
179 Remove all values except native, 8540 and 8548.
181 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
183 * config/i386/i386.c (ix86_output_function_return): Pass
184 INVALID_REGNUM, instead of -1, as invalid register number to
185 indirect_thunk_name and output_indirect_thunk.
187 2018-02-02 Julia Koval <julia.koval@intel.com>
189 * config.gcc: Add -march=icelake.
190 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
191 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
192 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
193 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
194 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
195 (processor_target_table): Add icelake.
196 (ix86_option_override_internal): Handle new PTAs.
197 (get_builtin_code_for_version): Handle icelake.
198 (M_INTEL_COREI7_ICELAKE): New.
199 (fold_builtin_cpu): Handle icelake.
200 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
201 * doc/invoke.texi: Add -march=icelake.
203 2018-02-02 Julia Koval <julia.koval@intel.com>
205 * config/i386/i386.c (ix86_option_override_internal): Change flags type
207 * wide-int-bitmask.h: New.
209 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
212 * config/i386/i386.md: Replace Pmode with word_mode in
213 builtin_setjmp_setup and builtin_longjmp to support x32.
215 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
219 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
221 (rs6000_supported_cpu_names): New static variable.
222 (linux_cpu_translation_table): Likewise.
223 (elf_platform) <cpu>: Define new static variable and use it.
224 Translate kernel AT_PLATFORM name to canonical name if needed.
225 Error if platform name is unknown.
227 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
230 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
232 2018-02-01 Jeff Law <law@redhat.com>
235 * config/i386/i386.c (release_scratch_register_on_entry): Add new
236 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
237 the scratch if RELEASE_VIA_POP is false.
238 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
239 If we have to save a temporary register, decrement SIZE appropriately.
240 Pass new arguments to release_scratch_register_on_entry.
241 (ix86_adjust_stack_and_probe): Likewise.
242 (ix86_emit_probe_stack_range): Pass new arguments to
243 release_scratch_register_on_entry.
245 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
247 PR rtl-optimization/84157
248 * combine.c (change_zero_ext): Use REG_P predicate in
249 front of HARD_REGISTER_P predicate.
251 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
253 * config/avr/avr.c (avr_option_override): Move disabling of
254 -fdelete-null-pointer-checks to...
255 * common/config/avr/avr-common.c (avr_option_optimization_table):
258 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
260 PR tree-optimization/81635
261 * tree-data-ref.c (split_constant_offset_1): For types that
262 wrap on overflow, try to use range info to prove that wrapping
265 2018-02-01 Renlin Li <renlin.li@arm.com>
268 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
270 (aarch64_register_move_cost): Likewise.
271 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
273 (REG_CLASS_NAMES): Likewise.
274 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
275 TAILCALL_ADDR_REGS. Remove IP registers.
276 * config/aarch64/aarch64.md (Ucs): Update register constraint.
278 2018-02-01 Richard Biener <rguenther@suse.de>
280 * domwalk.h (dom_walker::dom_walker): Add additional constructor
281 for specifying RPO order and allow NULL for that.
282 * domwalk.c (dom_walker::dom_walker): Likewise.
283 (dom_walker::walk): Handle NULL RPO order.
284 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
286 (rewrite_update_dom_walker): Likewise.
287 (mark_def_dom_walker): Likewise.
289 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
291 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
292 (aarch64_maybe_expand_sve_subreg_move): Declare.
293 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
294 * config/aarch64/predicates.md (aarch64_any_register_operand): New
296 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
297 that are semantically a reverse operation.
298 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
299 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
300 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
302 (aarch64_can_change_mode_class): For big-endian, forbid changes
303 between two SVE modes if they have different element sizes.
305 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
307 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
308 the TImode handling for big-endian targets.
310 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
312 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
313 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
315 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
316 Remove BSWAP handing for big-endian targets and use the form of
317 LD1RQ appropariate for the mode.
319 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
321 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
322 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
325 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
328 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
329 check for operands that need to go through aarch64_sve_reload_be.
331 2018-02-01 Jakub Jelinek <jakub@redhat.com>
333 PR tree-optimization/81661
334 PR tree-optimization/84117
335 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
336 * tree-eh.c: Include gimplify.h.
337 (find_trapping_overflow, replace_trapping_overflow,
338 rewrite_to_non_trapping_overflow): New functions.
339 * tree-vect-loop.c: Include tree-eh.h.
340 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
341 * tree-data-ref.c: Include tree-eh.h.
342 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
344 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
346 PR rtl-optimization/84123
347 * combine.c (change_zero_ext): Check if hard register satisfies
348 can_change_dest_mode before calling gen_lowpart_SUBREG.
350 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
353 * ira.c (ira_init_register_move_cost): Remove assert.
355 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
357 PR rtl-optimization/84071
358 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
359 * doc/tm.texi: Regenerate.
361 2018-01-31 Richard Biener <rguenther@suse.de>
363 PR tree-optimization/84132
364 * tree-data-ref.c (analyze_miv_subscript): Properly
365 check whether evolution_function_is_affine_multivariate_p
366 before calling gcd_of_steps_may_divide_p.
368 2018-01-31 Julia Koval <julia.koval@intel.com>
371 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
372 * config/i386/i386.md (rdpid_rex64) New.
373 (rdpid): Make 32bit only.
375 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
378 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
379 an IDENTIFIER_NODE for FUNCTION_TYPE's.
381 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
384 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
386 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
388 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
390 PR rtl-optimization/84071
391 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
392 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
394 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
396 * config/arc/arc.c (arc_handle_aux_attribute): New function.
397 (arc_attribute_table): Add 'aux' attribute.
398 (arc_in_small_data_p): Consider aux like variables.
399 (arc_is_aux_reg_p): New function.
400 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
401 (arc_get_aux_arg): New function.
402 (prepare_move_operands): Handle aux-register access.
403 (arc_handle_aux_attribute): New function.
404 * doc/extend.texi (ARC Variable attributes): Add subsection.
406 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
408 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
409 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
410 (arc_attribute_table): Add 'uncached' attribute.
411 (arc_print_operand): Print '.di' flag for uncached memory
413 (arc_in_small_data_p): Do not consider for small data the uncached
415 (arc_is_uncached_mem_p): New function.
416 * config/arc/predicates.md (compact_store_memory_operand): Check
417 for uncached memory accesses.
418 (nonvol_nonimm_operand): Likewise.
419 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
421 2018-01-31 Jakub Jelinek <jakub@redhat.com>
424 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
425 falign-loops=): Add Optimization flag.
427 2018-01-30 Jeff Law <law@redhat.com>
430 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
431 INT_REGISTERS_SAVED. Check it prior to calling
432 get_scratch_register_on_entry.
433 (ix86_adjust_stack_and_probe): Similarly.
434 (ix86_emit_probe_stack_range): Similarly.
435 (ix86_expand_prologue): Corresponding changes.
437 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
440 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
441 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
443 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
446 * lra-constraints.c (curr_insn_transform): Process AND in the
449 2018-01-30 Jakub Jelinek <jakub@redhat.com>
451 PR rtl-optimization/83986
452 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
453 dependence against last_pending_memory_flush in addition to
456 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
458 PR tree-optimization/81611
459 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
462 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
464 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
467 2018-01-30 Richard Biener <rguenther@suse.de>
468 Jakub Jelinek <jakub@redhat.com>
470 PR tree-optimization/84111
471 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
472 inner loops added during recursion, as they don't have up-to-date
475 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
478 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
479 (can_inline_edge_by_limits_p): ... here.
480 (can_early_inline_edge_p, check_callers,
481 update_caller_keys, update_callee_keys, recursive_inlining,
482 add_new_edges_to_heap, speculation_useful_p,
483 inline_small_functions,
484 inline_small_functions, flatten_function,
485 inline_to_all_callers_1): Update.
487 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
489 * profile-count.c (profile_count::combine_with_ipa_count): Handle
492 2018-01-30 Richard Biener <rguenther@suse.de>
494 PR tree-optimization/83008
495 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
496 invariant and constant vector uses in stmts when they need
499 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
502 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
503 * configure: Regenerate.
505 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
507 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
509 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
510 Use gen_rtx_REG rather than gen_lowpart.
512 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
514 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
515 rather than 0 when creating partial subregs.
517 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
519 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
522 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
525 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
526 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
527 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
528 flags. This restores the settings used before the 2017-07-24.
529 Turning off pre increment/decrement/modify allows IVOPTS to
530 optimize DF/SF loops where the index is an int.
532 2018-01-29 Richard Biener <rguenther@suse.de>
533 Kelvin Nilsen <kelvin@gcc.gnu.org>
536 * tree-vect-stmts.c (vectorizable_call): Don't call
537 targetm.vectorize_builtin_md_vectorized_function if callee is
540 2018-01-22 Carl Love <cel@us.ibm.com>
542 * doc/extend.tex: Fix typo in second arg in
543 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
545 2018-01-29 Richard Biener <rguenther@suse.de>
547 PR tree-optimization/84086
548 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
549 (flush_ssaname_freelist): When SSA names were released reset
552 2018-01-29 Richard Biener <rguenther@suse.de>
554 PR tree-optimization/84057
555 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
556 removed paths when removing edges.
558 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
560 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
561 -mfunction-return=@var{choice}.
563 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
566 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
568 (layout::print_source_line): Likewise.
569 (test_get_line_width_without_trailing_whitespace): Add test cases.
571 2018-01-27 Jakub Jelinek <jakub@redhat.com>
574 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
577 2018-01-26 Jim Wilson <jimw@sifive.com>
579 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
581 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
584 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
586 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
587 and CMP + SUB-immediate -> SUBS.
589 2018-01-26 Martin Sebor <msebor@redhat.com>
591 PR tree-optimization/83896
592 * tree-ssa-strlen.c (get_string_len): Rename...
593 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
594 Avoid assuming length is constant.
595 (handle_char_store): Use HOST_WIDE_INT for string length.
597 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
600 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
601 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
603 2018-01-26 Richard Biener <rguenther@suse.de>
605 PR rtl-optimization/84003
606 * dse.c (record_store): Only record redundant stores when
607 the earlier store aliases at least all accesses the later one does.
609 2018-01-26 Jakub Jelinek <jakub@redhat.com>
611 PR rtl-optimization/83985
612 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
613 REG_CFA_RESTORE insns.
614 (delete_unmarked_insns): Don't ignore separate shrink wrapping
615 REG_CFA_RESTORE insns here.
618 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
619 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
621 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
623 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
624 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
625 (arc_init): Likewise.
626 (arc_override_options): Likewise.
627 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
629 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
631 * config/arc/arc.h (TARGET_DBNZ): Define.
632 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
633 properly set the tune attribute.
634 (dbnz): Use TARGET_DBNZ guard.
635 * config/arc/arc.opt (mtune): Add core3 option.
637 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
639 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
640 recognize new pic like addresses.
641 (arc_delegitimize_address): Clean up.
643 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
645 * config/arc/arc-arches.def: Option mrf16 valid for all
647 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
648 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
649 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
650 * config/arc/arc-tables.opt: Regenerate.
651 * config/arc/arc.c (arc_conditional_register_usage): Handle
652 reduced register file case.
653 (arc_file_start): Set must have build attributes.
654 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
656 * config/arc/arc.opt (mrf16): Add new option.
657 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
658 * config/arc/genmultilib.awk: Handle new mrf16 option.
659 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
660 * config/arc/t-multilib: Regenerate.
661 * doc/invoke.texi (ARC Options): Document mrf16 option.
663 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
665 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
666 * config/arc/arc.c (arc_handle_secure_attribute): New function.
667 (arc_attribute_table): Add 'secure_call' attribute.
668 (arc_print_operand): Print secure call operand.
669 (arc_function_ok_for_sibcall): Don't optimize tail calls when
671 (arc_is_secure_call_p): New function. * config/arc/arc.md
672 (call_i): Add support for sjli instruction.
673 (call_value_i): Likewise.
674 * config/arc/constraints.md (Csc): New constraint.
676 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
677 John Eric Martin <John.Martin@emmicro-us.com>
679 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
680 * config/arc/arc.c (_arc_jli_section): New struct.
681 (arc_jli_section): New type.
682 (rc_jli_sections): New static variable.
683 (arc_handle_jli_attribute): New function.
684 (arc_attribute_table): Add jli_always and jli_fixed attribute.
685 (arc_file_end): New function.
686 (TARGET_ASM_FILE_END): Define.
687 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
688 (arc_add_jli_section): New function.
689 (jli_call_scan): Likewise.
690 (arc_reorg): Call jli_call_scan.
691 (arc_output_addsi): Remove 'S' from printing asm operand.
692 (arc_is_jli_call_p): New function.
693 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
695 (movhi_insn): Likewise.
696 (movsi_insn): Likewise.
697 (movsi_set_cc_insn): Likewise.
698 (loadqi_update): Likewise.
699 (load_zeroextendqisi_update): Likewise.
700 (load_signextendqisi_update): Likewise.
701 (loadhi_update): Likewise.
702 (load_zeroextendhisi_update): Likewise.
703 (load_signextendhisi_update): Likewise.
704 (loadsi_update): Likewise.
705 (loadsf_update): Likewise.
706 (movsicc_insn): Likewise.
707 (bset_insn): Likewise.
708 (bxor_insn): Likewise.
709 (bclr_insn): Likewise.
710 (bmsk_insn): Likewise.
711 (bicsi3_insn): Likewise.
712 (cmpsi_cc_c_insn): Likewise.
713 (movsi_ne): Likewise.
714 (movsi_cond_exec): Likewise.
715 (clrsbsi2): Likewise.
726 (call_i): Remove 'S' asm letter, add jli instruction.
727 (call_value_i): Likewise.
728 * config/arc/arc.op (mjli-always): New option.
729 * config/arc/constraints.md (Cji): New constraint.
730 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
732 (subsf3_fpx): Likewise.
733 (mulsf3_fpx): Likewise.
734 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
736 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
738 * doc/invoke.texi (ARC): Document mjli-always option.
740 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
742 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
743 and use incw and decw where possible
744 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
746 2018-01-26 Richard Biener <rguenther@suse.de>
748 PR tree-optimization/81082
749 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
750 association if it requires casting to unsigned.
751 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
752 from fold_plusminus_mult_expr to catch important cases late when
753 range info is available.
755 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
757 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
758 * configure.ac (hidden_linkonce): New test.
759 * configure: Regenerate.
760 * config.in: Regenerate.
762 2018-01-26 Julia Koval <julia.koval@intel.com>
764 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
765 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
766 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
767 _mm_mask_bitshuffle_epi64_mask): Fix type.
768 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
769 USI_FTYPE_V4DI_V4DI_USI): Remove.
770 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
771 __builtin_ia32_vpshufbitqmb256_mask,
772 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
773 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
774 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
776 2018-01-26 Alan Modra <amodra@gmail.com>
779 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
780 UNSPEC_VBPERMQ. Sort other unspecs.
782 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
784 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
786 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
789 * predict.c (drop_profile): Do not push/pop cfun; update also
791 (handle_missing_profiles): Fix logic looking for zero profiles.
793 2018-01-25 Jakub Jelinek <jakub@redhat.com>
796 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
797 on functions with #pragma omp declare simd or functions with simd
799 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
800 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
801 Remove trailing \n from warning_at calls.
803 2018-01-25 Tom de Vries <tom@codesourcery.com>
806 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
807 for neutered workers.
809 2018-01-24 Joseph Myers <joseph@codesourcery.com>
812 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
813 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
815 2018-01-24 Jeff Law <law@redhat.com>
818 * i386.c (get_probe_interval): Move to earlier point.
819 (ix86_compute_frame_layout): If -fstack-clash-protection and
820 the frame is larger than the probe interval, then use pushes
821 to save registers rather than reg->mem moves.
822 (ix86_expand_prologue): Remove conditional for int_registers_saved
825 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
828 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
829 min/max for never referenced object.
831 2018-01-24 Jakub Jelinek <jakub@redhat.com>
834 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
836 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
837 attributes from DECL_ATTRIBUTES (decl) without affecting
838 DECL_ATTRIBUTES (current_function_decl).
839 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
840 functions with non-NULL DECL_ABSTRACT_ORIGIN.
842 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
844 PR tree-optimization/83979
845 * fold-const.c (fold_comparison): Use constant_boolean_node
846 instead of boolean_{true,false}_node.
848 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
850 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
853 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
855 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
856 Simplify the clause that sets the length attribute.
857 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
858 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
859 clause that sets the length attribute.
860 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
862 2018-01-24 Tom de Vries <tom@codesourcery.com>
865 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
866 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
867 Add strict parameter.
868 (prevent_branch_around_nothing): Insert dummy insn between branch to
869 label and label with no ptx insn inbetween.
870 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
872 2018-01-24 Tom de Vries <tom@codesourcery.com>
875 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
876 for neutered threads in warp.
877 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
879 2018-01-24 Richard Biener <rguenther@suse.de>
881 PR tree-optimization/83176
882 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
885 2018-01-24 Richard Biener <rguenther@suse.de>
887 PR tree-optimization/82819
888 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
889 code generating pluses that are no-ops in the target precision.
891 2018-01-24 Richard Biener <rguenther@suse.de>
894 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
896 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
898 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
899 to merge probabilities.
900 * predict.c (probably_never_executed): Also mark as cold functions
901 with global 0 profile and guessed local profile.
902 * profile-count.c (profile_probability::combine_with_count): New
904 * profile-count.h (profile_probability::operator*,
905 profile_probability::operator*=, profile_probability::operator/,
906 profile_probability::operator/=): Reduce precision to adjusted
907 and set value to guessed on contradictory divisions.
908 (profile_probability::combine_with_freq): Remove.
909 (profile_probability::combine_wiht_count): Declare.
910 (profile_count::force_nonzero):: Set to adjusted.
911 (profile_count::probability_in):: Set quality to adjusted.
912 * tree-ssa-tail-merge.c (replace_block_by): Use
915 2018-01-23 Andrew Waterman <andrew@sifive.com>
916 Jim Wilson <jimw@sifive.com>
918 * config/riscv/riscv.c (riscv_stack_boundary): New.
919 (riscv_option_override): Set riscv_stack_boundary. Handle
920 riscv_preferred_stack_boundary_arg.
921 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
922 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
923 (STACK_BOUNDARY): Set to riscv_stack_boundary.
924 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
925 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
926 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
928 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
931 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
932 of struct ix86_frame.
933 (ix86_expand_epilogue): Likewise. Add a local variable for
934 the reg_save_offset field in struct ix86_frame.
936 2018-01-23 Bin Cheng <bin.cheng@arm.com>
938 PR tree-optimization/82604
939 * tree-loop-distribution.c (enum partition_kind): New enum item
940 PKIND_PARTIAL_MEMSET.
941 (partition_builtin_p): Support above new enum item.
942 (generate_code_for_partition): Ditto.
943 (compute_access_range): Differentiate cases that equality can be
944 proven at all loops, the innermost loops or no loops.
945 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
946 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
947 (finalize_partitions, distribute_loop): Don't fuse partition of
948 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
949 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
952 2018-01-23 Martin Liska <mliska@suse.cz>
954 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
955 order to ignore the predictor.
956 (PRED_POLYMORPHIC_CALL): Likewise.
957 (PRED_RECURSIVE_CALL): Likewise.
959 2018-01-23 Martin Liska <mliska@suse.cz>
961 * tree-profile.c (tree_profiling): Print function header to
962 aware reader which function we are working on.
963 * value-prof.c (gimple_find_values_to_profile): Do not print
964 not interesting value histograms.
966 2018-01-23 Martin Liska <mliska@suse.cz>
968 * profile-count.h (enum profile_quality): Add
969 profile_uninitialized as the first value. Do not number values
970 as they are zero based.
971 (profile_count::verify): Update sanity check.
972 (profile_probability::verify): Likewise.
974 2018-01-23 Nathan Sidwell <nathan@acm.org>
976 * doc/invoke.texi (ffor-scope): Deprecate.
978 2018-01-23 David Malcolm <dmalcolm@redhat.com>
980 PR tree-optimization/83510
981 * domwalk.c (set_all_edges_as_executable): New function.
982 (dom_walker::dom_walker): Convert bool param
983 "skip_unreachable_blocks" to enum reachability. Move setup of
984 edge flags to set_all_edges_as_executable and only do it when
985 reachability is REACHABLE_BLOCKS.
986 * domwalk.h (enum dom_walker::reachability): New enum.
987 (dom_walker::dom_walker): Convert bool param
988 "skip_unreachable_blocks" to enum reachability.
989 (set_all_edges_as_executable): New decl.
990 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
991 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
993 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
994 but converting true to REACHABLE_BLOCKS.
995 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
997 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
998 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
999 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1001 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1002 if check_all_array_refs will be called.
1004 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1006 * tree.c (selftest::test_location_wrappers): Add more test
1009 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1011 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1012 (selftest::test_bit_in_range): Likewise.
1014 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1017 * doc/sourcebuild.texi (vect_float): Say that the selector
1018 only describes the situation when -funsafe-math-optimizations is on.
1019 (vect_float_strict): Document.
1021 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1023 PR tree-optimization/83965
1024 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1025 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1026 instead of checking only for a reduction.
1027 (vect_recog_widen_sum_pattern): Likewise.
1029 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1031 * predict.c (probably_never_executed): Only use precise profile info.
1032 (compute_function_frequency): Skip after inlining hack since we now
1033 have quality checking.
