1 2018-03-07 Jakub Jelinek <jakub@redhat.com>
4 * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
5 aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
8 * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
10 * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
11 don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
14 * multiple_target.c: Include tree-inline.h and intl.h.
15 (expand_target_clones): Diagnose and fail if node->definition and
16 !tree_versionable_function_p (node->decl).
18 2018-03-06 John David Anglin <danglin@gcc.gnu.org>
20 * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
22 (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
23 (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
24 * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
26 2018-03-06 Jakub Jelinek <jakub@redhat.com>
29 * combine.c (try_combine): Use reg_or_subregno instead of handling
30 just paradoxical SUBREGs and REGs.
32 2018-03-06 Claudiu Zissulescu <claziss@synopsys.com>
34 * config/arc/arc.c (arc_finalize_pic): Remove function.
35 (arc_must_save_register): We use single base PIC register, remove
36 checks to save/restore the PIC register.
37 (arc_expand_prologue): Likewise.
38 * config/arc/arc-protos.h (arc_set_default_type_attributes):
40 (arc_verify_short): Likewise.
41 (arc_attr_type): Likewise.
42 * config/arc/arc.c (arc_set_default_type_attributes): Remove.
43 (walk_stores): Likewise.
44 (arc_address_cost): Make it static.
45 (arc_verify_short): Likewise.
46 (branch_dest): Likewise.
47 (arc_attr_type): Likewise.
48 * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
49 (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
50 (arc_final_prescan_insn): Remove inserting the nops due to
51 hardware hazards. It is done in reorg step.
52 (insn_length_variant_t): Remove.
53 (insn_length_parameters_t): Likewise.
54 (arc_insn_length_parameters): Likewise.
55 (arc_get_insn_variants): Likewise.
56 * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
58 2018-03-06 Jakub Jelinek <jakub@redhat.com>
61 * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
64 PR tree-optimization/84687
65 * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
67 * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
69 2018-03-05 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
71 * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
72 Rename to ppc_speculation_barrier.
73 * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
74 __builtin_ppc_speculation_barrier.
76 2018-03-05 Jakub Jelinek <jakub@redhat.com>
79 * combine.c (combine_simplify_rtx): Don't try to simplify if
80 if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
83 2018-03-05 Segher Boessenkool <segher@kernel.crashing.org>
85 * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
86 to 32 bytes when compiling for POWER9.
88 2018-03-05 Jakub Jelinek <jakub@redhat.com>
91 * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
92 regparm >= 3 with no arg reg available also for calls with
93 flag_force_indirect_call. Pass decl to ix86_function_regparm.
96 * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
98 (*<plusminus_insn><mode>3): Likewise. Remove <mask_operand3> uses.
100 2018-03-05 Peter Bergner <bergner@vnet.ibm.com>
103 * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
105 2018-03-05 Richard Biener <rguenther@suse.de>
107 PR tree-optimization/84486
108 * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
109 When inserting a __builtin_assume_aligned call set the LHS
110 SSA name alignment info accordingly.
112 2018-03-05 Wilco Dijkstra <wdijkstr@arm.com>
114 PR tree-optimization/84114
115 * config/aarch64/aarch64.c (aarch64_reassociation_width)
116 Avoid reassociation of FLOAT_MODE addition.
118 2018-03-05 Olga Makhotina <olga.makhotina@intel.com>
120 * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
121 OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
122 OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
123 (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
124 * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
125 * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
126 * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
128 * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
129 __builtin_ia32_wbinvd): New builtins.
130 (SPECIAL_ARGS2): New.
131 * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
132 (SPECIAL_ARGS2): New.
133 * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
134 (ix86_valid_target_attribute_inner_p): Ditto.
135 (ix86_init_mmx_sse_builtins): Add special_args2.
136 * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
137 TARGET_WBNOINVD_P): New.
138 * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
139 (define_insn "wbinvd", define_insn "wbnoinvd"): New.
140 * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
141 * config/i386/immintrin.h (_wbinvd): New intrinsic.
142 * config/i386/pconfigintrin.h: New file.
143 * config/i386/wbnoinvdintrin.h: Ditto.
144 * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
145 * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
147 2018-03-05 Richard Biener <rguenther@suse.de>
149 PR tree-optimization/84670
150 * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
152 (BB_VISITED_WITH_VISITED_SUCCS): New define.
153 (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
154 (compute_antic_aux): Only assert the number of values in ANTIC_IN
155 doesn't grow if all successors (recursively) were visited at least
158 2018-03-05 Richard Biener <rguenther@suse.de>
160 PR tree-optimization/84650
161 * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
162 if executed in the loop pipeline.
164 2018-03-05 Sandra Loosemore <sandra@codesourcery.com>
166 * doc/configfiles.texi (Configuration Files): Move info about
167 conditionalizing $target-protos.h to...
168 * doc/sourcebuild.texi (Back End): Here. Explain how $target.h
169 differs from $target-protos.h.
171 2018-03-05 Kito Cheng <kito.cheng@gmail.com>
172 Chung-Ju Wu <jasonwucj@gmail.com>
174 * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
175 * config/nds32/nds32-multiple.md (setmemsi): Define.
176 * config/nds32/nds32-memory-manipulation.c
177 (nds32_gen_dup_4_byte_to_word_value): New.
178 (emit_setmem_word_loop): New.
179 (emit_setmem_byte_loop): New.
180 (nds32_expand_setmem_loop): New.
181 (nds32_expand_setmem_loop_v3m): New.
182 (nds32_expand_setmem_unroll): New.
183 (nds32_expand_setmem): New.
185 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
186 Chung-Ju Wu <jasonwucj@gmail.com>
188 * config/nds32/nds32-memory-manipulation.c
189 (nds32_emit_load_store): New.
190 (nds32_emit_post_inc_load_store): New.
191 (nds32_emit_mem_move): New.
192 (nds32_emit_mem_move_block): New.
193 (nds32_expand_movmemsi_loop_unknown_size): New.
194 (nds32_expand_movmemsi_loop_known_size): New.
195 (nds32_expand_movmemsi_loop): New.
196 (nds32_expand_movmemsi_unroll): New.
197 (nds32_expand_movmemqi): Rename ...
198 (nds32_expand_movmemsi): ... to this.
199 * config/nds32/nds32-multiple.md (movmemqi): Rename ...
200 (movmemsi): ... to this.
201 * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
202 (nds32_expand_movmemsi): ... to this.
204 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
205 Monk Chiang <sh.chiang04@gmail.com>
206 Chung-Ju Wu <jasonwucj@gmail.com>
208 * config/nds32/nds32-protos.h
209 (nds32_expand_load_multiple): New arguments.
210 (nds32_expand_store_multiple): Ditto.
211 (nds32_valid_multiple_load_store): Rename ...
212 (nds32_valid_multiple_load_store_p): ... to this.
213 * config/nds32/nds32-memory-manipulation.c
214 (nds32_expand_load_multiple): Refine implementation.
215 (nds32_expand_store_multiple): Ditto.
216 * config/nds32/nds32-multiple.md
217 (load_multiple): Update nds32_expand_load_multiple interface.
218 (store_multiple): Update nds32_expand_store_multiple interface.
219 * config/nds32/nds32-predicates.c
220 (nds32_valid_multiple_load_store): Rename ...
221 (nds32_valid_multiple_load_store_p): ... to this and refine
223 * config/nds32/predicates.md
224 (nds32_load_multiple_and_update_address_operation): New predicate.
225 (nds32_store_multiple_and_update_address_operation): New predicate.
227 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
228 Chung-Ju Wu <jasonwucj@gmail.com>
230 * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
231 (combo): New attribute.
232 * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
234 2018-03-03 Chung-Ju Wu <jasonwucj@gmail.com>
236 * config/nds32/nds32.opt: Change -mcmodel= default value.
238 2018-03-03 Kito Cheng <kito.cheng@gmail.com>
239 Monk Chiang <sh.chiang04@gmail.com>
240 Chung-Ju Wu <jasonwucj@gmail.com>
242 * config/nds32/constants.md (unspec_element): New enum.
243 * config/nds32/constraints.md (Umw): New constraint.
244 * config/nds32/nds32-intrinsic.c: Add more builtin functions.
245 * config/nds32/nds32-intrinsic.md: Likewise.
246 * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
247 (nds32_valid_smw_lwm_base_p): New.
248 (nds32_output_smw_single_word): New.
249 (nds32_output_lmw_single_word): New.
250 (nds32_expand_unaligned_load): New.
251 (nds32_expand_unaligned_store): New.
252 * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
253 (nds32_output_smw_single_word): Declare.
254 (nds32_output_lmw_single_word): Declare.
255 (nds32_expand_unaligned_load): Declare.
256 (nds32_expand_unaligned_store): Declare.
257 * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
258 NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
259 NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
260 NDS32_BUILTIN_UASTORE_DW.
261 * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
264 2018-03-03 Monk Chiang <sh.chiang04@gmail.com>
265 Kito Cheng <kito.cheng@gmail.com>
266 Chung-Ju Wu <jasonwucj@gmail.com>
268 * config/nds32/nds32-intrinsic.c
269 (nds32_expand_builtin_null_ftype_reg): Delete.
270 (nds32_expand_builtin_reg_ftype_imm): Ditto.
271 (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
272 (nds32_read_argument): New.
273 (nds32_legitimize_target): Ditto.
274 (nds32_legitimize_argument): Ditto.
275 (nds32_check_constant_argument): Ditto.
276 (nds32_expand_unop_builtin): Ditto.
277 (nds32_expand_unopimm_builtin): Ditto.
278 (nds32_expand_binop_builtin): Ditto.
279 (nds32_builtin_decl_impl): Ditto.
280 (builtin_description): Ditto.
281 (nds32_expand_builtin_impl): Rewrite with new infrastructure.
282 (nds32_init_builtins_impl): Ditto.
283 * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
284 (nds32_builtin_decl): New.
285 * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
286 * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
288 2018-03-02 Jeff Law <law@redhat.com>
290 * reorg.c (stop_search_p): Handle DEBUG_INSN.
291 (redundant_insn, fill_simple_delay_slots): Likewise.
292 (fill_slots_from_thread): Likewise.
293 * resource.c (mark_referenced_resources): Likewise.
294 (mark_set_resources, find_dead_or_set_registers): Likewise.
296 2018-03-02 Jakub Jelinek <jakub@redhat.com>
298 * substring-locations.h (format_warning_va): Formatting fix for
300 (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
302 (format_warning_n_va, format_warning_at_substring_n): New prototypes.
303 * substring-locations.c: Include intl.h.
304 (format_warning_va): Turned into small wrapper around
305 format_warning_n_va, renamed to ...
306 (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
307 rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
309 (format_warning_at_substring_n): New function.
310 * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
311 (fmtwarn): Add ATTRIBUTE_GCC_DIAG. Turn into a copy of
312 format_warning_at_substring with just a shorter name instead of
313 const function pointer.
314 (fmtwarn_n): New function.
315 (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
316 appropriate, get rid of all the fmtstr temporaries, move conditionals
317 with G_() wrapped string literals directly into fmtwarn arguments,
318 cast dir.len to (int), formatting fixes.
320 2018-03-02 Thomas Schwinge <thomas@codesourcery.com>
322 * doc/invoke.texi: Remove "Cilk Plus" references.
324 2018-03-02 Jakub Jelinek <jakub@redhat.com>
325 Richard Biener <rguenther@suse.de>
328 * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
329 for error or warning attributes if CALL_FROM_THUNK_P is set.
332 2018-03-02 Jakub Jelinek <jakub@redhat.com>
335 * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
336 __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
339 * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
340 __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
342 * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
343 instead of -1U in last predictors element's probability member.
345 2018-03-02 Eric Botcazou <ebotcazou@adacore.com>
348 * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
349 arguments if they are comparable.
351 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
353 PR tree-optimization/84634
354 * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
355 masks and masked_loop_p with a single loop_masks, making sure it's
356 null for bb vectorization.
358 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
360 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
361 (vect_analyze_data_ref_access): Use loop->safe_len rather than
362 loop->force_vectorize to check whether there is no alias.
364 2018-03-02 Jakub Jelinek <jakub@redhat.com>
367 * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
369 * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
371 (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
372 * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
373 instead of a loop around prev_real_insn.
374 * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
378 * config/i386/i386.c (ix86_print_operand): Use conditional
379 output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
382 2018-03-02 Richard Biener <rguenther@suse.de>
384 PR tree-optimization/84427
385 * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
386 (bitmap_set_subtract_values): Rewrite to handle multiple
389 (prune_clobbered_mems): Likewise.
390 (phi_translate): Take edge instead of pred/phiblock.
391 (phi_translate_1): Likewise.
392 (phi_translate_set): Likewise. Insert all translated
393 exprs for a value into the set, keeping possibly multiple
394 expressions per value.
395 (compute_antic_aux): Adjust for phi_translate changes.
396 When intersecting union the expressions and prune those
397 not in the final value set, keeping possibly multiple
398 expressions per value. Do not use value-insertion
399 for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
400 all expressions. Add verification that the value-sets
401 only shrink during iteration.
402 (compute_partial_antic_aux): Adjust for the phi_translate changes.
403 (do_pre_regular_insertion): Likewise.
404 (do_pre_partial_partial_insertion): Likewise.
406 2018-03-02 Richard Biener <rguenther@suse.de>
409 * config/darwin.c (saved_debug_info_level): New static global.
410 (darwin_asm_lto_start): Disable debug info generation for LTO out.