1035 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1037 * profile-count.h (profile_probability::very_unlikely,
1038 profile_probability::unlikely, profile_probability::even): Set
1039 precision to guessed.
1041 2018-01-23 Richard Biener <rguenther@suse.de>
1043 PR tree-optimization/83963
1044 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1045 Properly terminate dominator walk when crossing the exit edge not
1046 when visiting its source block.
1048 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1051 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1052 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1054 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1056 PR tree-optimization/83957
1057 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1058 semicolon after for body surrounded by braces.
1060 PR tree-optimization/83081
1061 * profile-count.h (profile_probability::split): New method.
1062 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1063 Use profile_probability::split.
1064 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1065 when splitting a single conditional jump into 2.
1067 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1069 PR tree-optimization/69452
1070 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1073 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1075 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1076 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1077 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1079 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1081 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1082 * config/rl78/rl78.md: New define_expand "movdi"
1083 * config/rl78/rl78.c: New function definition rl78_split_movdi
1085 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1088 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1090 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1091 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1092 128-bit to produce an UNSPEC move to get the double word with the
1093 signbit and then a shift directly to do signbit.
1094 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1095 implementation with a new version that just does either a direct
1096 move or a regular move. Move memory interface to separate insns.
1097 Move insns so they are next to the expander.
1098 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1099 with signbit move. Split big and little endian case.
1100 (signbit<mode>2_dm_mem_le): Likewise.
1101 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1102 (signbit<mode>2_dm2): Likewise.
1104 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1106 * config/rl78/rl78.md: New define_expand "anddi3".
1108 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1110 * config/rl78/rl78.md: New define_expand "umindi3".
1112 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1114 * config/rl78/rl78.md: New define_expand "smindi3".
1116 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1118 * config/rl78/rl78.md: New define_expand "smaxdi3".
1120 2018-01-22 Carl Love <cel@us.ibm.com>
1122 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1123 LVX_V1TI): Add macro expansion.
1124 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1125 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1126 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1127 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1128 Change check to determine if the instruction is a byte reversing
1129 entry. Fix typo in comment.
1130 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1131 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1132 Add def_builtin calls for new builtins.
1133 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1134 Add define_insn expansion.
1136 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1138 * config/rl78/rl78.md: New define_expand "umaxdi3".
1140 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1142 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1143 for non-QImode registers
1145 2018-01-22 Richard Biener <rguenther@suse.de>
1147 PR tree-optimization/83963
1148 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1149 including the loop exit block.
1150 (scop_detection::merge_sese): Likewise.
1151 (scop_detection::add_scop): Do it here instead.
1153 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1155 * doc/sourcebuild.texi (arm_softfloat): Document.
1157 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1160 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1161 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1162 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1164 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1165 David Edelsohn <dje.gcc@gmail.com>
1168 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1169 Change "crset eq" to "crset 2".
1170 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1171 (*call_indirect_aix<mode>_nospec): Likewise.
1172 (*call_value_indirect_aix<mode>_nospec): Likewise.
1173 (*call_indirect_elfv2<mode>_nospec): Likewise.
1174 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1175 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1176 change assembly output from . to $.
1177 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1178 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1179 (*tablejump<mode>_internal1_nospec): Likewise.
1181 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1184 * config/sh/sh_optimize_sett_clrt.cc:
1185 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1187 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1189 PR tree-optimization/83940
1190 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1191 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1192 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1193 use it to pass back the definition type.
1194 (vect_check_store_rhs): Likewise rhs_dt_out.
1195 (vect_build_gather_load_calls): Add a mask_dt argument and use
1196 it instead of a call to vect_is_simple_use.
1197 (vectorizable_store): Update calls to vect_check_load_store_mask
1198 and vect_check_store_rhs. Use the dt returned by the latter instead
1199 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1200 instead of calls to vect_is_simple_use. Pass the scalar rather
1201 than the vector operand to vect_is_simple_use when handling
1202 second and subsequent copies of an rhs value.
1203 (vectorizable_load): Update calls to vect_check_load_store_mask
1204 and vect_build_gather_load_calls. Use the cached mask_dt and
1205 gs_info.offset_dt instead of calls to vect_is_simple_use.
1207 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1210 * tree-emutls.c: Include gimplify.h.
1211 (lower_emutls_2): New function.
1212 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1213 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1214 it before further processing.
1217 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1218 UINTVAL (trueop1) instead of INTVAL (op1).
1220 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1224 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1225 INCOMING_FRAME_SP_OFFSET if not defined.
1226 (scan_trace): Add ENTRY argument. If true and
1227 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1228 emit a note to adjust the CFA offset.
1229 (create_cfi_notes): Adjust scan_trace callers.
1230 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1231 INCOMING_FRAME_SP_OFFSET in the CIE.
1232 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1233 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1235 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1236 * doc/tm.texi: Regenerated.
1238 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1240 PR rtl-optimization/83147
1241 * lra-constraints.c (remove_inheritance_pseudos): Use
1242 lra_substitute_pseudo_within_insn.
1244 2018-01-19 Tom de Vries <tom@codesourcery.com>
1245 Cesar Philippidis <cesar@codesourcery.com>
1248 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1250 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1253 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1254 spaces for function labels.
1256 2018-01-19 Martin Liska <mliska@suse.cz>
1258 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1259 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1260 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1261 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1262 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1263 (PRED_CONST_RETURN): Change from 69 to 65.
1264 (PRED_NULL_RETURN): Change from 91 to 71.
1265 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1266 (PRED_LOOP_GUARD): Change from 66 to 73.
1268 2018-01-19 Martin Liska <mliska@suse.cz>
1270 * predict.c (predict_insn_def): Add new assert.
1271 (struct branch_predictor): Change type to signed integer.
1272 (test_prediction_value_range): Amend test to cover
1274 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1275 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1276 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1277 (PRED_LOOP_IV_COMPARE): Likewise.
1278 * predict.h (PROB_UNINITIALIZED): Define new constant.
1280 2018-01-19 Martin Liska <mliska@suse.cz>
1282 * predict.c (dump_prediction): Add new format for
1283 analyze_brprob.py script which is enabled with -details
1285 * profile-count.h (precise_p): New function.
1287 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1289 PR tree-optimization/83922
1290 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1291 there are no statements that need masking.
1292 (vect_active_double_reduction_p): New function.
1293 (vect_analyze_loop_operations): Use it when handling phis that
1294 are not in the loop header.
1296 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1298 PR tree-optimization/83914
1299 * tree-vect-loop.c (vectorizable_induction): Don't convert
1300 init_expr or apply the peeling adjustment for inductions
1301 that are nested within the vectorized loop.
1303 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1305 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1308 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1312 * function.h (gimplify_parameters): Add gimple_seq * argument.
1313 * function.c: Include gimple.h and options.h.
1314 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1315 for the added local temporaries if needed.
1316 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1317 if there are any parameter cleanups, wrap whole body into a
1318 try/finally with the cleanups.
1320 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1323 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1324 Use GET_MODE_CLASS for scalar floating point.
1326 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1330 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1331 Fix call of call_cgraph_insertion_hooks.
1333 2018-01-18 Martin Sebor <msebor@redhat.com>
1335 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1337 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1340 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1343 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1346 * common.opt: (-ffile-prefix-map): New option.
1347 * opts.c (common_handle_option): Defer it.
1348 * opts-global.c (handle_common_deferred_options): Handle it.
1349 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1350 * file-prefix-map.h: New file.
1351 (remap_debug_filename, add_debug_prefix_map): ...here.
1352 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1353 * final.c (debug_prefix_map, add_debug_prefix_map
1354 remap_debug_filename): Move to...
1355 * file-prefix-map.c: New file.
1356 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1357 generalize, get rid of alloca(), use strrchr() instead of strchr().
1358 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1359 Implement in terms of add_prefix_map().
1360 (remap_macro_filename, remap_debug_filename): Implement in term of
1362 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1363 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1364 * dbxout.c: Include file-prefix-map.h.
1365 * varasm.c: Likewise.
1366 * vmsdbgout.c: Likewise.
1367 * xcoffout.c: Likewise.
1368 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1369 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1370 * doc/invoke.texi (-ffile-prefix-map): Document.
1371 (-fdebug-prefix-map): Update description.
1373 2018-01-18 Martin Liska <mliska@suse.cz>
1375 * config/i386/i386.c (indirect_thunk_name): Document that also
1377 (output_indirect_thunk): Document why both instructions
1378 (pause and lfence) are generated.
1380 2018-01-18 Richard Biener <rguenther@suse.de>
1382 PR tree-optimization/83887
1383 * graphite-scop-detection.c
1384 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1385 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1386 (scop_detection::merge_sese): Re-implement with a flood-fill
1387 algorithm that properly finds a SESE region if it exists.
1389 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1392 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1393 pointer_diff optimizations use view_convert instead of convert.
1395 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1397 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1398 Generate different code for -mno-speculate-indirect-jumps.
1399 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1400 (*call_indirect_aix<mode>): Disable for
1401 -mno-speculate-indirect-jumps.
1402 (*call_indirect_aix<mode>_nospec): New define_insn.
1403 (*call_value_indirect_aix<mode>): Disable for
1404 -mno-speculate-indirect-jumps.
1405 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1406 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1407 -mno-speculate-indirect-jumps.
1408 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1410 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1412 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1413 long double type, set the flags for noting the default long double
1414 type, even if we don't pass or return a long double type.
1416 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1419 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1422 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1424 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1425 support for merge[hl].
1426 (fold_mergehl_helper): New helper function.
1427 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1428 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1429 (altivec_vmrglw_direct): Add xxmrglw insn.
1431 2018-01-17 Andrew Waterman <andrew@sifive.com>
1433 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1434 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1436 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1439 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1440 call the lto_location_cache before reading the
1441 DECL_SOURCE_LOCATION of the types.
1443 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1444 Richard Sandiford <richard.sandiford@linaro.org>
1446 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1447 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1448 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1449 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1450 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1452 * config/aarch64/constraints.md (aarch64_movti_operand):
1454 * config/aarch64/predicates.md (Uti): Add new constraint.
1456 2018-01-17 Carl Love <cel@us.ibm.com>
1457 * config/rs6000/vsx.md (define_expand xl_len_r,
1458 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1459 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1461 (define_expand, define_insn): Move the shift left from the
1462 define_insn to the define_expand for lxvl and stxvl instructions.
1463 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1464 and XL_LEN_R definitions to PURE.
1466 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1468 * config/i386/i386.c (indirect_thunk_name): Declare regno
1469 as unsigned int. Compare regno with INVALID_REGNUM.
1470 (output_indirect_thunk): Ditto.
1471 (output_indirect_thunk_function): Ditto.
1472 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1473 in the call to output_indirect_thunk_function.
1475 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1478 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1479 rather than the size of inner_type to determine the stack slot size
1480 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1482 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1485 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1488 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1490 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1491 endian Linux systems to optionally enable multilibs for selecting
1492 the long double type if the user configured an explicit type.
1493 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1494 have no long double multilibs if not defined.
1495 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1496 warn if the user used -mabi={ieee,ibm}longdouble and we built
1497 multilibs for long double.
1498 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1499 appropriate multilib option.
1500 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1502 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1503 for building long double multilibs.
1504 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1506 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1508 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1511 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1513 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1516 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1519 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1522 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1524 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1525 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1527 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1529 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1530 different rtl trees depending on TARGET_64BIT.
1531 (rs6000_gen_lvx): Likewise.
1533 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1535 * config/visium/visium.md (nop): Tweak comment.
1536 (hazard_nop): Likewise.
1538 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1540 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1541 -mspeculate-indirect-jumps.
1542 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1543 for -mno-speculate-indirect-jumps.
1544 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1545 (*call_value_indirect_elfv2<mode>): Disable for
1546 -mno-speculate-indirect-jumps.
1547 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1548 (indirect_jump): Emit different RTL for
1549 -mno-speculate-indirect-jumps.
1550 (*indirect_jump<mode>): Disable for
1551 -mno-speculate-indirect-jumps.
1552 (*indirect_jump<mode>_nospec): New define_insn.
1553 (tablejump): Emit different RTL for
1554 -mno-speculate-indirect-jumps.
1555 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1556 (tablejumpsi_nospec): New define_expand.
1557 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1558 (tablejumpdi_nospec): New define_expand.
1559 (*tablejump<mode>_internal1): Disable for
1560 -mno-speculate-indirect-jumps.
1561 (*tablejump<mode>_internal1_nospec): New define_insn.
1562 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1565 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1567 * caller-save.c (insert_save): Drop unnecessary parameter. All
1570 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1571 Richard Biener <rguenth@suse.de>
1574 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1575 return early, inline manually is_gimple_sizepos. Make sure if we
1576 call gimplify_expr we don't end up with a gimple constant.
1577 * tree.c (variably_modified_type_p): Don't return true for
1578 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1579 * gimplify.h (is_gimple_sizepos): Remove.
1581 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1583 PR tree-optimization/83857
1584 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1585 vectorizable_live_operation for pure SLP statements.
1586 (vectorizable_live_operation): Handle PHIs.
1588 2018-01-16 Richard Biener <rguenther@suse.de>
1590 PR tree-optimization/83867
1591 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1592 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1594 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1597 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1598 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1599 If off is not INTEGER_CST, issue a may not be aligned warning
1600 rather than isn't aligned. Use isn%'t rather than isn't.
1601 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1603 <case MULT_EXPR>: Improve the case when bottom and one of the
1604 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1605 operand, in that case check if the other operand is multiple of
1606 bottom divided by the INTEGER_CST operand.
1608 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1611 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1612 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1613 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1614 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1615 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1616 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1617 (pa_function_arg_size): New function.
1619 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1621 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1622 in a separate statement.
1624 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1626 PR tree-optimization/83847
1627 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1628 group gathers and scatters.
1630 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1632 PR rtl-optimization/86620
1633 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1635 PR rtl-optimization/83213
1636 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1637 to last if both are JUMP_INSNs.
1639 PR tree-optimization/83843
1640 * gimple-ssa-store-merging.c
1641 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1642 store_immediate_info for bswap/nop orig_stores.
1644 2018-01-15 Andrew Waterman <andrew@sifive.com>
1646 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1648 <UDIV>: Increase cost if !TARGET_DIV.
1650 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1652 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1653 (define_attr "cr_logical_3op"): New.
1654 (cceq_ior_compare): Adjust.
1655 (cceq_ior_compare_complement): Adjust.
1656 (*cceq_rev_compare): Adjust.
1657 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1658 (is_cracked_insn): Adjust.
1659 (insn_must_be_first_in_group): Adjust.
1660 * config/rs6000/40x.md: Adjust.
1661 * config/rs6000/440.md: Adjust.
1662 * config/rs6000/476.md: Adjust.
1663 * config/rs6000/601.md: Adjust.
1664 * config/rs6000/603.md: Adjust.
1665 * config/rs6000/6xx.md: Adjust.
1666 * config/rs6000/7450.md: Adjust.
1667 * config/rs6000/7xx.md: Adjust.
1668 * config/rs6000/8540.md: Adjust.
1669 * config/rs6000/cell.md: Adjust.
1670 * config/rs6000/e300c2c3.md: Adjust.
1671 * config/rs6000/e500mc.md: Adjust.
1672 * config/rs6000/e500mc64.md: Adjust.
1673 * config/rs6000/e5500.md: Adjust.
1674 * config/rs6000/e6500.md: Adjust.
1675 * config/rs6000/mpc.md: Adjust.
1676 * config/rs6000/power4.md: Adjust.
1677 * config/rs6000/power5.md: Adjust.
1678 * config/rs6000/power6.md: Adjust.
1679 * config/rs6000/power7.md: Adjust.
1680 * config/rs6000/power8.md: Adjust.
1681 * config/rs6000/power9.md: Adjust.
1682 * config/rs6000/rs64.md: Adjust.
1683 * config/rs6000/titan.md: Adjust.
1685 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1687 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1688 ix86_indirect_branch_register logic.
1690 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1692 * config/i386/constraints.md (Bs): Update
1693 ix86_indirect_branch_register check. Don't check
1694 ix86_indirect_branch_register with GOT_memory_operand.
1696 * config/i386/predicates.md (GOT_memory_operand): Don't check
1697 ix86_indirect_branch_register here.
1698 (GOT32_symbol_operand): Likewise.
1700 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1702 * config/i386/predicates.md (constant_call_address_operand):
1703 Rewrite ix86_indirect_branch_register logic.
1704 (sibcall_insn_operand): Likewise.
1706 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1708 * config/i386/constraints.md (Bs): Replace
1709 ix86_indirect_branch_thunk_register with
1710 ix86_indirect_branch_register.
1712 * config/i386/i386.md (indirect_jump): Likewise.
1713 (tablejump): Likewise.
1714 (*sibcall_memory): Likewise.
1715 (*sibcall_value_memory): Likewise.
1716 Peepholes of indirect call and jump via memory: Likewise.
1717 * config/i386/i386.opt: Likewise.
1718 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1719 (GOT_memory_operand): Likewise.
1720 (call_insn_operand): Likewise.
1721 (sibcall_insn_operand): Likewise.
1722 (GOT32_symbol_operand): Likewise.
1724 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1727 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1728 type rather than type addr's type points to.
1729 (expand_omp_atomic_mutex): Likewise.
1730 (expand_omp_atomic): Likewise.
1732 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1735 * config/i386/i386.c (output_indirect_thunk_function): Use
1736 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1737 for __x86_return_thunk.
1739 2018-01-15 Richard Biener <rguenther@suse.de>
1742 * expmed.c (extract_bit_field_1): Fix typo.
1744 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1747 * config/arm/iterators.md (VF): New mode iterator.
1748 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1749 Remove integer-related logic from pattern.
1750 (neon_vabd<mode>_3): Likewise.
1752 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1755 * common.opt (fstrict-overflow): No longer an alias.
1756 (fwrapv-pointer): New option.
1757 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1758 also for pointer types based on flag_wrapv_pointer.
1759 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1760 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1761 opts->x_flag_wrapv got set.
1762 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1763 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1764 POINTER_TYPE_OVERFLOW_UNDEFINED.
1765 * match.pd: Likewise in address comparison pattern.
1766 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1768 2018-01-15 Richard Biener <rguenther@suse.de>
1771 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1772 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1773 Reset type names to their identifier if their TYPE_DECL doesn't
1774 have linkage (and thus is used for ODR and devirt).
1775 (save_debug_info_for_decl): Remove.
1776 (save_debug_info_for_type): Likewise.
1777 (add_tree_to_fld_list): Adjust.
1778 * tree-pretty-print.c (dump_generic_node): Make dumping of
1779 type names more robust.
1781 2018-01-15 Richard Biener <rguenther@suse.de>
1783 * BASE-VER: Bump to 8.0.1.
1785 2018-01-14 Martin Sebor <msebor@redhat.com>
1788 * builtins.c (check_access): Avoid warning when the no-warning bit
1791 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1793 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1794 * ira-color (allocno_hard_regs_compare): Likewise.
1796 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1799 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1800 Use .pushsection/.popsection.
1802 2018-01-14 Martin Sebor <msebor@redhat.com>
1805 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1807 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1809 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1810 entry from extra_headers.
1811 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1812 extra_headers, make the list bitwise identical to the i?86-*-* one.
1814 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1816 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1817 -mcmodel=large with -mindirect-branch=thunk,
1818 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1819 -mfunction-return=thunk-extern.
1820 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1821 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1822 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1824 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1826 * config/i386/i386.c (print_reg): Print the name of the full
1827 integer register without '%'.
1828 (ix86_print_operand): Handle 'V'.
1829 * doc/extend.texi: Document 'V' modifier.
1831 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1833 * config/i386/constraints.md (Bs): Disallow memory operand for
1834 -mindirect-branch-register.
1836 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1837 (GOT_memory_operand): Likewise.
1838 (call_insn_operand): Likewise.
1839 (sibcall_insn_operand): Likewise.
1840 (GOT32_symbol_operand): Likewise.
1841 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1842 for -mindirect-branch-register.
1843 (tablejump): Likewise.
1844 (*sibcall_memory): Likewise.
1845 (*sibcall_value_memory): Likewise.
1846 Disallow peepholes of indirect call and jump via memory for
1847 -mindirect-branch-register.
1848 (*call_pop): Replace m with Bw.
1849 (*call_value_pop): Likewise.
1850 (*sibcall_pop_memory): Replace m with Bs.
1851 * config/i386/i386.opt (mindirect-branch-register): New option.
1852 * doc/invoke.texi: Document -mindirect-branch-register option.
1854 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1856 * config/i386/i386-protos.h (ix86_output_function_return): New.
1857 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1858 set function_return_type.
1859 (indirect_thunk_name): Add ret_p to indicate thunk for function
1861 (output_indirect_thunk_function): Pass false to
1862 indirect_thunk_name.
1863 (ix86_output_indirect_branch_via_reg): Likewise.
1864 (ix86_output_indirect_branch_via_push): Likewise.
1865 (output_indirect_thunk_function): Create alias for function
1866 return thunk if regno < 0.
1867 (ix86_output_function_return): New function.
1868 (ix86_handle_fndecl_attribute): Handle function_return.
1869 (ix86_attribute_table): Add function_return.
1870 * config/i386/i386.h (machine_function): Add
1871 function_return_type.
1872 * config/i386/i386.md (simple_return_internal): Use
1873 ix86_output_function_return.
1874 (simple_return_internal_long): Likewise.
1875 * config/i386/i386.opt (mfunction-return=): New option.
1876 (indirect_branch): Mention -mfunction-return=.
1877 * doc/extend.texi: Document function_return function attribute.
1878 * doc/invoke.texi: Document -mfunction-return= option.
1880 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1882 * config/i386/i386-opts.h (indirect_branch): New.
1883 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1884 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1885 with local indirect jump when converting indirect call and jump.
1886 (ix86_set_indirect_branch_type): New.
1887 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1888 (indirectlabelno): New.
1889 (indirect_thunk_needed): Likewise.
1890 (indirect_thunk_bnd_needed): Likewise.
1891 (indirect_thunks_used): Likewise.
1892 (indirect_thunks_bnd_used): Likewise.
1893 (INDIRECT_LABEL): Likewise.
1894 (indirect_thunk_name): Likewise.
1895 (output_indirect_thunk): Likewise.
1896 (output_indirect_thunk_function): Likewise.
1897 (ix86_output_indirect_branch_via_reg): Likewise.
1898 (ix86_output_indirect_branch_via_push): Likewise.
1899 (ix86_output_indirect_branch): Likewise.
1900 (ix86_output_indirect_jmp): Likewise.
1901 (ix86_code_end): Call output_indirect_thunk_function if needed.
1902 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1904 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1905 (ix86_attribute_table): Add indirect_branch.
1906 * config/i386/i386.h (machine_function): Add indirect_branch_type
1907 and has_local_indirect_jump.
1908 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1910 (tablejump): Likewise.
1911 (*indirect_jump): Use ix86_output_indirect_jmp.
1912 (*tablejump_1): Likewise.
1913 (simple_return_indirect_internal): Likewise.
1914 * config/i386/i386.opt (mindirect-branch=): New option.
1915 (indirect_branch): New.
1918 (thunk-inline): Likewise.
1919 (thunk-extern): Likewise.
1920 * doc/extend.texi: Document indirect_branch function attribute.
1921 * doc/invoke.texi: Document -mindirect-branch= option.
1923 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1926 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1928 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1930 * ipa-inline.c (want_inline_small_function_p): Return false if
1931 inlining has already failed with CIF_FINAL_ERROR.
1932 (update_caller_keys): Call want_inline_small_function_p before
1934 (update_callee_keys): Likewise.
1936 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1938 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1940 (rs6000_quadword_masked_address_p): Likewise.
1941 (quad_aligned_load_p): Likewise.
1942 (quad_aligned_store_p): Likewise.
1943 (const_load_sequence_p): Add comment to describe the outer-most loop.
1944 (mimic_memory_attributes_and_flags): New function.
1945 (rs6000_gen_stvx): Likewise.
1946 (replace_swapped_aligned_store): Likewise.
1947 (rs6000_gen_lvx): Likewise.
1948 (replace_swapped_aligned_load): Likewise.
1949 (replace_swapped_load_constant): Capitalize argument name in
1950 comment describing this function.
1951 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1952 and stores that access quad-word aligned addresses and replace
1953 with stvx or lvx instructions when appropriate.
1954 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1955 New function prototype.
1956 (rs6000_quadword_masked_address_p): Likewise.
1957 (rs6000_gen_lvx): Likewise.
1958 (rs6000_gen_stvx): Likewise.
1959 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1960 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1961 when memory address is aligned.
1962 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1963 this split to select lvx instruction when memory address is aligned.
1964 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1965 instruction when memory address is aligned.
1966 (*vsx_le_perm_load_v16qi): Likewise.
1967 (four unnamed splitters): Modify to select the stvx instruction
1968 when memory is aligned.
1970 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1972 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1973 which appears in the queue multiple times.
1975 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1976 Alan Hayward <alan.hayward@arm.com>
1977 David Sherwood <david.sherwood@arm.com>
1979 * tree-vectorizer.h (vec_lower_bound): New structure.
1980 (_loop_vec_info): Add check_nonzero and lower_bounds.
1981 (LOOP_VINFO_CHECK_NONZERO): New macro.
1982 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1983 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1984 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1985 fields. Make seg_len the distance travelled, not including the
1987 (dr_direction_indicator): Declare.
1988 (dr_zero_step_indicator): Likewise.
1989 (dr_known_forward_stride_p): Likewise.
1990 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1992 (runtime_alias_check_p): Allow runtime alias checks with
1994 (operator ==): Compare access_size and align.
1995 (prune_runtime_alias_test_list): Rework for new distinction between
1996 the access_size and seg_len.
1997 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1999 (get_segment_min_max): New function.
2000 (create_intersect_range_checks): Use it.
2001 (dr_step_indicator): New function.
2002 (dr_direction_indicator): Likewise.
2003 (dr_zero_step_indicator): Likewise.
2004 (dr_known_forward_stride_p): Likewise.
2005 * tree-loop-distribution.c (data_ref_segment_size): Return
2006 DR_STEP * (niters - 1).
2007 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2009 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2010 (vect_preserves_scalar_order_p): New function, split out from...
2011 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2012 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2013 (vect_vfa_access_size): New function.
2014 (vect_vfa_align): Likewise.
2015 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2016 (dump_lower_bound): New function.
2017 (vect_check_lower_bound): Likewise.
2018 (vect_small_gap_p): Likewise.
2019 (vectorizable_with_step_bound_p): Likewise.
2020 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2021 depencies if the vectorization factor is 1. Convert the checks
2022 for nonzero steps into checks on the bounds of DR_STEP. Try using
2023 a bunds check for variable steps if the minimum required step is
2024 relatively small. Update calls to the dr_with_seg_len
2025 constructor and to vect_compile_time_alias.
2026 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2028 (vect_loop_versioning): Call it.
2029 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2031 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2033 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2034 Alan Hayward <alan.hayward@arm.com>
2035 David Sherwood <david.sherwood@arm.com>
2037 * doc/sourcebuild.texi (vect_scatter_store): Document.
2038 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2040 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2042 * genopinit.c (main): Add supports_vec_scatter_store and
2043 supports_vec_scatter_store_cached to target_optabs.
2044 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2045 IFN_MASK_SCATTER_STORE.
2046 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2048 * internal-fn.h (internal_store_fn_p): Declare.
2049 (internal_fn_stored_value_index): Likewise.
2050 * internal-fn.c (scatter_store_direct): New macro.
2051 (expand_scatter_store_optab_fn): New function.
2052 (direct_scatter_store_optab_supported_p): New macro.
2053 (internal_store_fn_p): New function.
2054 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2055 IFN_MASK_SCATTER_STORE.
2056 (internal_fn_mask_index): Likewise.
2057 (internal_fn_stored_value_index): New function.
2058 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2060 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2061 * optabs-query.c (supports_vec_scatter_store_p): New function.
2062 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2063 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2064 true for scatter stores.
2065 (vect_gather_scatter_fn_p): Handle scatter stores too.
2066 (vect_check_gather_scatter): Consider using scatter stores if
2067 supports_vec_scatter_store_p.
2068 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2070 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2071 internal_fn_stored_value_index.
2072 (check_load_store_masking): Handle scatter stores too.
2073 (vect_get_store_rhs): Make public.
2074 (vectorizable_call): Use internal_store_fn_p.
2075 (vectorizable_store): Handle scatter store internal functions.
2076 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2077 when deciding whether the end of the group has been reached.
2078 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2079 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2080 (mask_scatter_store<mode>): New insns.
2082 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2083 Alan Hayward <alan.hayward@arm.com>
2084 David Sherwood <david.sherwood@arm.com>
2086 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2087 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2088 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2090 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2091 Use vect_truncate_gather_scatter_offset if we can't treat the
2092 operation as a normal gather load or scatter store.
2093 (get_group_load_store_type): Take the gather_scatter_info
2094 as argument. Try using a gather load or scatter store for
2095 single-element groups.
2096 (get_load_store_type): Update calls to get_group_load_store_type
2097 and vect_use_strided_gather_scatters_p.
2099 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2100 Alan Hayward <alan.hayward@arm.com>
2101 David Sherwood <david.sherwood@arm.com>
2103 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2104 optional tree argument.
2105 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2107 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2108 but continue to use the current value as a fallback.
2109 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2110 to compare the updates.
2111 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2112 (get_load_store_type): Use it when handling a strided access.
2113 (vect_get_strided_load_store_ops): New function.
2114 (vect_get_data_ptr_increment): Likewise.
2115 (vectorizable_load): Handle strided gather loads. Always pass
2116 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2118 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2119 Alan Hayward <alan.hayward@arm.com>
2120 David Sherwood <david.sherwood@arm.com>
2122 * doc/md.texi (gather_load@var{m}): Document.
2123 (mask_gather_load@var{m}): Likewise.
2124 * genopinit.c (main): Add supports_vec_gather_load and
2125 supports_vec_gather_load_cached to target_optabs.
2126 * optabs-tree.c (init_tree_optimization_optabs): Use
2127 ggc_cleared_alloc to allocate target_optabs.
2128 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2129 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2131 * internal-fn.h (internal_load_fn_p): Declare.
2132 (internal_gather_scatter_fn_p): Likewise.
2133 (internal_fn_mask_index): Likewise.
2134 (internal_gather_scatter_fn_supported_p): Likewise.
2135 * internal-fn.c (gather_load_direct): New macro.
2136 (expand_gather_load_optab_fn): New function.
2137 (direct_gather_load_optab_supported_p): New macro.
2138 (direct_internal_fn_optab): New function.
2139 (internal_load_fn_p): Likewise.
2140 (internal_gather_scatter_fn_p): Likewise.
2141 (internal_fn_mask_index): Likewise.
2142 (internal_gather_scatter_fn_supported_p): Likewise.
2143 * optabs-query.c (supports_at_least_one_mode_p): New function.
2144 (supports_vec_gather_load_p): Likewise.
2145 * optabs-query.h (supports_vec_gather_load_p): Declare.
2146 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2147 and memory_type field.
2148 (NUM_PATTERNS): Bump to 15.
2149 * tree-vect-data-refs.c: Include internal-fn.h.
2150 (vect_gather_scatter_fn_p): New function.
2151 (vect_describe_gather_scatter_call): Likewise.
2152 (vect_check_gather_scatter): Try using internal functions for
2153 gather loads. Recognize existing calls to a gather load function.
2154 (vect_analyze_data_refs): Consider using gather loads if
2155 supports_vec_gather_load_p.
2156 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2157 (vect_get_gather_scatter_offset_type): Likewise.
2158 (vect_convert_mask_for_vectype): Likewise.
2159 (vect_add_conversion_to_patterm): Likewise.
2160 (vect_try_gather_scatter_pattern): Likewise.
2161 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2162 (vect_vect_recog_func_ptrs): Add it.
2163 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2164 internal_fn_mask_index and internal_gather_scatter_fn_p.
2165 (check_load_store_masking): Take the gather_scatter_info as an
2166 argument and handle gather loads.
2167 (vect_get_gather_scatter_ops): New function.
2168 (vectorizable_call): Check internal_load_fn_p.
2169 (vectorizable_load): Likewise. Handle gather load internal
2171 (vectorizable_store): Update call to check_load_store_masking.
2172 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2173 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2174 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2175 (aarch64_gather_scale_operand_d): New predicates.
2176 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2177 (mask_gather_load<mode>): New insns.
2179 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2180 Alan Hayward <alan.hayward@arm.com>
2181 David Sherwood <david.sherwood@arm.com>
2183 * optabs.def (fold_left_plus_optab): New optab.
2184 * doc/md.texi (fold_left_plus_@var{m}): Document.
2185 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2186 * internal-fn.c (fold_left_direct): Define.
2187 (expand_fold_left_optab_fn): Likewise.
2188 (direct_fold_left_optab_supported_p): Likewise.
2189 * fold-const-call.c (fold_const_fold_left): New function.
2190 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2191 * tree-parloops.c (valid_reduction_p): New function.
2192 (gather_scalar_reductions): Use it.
2193 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2194 (vect_finish_replace_stmt): Declare.
2195 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2196 (needs_fold_left_reduction_p): New function, split out from...
2197 (vect_is_simple_reduction): ...here. Accept reductions that
2198 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2199 (vect_force_simple_reduction): Also store the reduction type in
2200 the assignment's STMT_VINFO_REDUC_TYPE.
2201 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2202 (merge_with_identity): New function.
2203 (vect_expand_fold_left): Likewise.
2204 (vectorize_fold_left_reduction): Likewise.
2205 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2206 scalar phi in place for it. Check for target support and reject
2207 cases that would reassociate the operation. Defer the transform
2208 phase to vectorize_fold_left_reduction.
2209 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2210 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2211 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2213 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2215 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2216 call to ifc_temp_var.
2218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2219 Alan Hayward <alan.hayward@arm.com>
2220 David Sherwood <david.sherwood@arm.com>
2222 * target.def (legitimize_address_displacement): Take the original
2223 offset as a poly_int.
2224 * targhooks.h (default_legitimize_address_displacement): Update
2226 * targhooks.c (default_legitimize_address_displacement): Likewise.
2227 * doc/tm.texi: Regenerate.
2228 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2229 as an argument, moving assert of ad->disp == ad->disp_term to...
2230 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2231 Try calling targetm.legitimize_address_displacement before expanding
2232 the address rather than afterwards, and adjust for the new interface.
2233 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2234 Match the new hook interface. Handle SVE addresses.
2235 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2238 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2240 * Makefile.in (OBJS): Add early-remat.o.
2241 * target.def (select_early_remat_modes): New hook.
2242 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2243 * doc/tm.texi: Regenerate.
2244 * targhooks.h (default_select_early_remat_modes): Declare.
2245 * targhooks.c (default_select_early_remat_modes): New function.
2246 * timevar.def (TV_EARLY_REMAT): New timevar.
2247 * passes.def (pass_early_remat): New pass.
2248 * tree-pass.h (make_pass_early_remat): Declare.
2249 * early-remat.c: New file.
2250 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2252 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2254 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2255 Alan Hayward <alan.hayward@arm.com>
2256 David Sherwood <david.sherwood@arm.com>
2258 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2259 vfm1 with a bound_epilog parameter.
2260 (vect_do_peeling): Update calls accordingly, and move the prologue
2261 call earlier in the function. Treat the base bound_epilog as 0 for
2262 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2263 this base when peeling for gaps.
2264 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2265 with fully-masked loops.
2266 (vect_estimate_min_profitable_iters): Handle the single peeled
2267 iteration in that case.
2269 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2270 Alan Hayward <alan.hayward@arm.com>
2271 David Sherwood <david.sherwood@arm.com>
2273 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2274 single-element interleaving even if the size is not a power of 2.
2275 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2276 accesses for single-element interleaving if the group size is
2279 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2280 Alan Hayward <alan.hayward@arm.com>
2281 David Sherwood <david.sherwood@arm.com>
2283 * doc/md.texi (fold_extract_last_@var{m}): Document.
2284 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2285 * optabs.def (fold_extract_last_optab): New optab.
2286 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2287 * internal-fn.c (fold_extract_direct): New macro.
2288 (expand_fold_extract_optab_fn): Likewise.
2289 (direct_fold_extract_optab_supported_p): Likewise.
2290 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2291 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2292 EXTRACT_LAST_REDUCTION.
2293 (get_initial_def_for_reduction): Do not create an initial vector
2294 for EXTRACT_LAST_REDUCTION reductions.
2295 (vectorizable_reduction): Leave the scalar phi in place for
2296 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2297 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2298 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2299 transform phase to vectorizable_condition.
2300 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2302 (vect_finish_stmt_generation): ...here.
2303 (vect_finish_replace_stmt): New function.
2304 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2305 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2307 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2309 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2310 Alan Hayward <alan.hayward@arm.com>
2311 David Sherwood <david.sherwood@arm.com>
2313 * doc/md.texi (extract_last_@var{m}): Document.
2314 * optabs.def (extract_last_optab): New optab.
2315 * internal-fn.def (EXTRACT_LAST): New internal function.
2316 * internal-fn.c (cond_unary_direct): New macro.
2317 (expand_cond_unary_optab_fn): Likewise.
2318 (direct_cond_unary_optab_supported_p): Likewise.
2319 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2320 loops using EXTRACT_LAST.
2321 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2322 (extract_last_<mode>): ...this optab.
2323 (vec_extract<mode><Vel>): Update accordingly.
2325 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2326 Alan Hayward <alan.hayward@arm.com>
2327 David Sherwood <david.sherwood@arm.com>
2329 * target.def (empty_mask_is_expensive): New hook.
2330 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2331 * doc/tm.texi: Regenerate.
2332 * targhooks.h (default_empty_mask_is_expensive): Declare.
2333 * targhooks.c (default_empty_mask_is_expensive): New function.
2334 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2335 if the target says that empty masks are expensive.
2336 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2338 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2340 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2341 Alan Hayward <alan.hayward@arm.com>
2342 David Sherwood <david.sherwood@arm.com>
2344 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2345 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2346 (vect_use_loop_mask_for_alignment_p): New function.
2347 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2348 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2349 niters_skip argument. Make sure that the first niters_skip elements
2350 of the first iteration are inactive.
2351 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2352 Update call to vect_set_loop_masks_directly.
2353 (get_misalign_in_elems): New function, split out from...
2354 (vect_gen_prolog_loop_niters): ...here.
2355 (vect_update_init_of_dr): Take a code argument that specifies whether
2356 the adjustment should be added or subtracted.
2357 (vect_update_init_of_drs): Likewise.
2358 (vect_prepare_for_masked_peels): New function.
2359 (vect_do_peeling): Skip prologue peeling if we're using a mask
2360 instead. Update call to vect_update_inits_of_drs.
2361 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2363 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2364 alignment. Do not include the number of peeled iterations in
2365 the minimum threshold in that case.
2366 (vectorizable_induction): Adjust the start value down by
2367 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2368 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2369 Take the number of skipped iterations into account when calculating
2371 * tree-vect-stmts.c (vect_gen_while_not): New function.
2373 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2374 Alan Hayward <alan.hayward@arm.com>
2375 David Sherwood <david.sherwood@arm.com>
2377 * doc/sourcebuild.texi (vect_fully_masked): Document.
2378 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2380 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2382 (vect_analyze_loop_2): ...here. Don't check the vectorization
2383 factor against the number of loop iterations if the loop is
2386 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2387 Alan Hayward <alan.hayward@arm.com>
2388 David Sherwood <david.sherwood@arm.com>
2390 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2391 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2392 (dump_groups): Update accordingly.
2393 (iv_use::mem_type): New member variable.
2394 (address_p): New function.
2395 (record_use): Add a mem_type argument and initialize the new
2397 (record_group_use): Add a mem_type argument. Use address_p.
2398 Remove obsolete null checks of base_object. Update call to record_use.
2399 (find_interesting_uses_op): Update call to record_group_use.
2400 (find_interesting_uses_cond): Likewise.
2401 (find_interesting_uses_address): Likewise.
2402 (get_mem_type_for_internal_fn): New function.
2403 (find_address_like_use): Likewise.
2404 (find_interesting_uses_stmt): Try find_address_like_use before
2405 calling find_interesting_uses_op.
2406 (addr_offset_valid_p): Use the iv mem_type field as the type
2407 of the addressed memory.
2408 (add_autoinc_candidates): Likewise.
2409 (get_address_cost): Likewise.
2410 (split_small_address_groups_p): Use address_p.
2411 (split_address_groups): Likewise.
2412 (add_iv_candidate_for_use): Likewise.
2413 (autoinc_possible_for_pair): Likewise.
2414 (rewrite_groups): Likewise.
2415 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2416 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2417 (get_alias_ptr_type_for_ptr_address): New function.
2418 (rewrite_use_address): Rewrite address uses in calls that were
2419 identified by find_address_like_use.
2421 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2422 Alan Hayward <alan.hayward@arm.com>
2423 David Sherwood <david.sherwood@arm.com>
2425 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2427 * gimple-expr.h (is_gimple_addressable: Likewise.
2428 * gimple-expr.c (is_gimple_address): Likewise.
2429 * internal-fn.c (expand_call_mem_ref): New function.
2430 (expand_mask_load_optab_fn): Use it.
2431 (expand_mask_store_optab_fn): Likewise.
2433 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2434 Alan Hayward <alan.hayward@arm.com>
2435 David Sherwood <david.sherwood@arm.com>
2437 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2438 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2439 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2440 (cond_umax@var{mode}): Document.
2441 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2442 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2443 (cond_umin_optab, cond_umax_optab): New optabs.
2444 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2445 (COND_IOR, COND_XOR): New internal functions.
2446 * internal-fn.h (get_conditional_internal_fn): Declare.
2447 * internal-fn.c (cond_binary_direct): New macro.
2448 (expand_cond_binary_optab_fn): Likewise.
2449 (direct_cond_binary_optab_supported_p): Likewise.
2450 (get_conditional_internal_fn): New function.
2451 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2452 Cope with reduction statements that are vectorized as calls rather
2454 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2455 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2456 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2457 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2458 (UNSPEC_COND_EOR): New unspecs.
2459 (optab): Add mappings for them.
2460 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2461 (sve_int_op, sve_fp_op): New int attributes.
2463 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2464 Alan Hayward <alan.hayward@arm.com>
2465 David Sherwood <david.sherwood@arm.com>
2467 * optabs.def (while_ult_optab): New optab.
2468 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2469 * internal-fn.def (WHILE_ULT): New internal function.
2470 * internal-fn.h (direct_internal_fn_supported_p): New override
2471 that takes two types as argument.
2472 * internal-fn.c (while_direct): New macro.
2473 (expand_while_optab_fn): New function.