411 (darwin_asm_lto_end): Restore debug info generation settings.
413 2018-03-01 Martin Liska <mliska@suse.cz>
416 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
419 2018-03-01 Richard Biener <rguenther@suse.de>
422 * dwarf2out.c (gen_variable_die): Properly handle late VLA
423 type annotation with LTO when debug was disabled at compile-time.
425 2018-03-01 Matthew Fortune <mfortune@gmail.com>
427 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
429 (mips_final_postscan_insn): Likewise.
431 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
433 PR rtl-optimization/84528
434 * alias.c (init_alias_target): Add commentary.
435 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
436 a unique base value if the frame pointer is not eliminated
437 to the stack pointer.
439 2018-03-01 Tom de Vries <tom@codesourcery.com>
441 PR rtl-optimization/83327
442 * lra-int.h (hard_regs_spilled_into): Declare.
443 * lra.c (hard_regs_spilled_into): Define.
444 (init_reg_info): Init hard_regs_spilled_into.
445 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
446 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
447 (process_bb_lives): Handle hard_regs_spilled_into.
448 (lra_create_live_ranges_1): Before doing liveness propagation, clear
449 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
451 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
453 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
454 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
455 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
456 * config/rs6000/aix72.h: New file.
458 2018-02-28 Jakub Jelinek <jakub@redhat.com>
460 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
461 instead of warning_at with conditional singular and plural messages
465 * stor-layout.c (update_alignment_for_field): For
466 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
467 && !DECL_PACKED (field), do the alignment update, just use
468 only desired_align instead of MAX (type_align, desired_align)
470 (place_field): Don't do known_align < desired_align handling
471 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
472 is non-NULL, instead do it after rli->prev_field handling and
473 only if not within a bitfield word. For DECL_PACKED (field)
474 use type_align of BITS_PER_UNIT.
476 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
478 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
479 superfluous parentheses and trailing spaces.
481 2018-02-28 Richard Biener <rguenther@suse.de>
483 PR tree-optimization/84584
484 * graphite-scop-detection.c (scop_detection::add_scop): Discard
485 SCoPs with fake exit edge.
487 2018-02-28 Martin Liska <mliska@suse.cz>
490 * timevar.c (timer::print): Fix format to properly print 100%
493 2018-02-28 Richard Biener <rguenther@suse.de>
496 * genmatch.c (capture_info::walk_match): Do not mark
497 captured expressions without operands as expr_p given
498 they act more like predicates and should be subject to
499 "lost tail" side-effect preserving.
501 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
503 PR rtl-optimization/81611
504 * auto-inc-dec.c (attempt_change): Move dead note from
505 mem_insn if it's the next use of regno
506 (find_address): Take address use of reg holding
507 non-incremented value. Add parm to limit search to the named
509 (merge_in_block): Attempt to use a mem insn that is the next
510 use of the original regno.
512 2018-02-27 Martin Sebor <msebor@redhat.com>
515 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
516 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
518 2018-02-27 Martin Sebor <msebor@redhat.com>
521 * diagnostic-core.h (warning_n, error_n, inform_n): Change
522 n argument to unsigned HOST_WIDE_INT.
523 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
524 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
525 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
526 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
528 2018-02-27 Richard Biener <rguenther@suse.de>
530 PR tree-optimization/84512
531 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
532 Do not use the estimate returned from record_stmt_cost for
533 the scalar iteration cost but sum properly using add_stmt_cost.
535 2018-02-27 Richard Biener <rguenther@suse.de>
537 PR tree-optimization/84466
538 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
539 Adjust last change to less strictly validate use operands.
541 2018-02-27 Martin Liska <mliska@suse.cz>
543 PR gcov-profile/84548
544 * gcov.c (process_file): Allow partial overlap and consider it
545 also as group functions.
546 (output_lines): Properly calculate range of lines for a group.
548 2018-02-27 Martin Liska <mliska@suse.cz>
550 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
551 'ggc' suffixes. Change first column width.
552 (timer::print): Fix formatting of the column.
554 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
556 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
557 preserve inline entry blocks for the sake of debug inline
558 entry point markers alone.
559 (remove_unused_locals): Suggest in comments a better place to
560 force the preservation of inline entry blocks that are
561 otherwise unused, but do not preserve them.
563 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
565 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
567 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
570 * config/i386/constraints.md (Bs): Replace
571 ix86_indirect_branch_register with
572 TARGET_INDIRECT_BRANCH_REGISTER.
574 * config/i386/i386.md (indirect_jump): Likewise.
575 (tablejump): Likewise.
576 (*sibcall_memory): Likewise.
577 (*sibcall_value_memory): Likewise.
578 Peepholes of indirect call and jump via memory: Likewise.
579 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
580 (*sibcall_value_GOT_32): Likewise.
581 * config/i386/predicates.md (indirect_branch_operand): Likewise.
582 (GOT_memory_operand): Likewise.
583 (call_insn_operand): Likewise.
584 (sibcall_insn_operand): Likewise.
585 (GOT32_symbol_operand): Likewise.
586 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
588 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
590 PR rtl-optimization/83496
591 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
592 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
593 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
594 redundant insn, if any.
595 (relax_delay_slots): Likewise.
596 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
598 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
600 PR tree-optimization/83965
601 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
602 that grouped statements are part of a reduction chain. Return
603 true if the statement is not marked as a reduction itself but
605 (vect_recog_dot_prod_pattern): Don't check whether the statement
606 is part of a group here.
607 (vect_recog_sad_pattern): Likewise.
608 (vect_recog_widen_sum_pattern): Likewise.
610 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
613 * final.c (rest_of_clean_state): Also look for calls inside sequences.
615 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
618 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
620 (ix86_output_indirect_function_return): New prototype.
621 (ix86_split_simple_return_pop_internal): Likewise.
622 * config/i386/i386.c (indirect_return_via_cx): New.
623 (indirect_return_via_cx_bnd): Likewise.
624 (indirect_thunk_name): Handle return va CX_REG.
625 (output_indirect_thunk_function): Create alias for
626 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
627 (ix86_output_indirect_jmp): Remove the bool argument.
628 (ix86_output_indirect_function_return): New function.
629 (ix86_split_simple_return_pop_internal): Likewise.
630 * config/i386/i386.md (*indirect_jump): Don't pass false
631 to ix86_output_indirect_jmp.
632 (*tablejump_1): Likewise.
633 (simple_return_pop_internal): Change it to define_insn_and_split.
634 Call ix86_split_simple_return_pop_internal to split it for
636 (simple_return_indirect_internal): Call
637 ix86_output_indirect_function_return instead of
638 ix86_output_indirect_jmp.
640 2018-02-26 Jakub Jelinek <jakub@redhat.com>
643 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
644 memset and value initialization afterwards.
646 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
648 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
650 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
653 * common/config/aarch64/aarch64-common.c
654 (aarch_option_optimization_table[]): Switch
655 off fomit-frame-pointer
657 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
658 Chung-Ju Wu <jasonwucj@gmail.com>
660 * config/nds32/nds32-multiple.md (load_multiple): Disallow
662 (store_multiple): Ditto.
664 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
666 * config.gcc: Add --with-cpu support for nds32 target.
667 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
668 * config/nds32/nds32.opt: Add -mcpu= option.
670 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
672 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
673 isel=yes): Warn for these deprecated options.
675 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
677 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
678 ISA_2_5_MASKS_EMBEDDED.
680 2018-02-23 Jakub Jelinek <jakub@redhat.com>
682 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
683 p->max as pointers rather than using iterative_hash_expr.
685 2018-02-23 Carl Love <cel@us.ibm.com>
687 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
688 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
690 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
691 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
692 P8V_BUILTIN_VEC_VUNSIGNED2.
694 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
697 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
698 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
700 * lra-constraints.c (curr_insn_transform): Set up
701 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
703 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
704 * lra-eliminations.c (spill_pseudos): Ditto.
705 (process_insn_for_elimination): Ditto.
706 * lra-lives.c (reg_early_clobber_p): Use the new macros.
707 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
708 LRA_NON_CLOBBERED_ALT.
710 2018-02-22 Martin Sebor <msebor@redhat.com>
712 PR tree-optimization/84480
713 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
714 to maybe_diag_stxncpy_trunc. Call it.
715 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
716 from gimple_fold_builtin_strcpy. Print inlining stack.
717 (handle_builtin_stxncpy): Print inlining stack.
718 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
720 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
723 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
724 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
725 and -fcheck-pointer-bounds are used together.
726 (indirect_thunk_prefix): New enum.
727 (indirect_thunk_need_prefix): New function.
728 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
729 "_nt" instead of "_bnd" for NOTRACK prefix.
730 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
731 (output_indirect_thunk_function): Likewise.
733 (ix86_code_end): Update output_indirect_thunk_function calls.
734 (ix86_output_indirect_branch_via_reg): Replace
735 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
736 (ix86_output_indirect_branch_via_push): Likewise.
737 (ix86_output_function_return): Likewise.
738 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
739 incompatible with -fcf-protection=branch and
740 -fcheck-pointer-bounds.
742 2018-02-22 Steve Ellcey <sellcey@cavium.com>
745 * config/aarch64/aarch64.c (aarch64_print_address_internal):
746 Change gcc_assert call to output_operand_lossage.
748 2018-02-22 Steve Ellcey <sellcey@cavium.com>
750 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
752 2018-02-22 DJ Delorie <dj@redhat.com>
753 Sebastian Perta <sebastian.perta@renesas.com>
754 Oleg Endo <olegendo@gcc.gnu.org>
756 * config/rx/rx.c (rx_rtx_costs): New function.
757 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
759 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
761 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
763 2018-02-22 Martin Liska <mliska@suse.cz>
766 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
767 Add "native" as a possible value.
769 2018-02-22 Martin Liska <mliska@suse.cz>
772 * config/i386/i386.c (ix86_option_override_internal):
773 Add "native" as a possible value for -march and -mtune.
775 2018-02-22 Jakub Jelinek <jakub@redhat.com>
778 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
779 to all type variants.
781 PR tree-optimization/84503
782 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
783 width as info->bitpos + info->bitsize - start.
784 (merged_store_group::merge_overlapping): Simplify width computation.
785 (check_no_overlap): New function.
786 (imm_store_chain_info::try_coalesce_bswap): Compute expected
787 start + width and last_order of the group, fail if check_no_overlap
789 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
790 to group if check_no_overlap fails.
792 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
794 * config/rs6000/altivec.md: Delete contraint arguments to
795 define_expand, define_split, and define_peephole2, and in
796 define_insn_and_split if always unused.
797 * config/rs6000/darwin.md: Ditto.
798 * config/rs6000/dfp.md: Ditto.
799 * config/rs6000/rs6000.md: Ditto.
800 * config/rs6000/sync.md: Ditto.
801 * config/rs6000/vector.md: Ditto.
802 * config/rs6000/vsx.md: Ditto.
804 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
806 * config/rs6000/altivec.md: Write output control strings as braced
807 blocks instead of double-quoted strings.
808 * config/rs6000/darwin.md: Ditto.
809 * config/rs6000/rs6000.md: Ditto.
810 * config/rs6000/vector.md: Ditto.
811 * config/rs6000/vsx.md: Ditto.
813 2018-02-21 Jason Merrill <jason@redhat.com>
815 PR c++/84314 - ICE with templates and fastcall attribute.
816 * attribs.c (build_type_attribute_qual_variant): Remove assert.
818 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
820 * ipa-cp.c (determine_versionability): Fix comment typos.
822 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
825 * ipa-cp.c (determine_versionability): Do not version functions caling
828 2018-02-21 Martin Liska <mliska@suse.cz>
831 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
832 Add "native" as a possible value.
833 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
834 the macro when native cpu detection is available.
836 2018-02-21 Martin Liska <mliska@suse.cz>
839 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
840 Add "native" as a possible value.
841 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
842 when native cpu detection is available.
844 2018-02-21 Jakub Jelinek <jakub@redhat.com>
845 Martin Sebor <msebor@redhat.com>
847 PR tree-optimization/84478
848 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
850 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
851 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
852 support which is conservatively correct, for 2 only stay conservative
853 for maxlen. Formatting and comment capitalization fixes. Add STRICT
854 argument to the 2 argument get_range_strlen, adjust 6 arg
855 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
857 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
858 (gimple_fold_builtin_strlen): Pass true as last argument to
861 2018-02-20 Martin Sebor <msebor@redhat.com>
864 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
865 (builtin_memref::set_base_and_offset): Same. Handle inner references.
866 (builtin_memref::builtin_memref): Factor out parts into
867 set_base_and_offset and call it.
869 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
872 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
873 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
874 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
875 search at the associated MODE_INT.
877 2018-02-20 Jeff Law <law@redhat.com>
880 PR tree-optimization/81592
882 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
883 for range data rather than using global data.
884 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
885 range data rather than using global data.
886 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
887 pass it to children as needed.
888 (struct directive::fmtresult): Similarly.
889 (struct directive::set_width): Similarly.
890 (struct directive::set_precision): Similarly.
891 (format_integer, format_directive, parse_directive): Similarly.
892 (format_none): Accept unnamed vr_values parameter.
893 (format_percent, format_floating, format_character): Similarly.
894 (format_string, format_plain): Similarly.
895 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
896 the EVRP range analyzer for range data rather than using global data.
897 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
898 gimple-ssa-evrp-analyze.h
899 (class sprintf_dom_walker): Add after_dom_children member function.
900 Add evrp_range_analyzer member.
901 (sprintf_dom_walker::before_dom_children): Call into the EVRP
902 range analyzer as needed.
903 (sprintf_dom_walker::after_dom_children): New member function.
904 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
906 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
907 (evrp_range_analyzer::pop_to_marker): Likewise.
909 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
911 PR tree-optimization/84419
912 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
913 with the required type if its current type is compatible but
916 2018-02-20 Jakub Jelinek <jakub@redhat.com>
919 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
922 2018-02-20 Martin Liska <mliska@suse.cz>
925 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
926 possible values if we don't have a hint.
928 2018-02-20 Martin Liska <mliska@suse.cz>
932 * final.c (shorten_branches): Build align_tab array with one
934 * opts.c (finish_options): Add alignment option limit check.
935 (MAX_CODE_ALIGN): Likewise.
936 (MAX_CODE_ALIGN_VALUE): Likewise.
937 * doc/invoke.texi: Document maximum allowed option value for
938 all -falign-* options.
940 2018-02-19 Jakub Jelinek <jakub@redhat.com>
943 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
944 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
945 * var-tracking.c (emit_note_insn_var_location): Remove all references
946 to NOTE_INSN_CALL_ARG_LOCATION.
947 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
948 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
949 Use copy_rtx_if_shared.
950 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
951 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
952 (dwarf2out_var_location): Remove handling of
953 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
955 * final.c (final_scan_insn): Remove all references to
956 NOTE_INSN_CALL_ARG_LOCATION.
957 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
958 before dumping final insns.
959 * except.c (emit_note_eh_region_end): Remove all references to
960 NOTE_INSN_CALL_ARG_LOCATION.
961 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
962 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
963 * config/arc/arc.c (hwloop_optimize): Likewise.
964 * config/arm/arm.c (create_fix_barrier): Likewise.
965 * config/s390/s390.c (s390_chunkify_start): Likewise.
966 * config/sh/sh.c (find_barrier): Likewise.
967 * config/i386/i386.c (rest_of_insert_endbranch,
968 ix86_seh_fixup_eh_fallthru): Likewise.
969 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
970 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
971 * config/frv/frv.c (frv_function_prologue): Likewise.
972 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
974 (note_outside_basic_block_p): Remove all references to
975 NOTE_INSN_CALL_ARG_LOCATION.
976 * gengtype.c (adjust_field_rtx_def): Likewise.
977 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
979 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
980 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
983 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
986 PR tree-optimization/84452
987 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
988 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
991 2018-02-19 Martin Liska <mliska@suse.cz>
994 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
996 2018-02-19 Martin Liska <mliska@suse.cz>
997 Richard Sandiford <richard.sandiford@linaro.org>
999 PR tree-optimization/82491
1000 * gimple-fold.c (get_base_constructor): Make earlier bail out
1003 2018-02-19 Carl Love <cel@us.ibm.com>
1005 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
1006 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
1008 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
1009 P8V_BUILTIN_VEC_NEG.
1011 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
1013 * config/rl78/rl78.md (movdf): New define expand.
1015 2018-02-19 Martin Liska <mliska@suse.cz>
1018 * doc/invoke.texi: Fix typo.
1019 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
1021 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
1023 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
1024 handle rs6000_single_float and rs6000_double_float specially for
1027 2018-02-16 Jeff Law <law@redhat.com>
1029 * config/rx/rx.c (add_pop_cfi_notes): New function.;
1032 2018-02-16 Jakub Jelinek <jakub@redhat.com>
1035 * ipa-inline.c (inline_small_functions): Fix a typo.
1037 2018-02-16 Nathan Sidwell <nathan@acm.org>
1039 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
1041 2018-02-16 Carl Love <cel@us.ibm.com>
1043 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
1044 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
1045 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
1046 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
1047 expansion to P8V_BUILTIN_VEC_FLOAT2.
1049 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
1051 PR rtl-optimization/70023
1052 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
1053 src_regno into account.
1055 2018-02-16 Carl Love <cel@us.ibm.com>
1057 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
1058 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
1059 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
1060 * config/rs6000/rs6000.c: Remove case statements for
1061 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
1062 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
1063 and P9V_BUILTIN_VEC_VINSERT4B.
1064 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
1065 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
1066 * config/rs6000/vsx.md:
1067 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
1070 2018-02-16 Carl Love <cel@us.ibm.com>
1072 * config/rs6000/altivec.h: Add builtin names vec_extract4b
1074 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
1076 * config/rs6000/rs6000-c.c: Add the definitions for
1077 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
1078 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
1079 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
1080 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
1081 definition for insert4b and define insn *insert3b_internal.
1082 * doc/extend.texi: Add documentation for vec_extract4b.
1084 2018-02-16 Nathan Sidwell <nathan@acm.org>
1086 * doc/extend.texi (Backwards Compatibility): Mention friend
1087 injection. Note for-scope is deprecated.
1088 * doc/invoke.texi (-ffriend-injection): Deprecate.
1090 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
1092 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
1093 that moved to I2, also allow destinations that are a paradoxical
1094 subreg (instead of a normal reg).
1096 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
1099 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
1102 2018-02-16 Richard Biener <rguenther@suse.de>
1104 PR tree-optimization/84037
1105 PR tree-optimization/84016
1107 * config/i386/i386.c (ix86_builtin_vectorization_cost):
1108 Adjust vec_construct for the fact we need additional higher latency
1109 128bit inserts for AVX256 and AVX512 vector builds.
1110 (ix86_add_stmt_cost): Scale vector construction cost for
1113 2018-02-16 Richard Biener <rguenther@suse.de>
1115 PR tree-optimization/84417
1116 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
1117 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
1118 (non_rewritable_lvalue_p): Likewise, use poly-ints.
1120 2018-02-16 Martin Liska <mliska@suse.cz>
1123 * internal-fn.def (ASAN_CHECK): Set proper flags.
1124 (ASAN_MARK): Likewise.
1126 2018-02-16 Julia Koval <julia.koval@intel.com>
1128 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
1129 from PTA_CANNONLAKE.
1131 2018-02-16 Jakub Jelinek <jakub@redhat.com>
1134 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
1135 Use ++iter rather than iter++ for std::list iterators.
1136 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
1137 defer deleting them until all nodes in the forest are processed. Do
1138 free even leaf nodes. Change to_process into auto_vec.
1141 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
1142 * vec.h (vec_default_construct): Use memset instead of placement new
1143 if BROKEN_VALUE_INITIALIZATION is defined.
1144 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
1145 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
1148 PR rtl-optimization/83723
1149 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
1150 * lra.c (lra_substitute_pseudo): Likewise. If true, use
1151 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
1153 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
1155 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
1157 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
1159 PR rtl-optimization/81443
1160 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
1161 from inner REGs to paradoxical SUBREGs.
1163 2018-02-16 Richard Biener <rguenther@suse.de>
1165 PR tree-optimization/84399
1166 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
1167 For operands we can analyze at their definition make sure we can
1168 analyze them at each use as well.
1170 2018-02-16 Richard Biener <rguenther@suse.de>
1172 PR tree-optimization/84190
1173 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
1174 volatile accesses if the decl isn't volatile.
1176 2018-02-15 Jason Merrill <jason@redhat.com>
1178 PR c++/84314 - ICE with templates and fastcall attribute.
1179 * attribs.c (build_type_attribute_qual_variant): Don't clobber
1180 TYPE_CANONICAL on an existing type.
1182 2018-02-15 Jakub Jelinek <jakub@redhat.com>
1184 PR tree-optimization/84383
1185 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
1186 dstoff nor call operand_equal_p if dstbase is NULL.
1188 PR tree-optimization/84334
1189 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
1190 also a CONSTANT_CLASS_P, punt.
1192 2018-02-14 Jim Wilson <jimw@sifive.com>
1194 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
1195 first SMALL_OPERAND check. New local min_second_step. Move assert
1196 to where locals are set. Add TARGET_RVC support.
1197 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
1199 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
1201 * doc/invoke.texi: Correct -Wformat-overflow code sample.
1203 2018-02-14 Martin Sebor <msebor@redhat.com>
1205 PR tree-optimization/83698
1206 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
1207 arrays constrain the offset range to their bounds.
1208 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
1209 (builtin_access::overlap): Avoid setting the size of overlap if it's
1211 (maybe_diag_overlap): Also consider arrays when deciding what values
1212 of offsets to include in diagnostics.
1214 2018-02-14 Martin Sebor <msebor@redhat.com>
1217 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
1218 that correspond to the kind of a declaration.
1220 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
1223 * config/pa/pa.md: Load address of PIC label using the linkage table
1224 if the label is nonlocal.
1226 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
1228 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
1229 warning message if user requests -maltivec=be.
1230 * doc/invoke.texi: Document deprecation of -maltivec=be.
1232 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
1235 * config/rs6000/rs6000-c.c: Update definitions for
1236 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
1237 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
1239 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1242 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
1243 add _get_ssp intrinsics. Remove argument from
1244 __builtin_ia32_rdssp[d|q].
1245 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
1246 * config/i386/i386-builtin.def: Remove argument from
1247 __builtin_ia32_rdssp[d|q].
1248 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
1249 ix86_expand_special_args_builtin for _rdssp[d|q].
1250 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
1251 Clear register before usage.
1252 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
1253 Add documentation for new _get_ssp and _inc_ssp intrinsics.
1255 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
1257 PR tree-optimization/84357
1258 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
1259 operand 1 of an ARRAY_REF too.
1261 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
1264 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
1265 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
1267 (set_of_reg): New struct.
1268 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
1269 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
1270 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
1272 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
1273 Split into bitclr, bitset, bitinvert patterns if appropriate.
1274 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
1275 use rx_fuse_in_memory_bitop.
1276 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
1277 to named insn, correct maximum insn length.
1279 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
1282 * machmode.def: Define a complex mode for PARTIAL_INT.
1283 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
1285 * doc/rtl.texi: Document CSPImode.
1286 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
1288 (msp430_hard_regno_nregs_with_padding): Likewise.
1290 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
1293 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
1295 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
1297 PR rtl-optimization/84169
1298 * combine.c (try_combine): New variable split_i2i3. Set it to true if
1299 we generated a parallel as new i3 and we split that to new i2 and i3
1300 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
1301 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
1302 those to i2, not i1. Partially rewrite this scan code.
1304 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1307 * stor-layout.c (place_field): For variable length fields, adjust
1308 offset_align afterwards not just based on the field's alignment,
1309 but also on the size.
1312 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
1313 of exps and logs in the use_exp2 case.
1315 2018-02-13 Jeff Law <law@redhat.com>
1317 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
1320 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
1323 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
1327 * common.opt (gas-loc-support, gas-locview-support): New.
1328 (ginline-points, ginternal-reset-location-views): New.
1329 * doc/invoke.texi: Document them. Use @itemx where intended.
1330 (gvariable-location-views): Adjust.
1331 * target.def (reset_location_view): New.
1332 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
1333 (TARGET_RESET_LOCATION_VIEW): New.
1334 * doc/tm.texi: Rebuilt.
1335 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
1336 (dwarf2out_default_as_locview_support): New.
1337 (output_asm_line_debug_info): Use option variables.
1338 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
1339 (output_loc_list): Likewise.
1340 (add_high_low_attributes): Check option variables.
1341 Don't output entry view attribute in strict mode.
1342 (gen_inlined_subroutine_die): Check option variables.
1343 (dwarf2out_inline_entry): Likewise.
1344 (init_sections_and_labels): Likewise.
1345 (dwarf2out_early_finish): Likewise.
1346 (maybe_reset_location_view): New, from...
1347 (dwarf2out_var_location): ... here. Call it.
1348 * debug.h (dwarf2out_default_as_loc_support): Declare.
1349 (dwarf2out_default_as_locview_support): Declare.
1350 * hooks.c (hook_int_rtx_insn_0): New.
1351 * hooks.h (hook_int_rtx_insn_0): Declare.
1352 * toplev.c (process_options): Take -gas-loc-support and
1353 -gas-locview-support from dwarf2out. Enable
1354 -gvariable-location-views by default only with locview
1355 assembler support. Enable -ginternal-reset-location-views by
1356 default only if the target defines the corresponding hook.
1357 Enable -ginline-points by default if location views are
1358 enabled; force it disabled if statement frontiers are
1360 * tree-inline.c (expand_call_inline): Check option variables.
1361 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
1363 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
1365 PR tree-optimization/84321
1366 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
1367 handling. Also check whether the anti-range contains any values
1368 that satisfy the mask; switch to a VR_RANGE if not.
1370 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
1373 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
1375 2018-02-13 Martin Jambor <mjambor@suse.cz>
1378 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
1379 of call statements, also set location of a load to a temporary.
1381 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
1383 * config/rl78/rl78.c (add_vector_labels): New function.
1384 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
1385 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
1386 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
1387 which checks that no arguments are passed.
1388 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
1389 * doc/extend.texi: Documentation for the new attribute.
1391 2018-02-13 Andreas Schwab <schwab@suse.de>
1393 * config/riscv/linux.h (CPP_SPEC): Define.
1395 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1398 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
1399 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
1400 OPTION_MASK_ISA_AES as first argument to def_builtin_const
1401 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
1402 instead of OPTION_MASK_ISA_PCLMUL as first argument to
1403 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
1404 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
1405 temporarily for AES and PCLMUL builtins.
1407 PR tree-optimization/84339
1408 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
1409 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1413 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1414 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1415 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1417 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1419 * omp-simd-clone.h: New file.