2474 (convert_optab_supported_p): Likewise.
2475 (direct_while_optab_supported_p): New macro.
2476 * wide-int.h (wi::udiv_ceil): New function.
2477 * tree-vectorizer.h (rgroup_masks): New structure.
2478 (vec_loop_masks): New typedef.
2479 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2481 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2482 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2483 (vect_max_vf): New function.
2484 (slpeel_make_loop_iterate_ntimes): Delete.
2485 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2486 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2487 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2488 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2489 internal-fn.h, stor-layout.h and optabs-query.h.
2490 (vect_set_loop_mask): New function.
2491 (add_preheader_seq): Likewise.
2492 (add_header_seq): Likewise.
2493 (interleave_supported_p): Likewise.
2494 (vect_maybe_permute_loop_masks): Likewise.
2495 (vect_set_loop_masks_directly): Likewise.
2496 (vect_set_loop_condition_masked): Likewise.
2497 (vect_set_loop_condition_unmasked): New function, split out from
2498 slpeel_make_loop_iterate_ntimes.
2499 (slpeel_make_loop_iterate_ntimes): Rename to..
2500 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2501 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2502 (vect_do_peeling): Update call accordingly.
2503 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2505 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2506 mask_compare_type, can_fully_mask_p and fully_masked_p.
2507 (release_vec_loop_masks): New function.
2508 (_loop_vec_info): Use it to free the loop masks.
2509 (can_produce_all_loop_masks_p): New function.
2510 (vect_get_max_nscalars_per_iter): Likewise.
2511 (vect_verify_full_masking): Likewise.
2512 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2513 retries, and free the mask rgroups before retrying. Check loop-wide
2514 reasons for disallowing fully-masked loops. Make the final decision
2515 about whether use a fully-masked loop or not.
2516 (vect_estimate_min_profitable_iters): Do not assume that peeling
2517 for the number of iterations will be needed for fully-masked loops.
2518 (vectorizable_reduction): Disable fully-masked loops.
2519 (vectorizable_live_operation): Likewise.
2520 (vect_halve_mask_nunits): New function.
2521 (vect_double_mask_nunits): Likewise.
2522 (vect_record_loop_mask): Likewise.
2523 (vect_get_loop_mask): Likewise.
2524 (vect_transform_loop): Handle the case in which the final loop
2525 iteration might handle a partial vector. Call vect_set_loop_condition
2526 instead of slpeel_make_loop_iterate_ntimes.
2527 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2528 (check_load_store_masking): New function.
2529 (prepare_load_store_mask): Likewise.
2530 (vectorizable_store): Handle fully-masked loops.
2531 (vectorizable_load): Likewise.
2532 (supportable_widening_operation): Use vect_halve_mask_nunits for
2534 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2535 (vect_gen_while): New function.
2536 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2537 (aarch64_uqdec<mode>): New insn.
2539 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2540 Alan Hayward <alan.hayward@arm.com>
2541 David Sherwood <david.sherwood@arm.com>
2543 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2544 (reduc_xor_scal_optab): New optabs.
2545 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2546 (reduc_xor_scal_@var{m}): Document.
2547 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2548 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2550 * fold-const-call.c (fold_const_call): Handle them.
2551 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2552 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2553 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2554 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2555 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2556 (UNSPEC_XORV): New unspecs.
2557 (optab): Add entries for them.
2558 (BITWISEV): New int iterator.
2559 (bit_reduc_op): New int attributes.
2561 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2562 Alan Hayward <alan.hayward@arm.com>
2563 David Sherwood <david.sherwood@arm.com>
2565 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2566 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2567 * optabs.def (vec_shl_insert_optab): New optab.
2568 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2569 (duplicate_and_interleave): Likewise.
2570 * tree-vect-loop.c: Include internal-fn.h.
2571 (neutral_op_for_slp_reduction): New function, split out from
2572 get_initial_defs_for_reduction.
2573 (get_initial_def_for_reduction): Handle option 2 for variable-length
2574 vectors by loading the neutral value into a vector and then shifting
2575 the initial value into element 0.
2576 (get_initial_defs_for_reduction): Replace the code argument with
2577 the neutral value calculated by neutral_op_for_slp_reduction.
2578 Use gimple_build_vector for constant-length vectors.
2579 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2580 but the first group_size elements have a neutral value.
2581 Use duplicate_and_interleave otherwise.
2582 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2583 Update call to get_initial_defs_for_reduction. Handle SLP
2584 reductions for variable-length vectors by creating one vector
2585 result for each scalar result, with the elements associated
2586 with other scalar results stubbed out with the neutral value.
2587 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2588 Require IFN_VEC_SHL_INSERT for double reductions on
2589 variable-length vectors, or SLP reductions that have
2590 a neutral value. Require can_duplicate_and_interleave_p
2591 support for variable-length unchained SLP reductions if there
2592 is no neutral value, such as for MIN/MAX reductions. Also require
2593 the number of vector elements to be a multiple of the number of
2594 SLP statements when doing variable-length unchained SLP reductions.
2595 Update call to vect_create_epilog_for_reduction.
2596 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2597 and remove initial values.
2598 (duplicate_and_interleave): Make public.
2599 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2600 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2602 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2603 Alan Hayward <alan.hayward@arm.com>
2604 David Sherwood <david.sherwood@arm.com>
2606 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2607 (can_duplicate_and_interleave_p): New function.
2608 (vect_get_and_check_slp_defs): Take the vector of statements
2609 rather than just the current one. Remove excess parentheses.
2610 Restriction rejectinon of vect_constant_def and vect_external_def
2611 for variable-length vectors to boolean types, or types for which
2612 can_duplicate_and_interleave_p is false.
2613 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2614 (duplicate_and_interleave): New function.
2615 (vect_get_constant_vectors): Use gimple_build_vector for
2616 constant-length vectors and suitable variable-length constant
2617 vectors. Use duplicate_and_interleave for other variable-length
2618 vectors. Don't defer the update when inserting new statements.
2620 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2621 Alan Hayward <alan.hayward@arm.com>
2622 David Sherwood <david.sherwood@arm.com>
2624 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2625 min_profitable_iters doesn't go negative.
2627 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2628 Alan Hayward <alan.hayward@arm.com>
2629 David Sherwood <david.sherwood@arm.com>
2631 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2632 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2633 * optabs.def (vec_mask_load_lanes_optab): New optab.
2634 (vec_mask_store_lanes_optab): Likewise.
2635 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2636 (MASK_STORE_LANES): Likewise.
2637 * internal-fn.c (mask_load_lanes_direct): New macro.
2638 (mask_store_lanes_direct): Likewise.
2639 (expand_mask_load_optab_fn): Handle masked operations.
2640 (expand_mask_load_lanes_optab_fn): New macro.
2641 (expand_mask_store_optab_fn): Handle masked operations.
2642 (expand_mask_store_lanes_optab_fn): New macro.
2643 (direct_mask_load_lanes_optab_supported_p): Likewise.
2644 (direct_mask_store_lanes_optab_supported_p): Likewise.
2645 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2647 (vect_load_lanes_supported): Likewise.
2648 * tree-vect-data-refs.c (strip_conversion): New function.
2649 (can_group_stmts_p): Likewise.
2650 (vect_analyze_data_ref_accesses): Use it instead of checking
2651 for a pair of assignments.
2652 (vect_store_lanes_supported): Take a masked_p parameter.
2653 (vect_load_lanes_supported): Likewise.
2654 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2655 vect_store_lanes_supported and vect_load_lanes_supported.
2656 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2657 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2658 parameter. Don't allow gaps for masked accesses.
2659 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2660 and vect_load_lanes_supported.
2661 (get_load_store_type): Take a masked_p parameter and update
2662 call to get_group_load_store_type.
2663 (vectorizable_store): Update call to get_load_store_type.
2664 Handle IFN_MASK_STORE_LANES.
2665 (vectorizable_load): Update call to get_load_store_type.
2666 Handle IFN_MASK_LOAD_LANES.
2668 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2669 Alan Hayward <alan.hayward@arm.com>
2670 David Sherwood <david.sherwood@arm.com>
2672 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2674 * config/aarch64/aarch64-protos.h
2675 (aarch64_sve_struct_memory_operand_p): Declare.
2676 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2677 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2678 (VPRED, vpred): Handle SVE structure modes.
2679 * config/aarch64/constraints.md (Utx): New constraint.
2680 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2681 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2682 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2683 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2684 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2685 structure modes. Split into pieces after RA.
2686 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2687 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2689 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2690 SVE structure modes.
2691 (aarch64_classify_address): Likewise.
2692 (sizetochar): Move earlier in file.
2693 (aarch64_print_operand): Handle SVE register lists.
2694 (aarch64_array_mode): New function.
2695 (aarch64_sve_struct_memory_operand_p): Likewise.
2696 (TARGET_ARRAY_MODE): Redefine.
2698 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2699 Alan Hayward <alan.hayward@arm.com>
2700 David Sherwood <david.sherwood@arm.com>
2702 * target.def (array_mode): New target hook.
2703 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2704 * doc/tm.texi: Regenerate.
2705 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2706 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2707 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2709 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2713 Alan Hayward <alan.hayward@arm.com>
2714 David Sherwood <david.sherwood@arm.com>
2716 * fold-const.c (fold_binary_loc): Check the argument types
2717 rather than the result type when testing for a vector operation.
2719 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2721 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2722 * doc/tm.texi: Regenerate.
2724 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2725 Alan Hayward <alan.hayward@arm.com>
2726 David Sherwood <david.sherwood@arm.com>
2728 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2729 (sve): Document new AArch64 extension.
2730 * doc/md.texi (w): Extend the description of the AArch64
2731 constraint to include SVE vectors.
2732 (Upl, Upa): Document new AArch64 predicate constraints.
2733 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2735 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2736 (msve-vector-bits=): New option.
2737 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2738 SVE when these are disabled.
2739 (sve): New extension.
2740 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2741 modes. Adjust their number of units based on aarch64_sve_vg.
2742 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2743 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2744 aarch64_addr_query_type.
2745 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2746 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2747 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2748 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2749 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2750 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2751 (aarch64_simd_imm_zero_p): Delete.
2752 (aarch64_check_zero_based_sve_index_immediate): Declare.
2753 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2754 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2755 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2756 (aarch64_sve_float_mul_immediate_p): Likewise.
2757 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2759 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2760 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2761 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2762 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2763 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2764 (aarch64_regmode_natural_size): Likewise.
2765 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2766 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2768 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2769 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2770 for VG and the SVE predicate registers.
2771 (V_ALIASES): Add a "z"-prefixed alias.
2772 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2773 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2774 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2775 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2776 (REG_CLASS_NAMES): Add entries for them.
2777 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2778 and the predicate registers.
2779 (aarch64_sve_vg): Declare.
2780 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2781 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2782 (REGMODE_NATURAL_SIZE): Define.
2783 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2785 * config/aarch64/aarch64.c: Include cfgrtl.h.
2786 (simd_immediate_info): Add a constructor for series vectors,
2787 and an associated step field.
2788 (aarch64_sve_vg): New variable.
2789 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2790 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2791 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2792 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2793 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2794 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2795 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2796 (aarch64_get_mask_mode): New functions.
2797 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2798 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2799 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2800 predicate modes and predicate registers. Explicitly restrict
2801 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2802 to store a vector mode if it is recognized by
2803 aarch64_classify_vector_mode.
2804 (aarch64_regmode_natural_size): New function.
2805 (aarch64_hard_regno_caller_save_mode): Return the original mode
2807 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2808 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2809 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2810 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2812 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2813 does not overlap dest if the function is frame-related. Handle
2815 (aarch64_split_add_offset): New function.
2816 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2817 them aarch64_add_offset.
2818 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2819 and update call to aarch64_sub_sp.
2820 (aarch64_add_cfa_expression): New function.
2821 (aarch64_expand_prologue): Pass extra temporary registers to the
2822 functions above. Handle the case in which we need to emit new
2823 DW_CFA_expressions for registers that were originally saved
2824 relative to the stack pointer, but now have to be expressed
2825 relative to the frame pointer.
2826 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2828 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2829 IP0 and IP1 values for SVE frames.
2830 (aarch64_expand_vec_series): New function.
2831 (aarch64_expand_sve_widened_duplicate): Likewise.
2832 (aarch64_expand_sve_const_vector): Likewise.
2833 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2834 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2835 into the register, rather than emitting a SET directly.
2836 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2837 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2838 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2839 (offset_9bit_signed_scaled_p): New functions.
2840 (aarch64_replicate_bitmask_imm): New function.
2841 (aarch64_bitmask_imm): Use it.
2842 (aarch64_cannot_force_const_mem): Reject expressions involving
2843 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2844 (aarch64_classify_index): Handle SVE indices, by requiring
2845 a plain register index with a scale that matches the element size.
2846 (aarch64_classify_address): Handle SVE addresses. Assert that
2847 the mode of the address is VOIDmode or an integer mode.
2848 Update call to aarch64_classify_symbol.
2849 (aarch64_classify_symbolic_expression): Update call to
2850 aarch64_classify_symbol.
2851 (aarch64_const_vec_all_in_range_p): New function.
2852 (aarch64_print_vector_float_operand): Likewise.
2853 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2854 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2855 and the FP immediates 1.0 and 0.5.
2856 (aarch64_print_address_internal): Handle SVE addresses.
2857 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2858 (aarch64_regno_regclass): Handle predicate registers.
2859 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2861 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2862 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2863 (aarch64_convert_sve_vector_bits): New function.
2864 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2865 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2867 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2868 Handle SVE vector and predicate modes. Accept VL-based constants
2869 that need only one temporary register, and VL offsets that require
2870 no temporary registers.
2871 (aarch64_conditional_register_usage): Mark the predicate registers
2872 as fixed if SVE isn't available.
2873 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2874 Return true for SVE vector and predicate modes.
2875 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2876 rather than an unsigned int. Handle SVE modes.
2877 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2879 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2881 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2882 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2883 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2884 (aarch64_sve_float_mul_immediate_p): New functions.
2885 (aarch64_sve_valid_immediate): New function.
2886 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2887 Explicitly reject structure modes. Check for INDEX constants.
2888 Handle PTRUE and PFALSE constants.
2889 (aarch64_check_zero_based_sve_index_immediate): New function.
2890 (aarch64_simd_imm_zero_p): Delete.
2891 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2892 vector modes. Accept constants in the range of CNT[BHWD].
2893 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2894 ask for an Advanced SIMD mode.
2895 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2896 (aarch64_simd_vector_alignment): Handle SVE predicates.
2897 (aarch64_vectorize_preferred_vector_alignment): New function.
2898 (aarch64_simd_vector_alignment_reachable): Use it instead of
2900 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2901 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2903 (MAX_VECT_LEN): Delete.
2904 (expand_vec_perm_d): Add a vec_flags field.
2905 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2906 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2907 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2909 (aarch64_evpc_rev): Rename to...
2910 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2911 (aarch64_evpc_rev_global): New function.
2912 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2913 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2915 (aarch64_evpc_sve_tbl): New function.
2916 (aarch64_expand_vec_perm_const_1): Update after rename of
2917 aarch64_evpc_rev. Handle SVE permutes too, trying
2918 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2919 than aarch64_evpc_tbl.
2920 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2921 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2922 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2923 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2924 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2925 (aarch64_expand_sve_vcond): New functions.
2926 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2927 of aarch64_vector_mode_p.
2928 (aarch64_dwarf_poly_indeterminate_value): New function.
2929 (aarch64_compute_pressure_classes): Likewise.
2930 (aarch64_can_change_mode_class): Likewise.
2931 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2932 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2933 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2934 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2935 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2936 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2937 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2938 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2940 (Dn, Dl, Dr): Accept const as well as const_vector.
2941 (Dz): Likewise. Compare against CONST0_RTX.
2942 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2943 of "vector" where appropriate.
2944 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2945 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2946 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2947 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2948 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2949 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2950 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2951 (v_int_equiv): Extend to SVE modes.
2952 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2954 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2955 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2956 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2957 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2958 (SVE_COND_FP_CMP): New int iterators.
2959 (perm_hilo): Handle the new unpack unspecs.
2960 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2962 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2963 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2964 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2965 (aarch64_equality_operator, aarch64_constant_vector_operand)
2966 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2967 (aarch64_sve_nonimmediate_operand): Likewise.
2968 (aarch64_sve_general_operand): Likewise.
2969 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2970 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2971 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2972 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2973 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2974 (aarch64_sve_float_arith_immediate): Likewise.
2975 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2976 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2977 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2978 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2979 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2980 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2981 (aarch64_sve_float_arith_operand): Likewise.
2982 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2983 (aarch64_sve_float_mul_operand): Likewise.
2984 (aarch64_sve_vec_perm_operand): Likewise.
2985 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2986 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2987 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2988 as well as const_vector.
2989 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2990 in file. Use CONST0_RTX and CONSTM1_RTX.
2991 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2992 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2993 Use aarch64_simd_imm_zero.
2994 * config/aarch64/aarch64-sve.md: New file.
2995 * config/aarch64/aarch64.md: Include it.
2996 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2997 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2998 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2999 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3000 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3001 (sve): New attribute.
3002 (enabled): Disable instructions with the sve attribute unless
3004 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3005 aarch64_expand_mov_immediate.
3006 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3007 CNT[BHSD] immediates.
3008 (movti): Split CONST_POLY_INT moves into two halves.
3009 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3010 Split additions that need a temporary here if the destination
3011 is the stack pointer.
3012 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3013 (*add<mode>3_poly_1): New instruction.
3014 (set_clobber_cc): New expander.
3016 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3018 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3019 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3020 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3021 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3022 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3023 Change innermode from fixed_mode_size to machine_mode.
3024 (simplify_subreg): Update call accordingly. Handle a constant-sized
3025 subreg of a variable-length CONST_VECTOR.
3027 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3028 Alan Hayward <alan.hayward@arm.com>
3029 David Sherwood <david.sherwood@arm.com>
3031 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3032 (add_offset_to_base): New function, split out from...
3033 (create_mem_ref): ...here. When handling a scale other than 1,
3034 check first whether the address is valid without the offset.
3035 Add it into the base if so, leaving the index and scale as-is.
3037 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3040 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3041 fold_for_warn before checking if arg2 is INTEGER_CST.
3043 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3045 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3046 (store_multiple_operation): Delete.
3047 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3048 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3049 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3050 guarded by TARGET_STRING.
3051 (rs6000_output_load_multiple): Delete.
3052 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3053 OPTION_MASK_STRING / TARGET_STRING handling.
3054 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3055 (const rs6000_opt_masks) <"string">: Change mask to 0.
3056 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3057 (MASK_STRING): Delete.
3058 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3060 (load_multiple): Delete.
3067 (store_multiple): Delete.
3074 (movmemsi_8reg): Delete.
3075 (corresponding unnamed define_insn): Delete.
3076 (movmemsi_6reg): Delete.
3077 (corresponding unnamed define_insn): Delete.
3078 (movmemsi_4reg): Delete.
3079 (corresponding unnamed define_insn): Delete.
3080 (movmemsi_2reg): Delete.
3081 (corresponding unnamed define_insn): Delete.
3082 (movmemsi_1reg): Delete.
3083 (corresponding unnamed define_insn): Delete.
3084 * config/rs6000/rs6000.opt (mno-string): New.
3085 (mstring): Replace by deprecation warning stub.
3086 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3088 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3090 * regrename.c (regrename_do_replace): If replacing the same
3091 reg multiple times, try to reuse last created gen_raw_REG.
3094 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3095 main to workaround a bug in GDB.
3097 2018-01-12 Tom de Vries <tom@codesourcery.com>
3100 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3102 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3104 PR rtl-optimization/80481
3105 * ira-color.c (get_cap_member): New function.
3106 (allocnos_conflict_by_live_ranges_p): Use it.
3107 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3108 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3110 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3113 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3114 (*saddl_se_1): Ditto.
3116 (*ssubl_se_1): Ditto.
3118 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3120 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3121 rather than wi::to_widest for DR_INITs.
3122 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3123 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3124 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3126 (vect_analyze_group_access_1): Note that here.
3128 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3130 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3131 polynomial type sizes.
3133 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3135 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3136 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3137 (gimple_add_tmp_var): Likewise.
3139 2018-01-12 Martin Liska <mliska@suse.cz>
3141 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3142 (gimple_alloc_sizes): Likewise.
3143 (dump_gimple_statistics): Use PRIu64 in printf format.
3144 * gimple.h: Change uint64_t to int.
3146 2018-01-12 Martin Liska <mliska@suse.cz>
3148 * tree-core.h: Use uint64_t instead of int.
3149 * tree.c (tree_node_counts): Likewise.
3150 (tree_node_sizes): Likewise.
3151 (dump_tree_statistics): Use PRIu64 in printf format.
3153 2018-01-12 Martin Liska <mliska@suse.cz>
3155 * Makefile.in: As qsort_chk is implemented in vec.c, add
3156 vec.o to linkage of gencfn-macros.
3157 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3158 passing the info to record_node_allocation_statistics.
3159 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3161 * ggc-common.c (struct ggc_usage): Add operator== and use
3162 it in operator< and compare function.
3163 * mem-stats.h (struct mem_usage): Likewise.
3164 * vec.c (struct vec_usage): Remove operator< and compare
3165 function. Can be simply inherited.
3167 2018-01-12 Martin Jambor <mjambor@suse.cz>
3170 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3171 * tree-ssa-math-opts.c: Include domwalk.h.
3172 (convert_mult_to_fma_1): New function.
3173 (fma_transformation_info): New type.
3174 (fma_deferring_state): Likewise.
3175 (cancel_fma_deferring): New function.
3176 (result_of_phi): Likewise.
3177 (last_fma_candidate_feeds_initial_phi): Likewise.
3178 (convert_mult_to_fma): Added deferring logic, split actual
3179 transformation to convert_mult_to_fma_1.
3180 (math_opts_dom_walker): New type.
3181 (math_opts_dom_walker::after_dom_children): New method, body moved
3182 here from pass_optimize_widening_mul::execute, added deferring logic
3184 (pass_optimize_widening_mul::execute): Moved most of code to
3185 math_opts_dom_walker::after_dom_children.
3186 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3187 * config/i386/i386.c (ix86_option_override_internal): Added
3188 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3190 2018-01-12 Richard Biener <rguenther@suse.de>
3193 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3194 inline instance vars.
3196 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3199 * config/rx/rx.c (rx_is_restricted_memory_address):
3202 2018-01-12 Richard Biener <rguenther@suse.de>
3204 PR tree-optimization/80846
3205 * target.def (split_reduction): New target hook.
3206 * targhooks.c (default_split_reduction): New function.
3207 * targhooks.h (default_split_reduction): Declare.
3208 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3209 target requests first reduce vectors by combining low and high
3211 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3212 (get_vectype_for_scalar_type_and_size): Export.
3213 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3214 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3215 * doc/tm.texi: Regenerate.
3216 * config/i386/i386.c (ix86_split_reduction): Implement
3217 TARGET_VECTORIZE_SPLIT_REDUCTION.
3219 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3222 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3223 in PIC mode except for TARGET_VXWORKS_RTP.
3224 * config/sparc/sparc.c: Include cfgrtl.h.
3225 (TARGET_INIT_PIC_REG): Define.
3226 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3227 (sparc_pic_register_p): New predicate.
3228 (sparc_legitimate_address_p): Use it.
3229 (sparc_legitimize_pic_address): Likewise.
3230 (sparc_delegitimize_address): Likewise.
3231 (sparc_mode_dependent_address_p): Likewise.
3232 (gen_load_pcrel_sym): Remove 4th parameter.
3233 (load_got_register): Adjust call to above. Remove obsolete stuff.
3234 (sparc_expand_prologue): Do not call load_got_register here.
3235 (sparc_flat_expand_prologue): Likewise.
3236 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3237 (sparc_use_pseudo_pic_reg): New function.
3238 (sparc_init_pic_reg): Likewise.
3239 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3240 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3242 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3244 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3245 Add item for branch_cost.
3247 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3249 PR rtl-optimization/83565
3250 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3251 not extend the result to a larger mode for rotate operations.
3252 (num_sign_bit_copies1): Likewise.
3254 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3257 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3259 Use values-Xc.o for -pedantic.
3260 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3262 2018-01-12 Martin Liska <mliska@suse.cz>
3265 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3267 (possible_polymorphic_call_targets): Use it.
3268 (ipa_devirt): Likewise.
3270 2018-01-12 Martin Liska <mliska@suse.cz>
3272 * profile-count.h (enum profile_quality): Use 0 as invalid
3273 enum value of profile_quality.
3275 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3277 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3278 -mext-string options.
3280 2018-01-12 Richard Biener <rguenther@suse.de>
3282 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3283 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3284 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3286 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3288 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3290 * configure.ac (--with-long-double-format): Add support for the
3291 configuration option to change the default long double format on
3293 * config.gcc (powerpc*-linux*-*): Likewise.
3294 * configure: Regenerate.
3295 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3296 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3297 used without modification.
3299 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3301 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3302 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3303 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3304 MISC_BUILTIN_SPEC_BARRIER.
3305 (rs6000_init_builtins): Likewise.
3306 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3308 (speculation_barrier): New define_insn.
3309 * doc/extend.texi: Document __builtin_speculation_barrier.
3311 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3314 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3315 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3316 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3318 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3319 integral modes instead of "ss" and "sd".
3320 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3321 vectors with 32-bit and 64-bit elements.
3322 (vecdupssescalarmodesuffix): New mode attribute.
3323 (vec_dup<mode>): Use it.
3325 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3328 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3329 frame if argument is passed on stack.
3331 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3334 * ree.c (combine_reaching_defs): Optimize also
3335 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3336 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3338 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3341 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3343 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3346 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3347 after they are computed.
3349 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3351 PR tree-optimization/83695
3352 * gimple-loop-linterchange.cc
3353 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3354 reset cached scev information after interchange.
3355 (pass_linterchange::execute): Remove call to scev_reset_htab.
3357 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3359 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3360 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3361 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3362 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3363 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3364 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3365 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3366 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3367 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3368 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3369 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3370 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3371 (V_lane_reg): Likewise.
3372 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3374 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3375 (vfmal_lane_low<mode>_intrinsic,
3376 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3377 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3378 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3379 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3380 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3381 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3383 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3385 * config/arm/arm-cpus.in (fp16fml): New feature.
3386 (ALL_SIMD): Add fp16fml.
3387 (armv8.2-a): Add fp16fml as an option.
3388 (armv8.3-a): Likewise.
3389 (armv8.4-a): Add fp16fml as part of fp16.
3390 * config/arm/arm.h (TARGET_FP16FML): Define.
3391 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3393 * config/arm/arm-modes.def (V2HF): Define.
3394 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3395 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3396 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3397 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3398 vfmsl_low, vfmsl_high): New set of builtins.
3399 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3400 (vfml_op): New code attribute.
3401 (VFMLHALVES): New int iterator.
3402 (VFML, VFMLSEL): New mode attributes.
3403 (V_reg): Define mapping for V2HF.
3404 (V_hi, V_lo): New mode attributes.
3405 (VF_constraint): Likewise.
3406 (vfml_half, vfml_half_selector): New int attributes.
3407 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3409 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3410 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3412 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3413 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3414 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3415 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3417 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3418 Document new effective target and option set.
3420 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3422 * config/arm/arm-cpus.in (armv8_4): New feature.
3423 (ARMv8_4a): New fgroup.
3424 (armv8.4-a): New arch.
3425 * config/arm/arm-tables.opt: Regenerate.
3426 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3427 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3428 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3429 Add matching rules for -march=armv8.4-a and extensions.
3430 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3432 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3435 * config/rx/rx.md (BW): New mode attribute.
3436 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3438 2018-01-11 Richard Biener <rguenther@suse.de>
3440 PR tree-optimization/83435
3441 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3442 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3443 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3445 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3446 Alan Hayward <alan.hayward@arm.com>
3447 David Sherwood <david.sherwood@arm.com>
3449 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3451 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3452 (aarch64_print_address_internal): Use it to check for a zero offset.
3454 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3455 Alan Hayward <alan.hayward@arm.com>
3456 David Sherwood <david.sherwood@arm.com>
3458 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3459 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3460 Return a poly_int64 rather than a HOST_WIDE_INT.
3461 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3462 rather than a HOST_WIDE_INT.
3463 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3464 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3465 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3466 final_offset from HOST_WIDE_INT to poly_int64.
3467 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3468 to_constant when getting the number of units in an Advanced SIMD
3470 (aarch64_builtin_vectorized_function): Check for a constant number
3472 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3474 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3475 attribute instead of GET_MODE_NUNITS.
3476 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3477 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3478 GET_MODE_SIZE for fixed-size registers.
3479 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3480 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3481 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3482 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3483 (aarch64_print_operand, aarch64_print_address_internal)
3484 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3485 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3486 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3487 Handle polynomial GET_MODE_SIZE.
3488 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3489 wider than SImode without modification.
3490 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3491 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3492 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3493 passing and returning SVE modes.
3494 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3495 rather than GEN_INT.
3496 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3497 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3498 (aarch64_allocate_and_probe_stack_space): Likewise.
3499 (aarch64_layout_frame): Cope with polynomial offsets.
3500 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3501 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3503 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3504 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3505 poly_int64 rather than a HOST_WIDE_INT.
3506 (aarch64_get_separate_components, aarch64_process_components)
3507 (aarch64_expand_prologue, aarch64_expand_epilogue)
3508 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3509 (aarch64_anchor_offset): New function, split out from...
3510 (aarch64_legitimize_address): ...here.
3511 (aarch64_builtin_vectorization_cost): Handle polynomial
3512 TYPE_VECTOR_SUBPARTS.
3513 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3515 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3516 number of elements from the PARALLEL rather than the mode.
3517 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3518 rather than GET_MODE_BITSIZE.
3519 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3520 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3521 (aarch64_expand_vec_perm_const_1): Handle polynomial
3522 d->perm.length () and d->perm elements.
3523 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3524 Apply to_constant to d->perm elements.
3525 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3526 polynomial CONST_VECTOR_NUNITS.
3527 (aarch64_move_pointer): Take amount as a poly_int64 rather
3529 (aarch64_progress_pointer): Avoid temporary variable.
3530 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3531 the mode attribute instead of GET_MODE.
3533 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3534 Alan Hayward <alan.hayward@arm.com>
3535 David Sherwood <david.sherwood@arm.com>
3537 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3538 x exists before using it.
3539 (aarch64_add_constant_internal): Rename to...
3540 (aarch64_add_offset_1): ...this. Replace regnum with separate
3541 src and dest rtxes. Handle the case in which they're different,
3542 including when the offset is zero. Replace scratchreg with an rtx.
3543 Use 2 additions if there is no spare register into which we can
3544 move a 16-bit constant.
3545 (aarch64_add_constant): Delete.
3546 (aarch64_add_offset): Replace reg with separate src and dest
3547 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3548 Use aarch64_add_offset_1.
3549 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3550 an rtx rather than an int. Take the delta as a poly_int64
3551 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3552 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3553 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3554 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3555 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3557 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3558 aarch64_add_constant.
3560 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3562 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3563 Use scalar_float_mode.
3565 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3567 * config/aarch64/aarch64-simd.md
3568 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3569 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3570 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3571 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3572 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3573 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3574 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3575 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3576 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3577 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3579 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3582 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3583 targ_options->x_arm_arch_string is non NULL.
3585 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3587 * config/aarch64/aarch64.h
3588 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3590 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3593 * expmed.c (emit_store_flag_force): Swap if const op0
3594 and change VOIDmode to mode of op0.
3596 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3598 PR rtl-optimization/83761
3599 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3600 than bytes to mode_for_size.
3602 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3605 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3606 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3609 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3612 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3613 when in layout mode.
3614 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3615 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3618 2018-01-10 Michael Collison <michael.collison@arm.com>
3620 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3621 * config/aarch64/aarch64-option-extension.def: Add
3622 AARCH64_OPT_EXTENSION of 'fp16fml'.
3623 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3624 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3625 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3626 * config/aarch64/constraints.md (Ui7): New constraint.
3627 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3628 (VFMLA_SEL_W): Ditto.
3631 (VFMLA16_LOW): New int iterator.
3632 (VFMLA16_HIGH): Ditto.
3633 (UNSPEC_FMLAL): New unspec.
3634 (UNSPEC_FMLSL): Ditto.
3635 (UNSPEC_FMLAL2): Ditto.
3636 (UNSPEC_FMLSL2): Ditto.
3637 (f16mac): New code attribute.
3638 * config/aarch64/aarch64-simd-builtins.def
3639 (aarch64_fmlal_lowv2sf): Ditto.
3640 (aarch64_fmlsl_lowv2sf): Ditto.
3641 (aarch64_fmlalq_lowv4sf): Ditto.
3642 (aarch64_fmlslq_lowv4sf): Ditto.
3643 (aarch64_fmlal_highv2sf): Ditto.
3644 (aarch64_fmlsl_highv2sf): Ditto.
3645 (aarch64_fmlalq_highv4sf): Ditto.
3646 (aarch64_fmlslq_highv4sf): Ditto.
3647 (aarch64_fmlal_lane_lowv2sf): Ditto.
3648 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3649 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3650 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3651 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3652 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3653 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3654 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3655 (aarch64_fmlal_lane_highv2sf): Ditto.
3656 (aarch64_fmlsl_lane_highv2sf): Ditto.
3657 (aarch64_fmlal_laneq_highv2sf): Ditto.
3658 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3659 (aarch64_fmlalq_lane_highv4sf): Ditto.
3660 (aarch64_fmlsl_lane_highv4sf): Ditto.
3661 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3662 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3663 * config/aarch64/aarch64-simd.md:
3664 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3665 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3666 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3667 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3668 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3669 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3670 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3671 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3672 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3673 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3674 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3675 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3676 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3677 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3678 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3679 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3680 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3681 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3682 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3683 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3684 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3685 (vfmlsl_low_u32): Ditto.
3686 (vfmlalq_low_u32): Ditto.
3687 (vfmlslq_low_u32): Ditto.
3688 (vfmlal_high_u32): Ditto.
3689 (vfmlsl_high_u32): Ditto.
3690 (vfmlalq_high_u32): Ditto.
3691 (vfmlslq_high_u32): Ditto.
3692 (vfmlal_lane_low_u32): Ditto.
3693 (vfmlsl_lane_low_u32): Ditto.
3694 (vfmlal_laneq_low_u32): Ditto.
3695 (vfmlsl_laneq_low_u32): Ditto.
3696 (vfmlalq_lane_low_u32): Ditto.
3697 (vfmlslq_lane_low_u32): Ditto.
3698 (vfmlalq_laneq_low_u32): Ditto.
3699 (vfmlslq_laneq_low_u32): Ditto.
3700 (vfmlal_lane_high_u32): Ditto.
3701 (vfmlsl_lane_high_u32): Ditto.
3702 (vfmlal_laneq_high_u32): Ditto.
3703 (vfmlsl_laneq_high_u32): Ditto.
3704 (vfmlalq_lane_high_u32): Ditto.
3705 (vfmlslq_lane_high_u32): Ditto.
3706 (vfmlalq_laneq_high_u32): Ditto.
3707 (vfmlslq_laneq_high_u32): Ditto.
3708 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3709 (AARCH64_FL_FOR_ARCH8_4): New.
3710 (AARCH64_ISA_F16FML): New ISA flag.
3711 (TARGET_F16FML): New feature flag for fp16fml.
3712 (doc/invoke.texi): Document new fp16fml option.
3714 2018-01-10 Michael Collison <michael.collison@arm.com>
3716 * config/aarch64/aarch64-builtins.c:
3717 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3718 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3719 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3720 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3721 (AARCH64_ISA_SHA3): New ISA flag.
3722 (TARGET_SHA3): New feature flag for sha3.
3723 * config/aarch64/iterators.md (sha512_op): New int attribute.
3724 (CRYPTO_SHA512): New int iterator.
3725 (UNSPEC_SHA512H): New unspec.
3726 (UNSPEC_SHA512H2): Ditto.
3727 (UNSPEC_SHA512SU0): Ditto.
3728 (UNSPEC_SHA512SU1): Ditto.
3729 * config/aarch64/aarch64-simd-builtins.def
3730 (aarch64_crypto_sha512hqv2di): New builtin.
3731 (aarch64_crypto_sha512h2qv2di): Ditto.
3732 (aarch64_crypto_sha512su0qv2di): Ditto.
3733 (aarch64_crypto_sha512su1qv2di): Ditto.
3734 (aarch64_eor3qv8hi): Ditto.
3735 (aarch64_rax1qv2di): Ditto.
3736 (aarch64_xarqv2di): Ditto.
3737 (aarch64_bcaxqv8hi): Ditto.
3738 * config/aarch64/aarch64-simd.md:
3739 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3740 (aarch64_crypto_sha512su0qv2di): Ditto.
3741 (aarch64_crypto_sha512su1qv2di): Ditto.
3742 (aarch64_eor3qv8hi): Ditto.
3743 (aarch64_rax1qv2di): Ditto.
3744 (aarch64_xarqv2di): Ditto.
3745 (aarch64_bcaxqv8hi): Ditto.
3746 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3747 (vsha512h2q_u64): Ditto.
3748 (vsha512su0q_u64): Ditto.
3749 (vsha512su1q_u64): Ditto.
3750 (veor3q_u16): Ditto.
3751 (vrax1q_u64): Ditto.
3753 (vbcaxq_u16): Ditto.
3754 * config/arm/types.md (crypto_sha512): New type attribute.
3755 (crypto_sha3): Ditto.
3756 (doc/invoke.texi): Document new sha3 option.
3758 2018-01-10 Michael Collison <michael.collison@arm.com>
3760 * config/aarch64/aarch64-builtins.c:
3761 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3762 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3763 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3764 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3765 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3766 (AARCH64_ISA_SM4): New ISA flag.
3767 (TARGET_SM4): New feature flag for sm4.
3768 * config/aarch64/aarch64-simd-builtins.def
3769 (aarch64_sm3ss1qv4si): Ditto.
3770 (aarch64_sm3tt1aq4si): Ditto.
3771 (aarch64_sm3tt1bq4si): Ditto.
3772 (aarch64_sm3tt2aq4si): Ditto.
3773 (aarch64_sm3tt2bq4si): Ditto.
3774 (aarch64_sm3partw1qv4si): Ditto.
3775 (aarch64_sm3partw2qv4si): Ditto.
3776 (aarch64_sm4eqv4si): Ditto.
3777 (aarch64_sm4ekeyqv4si): Ditto.
3778 * config/aarch64/aarch64-simd.md:
3779 (aarch64_sm3ss1qv4si): Ditto.
3780 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3781 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3782 (aarch64_sm4eqv4si): Ditto.
3783 (aarch64_sm4ekeyqv4si): Ditto.
3784 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3785 (sm3part_op): Ditto.
3786 (CRYPTO_SM3TT): Ditto.
3787 (CRYPTO_SM3PART): Ditto.
3788 (UNSPEC_SM3SS1): New unspec.
3789 (UNSPEC_SM3TT1A): Ditto.
3790 (UNSPEC_SM3TT1B): Ditto.
3791 (UNSPEC_SM3TT2A): Ditto.
3792 (UNSPEC_SM3TT2B): Ditto.
3793 (UNSPEC_SM3PARTW1): Ditto.
3794 (UNSPEC_SM3PARTW2): Ditto.
3795 (UNSPEC_SM4E): Ditto.
3796 (UNSPEC_SM4EKEY): Ditto.
3797 * config/aarch64/constraints.md (Ui2): New constraint.
3798 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3799 * config/arm/types.md (crypto_sm3): New type attribute.
3800 (crypto_sm4): Ditto.
3801 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3802 (vsm3tt1aq_u32): Ditto.
3803 (vsm3tt1bq_u32): Ditto.
3804 (vsm3tt2aq_u32): Ditto.
3805 (vsm3tt2bq_u32): Ditto.
3806 (vsm3partw1q_u32): Ditto.
3807 (vsm3partw2q_u32): Ditto.
3808 (vsm4eq_u32): Ditto.
3809 (vsm4ekeyq_u32): Ditto.
3810 (doc/invoke.texi): Document new sm4 option.
3812 2018-01-10 Michael Collison <michael.collison@arm.com>
3814 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3815 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3816 (AARCH64_FL_FOR_ARCH8_4): New.
3817 (AARCH64_FL_V8_4): New flag.
3818 (doc/invoke.texi): Document new armv8.4-a option.
3820 2018-01-10 Michael Collison <michael.collison@arm.com>
3822 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3823 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3824 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3825 * config/aarch64/aarch64-option-extension.def: Add
3826 AARCH64_OPT_EXTENSION of 'sha2'.
3827 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3828 (crypto): Disable sha2 and aes if crypto disabled.
3829 (crypto): Enable aes and sha2 if enabled.
3830 (simd): Disable sha2 and aes if simd disabled.
3831 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3833 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3834 (TARGET_SHA2): New feature flag for sha2.
3835 (TARGET_AES): New feature flag for aes.
3836 * config/aarch64/aarch64-simd.md:
3837 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3838 conditional on TARGET_AES.
3839 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3840 (aarch64_crypto_sha1hsi): Make pattern conditional
3842 (aarch64_crypto_sha1hv4si): Ditto.
3843 (aarch64_be_crypto_sha1hv4si): Ditto.
3844 (aarch64_crypto_sha1su1v4si): Ditto.
3845 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3846 (aarch64_crypto_sha1su0v4si): Ditto.
3847 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3848 (aarch64_crypto_sha256su0v4si): Ditto.
3849 (aarch64_crypto_sha256su1v4si): Ditto.
3850 (doc/invoke.texi): Document new aes and sha2 options.
3852 2018-01-10 Martin Sebor <msebor@redhat.com>
3854 PR tree-optimization/83781
3855 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3858 2018-01-11 Martin Sebor <msebor@gmail.com>
3859 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3861 PR tree-optimization/83501
3862 PR tree-optimization/81703
3864 * tree-ssa-strlen.c (get_string_cst): Rename...
3865 (get_string_len): ...to this. Handle global constants.
3866 (handle_char_store): Adjust.
3868 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3869 Jim Wilson <jimw@sifive.com>
3871 * config/riscv/riscv-protos.h (riscv_output_return): New.
3872 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3873 (riscv_attribute_table, riscv_output_return),
3874 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3875 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3876 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3878 (riscv_expand_prologue): Add early return for naked function.
3879 (riscv_expand_epilogue): Likewise.
3880 (riscv_function_ok_for_sibcall): Return false for naked function.
3881 (riscv_set_current_function): New.
3882 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3883 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3884 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3885 * doc/extend.texi (RISC-V Function Attributes): New.