1420 * omp-simd-clone.c: Include omp-simd-clone.h.
1421 (expand_simd_clones): No longer static.
1422 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1423 cgraph.h and omp-simd-clone.h.
1424 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1425 (vect_recog_widen_shift_pattern): Formatting fix.
1426 (vect_pattern_recog_1): Don't check optab for calls.
1429 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1430 operands[2] into a REG before using gen_lowpart on it.
1432 2018-02-12 Jeff Law <law@redhat.com>
1435 * config/sh/sh.c (find_barrier): Consider a sibling call
1438 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1439 successfully back substituting a reg.
1441 2018-02-12 Richard Biener <rguenther@suse.de>
1443 PR tree-optimization/84037
1444 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1445 parameter, move visited init to caller.
1446 (vect_slp_analyze_operations): Separate cost from validity
1447 check, initialize visited once for all instances.
1448 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1450 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1452 (vect_model_promotion_demotion_cost): Likewise.
1453 (vectorizable_bswap): Guard cost modeling with !slp_node
1454 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1456 (vectorizable_call): Likewise.
1457 (vectorizable_conversion): Likewise.
1458 (vectorizable_assignment): Likewise.
1459 (vectorizable_shift): Likewise.
1460 (vectorizable_operation): Likewise.
1461 (vectorizable_store): Likewise.
1462 (vectorizable_load): Likewise.
1463 (vectorizable_condition): Likewise.
1464 (vectorizable_comparison): Likewise.
1466 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1469 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1470 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1471 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1473 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1476 * params.def (inline-min-speedup): Increase from 8 to 15.
1477 (max-inline-insns-auto): Decrease from 40 to 30.
1478 * ipa-split.c (consider_split): Add some buffer for function to
1479 be considered inlining candidate.
1480 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1483 2018-02-12 Richard Biener <rguenther@suse.de>
1485 PR tree-optimization/84037
1486 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1487 matched stmts if we cannot swap the non-matched ones.
1489 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1491 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1492 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1493 _mm_maskz_scalef_round_ss): New intrinsics.
1494 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1495 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1496 __builtin_ia32_scalefss_round): Remove.
1497 (__builtin_ia32_scalefsd_mask_round,
1498 __builtin_ia32_scalefss_mask_round): New intrinsics.
1499 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1500 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1501 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1502 "<round_constraint>")): Changed to ...
1503 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1504 "<round_scalar_constraint>")): ... this.
1505 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1506 %0, %1, %2<round_op3>}"): Changed to ...
1507 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1508 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1509 %2<round_scalar_mask_op3>}"): ... this.
1510 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1512 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1514 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1515 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1516 (_mm_maskz_sqrt_round_ss): New intrinsics.
1517 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1518 (__builtin_ia32_sqrtsd_mask_round)
1519 (__builtin_ia32_sqrtss_mask_round): New builtins.
1520 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1521 (__builtin_ia32_sqrtss_round): Remove.
1522 (__builtin_ia32_sqrtsd_mask_round)
1523 (__builtin_ia32_sqrtss_mask_round): New builtins.
1524 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1525 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1526 ((match_operand:VF_128 1 "vector_operand"
1527 "xBm,<round_constraint>")): Changed to ...
1528 ((match_operand:VF_128 1 "vector_operand"
1529 "xBm,<round_scalar_constraint>")): ... this.
1530 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1531 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1532 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1533 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1534 %<iptr>1<round_scalar_mask_op3>}): ... this.
1535 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1536 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1538 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1541 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1542 Cast vec_cmpeq result to correct type.
1543 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1544 Cast vec_cmpgt result to correct type.
1546 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1548 * final.c (final_scan_insn_1): Renamed from...
1549 (final_scan_insn): ... this. New wrapper, to recover
1550 seen from the outermost call in recursive ones.
1551 * config/sparc/sparc.c (output_return): Drop seen from call.
1552 (output_sibcall): Likewise.
1553 * config/visium/visium.c (output_branch): Likewise.
1555 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1557 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1560 2018-02-10 Alan Modra <amodra@gmail.com>
1563 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1564 Specify LR as an input.
1566 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1569 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1570 remove_member_access_dummy_vars): New functions.
1571 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1572 lower_omp_1, execute_lower_omp): Use them.
1574 PR rtl-optimization/84308
1575 * shrink-wrap.c (spread_components): Release todo vector.
1577 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1579 PR rtl-optimization/57193
1580 * ira-color.c (struct allocno_color_data): Add member
1581 conflict_allocno_hard_prefs.
1582 (update_conflict_allocno_hard_prefs): New.
1583 (bucket_allocno_compare_func): Add a preference based on
1584 conflict_allocno_hard_prefs.
1585 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1586 (color_allocnos): Remove a dead code. Initiate
1587 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1589 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1592 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1593 constraint from =wa to wa. Avoid a subreg on the output operand,
1594 instead use a pseudo and subreg it in a move.
1595 (p9_xxbrd_<mode>): Changed to ...
1596 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1597 (p9_xxbrd_v2df): New expander.
1598 (p9_xxbrw_<mode>): Changed to ...
1599 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1600 (p9_xxbrw_v4sf): New expander.
1602 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1604 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1606 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1609 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1610 multiply in 32-bit mode.
1611 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1612 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1615 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1617 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1618 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1619 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1620 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1622 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1625 * dwarf2out.c (is_trivial_indirect_ref): New function.
1626 (dwarf2out_late_global_decl): Do not generate a location
1627 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1628 and that are not defined in the current unit.
1630 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1632 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1633 instead of a libcall for UNORDERED.
1635 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1638 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1639 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1641 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1644 * config/s390/s390.c (s390_set_current_function): Invoke
1645 s390_indirect_branch_settings also if fndecl didn't change.
1647 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1649 * config/rs6000/rs6000.md (blockage): Set length to zero.
1651 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1653 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1655 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1658 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1659 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1663 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1664 PARALLEL incoming that failed vt_get_decl_and_offset check.
1667 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1668 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1670 (get_variable_section): For decls in named .bss* sections pass true as
1671 second argument to bss_initializer_p.
1673 2018-02-09 Marek Polacek <polacek@redhat.com>
1674 Jakub Jelinek <jakub@redhat.com>
1677 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1678 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1679 Sync some changes from cxx_fold_indirect_ref.
1681 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1683 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1685 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1686 (BLOCK_INLINE_ENTRY_LABEL): New.
1687 (dwarf2out_var_location): Disregard inline entry markers.
1688 (inline_entry_data): New struct.
1689 (inline_entry_data_hasher): New hashtable type.
1690 (inline_entry_data_hasher::hash): New.
1691 (inline_entry_data_hasher::equal): New.
1692 (inline_entry_data_table): New variable.
1693 (add_high_low_attributes): Add DW_AT_entry_pc and
1694 DW_AT_GNU_entry_view attributes if a pending entry is found
1695 in inline_entry_data_table. Add old entry_pc attribute only
1696 if debug nonbinding markers are disabled.
1697 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1698 markers are enabled.
1699 (block_within_block_p, dwarf2out_inline_entry): New.
1700 (dwarf2out_finish): Check that no entries remained in
1701 inline_entry_data_table.
1702 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1703 (final_scan_insn, notice_source_line): Likewise.
1704 (rest_of_clean_state): Skip inline entry markers.
1705 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1707 * gimple.c (gimple_build_debug_inline_entry): New.
1708 * gimple.h (enum gimple_debug_subcode): Add
1709 GIMPLE_DEBUG_INLINE_ENTRY.
1710 (gimple_build_debug_inline_entry): Declare.
1711 (gimple_debug_inline_entry_p): New.
1712 (gimple_debug_nonbind_marker_p): Adjust.
1713 * insn-notes.def (INLINE_ENTRY): New.
1714 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1715 inline entry marker notes.
1716 (print_insn): Likewise.
1717 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1718 (INSN_DEBUG_MARKER_KIND): Likewise.
1719 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1720 * tree-inline.c (expand_call_inline): Build and insert
1721 debug_inline_entry stmt.
1722 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1723 inline entry blocks early, if nonbind markers are enabled.
1724 (dump_scope_block): Dump fragment info.
1725 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1726 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1727 (gimple_build_debug_inline_entry): New.
1728 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1729 Enable/disable inline entry points too.
1730 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1731 (DEBUG_INSN): Describe inline entry markers.
1733 * common.opt (gvariable-location-views): New.
1734 (gvariable-location-views=incompat5): New.
1735 * config.in: Rebuilt.
1736 * configure: Rebuilt.
1737 * configure.ac: Test assembler for view support.
1738 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1739 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1740 * dwarf2out.c (var_loc_view): New typedef.
1741 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1742 (dwarf2out_locviews_in_attribute): New.
1743 (dwarf2out_locviews_in_loclist): New.
1744 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1745 (enum dw_line_info_opcode): Add LI_adv_address.
1746 (struct dw_line_info_table): Add view.
1747 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1748 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1749 (zero_view_p): New variable.
1750 (ZERO_VIEW_P): New macro.
1751 (output_asm_line_debug_info): New.
1752 (struct var_loc_node): Add view.
1753 (add_AT_view_list, AT_loc_list): New.
1754 (add_var_loc_to_decl): Add view param. Test it against last.
1755 (new_loc_list): Add view params. Record them.
1756 (AT_loc_list_ptr): Handle loc and view lists.
1757 (view_list_to_loc_list_val_node): New.
1758 (print_dw_val): Handle dw_val_class_view_list.
1759 (size_of_die): Likewise.
1760 (value_format): Likewise.
1761 (loc_list_has_views): New.
1762 (gen_llsym): Set vl_symbol too.
1763 (maybe_gen_llsym, skip_loc_list_entry): New.
1764 (dwarf2out_maybe_output_loclist_view_pair): New.
1765 (output_loc_list): Output view list or entries too.
1766 (output_view_list_offset): New.
1767 (output_die): Handle dw_val_class_view_list.
1768 (output_dwarf_version): New.
1769 (output_compilation_unit_header): Use it.
1770 (output_skeleton_debug_sections): Likewise.
1771 (output_rnglists, output_line_info): Likewise.
1772 (output_pubnames, output_aranges): Update version comments.
1773 (output_one_line_info_table): Output view numbers in asm comments.
1774 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1775 Call maybe_gen_llsym.
1776 (loc_list_from_tree_1): Adjust.
1777 (add_AT_location_description): Create view list attribute if
1778 needed, check it's absent otherwise.
1779 (convert_cfa_to_fb_loc_list): Adjust.
1780 (maybe_emit_file): Call output_asm_line_debug_info for test.
1781 (dwarf2out_var_location): Reset views as needed. Precompute
1782 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1783 attribute. Set view.
1784 (new_line_info_table): Reset next view.
1785 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1786 (dwarf2out_source_line): Likewise. Output view resets and labels to
1787 the assembler, or select appropriate line info opcodes.
1788 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1789 (optimize_string_length): Catch it. Adjust.
1790 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1791 dw_val_class_view_list, and remove it if no longer needed.
1792 (hash_loc_list): Hash view numbers.
1793 (loc_list_hasher::equal): Compare them.
1794 (optimize_location_lists): Check whether a view list symbol is
1795 needed, and whether the locview attribute is present, and
1796 whether they match. Remove the locview attribute if no longer
1798 (index_location_lists): Call skip_loc_list_entry for test.
1799 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1800 Use output_dwarf_version.
1801 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1802 (struct dw_val_node): Add val_view_list.
1803 * final.c (SEEN_NEXT_VIEW): New.
1804 (set_next_view_needed): New.
1805 (clear_next_view_needed): New.
1806 (maybe_output_next_view): New.
1807 (final_start_function): Rename to...
1808 (final_start_function_1): ... this. Take pointer to FIRST,
1809 add SEEN parameter. Emit param bindings in the initial view.
1810 (final_start_function): Reintroduce SEEN-less interface.
1811 (final): Rename to...
1812 (final_1): ... this. Take SEEN parameter. Output final pending
1813 next view at the end.
1814 (final): Reintroduce seen-less interface.
1815 (final_scan_insn): Output pending next view before switching
1816 sections or ending a block. Mark the next view as needed when
1817 outputting variable locations. Notify debug backend of section
1818 changes, and of location view changes.
1819 (rest_of_handle_final): Adjust.
1820 * toplev.c (process_options): Autodetect value for debug variable
1821 location views option. Warn on incompat5 without -gdwarf-5.
1822 * doc/invoke.texi (gvariable-location-views): New.
1823 (gvariable-location-views=incompat5): New.
1824 (gno-variable-location-views): New.
1826 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1828 PR tree-optimization/84136
1829 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1830 that the result of find_edge is non-NULL.
1832 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1835 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1836 storing integer register in SImode. Fix cost of 256 and 512
1837 byte aligned SSE register store.
1839 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1841 * config/i386/i386.c (ix86_multiplication_cost): Fix
1842 multiplication cost for TARGET_AVX512DQ.
1844 2018-02-08 Marek Polacek <polacek@redhat.com>
1846 PR tree-optimization/84238
1847 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1850 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1852 PR tree-optimization/84265
1853 * tree-vect-stmts.c (vectorizable_store): Don't treat
1854 VMAT_CONTIGUOUS accesses as grouped.
1855 (vectorizable_load): Likewise.
1857 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1859 PR tree-optimization/81635
1860 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1861 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1862 (test_round_for_mask): New functions.
1863 (wide_int_cc_tests): Call test_round_for_mask.