3887 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3889 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3890 check for 128-bit long double before checking TCmode.
3891 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3892 128-bit long doubles before checking TFmode or TCmode.
3893 (FLOAT128_IBM_P): Likewise.
3895 2018-01-10 Martin Sebor <msebor@redhat.com>
3897 PR tree-optimization/83671
3898 * builtins.c (c_strlen): Unconditionally return zero for the empty
3900 Use -Warray-bounds for warnings.
3901 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3902 for non-constant array indices with COMPONENT_REF, arrays of
3903 arrays, and pointers to arrays.
3904 (gimple_fold_builtin_strlen): Determine and set length range for
3905 non-constant character arrays.
3907 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3910 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3913 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3915 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3917 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3920 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3921 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3922 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3923 indexed_or_indirect_operand predicate.
3924 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3925 (*vsx_le_perm_load_v8hi): Likewise.
3926 (*vsx_le_perm_load_v16qi): Likewise.
3927 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3928 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3929 (*vsx_le_perm_store_v8hi): Likewise.
3930 (*vsx_le_perm_store_v16qi): Likewise.
3931 (eight unnamed splitters): Likewise.
3933 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3935 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3936 * config/rs6000/emmintrin.h: Likewise.
3937 * config/rs6000/mmintrin.h: Likewise.
3938 * config/rs6000/xmmintrin.h: Likewise.
3940 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3943 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3945 * tree.c (tree_nop_conversion): Return true for location wrapper
3947 (maybe_wrap_with_location): New function.
3948 (selftest::check_strip_nops): New function.
3949 (selftest::test_location_wrappers): New function.
3950 (selftest::tree_c_tests): Call it.
3951 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3952 (maybe_wrap_with_location): New decl.
3953 (EXPR_LOCATION_WRAPPER_P): New macro.
3954 (location_wrapper_p): New inline function.
3955 (tree_strip_any_location_wrapper): New inline function.
3957 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3960 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3961 stack_realign_offset for the largest alignment of stack slot
3963 (ix86_find_max_used_stack_alignment): New function.
3964 (ix86_finalize_stack_frame_flags): Use it. Set
3965 max_used_stack_alignment if we don't realign stack.
3966 * config/i386/i386.h (machine_function): Add
3967 max_used_stack_alignment.
3969 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3971 * config/arm/arm.opt (-mbranch-cost): New option.
3972 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3975 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3978 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3979 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3981 2018-01-10 Richard Biener <rguenther@suse.de>
3984 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3985 early out so it also covers the case where we have a non-NULL
3988 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3990 PR tree-optimization/83753
3991 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3992 for non-strided grouped accesses if the number of elements is 1.
3994 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3997 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3998 * i386.h (TARGET_USE_GATHER): Define.
3999 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4001 2018-01-10 Martin Liska <mliska@suse.cz>
4004 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4005 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4007 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4008 CLEANUP_NO_PARTITIONING is not set.
4010 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4012 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4013 for vectors, as a partial revert of r254296.
4014 * rtl.h (const_vec_p): Delete.
4015 (const_vec_duplicate_p): Don't test for vector CONSTs.
4016 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4017 * expmed.c (make_tree): Likewise.
4020 * common.md (E, F): Use CONSTANT_P instead of checking for
4022 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4023 checking for CONST_VECTOR.
4025 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4028 * predict.c (force_edge_cold): Handle in more sane way edges
4031 2018-01-09 Carl Love <cel@us.ibm.com>
4033 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4035 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4036 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4037 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4038 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4039 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4040 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4041 * config/rs6000/rs6000-protos.h: Add extern defition for
4042 rs6000_generate_float2_double_code.
4043 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4045 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4046 (float2_v2df): Add define_expand.
4048 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4051 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4052 op_mode in the force_to_mode call.
4054 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4056 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4057 instead of checking each element individually.
4058 (aarch64_evpc_uzp): Likewise.
4059 (aarch64_evpc_zip): Likewise.
4060 (aarch64_evpc_ext): Likewise.
4061 (aarch64_evpc_rev): Likewise.
4062 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4063 instead of checking each element individually. Return true without
4065 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4066 whether all selected elements come from the same input, instead of
4067 checking each element individually. Remove calls to gen_rtx_REG,
4068 start_sequence and end_sequence and instead assert that no rtl is
4071 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4073 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4074 order of HIGH and CONST checks.
4076 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4078 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4079 if the destination isn't an SSA_NAME.
4081 2018-01-09 Richard Biener <rguenther@suse.de>
4083 PR tree-optimization/83668
4084 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4086 (canonicalize_loop_form): ... here, renamed from ...
4087 (canonicalize_loop_closed_ssa_form): ... this and amended to
4088 swap successor edges for loop exit blocks to make us use
4089 the RPO order we need for initial schedule generation.
4091 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4093 PR tree-optimization/64811
4094 * match.pd: When optimizing comparisons with Inf, avoid
4095 introducing or losing exceptions from comparisons with NaN.
4097 2018-01-09 Martin Liska <mliska@suse.cz>
4100 * asan.c (shadow_mem_size): Add gcc_assert.
4102 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4104 Don't save registers in main().
4107 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4108 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4109 * config/avr/avr.c (avr_set_current_function): Don't error if
4110 naked, OS_task or OS_main are specified at the same time.
4111 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4113 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4115 * common/config/avr/avr-common.c (avr_option_optimization_table):
4116 Switch on -mmain-is-OS_task for optimizing compilations.
4118 2018-01-09 Richard Biener <rguenther@suse.de>
4120 PR tree-optimization/83572
4121 * graphite.c: Include cfganal.h.
4122 (graphite_transform_loops): Connect infinite loops to exit
4123 and remove fake edges at the end.
4125 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4127 * ipa-inline.c (edge_badness): Revert accidental checkin.
4129 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4132 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4133 symbols; not inline clones.
4135 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4138 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4139 hard registers. Formatting fixes.
4141 PR preprocessor/83722
4142 * gcc.c (try_generate_repro): Pass
4143 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4144 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4147 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4148 Kito Cheng <kito.cheng@gmail.com>
4150 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4151 (riscv_leaf_function_p): Delete.
4152 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4154 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4156 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4158 (do_ifelse): New function.
4159 (do_isel): New function.
4160 (do_sub3): New function.
4161 (do_add3): New function.
4162 (do_load_mask_compare): New function.
4163 (do_overlap_load_compare): New function.
4164 (expand_compare_loop): New function.
4165 (expand_block_compare): Call expand_compare_loop() when appropriate.
4166 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4168 (-mblock-compare-inline-loop-limit): New option.
4170 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4173 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4174 Reverse order of second and third operands in first alternative.
4175 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4176 of first and second elements in UNSPEC_VPERMR vector.
4177 (altivec_expand_vec_perm_le): Likewise.
4179 2018-01-08 Jeff Law <law@redhat.com>
4181 PR rtl-optimizatin/81308
4182 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4183 (process_switch): If group_case_labels makes a change, then set
4185 (pass_convert_switch::execute): If a switch is converted, then
4186 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4188 PR rtl-optimization/81308
4189 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4192 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4194 PR target/83663 - Revert r255946
4195 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4196 generation for cases where splatting a value is not useful.
4197 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4198 across a vec_duplicate and a paradoxical subreg forming a vector
4199 mode to a vec_concat.
4201 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4203 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4204 -march=armv8.3-a variants.
4205 * config/arm/t-multilib: Likewise.
4206 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4208 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4210 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4212 (cceq_ior_compare_complement): Give it a name so I can use it, and
4213 change boolean_or_operator predicate to boolean_operator so it can
4214 be used to generate a crand.
4215 (eqne): New code iterator.
4216 (bd/bd_neg): New code_attrs.
4217 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4218 a single define_insn.
4219 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4220 decrement (bdnzt/bdnzf/bdzt/bdzf).
4221 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4222 with the new names of the branch decrement patterns, and added the
4223 names of the branch decrement conditional patterns.
4225 2018-01-08 Richard Biener <rguenther@suse.de>
4227 PR tree-optimization/83563
4228 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4231 2018-01-08 Richard Biener <rguenther@suse.de>
4234 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4236 2018-01-08 Richard Biener <rguenther@suse.de>
4238 PR tree-optimization/83685
4239 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4240 references to abnormals.
4242 2018-01-08 Richard Biener <rguenther@suse.de>
4245 * dwarf2out.c (output_indirect_strings): Handle empty
4246 skeleton_debug_str_hash.
4247 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4249 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4251 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4252 (emit_store_direct): Likewise.
4253 (arc_trampoline_adjust_address): Likewise.
4254 (arc_asm_trampoline_template): New function.
4255 (arc_initialize_trampoline): Use asm_trampoline_template.
4256 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4257 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4258 * config/arc/arc.md (flush_icache): Delete pattern.
4260 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4262 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4263 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4266 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4269 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4270 by not USED_FOR_TARGET.
4271 (make_pass_resolve_sw_modes): Likewise.
4273 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4275 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4278 2018-01-08 Richard Biener <rguenther@suse.de>
4281 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4283 2018-01-08 Richard Biener <rguenther@suse.de>
4286 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4288 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4291 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4292 basic blocks with a small number of successors.
4293 (convert_control_dep_chain_into_preds): Improve handling of
4295 (dump_predicates): Split apart into...
4296 (dump_pred_chain): ...here...
4297 (dump_pred_info): ...and here.
4298 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4299 (can_chain_union_be_invalidated_p): Improve check for invalidation
4301 (uninit_uses_cannot_happen): Avoid unnecessary if
4302 convert_control_dep_chain_into_preds yielded nothing.
4304 2018-01-06 Martin Sebor <msebor@redhat.com>
4306 PR tree-optimization/83640
4307 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4308 subtracting negative offset from size.
4309 (builtin_access::overlap): Adjust offset bounds of the access to fall
4310 within the size of the object if possible.
4312 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4314 PR rtl-optimization/83699
4315 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4316 extract_bit_field_as_subreg to cases in which the extracted
4317 value is also a vector.
4319 * lra-constraints.c (process_alt_operands): Test for the equivalence
4320 substitutions when detecting a possible reload cycle.
4322 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4325 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4326 by default if flag_selective_schedling{,2}. Formatting fixes.
4328 PR rtl-optimization/83682
4329 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4330 if it has non-VECTOR_MODE element mode.
4331 (vec_duplicate_p): Likewise.
4334 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4335 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4337 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4340 * config/i386/i386-builtin.def
4341 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4342 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4343 Require also OPTION_MASK_ISA_AVX512F in addition to
4344 OPTION_MASK_ISA_GFNI.
4345 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4346 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4347 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4348 to OPTION_MASK_ISA_GFNI.
4349 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4350 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4351 OPTION_MASK_ISA_AVX512BW.
4352 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4353 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4354 addition to OPTION_MASK_ISA_GFNI.
4355 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4356 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4357 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4358 to OPTION_MASK_ISA_GFNI.
4359 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4360 a requirement for all ISAs rather than any of them with a few
4362 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4364 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4365 bitmasks to be enabled with 3 exceptions, instead of requiring any
4366 enabled ISA with lots of exceptions.
4367 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4368 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4369 Change avx512bw in isa attribute to avx512f.
4370 * config/i386/sgxintrin.h: Add license boilerplate.
4371 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4372 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4373 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4374 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4376 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4377 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4378 temporarily sse2 rather than sse if not enabled already.
4381 * config/i386/sse.md (VI248_VLBW): Rename to ...
4382 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4383 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4384 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4385 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4386 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4387 mode iterator instead of VI248_VLBW.
4389 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4391 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4392 (record_modified): Skip clobbers; add debug output.
4393 (param_change_prob): Use sreal frequencies.
4395 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4397 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4398 punt for user-aligned variables.
4400 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4402 * tree-chrec.c (chrec_contains_symbols): Return true for
4405 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4408 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4409 of (x|y) == x for BICS pattern.
4411 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4413 PR tree-optimization/83605
4414 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4415 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4418 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4420 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4421 * config/epiphany/rtems.h: New file.
4423 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4424 Uros Bizjak <ubizjak@gmail.com>
4427 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4428 QIreg_operand instead of register_operand predicate.
4429 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4430 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4431 comments instead of -fmitigate[-_]rop.
4433 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4436 * cgraphunit.c (symbol_table::compile): Switch to text_section
4437 before calling assembly_start debug hook.
4438 * run-rtl-passes.c (run_rtl_passes): Likewise.
4441 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4443 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4444 range_int_cst_p rather than !symbolic_range_p before calling
4445 extract_range_from_multiplicative_op_1.
4447 2018-01-04 Jeff Law <law@redhat.com>
4449 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4450 redundant test in assertion.
4452 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4454 * doc/rtl.texi: Document machine_mode wrapper classes.
4456 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4458 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4461 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4463 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4464 the VEC_PERM_EXPR fold to fail.
4466 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4469 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4470 to switched_sections.
4472 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4475 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4478 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4481 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4482 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4484 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4487 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4488 is BLKmode and bitpos not zero or mode change is needed.
4490 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4493 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4496 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4499 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4500 instead of MULT rtx. Update all corresponding splitters.
4502 (*ssub<modesuffix>): Ditto.
4504 (*cmp_sadd_di): Update split patterns.
4505 (*cmp_sadd_si): Ditto.
4506 (*cmp_sadd_sidi): Ditto.
4507 (*cmp_ssub_di): Ditto.
4508 (*cmp_ssub_si): Ditto.
4509 (*cmp_ssub_sidi): Ditto.
4510 * config/alpha/predicates.md (const23_operand): New predicate.
4511 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4512 Look for ASHIFT, not MULT inner operand.
4513 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4515 2018-01-04 Martin Liska <mliska@suse.cz>
4517 PR gcov-profile/83669
4518 * gcov.c (output_intermediate_file): Add version to intermediate
4520 * doc/gcov.texi: Document new field 'version' in intermediate
4521 file format. Fix location of '-k' option of gcov command.
4523 2018-01-04 Martin Liska <mliska@suse.cz>
4526 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4528 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4530 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4532 2018-01-03 Martin Sebor <msebor@redhat.com>
4534 PR tree-optimization/83655
4535 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4536 checking calls with invalid arguments.
4538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4540 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4541 (vectorizable_mask_load_store): Delete.
4542 (vectorizable_call): Return false for masked loads and stores.
4543 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4544 instead of gimple_assign_rhs1.
4545 (vectorizable_load): Handle IFN_MASK_LOAD.
4546 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4548 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4550 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4552 (vectorizable_mask_load_store): ...here.
4553 (vectorizable_load): ...and here.
4555 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4557 * tree-vect-stmts.c (vect_build_all_ones_mask)
4558 (vect_build_zero_merge_argument): New functions, split out from...
4559 (vectorizable_load): ...here.
4561 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4563 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4565 (vectorizable_mask_load_store): ...here.
4566 (vectorizable_store): ...and here.
4568 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4570 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4572 (vectorizable_mask_load_store): ...here.
4574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4576 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4577 (vect_model_store_cost): Take a vec_load_store_type instead of a
4579 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4580 (vect_model_store_cost): Take a vec_load_store_type instead of a
4582 (vectorizable_mask_load_store): Update accordingly.
4583 (vectorizable_store): Likewise.
4584 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4586 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4588 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4589 IFN_MASK_LOAD calls here rather than...
4590 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4593 Alan Hayward <alan.hayward@arm.com>
4594 David Sherwood <david.sherwood@arm.com>
4596 * expmed.c (extract_bit_field_1): For vector extracts,
4597 fall back to extract_bit_field_as_subreg if vec_extract
4600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4601 Alan Hayward <alan.hayward@arm.com>
4602 David Sherwood <david.sherwood@arm.com>
4604 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4605 they are variable or constant sized.
4606 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4607 slots for constant-sized data.
4609 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4610 Alan Hayward <alan.hayward@arm.com>
4611 David Sherwood <david.sherwood@arm.com>
4613 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4614 handling COND_EXPRs with boolean comparisons, try to find a better
4615 basis for the mask type than the boolean itself.
4617 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4619 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4620 is calculated and how it can be overridden.
4621 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4622 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4624 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4627 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4628 Alan Hayward <alan.hayward@arm.com>
4629 David Sherwood <david.sherwood@arm.com>
4631 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4632 Remove the mode argument.
4633 (aarch64_simd_valid_immediate): Remove the mode and inverse
4635 * config/aarch64/iterators.md (bitsize): New iterator.
4636 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4637 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4638 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4639 aarch64_simd_valid_immediate.
4640 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4641 (aarch64_reg_or_bic_imm): Likewise.
4642 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4643 with an insn_type enum and msl with a modifier_type enum.
4644 Replace element_width with a scalar_mode. Change the shift
4645 to unsigned int. Add constructors for scalar_float_mode and
4646 scalar_int_mode elements.
4647 (aarch64_vect_float_const_representable_p): Delete.
4648 (aarch64_can_const_movi_rtx_p)
4649 (aarch64_simd_scalar_immediate_valid_for_move)
4650 (aarch64_simd_make_constant): Update call to
4651 aarch64_simd_valid_immediate.
4652 (aarch64_advsimd_valid_immediate_hs): New function.
4653 (aarch64_advsimd_valid_immediate): Likewise.
4654 (aarch64_simd_valid_immediate): Remove mode and inverse
4655 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4656 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4657 and aarch64_float_const_representable_p on the result.
4658 (aarch64_output_simd_mov_immediate): Remove mode argument.
4659 Update call to aarch64_simd_valid_immediate and use of
4660 simd_immediate_info.
4661 (aarch64_output_scalar_simd_mov_immediate): Update call
4664 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4665 Alan Hayward <alan.hayward@arm.com>
4666 David Sherwood <david.sherwood@arm.com>
4668 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4669 (mode_nunits): Likewise CONST_MODE_NUNITS.
4670 * machmode.def (ADJUST_NUNITS): Document.
4671 * genmodes.c (mode_data::need_nunits_adj): New field.
4672 (blank_mode): Update accordingly.
4673 (adj_nunits): New variable.
4674 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4676 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4677 listed in adj_nunits.
4678 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4679 listed in adj_nunits. Don't emit case statements for such modes.
4680 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4681 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4682 nothing if adj_nunits is nonnull.
4683 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4684 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4685 (emit_mode_fbit): Update use of print_maybe_const_decl.
4686 (emit_move_size): Likewise. Treat the array as non-const
4688 (emit_mode_adjustments): Handle adj_nunits.
4690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4692 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4693 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4694 (VECTOR_MODES): Use it.
4695 (make_vector_modes): Take the prefix as an argument.
4697 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4698 Alan Hayward <alan.hayward@arm.com>
4699 David Sherwood <david.sherwood@arm.com>
4701 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4702 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4703 for MODE_VECTOR_BOOL.
4704 * machmode.def (VECTOR_BOOL_MODE): Document.
4705 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4706 (make_vector_bool_mode): New function.
4707 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4709 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4710 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4712 * stor-layout.c (int_mode_for_mode): Likewise.
4713 * tree.c (build_vector_type_for_mode): Likewise.
4714 * varasm.c (output_constant_pool_2): Likewise.
4715 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4716 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4717 for MODE_VECTOR_BOOL.
4718 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4719 of mode class checks.
4720 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4721 instead of a list of mode class checks.
4722 (expand_vector_scalar_condition): Likewise.
4723 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4726 Alan Hayward <alan.hayward@arm.com>
4727 David Sherwood <david.sherwood@arm.com>
4729 * machmode.h (mode_size): Change from unsigned short to
4731 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4732 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4733 or if measurement_type is not polynomial.
4734 (fixed_size_mode::includes_p): Check for constant-sized modes.
4735 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4736 return a poly_uint16 rather than an unsigned short.
4737 (emit_mode_size): Change the type of mode_size from unsigned short
4738 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4739 (emit_mode_adjustments): Cope with polynomial vector sizes.
4740 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4742 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4744 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4745 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4746 * caller-save.c (setup_save_areas): Likewise.
4747 (replace_reg_with_saved_mem): Likewise.
4748 * calls.c (emit_library_call_value_1): Likewise.
4749 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4750 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4751 (gen_lowpart_for_combine): Likewise.
4752 * convert.c (convert_to_integer_1): Likewise.
4753 * cse.c (equiv_constant, cse_insn): Likewise.
4754 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4755 (cselib_subst_to_values): Likewise.
4756 * dce.c (word_dce_process_block): Likewise.
4757 * df-problems.c (df_word_lr_mark_ref): Likewise.
4758 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4759 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4760 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4761 (rtl_for_decl_location): Likewise.
4762 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4763 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4764 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4765 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4766 (expand_expr_real_1): Likewise.
4767 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4768 (pad_below): Likewise.
4769 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4770 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4771 * ira.c (get_subreg_tracking_sizes): Likewise.
4772 * ira-build.c (ira_create_allocno_objects): Likewise.
4773 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4774 (ira_sort_regnos_for_alter_reg): Likewise.
4775 * ira-costs.c (record_operand_costs): Likewise.
4776 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4777 (resolve_simple_move): Likewise.
4778 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4779 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4780 (lra_constraints): Likewise.
4781 (CONST_POOL_OK_P): Reject variable-sized modes.
4782 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4783 (add_pseudo_to_slot, lra_spill): Likewise.
4784 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4785 * optabs-query.c (get_best_extraction_insn): Likewise.
4786 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4787 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4788 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4789 * recog.c (offsettable_address_addr_space_p): Likewise.
4790 * regcprop.c (maybe_mode_change): Likewise.
4791 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4792 * regrename.c (build_def_use): Likewise.
4793 * regstat.c (dump_reg_info): Likewise.
4794 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4795 (find_reloads, find_reloads_subreg_address): Likewise.
4796 * reload1.c (eliminate_regs_1): Likewise.
4797 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4798 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4799 (simplify_binary_operation_1, simplify_subreg): Likewise.
4800 * targhooks.c (default_function_arg_padding): Likewise.
4801 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4802 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4803 (verify_gimple_assign_ternary): Likewise.
4804 * tree-inline.c (estimate_move_cost): Likewise.
4805 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4806 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4807 (get_address_cost_ainc): Likewise.
4808 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4809 (vect_supportable_dr_alignment): Likewise.
4810 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4811 (vectorizable_reduction): Likewise.
4812 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4813 (vectorizable_operation, vectorizable_load): Likewise.
4814 * tree.c (build_same_sized_truth_vector_type): Likewise.
4815 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4816 * var-tracking.c (emit_note_insn_var_location): Likewise.
4817 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4818 (ADDR_VEC_ALIGN): Likewise.
4820 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4821 Alan Hayward <alan.hayward@arm.com>
4822 David Sherwood <david.sherwood@arm.com>
4824 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4826 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4827 or if measurement_type is polynomial.
4828 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4829 * combine.c (make_extraction): Likewise.
4830 * dse.c (find_shift_sequence): Likewise.
4831 * dwarf2out.c (mem_loc_descriptor): Likewise.
4832 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4833 (extract_bit_field, extract_low_bits): Likewise.
4834 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4835 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4836 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4837 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4838 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4839 * reload.c (find_reloads): Likewise.
4840 * reload1.c (alter_reg): Likewise.
4841 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4842 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4843 * tree-if-conv.c (predicate_mem_writes): Likewise.
4844 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4845 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4846 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4847 * valtrack.c (dead_debug_insert_temp): Likewise.
4848 * varasm.c (mergeable_constant_section): Likewise.
4849 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4851 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4852 Alan Hayward <alan.hayward@arm.com>
4853 David Sherwood <david.sherwood@arm.com>
4855 * expr.c (expand_assignment): Cope with polynomial mode sizes
4856 when assigning to a CONCAT.
4858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4859 Alan Hayward <alan.hayward@arm.com>
4860 David Sherwood <david.sherwood@arm.com>
4862 * machmode.h (mode_precision): Change from unsigned short to
4864 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4866 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4867 or if measurement_type is not polynomial.
4868 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4869 in which the mode is already known to be a scalar_int_mode.
4870 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4871 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4873 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4874 for GET_MODE_PRECISION.
4875 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4876 for GET_MODE_PRECISION.
4877 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4879 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4880 (expand_field_assignment, make_extraction): Likewise.
4881 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4882 (get_last_value): Likewise.
4883 * convert.c (convert_to_integer_1): Likewise.
4884 * cse.c (cse_insn): Likewise.
4885 * expr.c (expand_expr_real_1): Likewise.
4886 * lra-constraints.c (simplify_operand_subreg): Likewise.
4887 * optabs-query.c (can_atomic_load_p): Likewise.
4888 * optabs.c (expand_atomic_load): Likewise.
4889 (expand_atomic_store): Likewise.
4890 * ree.c (combine_reaching_defs): Likewise.
4891 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4892 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4893 * tree.h (type_has_mode_precision_p): Likewise.
4894 * ubsan.c (instrument_si_overflow): Likewise.
4896 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4897 Alan Hayward <alan.hayward@arm.com>
4898 David Sherwood <david.sherwood@arm.com>
4900 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4901 polynomial numbers of units.
4902 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4903 (valid_vector_subparts_p): New function.
4904 (build_vector_type): Remove temporary shim and take the number
4905 of units as a poly_uint64 rather than an int.
4906 (build_opaque_vector_type): Take the number of units as a
4907 poly_uint64 rather than an int.
4908 * tree.c (build_vector_from_ctor): Handle polynomial
4909 TYPE_VECTOR_SUBPARTS.
4910 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4911 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4912 (build_vector_from_val): If the number of units is variable,
4913 use build_vec_duplicate_cst for constant operands and
4914 VEC_DUPLICATE_EXPR otherwise.
4915 (make_vector_type): Remove temporary is_constant ().
4916 (build_vector_type, build_opaque_vector_type): Take the number of
4917 units as a poly_uint64 rather than an int.
4918 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4920 * cfgexpand.c (expand_debug_expr): Likewise.
4921 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4922 (store_constructor, expand_expr_real_1): Likewise.
4923 (const_scalar_mask_from_tree): Likewise.
4924 * fold-const-call.c (fold_const_reduction): Likewise.
4925 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4926 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4927 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4928 (fold_relational_const): Likewise.
4929 (native_interpret_vector): Likewise. Change the size from an
4930 int to an unsigned int.
4931 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4932 TYPE_VECTOR_SUBPARTS.
4933 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4934 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4935 duplicating a non-constant operand into a variable-length vector.
4936 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4937 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4938 * ipa-icf.c (sem_variable::equals): Likewise.
4939 * match.pd: Likewise.
4940 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4941 * print-tree.c (print_node): Likewise.
4942 * stor-layout.c (layout_type): Likewise.
4943 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4944 * tree-cfg.c (verify_gimple_comparison): Likewise.
4945 (verify_gimple_assign_binary): Likewise.
4946 (verify_gimple_assign_ternary): Likewise.
4947 (verify_gimple_assign_single): Likewise.
4948 * tree-pretty-print.c (dump_generic_node): Likewise.
4949 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4950 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4951 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4952 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4953 (vect_shift_permute_load_chain): Likewise.
4954 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4955 (expand_vector_condition, optimize_vector_constructor): Likewise.
4956 (lower_vec_perm, get_compute_type): Likewise.
4957 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4958 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4959 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4960 (vect_recog_mask_conversion_pattern): Likewise.
4961 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4962 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4963 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4964 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4965 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4966 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4967 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4968 (supportable_widening_operation): Likewise.
4969 (supportable_narrowing_operation): Likewise.
4970 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4972 * varasm.c (output_constant): Likewise.
4974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4975 Alan Hayward <alan.hayward@arm.com>
4976 David Sherwood <david.sherwood@arm.com>
4978 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4979 so that both the length == 3 and length != 3 cases set up their
4980 own permute vectors. Add comments explaining why we know the
4981 number of elements is constant.
4982 (vect_permute_load_chain): Likewise.
4984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4985 Alan Hayward <alan.hayward@arm.com>
4986 David Sherwood <david.sherwood@arm.com>
4988 * machmode.h (mode_nunits): Change from unsigned char to
4990 (ONLY_FIXED_SIZE_MODES): New macro.
4991 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4992 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4993 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4995 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4996 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4997 or if measurement_type is not polynomial.
4998 * genmodes.c (ZERO_COEFFS): New macro.
4999 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5001 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5002 Use ZERO_COEFFS when emitting initializers.
5003 * data-streamer.h (bp_pack_poly_value): New function.
5004 (bp_unpack_poly_value): Likewise.
5005 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5006 for GET_MODE_NUNITS.
5007 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5008 for GET_MODE_NUNITS.
5009 * tree.c (make_vector_type): Remove temporary shim and make
5010 the real function take the number of units as a poly_uint64
5012 (build_vector_type_for_mode): Handle polynomial nunits.
5013 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5014 * emit-rtl.c (const_vec_series_p_1): Likewise.
5015 (gen_rtx_CONST_VECTOR): Likewise.
5016 * fold-const.c (test_vec_duplicate_folding): Likewise.
5017 * genrecog.c (validate_pattern): Likewise.
5018 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5019 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5020 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5021 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5022 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5023 * rtlanal.c (subreg_get_info): Likewise.
5024 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5025 (vect_grouped_load_supported): Likewise.
5026 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5027 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5028 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5029 (simplify_const_unary_operation, simplify_binary_operation_1)
5030 (simplify_const_binary_operation, simplify_ternary_operation)
5031 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5032 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5033 instead of CONST_VECTOR_NUNITS.
5034 * varasm.c (output_constant_pool_2): Likewise.
5035 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5036 explicit-encoded elements in the XVEC for variable-length vectors.
5038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5040 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5043 Alan Hayward <alan.hayward@arm.com>
5044 David Sherwood <david.sherwood@arm.com>
5046 * coretypes.h (fixed_size_mode): Declare.
5047 (fixed_size_mode_pod): New typedef.
5048 * builtins.h (target_builtins::x_apply_args_mode)
5049 (target_builtins::x_apply_result_mode): Change type to
5050 fixed_size_mode_pod.
5051 * builtins.c (apply_args_size, apply_result_size, result_vector)
5052 (expand_builtin_apply_args_1, expand_builtin_apply)
5053 (expand_builtin_return): Update accordingly.
5055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5057 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5058 * cselib.c (cselib_hash_rtx): Likewise.
5059 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5060 CONST_VECTOR encoding.
5062 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5063 Jeff Law <law@redhat.com>
5066 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5067 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5068 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5069 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5072 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5073 explicitly probe *sp in a noreturn function if there were any callee
5074 register saves or frame pointer is needed.
5076 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5079 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5080 BLKmode for ternary, binary or unary expressions.
5083 * var-tracking.c (delete_vta_debug_insn): New inline function.
5084 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5085 insns from get_insns () to NULL instead of each bb separately.
5086 Use delete_vta_debug_insn. No longer static.
5087 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5088 delete_vta_debug_insns callers.
5089 * rtl.h (delete_vta_debug_insns): Declare.
5090 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5091 instead of variable_tracking_main.
5093 2018-01-03 Martin Sebor <msebor@redhat.com>
5095 PR tree-optimization/83603
5096 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5097 arguments past the endof the argument list in functions declared
5098 without a prototype.
5099 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5100 Avoid checking when arguments are null.
5102 2018-01-03 Martin Sebor <msebor@redhat.com>
5105 * doc/extend.texi (attribute const): Fix a typo.
5106 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5107 issuing -Wsuggest-attribute for void functions.
5109 2018-01-03 Martin Sebor <msebor@redhat.com>
5111 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5112 offset_int::from instead of wide_int::to_shwi.
5113 (maybe_diag_overlap): Remove assertion.
5114 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5115 * gimple-ssa-sprintf.c (format_directive): Same.
5116 (parse_directive): Same.
5117 (sprintf_dom_walker::compute_format_length): Same.
5118 (try_substitute_return_value): Same.
5120 2018-01-03 Jeff Law <law@redhat.com>
5123 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5124 non-constant residual for zero at runtime and avoid probing in
5125 that case. Reorganize code for trailing problem to mirror handling
5128 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5130 PR tree-optimization/83501
5131 * tree-ssa-strlen.c (get_string_cst): New.
5132 (handle_char_store): Call get_string_cst.
5134 2018-01-03 Martin Liska <mliska@suse.cz>
5136 PR tree-optimization/83593
5137 * tree-ssa-strlen.c: Include tree-cfg.h.
5138 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5139 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5140 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5142 (strlen_dom_walker::before_dom_children): Call
5143 gimple_purge_dead_eh_edges. Dump tranformation with details
5145 (strlen_dom_walker::before_dom_children): Update call by adding
5146 new argument cleanup_eh.
5147 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5149 2018-01-03 Martin Liska <mliska@suse.cz>
5152 * cif-code.def (VARIADIC_THUNK): New enum value.
5153 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5156 2018-01-03 Jan Beulich <jbeulich@suse.com>
5158 * sse.md (mov<mode>_internal): Tighten condition for when to use
5159 vmovdqu<ssescalarsize> for TI and OI modes.
5161 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5163 Update copyright years.
5165 2018-01-03 Martin Liska <mliska@suse.cz>
5168 * ipa-visibility.c (function_and_variable_visibility): Skip
5169 functions with noipa attribure.
5171 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5173 * gcc.c (process_command): Update copyright notice dates.
5174 * gcov-dump.c (print_version): Ditto.
5175 * gcov.c (print_version): Ditto.
5176 * gcov-tool.c (print_version): Ditto.
5177 * gengtype.c (create_file): Ditto.
5178 * doc/cpp.texi: Bump @copying's copyright year.
5179 * doc/cppinternals.texi: Ditto.
5180 * doc/gcc.texi: Ditto.
5181 * doc/gccint.texi: Ditto.
5182 * doc/gcov.texi: Ditto.
5183 * doc/install.texi: Ditto.
5184 * doc/invoke.texi: Ditto.
5186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5188 * vector-builder.h (vector_builder::m_full_nelts): Change from
5189 unsigned int to poly_uint64.
5190 (vector_builder::full_nelts): Update prototype accordingly.
5191 (vector_builder::new_vector): Likewise.
5192 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5193 (vector_builder::operator ==): Likewise.
5194 (vector_builder::finalize): Likewise.
5195 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5196 Take the number of elements as a poly_uint64 rather than an
5198 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5199 from unsigned int to poly_uint64.
5200 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5201 (vec_perm_indices::new_vector): Likewise.
5202 (vec_perm_indices::length): Likewise.
5203 (vec_perm_indices::nelts_per_input): Likewise.
5204 (vec_perm_indices::input_nelts): Likewise.
5205 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5206 number of elements per input as a poly_uint64 rather than an
5207 unsigned int. Use the original encoding for variable-length
5208 vectors, rather than clamping each individual element.
5209 For the second and subsequent elements in each pattern,
5210 clamp the step and base before clamping their sum.
5211 (vec_perm_indices::series_p): Handle polynomial element counts.
5212 (vec_perm_indices::all_in_range_p): Likewise.
5213 (vec_perm_indices_to_tree): Likewise.
5214 (vec_perm_indices_to_rtx): Likewise.
5215 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5216 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5217 (tree_vector_builder::new_binary_operation): Handle polynomial
5218 element counts. Return false if we need to know the number
5219 of elements at compile time.
5220 * fold-const.c (fold_vec_perm): Punt if the number of elements
5221 isn't known at compile time.
5223 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5225 * vec-perm-indices.h (vec_perm_builder): Change element type
5226 from HOST_WIDE_INT to poly_int64.
5227 (vec_perm_indices::element_type): Update accordingly.
5228 (vec_perm_indices::clamp): Handle polynomial element_types.
5229 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5230 (vec_perm_indices::all_in_range_p): Likewise.
5231 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5233 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5234 polynomial vec_perm_indices element types.
5235 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5236 * fold-const.c (fold_vec_perm): Likewise.
5237 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5238 * tree-vect-generic.c (lower_vec_perm): Likewise.
5239 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5240 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5241 element type to HOST_WIDE_INT.
5243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5244 Alan Hayward <alan.hayward@arm.com>
5245 David Sherwood <david.sherwood@arm.com>
5247 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5248 rather than an int. Use plus_constant.
5249 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5250 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5253 Alan Hayward <alan.hayward@arm.com>
5254 David Sherwood <david.sherwood@arm.com>
5256 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5257 a HOST_WIDE_INT to a poly_int64.
5259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5260 Alan Hayward <alan.hayward@arm.com>
5261 David Sherwood <david.sherwood@arm.com>
5263 * calls.c (load_register_parameters): Cope with polynomial
5264 mode sizes. Require a constant size for BLKmode parameters
5265 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5266 forces a parameter to be padded at the lsb end in order to
5267 fill a complete number of words, require the parameter size
5268 to be ordered wrt UNITS_PER_WORD.
5270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5271 Alan Hayward <alan.hayward@arm.com>
5272 David Sherwood <david.sherwood@arm.com>
5274 * reload1.c (spill_stack_slot_width): Change element type
5275 from unsigned int to poly_uint64_pod.
5276 (alter_reg): Treat mode sizes as polynomial.
5278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5279 Alan Hayward <alan.hayward@arm.com>
5280 David Sherwood <david.sherwood@arm.com>
5282 * reload.c (complex_word_subreg_p): New function.
5283 (reload_inner_reg_of_subreg, push_reload): Use it.
5285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5286 Alan Hayward <alan.hayward@arm.com>
5287 David Sherwood <david.sherwood@arm.com>
5289 * lra-constraints.c (process_alt_operands): Reject matched
5290 operands whose sizes aren't ordered.
5291 (match_reload): Refer to this check here.
5293 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5294 Alan Hayward <alan.hayward@arm.com>
5295 David Sherwood <david.sherwood@arm.com>
5297 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5298 that the mode size is in the set {1, 2, 4, 8, 16}.
5300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5301 Alan Hayward <alan.hayward@arm.com>
5302 David Sherwood <david.sherwood@arm.com>
5304 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5305 Use plus_constant instead of gen_rtx_PLUS.
5307 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5308 Alan Hayward <alan.hayward@arm.com>
5309 David Sherwood <david.sherwood@arm.com>
5311 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5312 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5313 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5314 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5315 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5316 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5317 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5318 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5319 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5320 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5322 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5323 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5324 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5325 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5326 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5327 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5328 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5329 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5330 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5331 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5333 * expr.c (emit_move_resolve_push): Treat the input and result
5334 of PUSH_ROUNDING as a poly_int64.
5335 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5336 (emit_push_insn): Likewise.
5337 * lra-eliminations.c (mark_not_eliminable): Likewise.
5338 * recog.c (push_operand): Likewise.
5339 * reload1.c (elimination_effects): Likewise.
5340 * rtlanal.c (nonzero_bits1): Likewise.
5341 * calls.c (store_one_arg): Likewise. Require the padding to be
5342 known at compile time.
5344 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5345 Alan Hayward <alan.hayward@arm.com>
5346 David Sherwood <david.sherwood@arm.com>
5348 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5349 Use plus_constant instead of gen_rtx_PLUS.
5351 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5352 Alan Hayward <alan.hayward@arm.com>
5353 David Sherwood <david.sherwood@arm.com>
5355 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5358 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5359 Alan Hayward <alan.hayward@arm.com>
5360 David Sherwood <david.sherwood@arm.com>
5362 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5363 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5364 via stack temporaries. Treat the mode size as polynomial too.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * expr.c (expand_expr_real_2): When handling conversions involving
5371 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5372 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5373 as a poly_uint64 too.
5375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5376 Alan Hayward <alan.hayward@arm.com>
5377 David Sherwood <david.sherwood@arm.com>
5379 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5382 Alan Hayward <alan.hayward@arm.com>
5383 David Sherwood <david.sherwood@arm.com>
5385 * combine.c (can_change_dest_mode): Handle polynomial
5386 REGMODE_NATURAL_SIZE.
5387 * expmed.c (store_bit_field_1): Likewise.
5388 * expr.c (store_constructor): Likewise.
5389 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5390 and polynomial REGMODE_NATURAL_SIZE.
5391 (gen_lowpart_common): Likewise.
5392 * reginfo.c (record_subregs_of_mode): Likewise.
5393 * rtlanal.c (read_modify_subreg_p): Likewise.
5395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5396 Alan Hayward <alan.hayward@arm.com>
5397 David Sherwood <david.sherwood@arm.com>
5399 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5400 numbers of elements.
5402 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5403 Alan Hayward <alan.hayward@arm.com>
5404 David Sherwood <david.sherwood@arm.com>
5406 * match.pd: Cope with polynomial numbers of vector elements.
5408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5409 Alan Hayward <alan.hayward@arm.com>
5410 David Sherwood <david.sherwood@arm.com>
5412 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5413 in a POINTER_PLUS_EXPR.
5415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5416 Alan Hayward <alan.hayward@arm.com>
5417 David Sherwood <david.sherwood@arm.com>
5419 * omp-simd-clone.c (simd_clone_subparts): New function.
5420 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5421 (ipa_simd_modify_function_body): Likewise.
5423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5424 Alan Hayward <alan.hayward@arm.com>
5425 David Sherwood <david.sherwood@arm.com>
5427 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5428 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5429 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5430 (expand_vector_condition, vector_element): Likewise.
5431 (subparts_gt): New function.
5432 (get_compute_type): Use subparts_gt.
5433 (count_type_subparts): Delete.
5434 (expand_vector_operations_1): Use subparts_gt instead of
5435 count_type_subparts.
5437 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5438 Alan Hayward <alan.hayward@arm.com>
5439 David Sherwood <david.sherwood@arm.com>
5441 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5442 (vect_compile_time_alias): ...this new function. Do the calculation
5443 on poly_ints rather than trees.
5444 (vect_prune_runtime_alias_test_list): Update call accordingly.
5446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5447 Alan Hayward <alan.hayward@arm.com>
5448 David Sherwood <david.sherwood@arm.com>
5450 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5452 (vect_schedule_slp_instance): Likewise.
5454 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5455 Alan Hayward <alan.hayward@arm.com>
5456 David Sherwood <david.sherwood@arm.com>
5458 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5459 constant and extern definitions for variable-length vectors.
5460 (vect_get_constant_vectors): Note that the number of units
5461 is known to be constant.
5463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5464 Alan Hayward <alan.hayward@arm.com>
5465 David Sherwood <david.sherwood@arm.com>
5467 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5468 of units as polynomial. Choose between WIDE and NARROW based
5471 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5472 Alan Hayward <alan.hayward@arm.com>
5473 David Sherwood <david.sherwood@arm.com>
5475 * tree-vect-stmts.c (simd_clone_subparts): New function.
5476 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5479 Alan Hayward <alan.hayward@arm.com>
5480 David Sherwood <david.sherwood@arm.com>
5482 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5483 vectors as polynomial. Use build_index_vector for
5486 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5487 Alan Hayward <alan.hayward@arm.com>
5488 David Sherwood <david.sherwood@arm.com>
5490 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5491 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5492 for variable-length vectors.