1864 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1865 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1866 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1867 range returned by get_range_info.
1869 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1872 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1873 * symtab.c: Include builtins.h
1874 (symtab_node::output_to_lto_symbol_table_p): Move here
1875 from lto-streamer-out.c:output_symbol_p.
1876 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1877 (output_symbol_p): Move all logic to symtab.c
1878 (produce_symtab): Update.
1880 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1882 * config/s390/s390-opts.h (enum indirect_branch): Define.
1883 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1884 (s390_indirect_branch_via_thunk)
1885 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1886 (enum s390_indirect_branch_type): Define.
1887 * config/s390/s390.c (struct s390_frame_layout, struct
1888 machine_function): Remove.
1889 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1890 (indirect_branch_table_label_no, indirect_branch_table_name):
1892 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1893 (enum s390_indirect_branch_option): Define.
1894 (s390_return_addr_from_memory): New function.
1895 (s390_handle_string_attribute): New function.
1896 (s390_attribute_table): Add new attribute handler.
1897 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1898 (s390_indirect_branch_via_thunk): New function.
1899 (s390_indirect_branch_via_inline_thunk): New function.
1900 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1901 sibling call optimization for non z10 compiles.
1902 (s390_emit_call): Force indirect branch target to be a single
1903 register. Add r1 clobber for non-z10 compiles.
1904 (s390_emit_epilogue): Emit return jump via return_use expander.
1905 (s390_reorg): Handle JUMP_INSNs as execute targets.
1906 (s390_option_override_internal): Perform validity checks for the
1907 new command line options.
1908 (s390_indirect_branch_attrvalue): New function.
1909 (s390_indirect_branch_settings): New function.
1910 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1911 (s390_output_indirect_thunk_function): New function.
1912 (s390_code_end): Implement target hook.
1913 (s390_case_values_threshold): Implement target hook.
1914 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1916 * config/s390/s390.h (struct s390_frame_layout)
1917 (struct machine_function): Move here from s390.c.
1918 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1919 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1920 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1921 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1922 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1923 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1924 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1925 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1926 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1927 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1928 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1929 (mnemonic attribute): Add values which aren't recognized
1931 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1932 pattern for branch conversion. Fix mnemonic attribute.
1933 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1934 indirect branch via thunk if requested.
1935 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1936 ("*indirect_jump"): Disable for branch conversion using out of
1938 ("indirect_jump_via_thunk<mode>_z10")
1939 ("indirect_jump_via_thunk<mode>")
1940 ("indirect_jump_via_inlinethunk<mode>_z10")
1941 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1942 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1943 ("casesi_jump_via_inlinethunk<mode>_z10")
1944 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1945 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1946 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1947 ("*indirect2_jump"): Disable for branch conversion.
1948 ("casesi_jump"): Turn into expander and expand patterns for branch
1950 ("return_use"): New expander.
1951 ("*return"): Emit return via thunk and rename it to ...
1952 ("*return<mode>"): ... this one.
1953 * config/s390/s390.opt: Add new options and and enum for the
1956 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1958 * lra-constraints.c (match_reload): Unconditionally use
1959 gen_lowpart_SUBREG, rather than selecting between that
1960 and equivalent gen_rtx_SUBREG code.
1962 2018-02-08 Richard Biener <rguenther@suse.de>
1964 PR tree-optimization/84233
1965 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1966 changed flag instead of boguously re-using phi_inserted.
1968 2018-02-08 Martin Jambor <mjambor@suse.cz>
1970 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1971 static local variables.
1973 2018-02-08 Richard Biener <rguenther@suse.de>
1975 PR tree-optimization/84278
1976 * tree-vect-stmts.c (vectorizable_store): When looking for
1977 smaller vector types to perform grouped strided loads/stores
1978 make sure the mode is supported by the target.
1979 (vectorizable_load): Likewise.
1981 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1983 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1984 Increase LDP/STP opportunities by adding adjacent callee-saves.
1986 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1988 PR rtl-optimization/84068
1989 PR rtl-optimization/83459
1990 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1992 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1994 PR tree-optimization/84224
1995 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1996 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1999 2018-02-07 Iain Sandoe <iain@codesourcery.com>
2002 * config/rs6000/altivec.md (*restore_world): Remove LR use.
2003 * config/rs6000/predicates.md (restore_world_operation): Adjust op
2004 count, remove one USE.
2006 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
2008 * doc/install.texi (Configuration): Document the
2009 --with-long-double-format={ibm,ieee} PowerPC configuration
2013 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
2014 Convert from define_expand to be define_insn_and_split. Rework
2015 float/double/_Float128 conversions to QI/HI/SImode to work with
2016 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
2017 conversions to QI/HImode types did a store and then a load to
2018 truncate the value. For conversions to VSX registers, don't split
2019 the insn, instead emit the code directly. Use the code iterator
2020 any_fix to combine signed and unsigned conversions.
2021 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
2022 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
2023 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
2024 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
2025 (fix_<mode>di2_hw): Likewise.
2026 (fixuns_<mode>di2_hw): Likewise.
2027 (fix_<mode>si2_hw): Likewise.
2028 (fixuns_<mode>si2_hw): Likewise.
2029 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
2030 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
2031 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
2032 fix<uns>_trunc<SFDF:mode>si2_p8.
2033 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
2035 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
2036 (fix<uns>_<mode>_mem): Likewise.
2037 (fctiw<u>z_<mode>_mem): Likewise.
2038 (fix<uns>_<mode>_mem): Likewise.
2039 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
2040 the register allocator from doing a direct move to the GPRs to do
2041 a store, and instead use the ISA 3.0 store byte/half-word from
2042 vector register instruction. For IEEE 128-bit floating point,
2043 also optimize stores of 32-bit ints.
2044 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
2046 2018-02-07 Alan Hayward <alan.hayward@arm.com>
2048 * genextract.c (push_pathstr_operand): New function to support
2050 (walk_rtx): Call push_pathstr_operand.
2051 (print_path): Support [a-zA-Z].
2053 2018-02-07 Richard Biener <rguenther@suse.de>
2055 PR tree-optimization/84037
2056 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
2057 (cse_and_gimplify_to_preheader): Declare.
2058 (vect_get_place_in_interleaving_chain): Likewise.
2059 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2061 (_loop_vec_info::~_loop_vec_info): Delete it.
2062 (cse_and_gimplify_to_preheader): New function.
2063 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
2064 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
2065 (vectorizable_load): Likewise. For grouped stores always base
2066 the IV on the first element.
2067 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
2068 condition before gimplifying.
2070 2018-02-07 Jakub Jelinek <jakub@redhat.com>
2072 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
2073 *DIV_EXPR and *MOD_EXPR.
2075 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
2078 * config/i386/i386.c (ix86_option_override_internal): Mask out
2079 the CF_SET bit when checking -fcf-protection.
2081 2018-02-07 Tom de Vries <tom@codesourcery.com>
2084 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
2087 2018-02-07 Richard Biener <rguenther@suse.de>
2089 PR tree-optimization/84204
2090 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
2093 PR tree-optimization/84205
2094 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
2095 special-case isl_ast_op_zdiv_r.
2097 PR tree-optimization/84223
2098 * graphite-scop-detection.c (gather_bbs::before_dom_children):
2099 Only add conditions from within the region.
2100 (gather_bbs::after_dom_children): Adjust.
2102 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
2105 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
2106 * config/avr/avr.md: Only post-reload split REG-REG moves if
2107 either register is GENERAL_REG_P.
2109 2018-02-07 Jakub Jelinek <jakub@redhat.com>
2111 PR tree-optimization/84235
2112 * tree-ssa-scopedtables.c
2113 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
2114 if the subtraction is performed in floating point type where NaNs are
2115 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
2116 build 1. Formatting fix.
2118 2018-02-06 Jakub Jelinek <jakub@redhat.com>
2121 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
2122 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
2123 and skip it regardless of bb boundaries. Use CALL_P macro,
2124 don't test INSN_P (insn) together with CALL_P or JUMP_P check
2125 unnecessarily, formatting fix.
2127 2018-02-06 Michael Collison <michael.collison@arm.com>
2129 * config/arm/thumb2.md:
2130 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
2131 (*thumb_mov_notscc): Ditto.
2133 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
2136 * config/rs6000/rs6000.md (su code attribute): Use "u" for
2137 unsigned_fix, not "s".
2139 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2141 * configure.ac (gcc_fn_eh_frame_ro): New function.
2142 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
2143 correct .eh_frame permissions.
2144 * configure: Regenerate.
2146 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
2148 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
2151 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2153 * config/rs6000/rs6000.c (rs6000_option_override_internal):
2154 Display warning message for -mno-speculate-indirect-jumps.
2156 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
2158 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
2160 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
2162 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
2164 PR tree-optimization/84225
2165 * tree-eh.c (find_trapping_overflow): Only call
2166 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
2168 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
2171 * config/i386/i386.c: Reimplement the check of possible options
2172 -mibt/-mshstk conbination. Change error messages.
2173 * doc/invoke.texi: Fix a typo: remove extra '='.
2175 2018-02-06 Marek Polacek <polacek@redhat.com>
2177 PR tree-optimization/84228
2178 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
2180 2018-02-06 Tamar Christina <tamar.christina@arm.com>
2183 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
2184 emitted arch directives.
2185 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
2186 __ARM_FEATURE_COPROC before changing architectures.
2188 2018-02-06 Richard Biener <rguenther@suse.de>
2190 * config/i386/i386.c (print_reg): Fix typo.
2191 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
2193 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
2195 * configure: Regenerate.
2197 2018-02-05 Martin Sebor <msebor@redhat.com>
2199 PR tree-optimization/83369
2200 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
2203 2018-02-05 Martin Liska <mliska@suse.cz>
2205 * doc/invoke.texi: Cherry-pick upstream r323995.
2207 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
2209 * ira.c (ira_init_register_move_cost): Adjust comment.
2211 2018-02-05 Martin Liska <mliska@suse.cz>
2213 PR gcov-profile/84137
2214 * doc/gcov.texi: Fix typo in documentation.
2216 2018-02-05 Martin Liska <mliska@suse.cz>
2218 PR gcov-profile/83879
2219 * doc/gcov.texi: Document necessity of --dynamic-list-data when
2220 using dlopen functionality.
2222 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
2224 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
2225 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
2226 _mm_maskz_range_ss, _mm_mask_range_round_ss,
2227 _mm_maskz_range_round_ss): New intrinsics.
2228 (__builtin_ia32_rangesd128_round)
2229 (__builtin_ia32_rangess128_round): Remove.
2230 (__builtin_ia32_rangesd128_mask_round,
2231 __builtin_ia32_rangess128_mask_round): New builtins.
2232 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
2233 __builtin_ia32_rangess128_round): Remove.
2234 (__builtin_ia32_rangesd128_mask_round,
2235 __builtin_ia32_rangess128_mask_round): New builtins.
2236 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
2237 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
2238 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
2239 "<round_saeonly_constraint>")): Changed to ...
2240 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
2241 "<round_saeonly_scalar_constraint>")): ... this.
2242 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
2243 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
2244 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
2245 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
2246 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
2248 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
2250 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
2252 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
2253 Remove all values except native, 8540 and 8548.
2255 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
2257 * config/i386/i386.c (ix86_output_function_return): Pass
2258 INVALID_REGNUM, instead of -1, as invalid register number to
2259 indirect_thunk_name and output_indirect_thunk.
2261 2018-02-02 Julia Koval <julia.koval@intel.com>
2263 * config.gcc: Add -march=icelake.
2264 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
2265 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
2266 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
2267 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
2268 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
2269 (processor_target_table): Add icelake.
2270 (ix86_option_override_internal): Handle new PTAs.
2271 (get_builtin_code_for_version): Handle icelake.
2272 (M_INTEL_COREI7_ICELAKE): New.
2273 (fold_builtin_cpu): Handle icelake.
2274 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
2275 * doc/invoke.texi: Add -march=icelake.
2277 2018-02-02 Julia Koval <julia.koval@intel.com>
2279 * config/i386/i386.c (ix86_option_override_internal): Change flags type
2280 to wide_int_bitmask.
2281 * wide-int-bitmask.h: New.
2283 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
2286 * config/i386/i386.md: Replace Pmode with word_mode in
2287 builtin_setjmp_setup and builtin_longjmp to support x32.
2289 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
2293 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
2295 (rs6000_supported_cpu_names): New static variable.
2296 (linux_cpu_translation_table): Likewise.
2297 (elf_platform) <cpu>: Define new static variable and use it.
2298 Translate kernel AT_PLATFORM name to canonical name if needed.
2299 Error if platform name is unknown.
2301 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
2304 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
2306 2018-02-01 Jeff Law <law@redhat.com>
2309 * config/i386/i386.c (release_scratch_register_on_entry): Add new
2310 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
2311 the scratch if RELEASE_VIA_POP is false.
2312 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
2313 If we have to save a temporary register, decrement SIZE appropriately.
2314 Pass new arguments to release_scratch_register_on_entry.
2315 (ix86_adjust_stack_and_probe): Likewise.
2316 (ix86_emit_probe_stack_range): Pass new arguments to
2317 release_scratch_register_on_entry.
2319 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
2321 PR rtl-optimization/84157
2322 * combine.c (change_zero_ext): Use REG_P predicate in
2323 front of HARD_REGISTER_P predicate.
2325 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
2327 * config/avr/avr.c (avr_option_override): Move disabling of
2328 -fdelete-null-pointer-checks to...