5493 (vectorizable_mask_load_store): Treat the number of units as
5494 polynomial, asserting that it is constant if the condition has
5495 already been enforced.
5496 (vectorizable_store, vectorizable_load): Likewise.
5498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5499 Alan Hayward <alan.hayward@arm.com>
5500 David Sherwood <david.sherwood@arm.com>
5502 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5503 of units as polynomial. Punt if we can't tell at compile time
5504 which vector contains the final result.
5506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5507 Alan Hayward <alan.hayward@arm.com>
5508 David Sherwood <david.sherwood@arm.com>
5510 * tree-vect-loop.c (vectorizable_induction): Treat the number
5511 of units as polynomial. Punt on SLP inductions. Use an integer
5512 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5513 cast of such a series for variable-length floating-point
5516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5517 Alan Hayward <alan.hayward@arm.com>
5518 David Sherwood <david.sherwood@arm.com>
5520 * tree.h (build_index_vector): Declare.
5521 * tree.c (build_index_vector): New function.
5522 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5523 of units as polynomial, forcibly converting it to a constant if
5524 vectorizable_reduction has already enforced the condition.
5525 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5526 to create a {1,2,3,...} vector.
5527 (vectorizable_reduction): Treat the number of units as polynomial.
5528 Choose vectype_in based on the largest scalar element size rather
5529 than the smallest number of units. Enforce the restrictions
5532 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5533 Alan Hayward <alan.hayward@arm.com>
5534 David Sherwood <david.sherwood@arm.com>
5536 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5537 number of units as polynomial.
5539 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5540 Alan Hayward <alan.hayward@arm.com>
5541 David Sherwood <david.sherwood@arm.com>
5543 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5544 * target.def (autovectorize_vector_sizes): Return the vector sizes
5545 by pointer, using vector_sizes rather than a bitmask.
5546 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5547 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5548 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5550 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5551 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5552 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5553 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5554 * omp-general.c (omp_max_vf): Likewise.
5555 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5556 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5557 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5558 * tree-vect-slp.c (vect_slp_bb): Likewise.
5559 * doc/tm.texi: Regenerate.
5560 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5562 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5563 the vector size as a poly_uint64 rather than an unsigned int.
5564 (current_vector_size): Change from an unsigned int to a poly_uint64.
5565 (get_vectype_for_scalar_type): Update accordingly.
5566 * tree.h (build_truth_vector_type): Take the size and number of
5567 units as a poly_uint64 rather than an unsigned int.
5568 (build_vector_type): Add a temporary overload that takes
5569 the number of units as a poly_uint64 rather than an unsigned int.
5570 * tree.c (make_vector_type): Likewise.
5571 (build_truth_vector_type): Take the number of units as a poly_uint64
5572 rather than an unsigned int.
5574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5575 Alan Hayward <alan.hayward@arm.com>
5576 David Sherwood <david.sherwood@arm.com>
5578 * target.def (get_mask_mode): Take the number of units and length
5579 as poly_uint64s rather than unsigned ints.
5580 * targhooks.h (default_get_mask_mode): Update accordingly.
5581 * targhooks.c (default_get_mask_mode): Likewise.
5582 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5583 * doc/tm.texi: Regenerate.
5585 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5586 Alan Hayward <alan.hayward@arm.com>
5587 David Sherwood <david.sherwood@arm.com>
5589 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5590 * omp-general.c (omp_max_vf): Likewise.
5591 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5592 (expand_omp_simd): Handle polynomial safelen.
5593 * omp-low.c (omplow_simd_context): Add a default constructor.
5594 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5595 (lower_rec_simd_input_clauses): Update accordingly.
5596 (lower_rec_input_clauses): Likewise.
5598 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5599 Alan Hayward <alan.hayward@arm.com>
5600 David Sherwood <david.sherwood@arm.com>
5602 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5603 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5604 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5605 (vect_analyze_slp_cost): Likewise.
5606 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5607 (vect_model_load_cost): Likewise.
5609 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5610 Alan Hayward <alan.hayward@arm.com>
5611 David Sherwood <david.sherwood@arm.com>
5613 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5614 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5615 from an unsigned int * to a poly_uint64_pod *.
5616 (calculate_unrolling_factor): New function.
5617 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5620 Alan Hayward <alan.hayward@arm.com>
5621 David Sherwood <david.sherwood@arm.com>
5623 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5624 from an unsigned int to a poly_uint64.
5625 (_loop_vec_info::slp_unrolling_factor): Likewise.
5626 (_loop_vec_info::vectorization_factor): Change from an int
5628 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5629 (vect_get_num_vectors): New function.
5630 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5631 (vect_get_num_copies): Use vect_get_num_vectors.
5632 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5633 to an unsigned int *.
5634 (vect_analyze_data_refs): Change min_vf from an int * to a
5636 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5637 than an unsigned HOST_WIDE_INT.
5638 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5639 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5640 to an unsigned int *.
5641 (vect_analyze_data_ref_dependences): Likewise.
5642 (vect_compute_data_ref_alignment): Handle polynomial vf.
5643 (vect_enhance_data_refs_alignment): Likewise.
5644 (vect_prune_runtime_alias_test_list): Likewise.
5645 (vect_shift_permute_load_chain): Likewise.
5646 (vect_supportable_dr_alignment): Likewise.
5647 (dependence_distance_ge_vf): Take the vectorization factor as a
5648 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5649 (vect_analyze_data_refs): Change min_vf from an int * to a
5651 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5652 vfm1 as a poly_uint64 rather than an int. Make the same change
5653 for the returned bound_scalar.
5654 (vect_gen_vector_loop_niters): Handle polynomial vf.
5655 (vect_do_peeling): Likewise. Update call to
5656 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5657 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5659 * tree-vect-loop.c (vect_determine_vectorization_factor)
5660 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5661 (vect_get_known_peeling_cost): Likewise.
5662 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5663 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5664 (vect_transform_loop): Likewise. Use the lowest possible VF when
5665 updating the upper bounds of the loop.
5666 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5668 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5669 polynomial unroll factors.
5670 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5671 (vect_make_slp_decision): Likewise.
5672 (vect_supported_load_permutation_p): Likewise, and polynomial
5674 (vect_analyze_slp_cost): Handle polynomial vf.
5675 (vect_slp_analyze_node_operations): Likewise.
5676 (vect_slp_analyze_bb_1): Likewise.
5677 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5678 than an unsigned HOST_WIDE_INT.
5679 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5680 (vectorizable_load): Handle polynomial vf.
5681 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5683 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5685 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5686 Alan Hayward <alan.hayward@arm.com>
5687 David Sherwood <david.sherwood@arm.com>
5689 * match.pd: Handle bit operations involving three constants
5690 and try to fold one pair.
5692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5694 * tree-vect-loop-manip.c: Include gimple-fold.h.
5695 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5696 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5697 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5698 Add a path that uses a step of VF instead of 1, but disable it
5700 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5701 and niters_no_overflow parameters. Update calls to
5702 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5703 Create a new SSA name if the latter choses to use a ste other
5704 than zero, and return it via niters_vector_mult_vf_var.
5705 * tree-vect-loop.c (vect_transform_loop): Update calls to
5706 vect_do_peeling, vect_gen_vector_loop_niters and
5707 slpeel_make_loop_iterate_ntimes.
5708 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5709 (vect_gen_vector_loop_niters): Update declarations after above changes.
5711 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5713 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5714 128-bit round to integer instructions.
5715 (ceil<mode>2): Likewise.
5716 (btrunc<mode>2): Likewise.
5717 (round<mode>2): Likewise.
5719 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5721 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5722 unaligned VSX load/store on P8/P9.
5723 (expand_block_clear): Allow the use of unaligned VSX
5724 load/store on P8/P9.
5726 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5728 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5730 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5731 swap associated with both a load and a store.
5733 2018-01-02 Andrew Waterman <andrew@sifive.com>
5735 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5736 * config/riscv/riscv.md (clear_cache): Use it.
5738 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5740 * web.c: Remove out-of-date comment.
5742 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5744 * expr.c (fixup_args_size_notes): Check that any existing
5745 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5746 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5747 (emit_single_push_insn): ...here.
5749 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5751 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5752 (const_vector_encoded_nelts): New function.
5753 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5754 (const_vector_int_elt, const_vector_elt): Declare.
5755 * emit-rtl.c (const_vector_int_elt_1): New function.
5756 (const_vector_elt): Likewise.
5757 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5758 of CONST_VECTOR_ELT.
5760 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5762 * expr.c: Include rtx-vector-builder.h.
5763 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5764 directly on the tree encoding.
5765 (const_vector_from_tree): Likewise.
5766 * optabs.c: Include rtx-vector-builder.h.
5767 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5768 sequence of "u" values.
5769 * vec-perm-indices.c: Include rtx-vector-builder.h.
5770 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5771 directly on the vec_perm_indices encoding.
5773 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5775 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5776 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5777 * rtx-vector-builder.h: New file.
5778 * rtx-vector-builder.c: Likewise.
5779 * rtl.h (rtx_def::u2): Add a const_vector field.
5780 (CONST_VECTOR_NPATTERNS): New macro.
5781 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5782 (CONST_VECTOR_DUPLICATE_P): Likewise.
5783 (CONST_VECTOR_STEPPED_P): Likewise.
5784 (CONST_VECTOR_ENCODED_ELT): Likewise.
5785 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5786 (unwrap_const_vec_duplicate): Likewise.
5787 (const_vec_series_p): Check for a non-duplicated vector encoding.
5788 Say that the function only returns true for integer vectors.
5789 * emit-rtl.c: Include rtx-vector-builder.h.
5790 (gen_const_vec_duplicate_1): Delete.
5791 (gen_const_vector): Call gen_const_vec_duplicate instead of
5792 gen_const_vec_duplicate_1.
5793 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5794 (gen_const_vec_duplicate): Use rtx_vector_builder.
5795 (gen_const_vec_series): Likewise.
5796 (gen_rtx_CONST_VECTOR): Likewise.
5797 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5798 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5799 Build a new vector rather than modifying a CONST_VECTOR in-place.
5800 (handle_special_swappables): Update call accordingly.
5801 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5802 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5803 Build a new vector rather than modifying a CONST_VECTOR in-place.
5804 (handle_special_swappables): Update call accordingly.
5806 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5808 * simplify-rtx.c (simplify_const_binary_operation): Use
5809 CONST_VECTOR_ELT instead of XVECEXP.
5811 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5813 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5814 the selector elements to be different from the data elements
5815 if the selector is a VECTOR_CST.
5816 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5817 ssizetype for the selector.
5819 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5821 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5822 before testing each element individually.
5823 * tree-vect-generic.c (lower_vec_perm): Likewise.
5825 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5827 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5828 * selftest-run-tests.c (selftest::run_tests): Call it.
5829 * vector-builder.h (vector_builder::operator ==): New function.
5830 (vector_builder::operator !=): Likewise.
5831 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5832 (vec_perm_indices::all_from_input_p): New function.
5833 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5834 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5835 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5836 instead of reading the VECTOR_CST directly. Detect whether both
5837 vector inputs are the same before constructing the vec_perm_indices,
5838 and update the number of inputs argument accordingly. Use the
5839 utility functions added above. Only construct sel2 if we need to.
5841 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5843 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5844 the broadcast of the low byte.
5845 (expand_mult_highpart): Use an explicit encoding for the permutes.
5846 * optabs-query.c (can_mult_highpart_p): Likewise.
5847 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5848 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5849 (vectorizable_bswap): Likewise.
5850 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5851 explicit encoding for the power-of-2 permutes.
5852 (vect_permute_store_chain): Likewise.
5853 (vect_grouped_load_supported): Likewise.
5854 (vect_permute_load_chain): Likewise.
5856 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5858 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5859 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5860 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5861 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5862 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5863 (vect_gen_perm_mask_any): Likewise.
5865 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5867 * int-vector-builder.h: New file.
5868 * vec-perm-indices.h: Include int-vector-builder.h.
5869 (vec_perm_indices): Redefine as an int_vector_builder.
5870 (auto_vec_perm_indices): Delete.
5871 (vec_perm_builder): Redefine as a stand-alone class.
5872 (vec_perm_indices::vec_perm_indices): New function.
5873 (vec_perm_indices::clamp): Likewise.
5874 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5875 (vec_perm_indices::new_vector): New function.
5876 (vec_perm_indices::new_expanded_vector): Update for new
5877 vec_perm_indices class.
5878 (vec_perm_indices::rotate_inputs): New function.
5879 (vec_perm_indices::all_in_range_p): Operate directly on the
5880 encoded form, without computing elided elements.
5881 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5882 encoding. Update for new vec_perm_indices class.
5883 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5884 the given vec_perm_builder.
5885 (expand_vec_perm_var): Update vec_perm_builder constructor.
5886 (expand_mult_highpart): Use vec_perm_builder instead of
5887 auto_vec_perm_indices.
5888 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5889 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5890 or double series encoding as appropriate.
5891 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5892 vec_perm_indices instead of auto_vec_perm_indices.
5893 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5894 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5895 (vect_permute_store_chain): Likewise.
5896 (vect_grouped_load_supported): Likewise.
5897 (vect_permute_load_chain): Likewise.
5898 (vect_shift_permute_load_chain): Likewise.
5899 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5900 (vect_transform_slp_perm_load): Likewise.
5901 (vect_schedule_slp_instance): Likewise.
5902 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5903 (vectorizable_mask_load_store): Likewise.
5904 (vectorizable_bswap): Likewise.
5905 (vectorizable_store): Likewise.
5906 (vectorizable_load): Likewise.
5907 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5908 vec_perm_indices instead of auto_vec_perm_indices. Use
5909 tree_to_vec_perm_builder to read the vector from a tree.
5910 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5911 vec_perm_builder instead of a vec_perm_indices.
5912 (have_whole_vector_shift): Use vec_perm_builder and
5913 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5914 truncation to calc_vec_perm_mask_for_shift.
5915 (vect_create_epilog_for_reduction): Likewise.
5916 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5917 from auto_vec_perm_indices to vec_perm_indices.
5918 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5919 instead of changing individual elements.
5920 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5921 the vector in d.perm.
5922 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5923 from auto_vec_perm_indices to vec_perm_indices.
5924 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5925 instead of changing individual elements.
5926 (arm_vectorize_vec_perm_const): Use new_vector to install
5927 the vector in d.perm.
5928 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5929 Update vec_perm_builder constructor.
5930 (rs6000_expand_interleave): Likewise.
5931 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5932 (rs6000_expand_interleave): Likewise.
5934 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5936 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5937 to qimode could truncate the indices.
5938 * optabs.c (expand_vec_perm_var): Likewise.
5940 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5942 * Makefile.in (OBJS): Add vec-perm-indices.o.
5943 * vec-perm-indices.h: New file.
5944 * vec-perm-indices.c: Likewise.
5945 * target.h (vec_perm_indices): Replace with a forward class
5947 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5948 * optabs.h: Include vec-perm-indices.h.
5949 (expand_vec_perm): Delete.
5950 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5951 (expand_vec_perm_const): Declare.
5952 * target.def (vec_perm_const_ok): Replace with...
5953 (vec_perm_const): ...this new hook.
5954 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5955 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5956 * doc/tm.texi: Regenerate.
5957 * optabs.def (vec_perm_const): Delete.
5958 * doc/md.texi (vec_perm_const): Likewise.
5959 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5960 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5961 expand_vec_perm for constant permutation vectors. Assert that
5962 the mode of variable permutation vectors is the integer equivalent
5963 of the mode that is being permuted.
5964 * optabs-query.h (selector_fits_mode_p): Declare.
5965 * optabs-query.c: Include vec-perm-indices.h.
5966 (selector_fits_mode_p): New function.
5967 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5968 is defined, instead of checking whether the vec_perm_const_optab
5969 exists. Use targetm.vectorize.vec_perm_const instead of
5970 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5971 fit in the vector mode before using a variable permute.
5972 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5973 vec_perm_indices instead of an rtx.
5974 (expand_vec_perm): Replace with...
5975 (expand_vec_perm_const): ...this new function. Take the selector
5976 as a vec_perm_indices rather than an rtx. Also take the mode of
5977 the selector. Update call to shift_amt_for_vec_perm_mask.
5978 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5979 Use vec_perm_indices::new_expanded_vector to expand the original
5980 selector into bytes. Check whether the indices fit in the vector
5981 mode before using a variable permute.
5982 (expand_vec_perm_var): Make global.
5983 (expand_mult_highpart): Use expand_vec_perm_const.
5984 * fold-const.c: Includes vec-perm-indices.h.
5985 * tree-ssa-forwprop.c: Likewise.
5986 * tree-vect-data-refs.c: Likewise.
5987 * tree-vect-generic.c: Likewise.
5988 * tree-vect-loop.c: Likewise.
5989 * tree-vect-slp.c: Likewise.
5990 * tree-vect-stmts.c: Likewise.
5991 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5993 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5994 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5995 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5996 (aarch64_vectorize_vec_perm_const): ...this new function.
5997 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5998 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5999 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6000 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6001 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6002 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6003 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6005 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6006 check for NEON modes.
6007 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6008 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6009 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6010 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6012 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6013 the old VEC_PERM_CONST conditions.
6014 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6015 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6016 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6017 (ia64_vectorize_vec_perm_const_ok): Merge into...
6018 (ia64_vectorize_vec_perm_const): ...this new function.
6019 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6020 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6021 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6022 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6023 * config/mips/mips.c (mips_expand_vec_perm_const)
6024 (mips_vectorize_vec_perm_const_ok): Merge into...
6025 (mips_vectorize_vec_perm_const): ...this new function.
6026 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6027 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6028 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6029 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6030 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6031 (rs6000_expand_vec_perm_const): Delete.
6032 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6034 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6035 (altivec_expand_vec_perm_const_le): Take each operand individually.
6036 Operate on constant selectors rather than rtxes.
6037 (altivec_expand_vec_perm_const): Likewise. Update call to
6038 altivec_expand_vec_perm_const_le.
6039 (rs6000_expand_vec_perm_const): Delete.
6040 (rs6000_vectorize_vec_perm_const_ok): Delete.
6041 (rs6000_vectorize_vec_perm_const): New function.
6042 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6043 an element count and rtx array.
6044 (rs6000_expand_extract_even): Update call accordingly.
6045 (rs6000_expand_interleave): Likewise.
6046 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6047 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6048 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6049 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6050 (rs6000_expand_vec_perm_const): Delete.
6051 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6052 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6053 (altivec_expand_vec_perm_const_le): Take each operand individually.
6054 Operate on constant selectors rather than rtxes.
6055 (altivec_expand_vec_perm_const): Likewise. Update call to
6056 altivec_expand_vec_perm_const_le.
6057 (rs6000_expand_vec_perm_const): Delete.
6058 (rs6000_vectorize_vec_perm_const_ok): Delete.
6059 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6060 reference to the SPE evmerge intructions.
6061 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6062 an element count and rtx array.
6063 (rs6000_expand_extract_even): Update call accordingly.
6064 (rs6000_expand_interleave): Likewise.
6065 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6066 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6068 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6070 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6072 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6073 vector mode and that that mode matches the mode of the data
6075 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6076 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6077 directly using expand_vec_perm_1 when forcing selectors into
6079 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6081 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6083 * optabs-query.h (can_vec_perm_p): Delete.
6084 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6085 * optabs-query.c (can_vec_perm_p): Split into...
6086 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6087 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6088 particular selector is valid.
6089 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6090 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6091 (vect_grouped_load_supported): Likewise.
6092 (vect_shift_permute_load_chain): Likewise.
6093 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6094 (vect_transform_slp_perm_load): Likewise.
6095 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6096 (vectorizable_bswap): Likewise.
6097 (vect_gen_perm_mask_checked): Likewise.
6098 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6099 implementations of variable permutation vectors into account
6100 when deciding which selector to use.
6101 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6102 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6103 with a false third argument.
6104 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6105 to test whether the constant selector is valid and can_vec_perm_var_p
6106 to test whether a variable selector is valid.
6108 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6110 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6111 * optabs-query.c (can_vec_perm_p): Likewise.
6112 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6113 instead of vec_perm_indices.
6114 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6115 (vect_gen_perm_mask_checked): Likewise,
6116 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6117 (vect_gen_perm_mask_checked): Likewise,
6119 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6121 * optabs-query.h (qimode_for_vec_perm): Declare.
6122 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6123 (qimode_for_vec_perm): ...this new function.
6124 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6126 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6128 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6129 does not have a conditional at the top.
6131 2018-01-02 Richard Biener <rguenther@suse.de>
6133 * ipa-inline.c (big_speedup_p): Fix expression.
6135 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6138 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6141 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6145 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6146 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6147 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6148 cond_taken_branch_cost 3->4.
6150 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6152 PR tree-optimization/83581
6153 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6154 TODO_cleanup_cfg if any changes have been made.
6157 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6158 convert_modes if target mode has the right side, but different mode
6162 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6163 last argument when extracting from CONCAT. If either from_real or
6164 from_imag is NULL, use expansion through memory. If result is not
6165 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6166 the parts directly to inner mode, if even that fails, use expansion
6170 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6171 check for bswap in mode rather than HImode and use that in expand_unop
6174 Copyright (C) 2018 Free Software Foundation, Inc.
6176 Copying and distribution of this file, with or without modification,
6177 are permitted in any medium without royalty provided the copyright
6178 notice and this notice are preserved.