2329 * common/config/avr/avr-common.c (avr_option_optimization_table):
2332 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2334 PR tree-optimization/81635
2335 * tree-data-ref.c (split_constant_offset_1): For types that
2336 wrap on overflow, try to use range info to prove that wrapping
2339 2018-02-01 Renlin Li <renlin.li@arm.com>
2342 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
2344 (aarch64_register_move_cost): Likewise.
2345 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
2347 (REG_CLASS_NAMES): Likewise.
2348 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
2349 TAILCALL_ADDR_REGS. Remove IP registers.
2350 * config/aarch64/aarch64.md (Ucs): Update register constraint.
2352 2018-02-01 Richard Biener <rguenther@suse.de>
2354 * domwalk.h (dom_walker::dom_walker): Add additional constructor
2355 for specifying RPO order and allow NULL for that.
2356 * domwalk.c (dom_walker::dom_walker): Likewise.
2357 (dom_walker::walk): Handle NULL RPO order.
2358 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
2360 (rewrite_update_dom_walker): Likewise.
2361 (mark_def_dom_walker): Likewise.
2363 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2365 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
2366 (aarch64_maybe_expand_sve_subreg_move): Declare.
2367 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
2368 * config/aarch64/predicates.md (aarch64_any_register_operand): New
2370 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
2371 that are semantically a reverse operation.
2372 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
2373 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
2374 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
2376 (aarch64_can_change_mode_class): For big-endian, forbid changes
2377 between two SVE modes if they have different element sizes.
2379 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2381 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
2382 the TImode handling for big-endian targets.
2384 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2386 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
2387 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
2389 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
2390 Remove BSWAP handing for big-endian targets and use the form of
2391 LD1RQ appropariate for the mode.
2393 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2395 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
2396 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
2399 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2402 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
2403 check for operands that need to go through aarch64_sve_reload_be.
2405 2018-02-01 Jakub Jelinek <jakub@redhat.com>
2407 PR tree-optimization/81661
2408 PR tree-optimization/84117
2409 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
2410 * tree-eh.c: Include gimplify.h.
2411 (find_trapping_overflow, replace_trapping_overflow,
2412 rewrite_to_non_trapping_overflow): New functions.
2413 * tree-vect-loop.c: Include tree-eh.h.
2414 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2415 * tree-data-ref.c: Include tree-eh.h.
2416 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2418 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2420 PR rtl-optimization/84123
2421 * combine.c (change_zero_ext): Check if hard register satisfies
2422 can_change_dest_mode before calling gen_lowpart_SUBREG.
2424 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2427 * ira.c (ira_init_register_move_cost): Remove assert.
2429 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2431 PR rtl-optimization/84071
2432 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2433 * doc/tm.texi: Regenerate.
2435 2018-01-31 Richard Biener <rguenther@suse.de>
2437 PR tree-optimization/84132
2438 * tree-data-ref.c (analyze_miv_subscript): Properly
2439 check whether evolution_function_is_affine_multivariate_p
2440 before calling gcd_of_steps_may_divide_p.
2442 2018-01-31 Julia Koval <julia.koval@intel.com>
2445 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2446 * config/i386/i386.md (rdpid_rex64) New.
2447 (rdpid): Make 32bit only.
2449 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2452 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2453 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2455 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2458 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2460 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2462 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2464 PR rtl-optimization/84071
2465 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2466 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2468 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2470 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2471 (arc_attribute_table): Add 'aux' attribute.
2472 (arc_in_small_data_p): Consider aux like variables.
2473 (arc_is_aux_reg_p): New function.
2474 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2475 (arc_get_aux_arg): New function.
2476 (prepare_move_operands): Handle aux-register access.
2477 (arc_handle_aux_attribute): New function.
2478 * doc/extend.texi (ARC Variable attributes): Add subsection.
2480 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2482 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2483 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2484 (arc_attribute_table): Add 'uncached' attribute.
2485 (arc_print_operand): Print '.di' flag for uncached memory
2487 (arc_in_small_data_p): Do not consider for small data the uncached
2489 (arc_is_uncached_mem_p): New function.
2490 * config/arc/predicates.md (compact_store_memory_operand): Check
2491 for uncached memory accesses.
2492 (nonvol_nonimm_operand): Likewise.
2493 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2495 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2498 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2499 falign-loops=): Add Optimization flag.
2501 2018-01-30 Jeff Law <law@redhat.com>
2504 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2505 INT_REGISTERS_SAVED. Check it prior to calling
2506 get_scratch_register_on_entry.
2507 (ix86_adjust_stack_and_probe): Similarly.
2508 (ix86_emit_probe_stack_range): Similarly.
2509 (ix86_expand_prologue): Corresponding changes.
2511 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2514 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2515 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2517 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2520 * lra-constraints.c (curr_insn_transform): Process AND in the
2523 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2525 PR rtl-optimization/83986
2526 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2527 dependence against last_pending_memory_flush in addition to
2530 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2532 PR tree-optimization/81611
2533 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2536 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2539 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2542 2018-01-30 Richard Biener <rguenther@suse.de>
2543 Jakub Jelinek <jakub@redhat.com>
2545 PR tree-optimization/84111
2546 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2547 inner loops added during recursion, as they don't have up-to-date
2550 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2553 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2554 (can_inline_edge_by_limits_p): ... here.
2555 (can_early_inline_edge_p, check_callers,
2556 update_caller_keys, update_callee_keys, recursive_inlining,
2557 add_new_edges_to_heap, speculation_useful_p,
2558 inline_small_functions,
2559 inline_small_functions, flatten_function,
2560 inline_to_all_callers_1): Update.
2562 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2564 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2567 2018-01-30 Richard Biener <rguenther@suse.de>
2569 PR tree-optimization/83008
2570 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2571 invariant and constant vector uses in stmts when they need
2574 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2577 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2578 * configure: Regenerate.
2580 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2582 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2584 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2585 Use gen_rtx_REG rather than gen_lowpart.
2587 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2589 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2590 rather than 0 when creating partial subregs.
2592 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2594 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2597 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2600 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2601 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2602 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2603 flags. This restores the settings used before the 2017-07-24.
2604 Turning off pre increment/decrement/modify allows IVOPTS to
2605 optimize DF/SF loops where the index is an int.
2607 2018-01-29 Richard Biener <rguenther@suse.de>
2608 Kelvin Nilsen <kelvin@gcc.gnu.org>
2611 * tree-vect-stmts.c (vectorizable_call): Don't call
2612 targetm.vectorize_builtin_md_vectorized_function if callee is
2615 2018-01-22 Carl Love <cel@us.ibm.com>
2617 * doc/extend.tex: Fix typo in second arg in
2618 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2620 2018-01-29 Richard Biener <rguenther@suse.de>
2622 PR tree-optimization/84086
2623 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2624 (flush_ssaname_freelist): When SSA names were released reset
2625 the SCEV hash table.
2627 2018-01-29 Richard Biener <rguenther@suse.de>
2629 PR tree-optimization/84057
2630 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2631 removed paths when removing edges.
2633 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2635 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2636 -mfunction-return=@var{choice}.
2638 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2641 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2643 (layout::print_source_line): Likewise.
2644 (test_get_line_width_without_trailing_whitespace): Add test cases.
2646 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2649 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2652 2018-01-26 Jim Wilson <jimw@sifive.com>
2654 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2656 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2659 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2661 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2662 and CMP + SUB-immediate -> SUBS.
2664 2018-01-26 Martin Sebor <msebor@redhat.com>
2666 PR tree-optimization/83896
2667 * tree-ssa-strlen.c (get_string_len): Rename...
2668 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2669 Avoid assuming length is constant.
2670 (handle_char_store): Use HOST_WIDE_INT for string length.
2672 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2675 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2676 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2678 2018-01-26 Richard Biener <rguenther@suse.de>
2680 PR rtl-optimization/84003
2681 * dse.c (record_store): Only record redundant stores when
2682 the earlier store aliases at least all accesses the later one does.
2684 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2686 PR rtl-optimization/83985
2687 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2688 REG_CFA_RESTORE insns.
2689 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2690 REG_CFA_RESTORE insns here.
2693 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2694 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2696 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2698 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2699 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2700 (arc_init): Likewise.
2701 (arc_override_options): Likewise.
2702 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2704 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2706 * config/arc/arc.h (TARGET_DBNZ): Define.
2707 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2708 properly set the tune attribute.
2709 (dbnz): Use TARGET_DBNZ guard.
2710 * config/arc/arc.opt (mtune): Add core3 option.
2712 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2714 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2715 recognize new pic like addresses.
2716 (arc_delegitimize_address): Clean up.
2718 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2720 * config/arc/arc-arches.def: Option mrf16 valid for all
2722 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2723 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2724 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2725 * config/arc/arc-tables.opt: Regenerate.
2726 * config/arc/arc.c (arc_conditional_register_usage): Handle
2727 reduced register file case.
2728 (arc_file_start): Set must have build attributes.
2729 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2731 * config/arc/arc.opt (mrf16): Add new option.
2732 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2733 * config/arc/genmultilib.awk: Handle new mrf16 option.
2734 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2735 * config/arc/t-multilib: Regenerate.
2736 * doc/invoke.texi (ARC Options): Document mrf16 option.
2738 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2740 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2741 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2742 (arc_attribute_table): Add 'secure_call' attribute.
2743 (arc_print_operand): Print secure call operand.
2744 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2746 (arc_is_secure_call_p): New function. * config/arc/arc.md
2747 (call_i): Add support for sjli instruction.
2748 (call_value_i): Likewise.
2749 * config/arc/constraints.md (Csc): New constraint.
2751 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2752 John Eric Martin <John.Martin@emmicro-us.com>
2754 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2755 * config/arc/arc.c (_arc_jli_section): New struct.
2756 (arc_jli_section): New type.
2757 (rc_jli_sections): New static variable.
2758 (arc_handle_jli_attribute): New function.
2759 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2760 (arc_file_end): New function.
2761 (TARGET_ASM_FILE_END): Define.
2762 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2763 (arc_add_jli_section): New function.
2764 (jli_call_scan): Likewise.
2765 (arc_reorg): Call jli_call_scan.
2766 (arc_output_addsi): Remove 'S' from printing asm operand.
2767 (arc_is_jli_call_p): New function.
2768 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2770 (movhi_insn): Likewise.
2771 (movsi_insn): Likewise.
2772 (movsi_set_cc_insn): Likewise.
2773 (loadqi_update): Likewise.
2774 (load_zeroextendqisi_update): Likewise.
2775 (load_signextendqisi_update): Likewise.
2776 (loadhi_update): Likewise.
2777 (load_zeroextendhisi_update): Likewise.
2778 (load_signextendhisi_update): Likewise.
2779 (loadsi_update): Likewise.
2780 (loadsf_update): Likewise.
2781 (movsicc_insn): Likewise.
2782 (bset_insn): Likewise.
2783 (bxor_insn): Likewise.
2784 (bclr_insn): Likewise.
2785 (bmsk_insn): Likewise.
2786 (bicsi3_insn): Likewise.
2787 (cmpsi_cc_c_insn): Likewise.
2788 (movsi_ne): Likewise.
2789 (movsi_cond_exec): Likewise.
2790 (clrsbsi2): Likewise.
2801 (call_i): Remove 'S' asm letter, add jli instruction.
2802 (call_value_i): Likewise.
2803 * config/arc/arc.op (mjli-always): New option.
2804 * config/arc/constraints.md (Cji): New constraint.
2805 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2807 (subsf3_fpx): Likewise.
2808 (mulsf3_fpx): Likewise.
2809 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2811 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2813 * doc/invoke.texi (ARC): Document mjli-always option.
2815 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2817 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2818 avoid addition with 0 and use incw and decw where possible.
2820 2018-01-26 Richard Biener <rguenther@suse.de>
2822 PR tree-optimization/81082
2823 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2824 association if it requires casting to unsigned.
2825 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2826 from fold_plusminus_mult_expr to catch important cases late when
2827 range info is available.
2829 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2831 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2832 * configure.ac (hidden_linkonce): New test.
2833 * configure: Regenerate.
2834 * config.in: Regenerate.
2836 2018-01-26 Julia Koval <julia.koval@intel.com>
2838 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2839 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2840 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2841 _mm_mask_bitshuffle_epi64_mask): Fix type.
2842 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2843 USI_FTYPE_V4DI_V4DI_USI): Remove.
2844 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2845 __builtin_ia32_vpshufbitqmb256_mask,
2846 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2847 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2848 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2850 2018-01-26 Alan Modra <amodra@gmail.com>
2853 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2854 UNSPEC_VBPERMQ. Sort other unspecs.
2856 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2858 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2860 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2863 * predict.c (drop_profile): Do not push/pop cfun; update also
2865 (handle_missing_profiles): Fix logic looking for zero profiles.
2867 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2870 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2871 on functions with #pragma omp declare simd or functions with simd
2873 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2874 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2875 Remove trailing \n from warning_at calls.
2877 2018-01-25 Tom de Vries <tom@codesourcery.com>
2880 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2881 for neutered workers.
2883 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2886 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2887 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2889 2018-01-24 Jeff Law <law@redhat.com>
2892 * i386.c (get_probe_interval): Move to earlier point.
2893 (ix86_compute_frame_layout): If -fstack-clash-protection and
2894 the frame is larger than the probe interval, then use pushes
2895 to save registers rather than reg->mem moves.
2896 (ix86_expand_prologue): Remove conditional for int_registers_saved
2899 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2902 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2903 min/max for never referenced object.
2905 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2908 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2910 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2911 attributes from DECL_ATTRIBUTES (decl) without affecting
2912 DECL_ATTRIBUTES (current_function_decl).
2913 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2914 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2916 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2918 PR tree-optimization/83979
2919 * fold-const.c (fold_comparison): Use constant_boolean_node
2920 instead of boolean_{true,false}_node.
2922 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2924 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2927 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2929 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2930 Simplify the clause that sets the length attribute.
2931 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2932 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2933 clause that sets the length attribute.
2934 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2936 2018-01-24 Tom de Vries <tom@codesourcery.com>
2939 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2940 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2941 Add strict parameter.
2942 (prevent_branch_around_nothing): Insert dummy insn between branch to
2943 label and label with no ptx insn inbetween.
2944 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2946 2018-01-24 Tom de Vries <tom@codesourcery.com>
2949 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2950 for neutered threads in warp.
2951 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2953 2018-01-24 Richard Biener <rguenther@suse.de>
2955 PR tree-optimization/83176
2956 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2959 2018-01-24 Richard Biener <rguenther@suse.de>
2961 PR tree-optimization/82819
2962 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2963 code generating pluses that are no-ops in the target precision.
2965 2018-01-24 Richard Biener <rguenther@suse.de>
2968 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2970 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2972 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2973 to merge probabilities.
2974 * predict.c (probably_never_executed): Also mark as cold functions
2975 with global 0 profile and guessed local profile.
2976 * profile-count.c (profile_probability::combine_with_count): New
2978 * profile-count.h (profile_probability::operator*,
2979 profile_probability::operator*=, profile_probability::operator/,
2980 profile_probability::operator/=): Reduce precision to adjusted
2981 and set value to guessed on contradictory divisions.
2982 (profile_probability::combine_with_freq): Remove.
2983 (profile_probability::combine_wiht_count): Declare.
2984 (profile_count::force_nonzero):: Set to adjusted.
2985 (profile_count::probability_in):: Set quality to adjusted.
2986 * tree-ssa-tail-merge.c (replace_block_by): Use
2989 2018-01-23 Andrew Waterman <andrew@sifive.com>
2990 Jim Wilson <jimw@sifive.com>
2992 * config/riscv/riscv.c (riscv_stack_boundary): New.
2993 (riscv_option_override): Set riscv_stack_boundary. Handle
2994 riscv_preferred_stack_boundary_arg.
2995 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2996 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2997 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2998 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2999 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
3000 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
3002 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
3005 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
3006 of struct ix86_frame.
3007 (ix86_expand_epilogue): Likewise. Add a local variable for
3008 the reg_save_offset field in struct ix86_frame.
3010 2018-01-23 Bin Cheng <bin.cheng@arm.com>
3012 PR tree-optimization/82604
3013 * tree-loop-distribution.c (enum partition_kind): New enum item
3014 PKIND_PARTIAL_MEMSET.
3015 (partition_builtin_p): Support above new enum item.
3016 (generate_code_for_partition): Ditto.
3017 (compute_access_range): Differentiate cases that equality can be
3018 proven at all loops, the innermost loops or no loops.
3019 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
3020 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
3021 (finalize_partitions, distribute_loop): Don't fuse partition of
3022 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
3023 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
3026 2018-01-23 Martin Liska <mliska@suse.cz>
3028 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
3029 order to ignore the predictor.
3030 (PRED_POLYMORPHIC_CALL): Likewise.
3031 (PRED_RECURSIVE_CALL): Likewise.
3033 2018-01-23 Martin Liska <mliska@suse.cz>
3035 * tree-profile.c (tree_profiling): Print function header to
3036 aware reader which function we are working on.
3037 * value-prof.c (gimple_find_values_to_profile): Do not print
3038 not interesting value histograms.
3040 2018-01-23 Martin Liska <mliska@suse.cz>
3042 * profile-count.h (enum profile_quality): Add
3043 profile_uninitialized as the first value. Do not number values
3044 as they are zero based.
3045 (profile_count::verify): Update sanity check.
3046 (profile_probability::verify): Likewise.
3048 2018-01-23 Nathan Sidwell <nathan@acm.org>
3050 * doc/invoke.texi (ffor-scope): Deprecate.
3052 2018-01-23 David Malcolm <dmalcolm@redhat.com>
3054 PR tree-optimization/83510
3055 * domwalk.c (set_all_edges_as_executable): New function.
3056 (dom_walker::dom_walker): Convert bool param
3057 "skip_unreachable_blocks" to enum reachability. Move setup of
3058 edge flags to set_all_edges_as_executable and only do it when
3059 reachability is REACHABLE_BLOCKS.
3060 * domwalk.h (enum dom_walker::reachability): New enum.
3061 (dom_walker::dom_walker): Convert bool param
3062 "skip_unreachable_blocks" to enum reachability.
3063 (set_all_edges_as_executable): New decl.
3064 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
3065 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
3067 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
3068 but converting true to REACHABLE_BLOCKS.
3069 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
3071 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
3072 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
3073 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
3075 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
3076 if check_all_array_refs will be called.
3078 2018-01-23 David Malcolm <dmalcolm@redhat.com>
3080 * tree.c (selftest::test_location_wrappers): Add more test
3083 2018-01-23 David Malcolm <dmalcolm@redhat.com>
3085 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
3086 (selftest::test_bit_in_range): Likewise.
3088 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
3091 * doc/sourcebuild.texi (vect_float): Say that the selector
3092 only describes the situation when -funsafe-math-optimizations is on.
3093 (vect_float_strict): Document.
3095 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
3097 PR tree-optimization/83965
3098 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
3099 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
3100 instead of checking only for a reduction.
3101 (vect_recog_widen_sum_pattern): Likewise.
3103 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3105 * predict.c (probably_never_executed): Only use precise profile info.
3106 (compute_function_frequency): Skip after inlining hack since we now
3107 have quality checking.
3109 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3111 * profile-count.h (profile_probability::very_unlikely,
3112 profile_probability::unlikely, profile_probability::even): Set
3113 precision to guessed.
3115 2018-01-23 Richard Biener <rguenther@suse.de>
3117 PR tree-optimization/83963
3118 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
3119 Properly terminate dominator walk when crossing the exit edge not
3120 when visiting its source block.
3122 2018-01-23 Jakub Jelinek <jakub@redhat.com>
3125 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
3126 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
3128 2018-01-22 Jakub Jelinek <jakub@redhat.com>
3130 PR tree-optimization/83957
3131 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
3132 semicolon after for body surrounded by braces.
3134 PR tree-optimization/83081
3135 * profile-count.h (profile_probability::split): New method.
3136 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
3137 Use profile_probability::split.
3138 (do_compare_rtx_and_jump): Fix adjustment of probabilities
3139 when splitting a single conditional jump into 2.
3141 2018-01-22 David Malcolm <dmalcolm@redhat.com>
3143 PR tree-optimization/69452
3144 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
3147 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3149 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
3150 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
3151 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
3153 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3155 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
3156 * config/rl78/rl78.md (movdi): New define_expand.
3157 * config/rl78/rl78.c (rl78_split_movdi): New function.
3159 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
3162 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
3164 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
3165 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
3166 128-bit to produce an UNSPEC move to get the double word with the
3167 signbit and then a shift directly to do signbit.
3168 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
3169 implementation with a new version that just does either a direct
3170 move or a regular move. Move memory interface to separate insns.
3171 Move insns so they are next to the expander.
3172 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
3173 with signbit move. Split big and little endian case.
3174 (signbit<mode>2_dm_mem_le): Likewise.
3175 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
3176 (signbit<mode>2_dm2): Likewise.
3178 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3180 * config/rl78/rl78.md (anddi3): New define_expand.
3182 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3184 * config/rl78/rl78.md (umindi3): New define_expand.
3186 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3188 * config/rl78/rl78.md (smindi3): New define_expand.
3190 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3192 * config/rl78/rl78.md (smaxdi3): New define_expand.
3194 2018-01-22 Carl Love <cel@us.ibm.com>
3196 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
3197 LVX_V1TI): Add macro expansion.
3198 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
3199 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
3200 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
3201 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
3202 Change check to determine if the instruction is a byte reversing
3203 entry. Fix typo in comment.
3204 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
3205 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
3206 Add def_builtin calls for new builtins.
3207 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
3208 Add define_insn expansion.
3210 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3212 * config/rl78/rl78.md (umaxdi3): New define_expand.
3214 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3216 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
3217 for non-QImode registers.
3219 2018-01-22 Richard Biener <rguenther@suse.de>
3221 PR tree-optimization/83963
3222 * graphite-scop-detection.c (scop_detection::get_sese): Delay
3223 including the loop exit block.
3224 (scop_detection::merge_sese): Likewise.
3225 (scop_detection::add_scop): Do it here instead.
3227 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3229 * doc/sourcebuild.texi (arm_softfloat): Document.
3231 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
3234 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
3235 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
3236 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
3238 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3239 David Edelsohn <dje.gcc@gmail.com>
3242 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3243 Change "crset eq" to "crset 2".
3244 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3245 (*call_indirect_aix<mode>_nospec): Likewise.
3246 (*call_value_indirect_aix<mode>_nospec): Likewise.
3247 (*call_indirect_elfv2<mode>_nospec): Likewise.
3248 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
3249 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
3250 change assembly output from . to $.
3251 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3252 (indirect_jump<mode>_nospec): Change assembly output from . to $.
3253 (*tablejump<mode>_internal1_nospec): Likewise.
3255 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
3258 * config/sh/sh_optimize_sett_clrt.cc:
3259 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
3261 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
3263 PR tree-optimization/83940
3264 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
3265 offset_dt to vect_constant_def rather than vect_unknown_def_type.
3266 (vect_check_load_store_mask): Add a mask_dt_out parameter and
3267 use it to pass back the definition type.
3268 (vect_check_store_rhs): Likewise rhs_dt_out.
3269 (vect_build_gather_load_calls): Add a mask_dt argument and use
3270 it instead of a call to vect_is_simple_use.
3271 (vectorizable_store): Update calls to vect_check_load_store_mask
3272 and vect_check_store_rhs. Use the dt returned by the latter instead
3273 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
3274 instead of calls to vect_is_simple_use. Pass the scalar rather
3275 than the vector operand to vect_is_simple_use when handling
3276 second and subsequent copies of an rhs value.
3277 (vectorizable_load): Update calls to vect_check_load_store_mask
3278 and vect_build_gather_load_calls. Use the cached mask_dt and
3279 gs_info.offset_dt instead of calls to vect_is_simple_use.
3281 2018-01-20 Jakub Jelinek <jakub@redhat.com>
3284 * tree-emutls.c: Include gimplify.h.
3285 (lower_emutls_2): New function.
3286 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
3287 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
3288 it before further processing.
3291 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
3292 UINTVAL (trueop1) instead of INTVAL (op1).
3294 2018-01-19 Jakub Jelinek <jakub@redhat.com>
3298 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
3299 INCOMING_FRAME_SP_OFFSET if not defined.
3300 (scan_trace): Add ENTRY argument. If true and
3301 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
3302 emit a note to adjust the CFA offset.
3303 (create_cfi_notes): Adjust scan_trace callers.
3304 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
3305 INCOMING_FRAME_SP_OFFSET in the CIE.
3306 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
3307 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
3309 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
3310 * doc/tm.texi: Regenerated.
3312 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
3314 PR rtl-optimization/83147
3315 * lra-constraints.c (remove_inheritance_pseudos): Use
3316 lra_substitute_pseudo_within_insn.
3318 2018-01-19 Tom de Vries <tom@codesourcery.com>
3319 Cesar Philippidis <cesar@codesourcery.com>
3322 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
3324 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
3327 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
3328 spaces for function labels.
3330 2018-01-19 Martin Liska <mliska@suse.cz>
3332 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
3333 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
3334 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
3335 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
3336 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
3337 (PRED_CONST_RETURN): Change from 69 to 65.
3338 (PRED_NULL_RETURN): Change from 91 to 71.
3339 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
3340 (PRED_LOOP_GUARD): Change from 66 to 73.
3342 2018-01-19 Martin Liska <mliska@suse.cz>
3344 * predict.c (predict_insn_def): Add new assert.
3345 (struct branch_predictor): Change type to signed integer.
3346 (test_prediction_value_range): Amend test to cover
3348 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
3349 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
3350 (PRED_LOOP_ITERATIONS_MAX): Likewise.
3351 (PRED_LOOP_IV_COMPARE): Likewise.
3352 * predict.h (PROB_UNINITIALIZED): Define new constant.
3354 2018-01-19 Martin Liska <mliska@suse.cz>
3356 * predict.c (dump_prediction): Add new format for
3357 analyze_brprob.py script which is enabled with -details
3359 * profile-count.h (precise_p): New function.
3361 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3363 PR tree-optimization/83922
3364 * tree-vect-loop.c (vect_verify_full_masking): Return false if
3365 there are no statements that need masking.
3366 (vect_active_double_reduction_p): New function.
3367 (vect_analyze_loop_operations): Use it when handling phis that
3368 are not in the loop header.
3370 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3372 PR tree-optimization/83914
3373 * tree-vect-loop.c (vectorizable_induction): Don't convert
3374 init_expr or apply the peeling adjustment for inductions
3375 that are nested within the vectorized loop.
3377 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3379 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
3382 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3386 * function.h (gimplify_parameters): Add gimple_seq * argument.
3387 * function.c: Include gimple.h and options.h.
3388 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
3389 for the added local temporaries if needed.
3390 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
3391 if there are any parameter cleanups, wrap whole body into a
3392 try/finally with the cleanups.
3394 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
3397 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
3398 Use GET_MODE_CLASS for scalar floating point.
3400 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3404 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
3405 Fix call of call_cgraph_insertion_hooks.
3407 2018-01-18 Martin Sebor <msebor@redhat.com>
3409 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
3411 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3414 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3417 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3420 * common.opt: (-ffile-prefix-map): New option.
3421 * opts.c (common_handle_option): Defer it.
3422 * opts-global.c (handle_common_deferred_options): Handle it.
3423 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3424 * file-prefix-map.h: New file.
3425 (remap_debug_filename, add_debug_prefix_map): ...here.
3426 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3427 * final.c (debug_prefix_map, add_debug_prefix_map
3428 remap_debug_filename): Move to...
3429 * file-prefix-map.c: New file.
3430 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3431 generalize, get rid of alloca(), use strrchr() instead of strchr().
3432 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3433 Implement in terms of add_prefix_map().
3434 (remap_macro_filename, remap_debug_filename): Implement in term of
3436 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3437 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3438 * dbxout.c: Include file-prefix-map.h.
3439 * varasm.c: Likewise.
3440 * vmsdbgout.c: Likewise.
3441 * xcoffout.c: Likewise.
3442 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3443 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3444 * doc/invoke.texi (-ffile-prefix-map): Document.
3445 (-fdebug-prefix-map): Update description.
3447 2018-01-18 Martin Liska <mliska@suse.cz>
3449 * config/i386/i386.c (indirect_thunk_name): Document that also
3451 (output_indirect_thunk): Document why both instructions
3452 (pause and lfence) are generated.
3454 2018-01-18 Richard Biener <rguenther@suse.de>
3456 PR tree-optimization/83887
3457 * graphite-scop-detection.c
3458 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3459 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3460 (scop_detection::merge_sese): Re-implement with a flood-fill
3461 algorithm that properly finds a SESE region if it exists.
3463 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3466 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3467 pointer_diff optimizations use view_convert instead of convert.
3469 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3471 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3472 Generate different code for -mno-speculate-indirect-jumps.
3473 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3474 (*call_indirect_aix<mode>): Disable for
3475 -mno-speculate-indirect-jumps.
3476 (*call_indirect_aix<mode>_nospec): New define_insn.
3477 (*call_value_indirect_aix<mode>): Disable for
3478 -mno-speculate-indirect-jumps.
3479 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3480 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3481 -mno-speculate-indirect-jumps.
3482 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3484 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3486 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3487 long double type, set the flags for noting the default long double
3488 type, even if we don't pass or return a long double type.
3490 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3493 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3496 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3498 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3499 support for merge[hl].
3500 (fold_mergehl_helper): New helper function.
3501 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3502 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3503 (altivec_vmrglw_direct): Add xxmrglw insn.
3505 2018-01-17 Andrew Waterman <andrew@sifive.com>
3507 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3508 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3510 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3513 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3514 call the lto_location_cache before reading the
3515 DECL_SOURCE_LOCATION of the types.
3517 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3518 Richard Sandiford <richard.sandiford@linaro.org>
3520 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3521 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3522 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3523 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3524 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3526 * config/aarch64/constraints.md (aarch64_movti_operand):
3528 * config/aarch64/predicates.md (Uti): Add new constraint.
3530 2018-01-17 Carl Love <cel@us.ibm.com>
3532 * config/rs6000/vsx.md (define_expand xl_len_r,
3533 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3534 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3536 (define_expand, define_insn): Move the shift left from the
3537 define_insn to the define_expand for lxvl and stxvl instructions.
3538 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3539 and XL_LEN_R definitions to PURE.
3541 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3543 * config/i386/i386.c (indirect_thunk_name): Declare regno
3544 as unsigned int. Compare regno with INVALID_REGNUM.
3545 (output_indirect_thunk): Ditto.
3546 (output_indirect_thunk_function): Ditto.
3547 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3548 in the call to output_indirect_thunk_function.
3550 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3553 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3554 rather than the size of inner_type to determine the stack slot size
3555 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3557 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3560 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3563 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3565 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3566 endian Linux systems to optionally enable multilibs for selecting
3567 the long double type if the user configured an explicit type.
3568 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3569 have no long double multilibs if not defined.
3570 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3571 warn if the user used -mabi={ieee,ibm}longdouble and we built
3572 multilibs for long double.
3573 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3574 appropriate multilib option.
3575 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3577 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3578 for building long double multilibs.
3579 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3581 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3583 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3586 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3588 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3591 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3594 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3597 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3599 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3600 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3602 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3604 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3605 different rtl trees depending on TARGET_64BIT.
3606 (rs6000_gen_lvx): Likewise.
3608 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3610 * config/visium/visium.md (nop): Tweak comment.
3611 (hazard_nop): Likewise.
3613 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3615 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3616 -mspeculate-indirect-jumps.
3617 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3618 for -mno-speculate-indirect-jumps.
3619 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3620 (*call_value_indirect_elfv2<mode>): Disable for
3621 -mno-speculate-indirect-jumps.
3622 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3623 (indirect_jump): Emit different RTL for
3624 -mno-speculate-indirect-jumps.
3625 (*indirect_jump<mode>): Disable for
3626 -mno-speculate-indirect-jumps.
3627 (*indirect_jump<mode>_nospec): New define_insn.
3628 (tablejump): Emit different RTL for
3629 -mno-speculate-indirect-jumps.
3630 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3631 (tablejumpsi_nospec): New define_expand.
3632 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3633 (tablejumpdi_nospec): New define_expand.
3634 (*tablejump<mode>_internal1): Disable for
3635 -mno-speculate-indirect-jumps.
3636 (*tablejump<mode>_internal1_nospec): New define_insn.
3637 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3640 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3642 * caller-save.c (insert_save): Drop unnecessary parameter. All
3645 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3646 Richard Biener <rguenth@suse.de>
3649 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3650 return early, inline manually is_gimple_sizepos. Make sure if we
3651 call gimplify_expr we don't end up with a gimple constant.
3652 * tree.c (variably_modified_type_p): Don't return true for
3653 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3654 * gimplify.h (is_gimple_sizepos): Remove.
3656 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3658 PR tree-optimization/83857
3659 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3660 vectorizable_live_operation for pure SLP statements.
3661 (vectorizable_live_operation): Handle PHIs.
3663 2018-01-16 Richard Biener <rguenther@suse.de>
3665 PR tree-optimization/83867
3666 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3667 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3669 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3672 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3673 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3674 If off is not INTEGER_CST, issue a may not be aligned warning
3675 rather than isn't aligned. Use isn%'t rather than isn't.
3676 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3678 <case MULT_EXPR>: Improve the case when bottom and one of the
3679 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3680 operand, in that case check if the other operand is multiple of
3681 bottom divided by the INTEGER_CST operand.
3683 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3686 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3687 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3688 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3689 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3690 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3691 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3692 (pa_function_arg_size): New function.
3694 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3696 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3697 in a separate statement.
3699 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3701 PR tree-optimization/83847
3702 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3703 group gathers and scatters.
3705 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3707 PR rtl-optimization/86620
3708 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3710 PR rtl-optimization/83213
3711 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3712 to last if both are JUMP_INSNs.
3714 PR tree-optimization/83843
3715 * gimple-ssa-store-merging.c
3716 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3717 store_immediate_info for bswap/nop orig_stores.
3719 2018-01-15 Andrew Waterman <andrew@sifive.com>
3721 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3723 <UDIV>: Increase cost if !TARGET_DIV.
3725 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3727 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3728 (define_attr "cr_logical_3op"): New.
3729 (cceq_ior_compare): Adjust.
3730 (cceq_ior_compare_complement): Adjust.
3731 (*cceq_rev_compare): Adjust.
3732 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3733 (is_cracked_insn): Adjust.
3734 (insn_must_be_first_in_group): Adjust.
3735 * config/rs6000/40x.md: Adjust.
3736 * config/rs6000/440.md: Adjust.
3737 * config/rs6000/476.md: Adjust.
3738 * config/rs6000/601.md: Adjust.
3739 * config/rs6000/603.md: Adjust.
3740 * config/rs6000/6xx.md: Adjust.
3741 * config/rs6000/7450.md: Adjust.
3742 * config/rs6000/7xx.md: Adjust.
3743 * config/rs6000/8540.md: Adjust.
3744 * config/rs6000/cell.md: Adjust.
3745 * config/rs6000/e300c2c3.md: Adjust.
3746 * config/rs6000/e500mc.md: Adjust.
3747 * config/rs6000/e500mc64.md: Adjust.
3748 * config/rs6000/e5500.md: Adjust.
3749 * config/rs6000/e6500.md: Adjust.
3750 * config/rs6000/mpc.md: Adjust.
3751 * config/rs6000/power4.md: Adjust.
3752 * config/rs6000/power5.md: Adjust.
3753 * config/rs6000/power6.md: Adjust.
3754 * config/rs6000/power7.md: Adjust.
3755 * config/rs6000/power8.md: Adjust.
3756 * config/rs6000/power9.md: Adjust.
3757 * config/rs6000/rs64.md: Adjust.
3758 * config/rs6000/titan.md: Adjust.
3760 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3762 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3763 ix86_indirect_branch_register logic.
3765 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3767 * config/i386/constraints.md (Bs): Update
3768 ix86_indirect_branch_register check. Don't check
3769 ix86_indirect_branch_register with GOT_memory_operand.
3771 * config/i386/predicates.md (GOT_memory_operand): Don't check
3772 ix86_indirect_branch_register here.
3773 (GOT32_symbol_operand): Likewise.
3775 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3777 * config/i386/predicates.md (constant_call_address_operand):
3778 Rewrite ix86_indirect_branch_register logic.
3779 (sibcall_insn_operand): Likewise.
3781 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3783 * config/i386/constraints.md (Bs): Replace
3784 ix86_indirect_branch_thunk_register with
3785 ix86_indirect_branch_register.
3787 * config/i386/i386.md (indirect_jump): Likewise.
3788 (tablejump): Likewise.
3789 (*sibcall_memory): Likewise.
3790 (*sibcall_value_memory): Likewise.
3791 Peepholes of indirect call and jump via memory: Likewise.
3792 * config/i386/i386.opt: Likewise.
3793 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3794 (GOT_memory_operand): Likewise.
3795 (call_insn_operand): Likewise.
3796 (sibcall_insn_operand): Likewise.
3797 (GOT32_symbol_operand): Likewise.
3799 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3802 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3803 type rather than type addr's type points to.
3804 (expand_omp_atomic_mutex): Likewise.
3805 (expand_omp_atomic): Likewise.
3807 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3810 * config/i386/i386.c (output_indirect_thunk_function): Use
3811 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3812 for __x86_return_thunk.
3814 2018-01-15 Richard Biener <rguenther@suse.de>
3817 * expmed.c (extract_bit_field_1): Fix typo.
3819 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3822 * config/arm/iterators.md (VF): New mode iterator.
3823 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3824 Remove integer-related logic from pattern.
3825 (neon_vabd<mode>_3): Likewise.
3827 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3830 * common.opt (fstrict-overflow): No longer an alias.
3831 (fwrapv-pointer): New option.
3832 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3833 also for pointer types based on flag_wrapv_pointer.
3834 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3835 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3836 opts->x_flag_wrapv got set.
3837 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3838 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3839 POINTER_TYPE_OVERFLOW_UNDEFINED.
3840 * match.pd: Likewise in address comparison pattern.
3841 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3843 2018-01-15 Richard Biener <rguenther@suse.de>
3846 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3847 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3848 Reset type names to their identifier if their TYPE_DECL doesn't
3849 have linkage (and thus is used for ODR and devirt).
3850 (save_debug_info_for_decl): Remove.
3851 (save_debug_info_for_type): Likewise.
3852 (add_tree_to_fld_list): Adjust.
3853 * tree-pretty-print.c (dump_generic_node): Make dumping of
3854 type names more robust.
3856 2018-01-15 Richard Biener <rguenther@suse.de>
3858 * BASE-VER: Bump to 8.0.1.
3860 2018-01-14 Martin Sebor <msebor@redhat.com>
3863 * builtins.c (check_access): Avoid warning when the no-warning bit
3866 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3868 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3869 * ira-color (allocno_hard_regs_compare): Likewise.
3871 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3874 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3875 Use .pushsection/.popsection.
3877 2018-01-14 Martin Sebor <msebor@redhat.com>
3880 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3882 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3884 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3885 entry from extra_headers.
3886 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3887 extra_headers, make the list bitwise identical to the i?86-*-* one.
3889 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3891 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3892 -mcmodel=large with -mindirect-branch=thunk,
3893 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3894 -mfunction-return=thunk-extern.
3895 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3896 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3897 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3899 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3901 * config/i386/i386.c (print_reg): Print the name of the full
3902 integer register without '%'.
3903 (ix86_print_operand): Handle 'V'.
3904 * doc/extend.texi: Document 'V' modifier.
3906 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3908 * config/i386/constraints.md (Bs): Disallow memory operand for
3909 -mindirect-branch-register.
3911 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3912 (GOT_memory_operand): Likewise.
3913 (call_insn_operand): Likewise.
3914 (sibcall_insn_operand): Likewise.
3915 (GOT32_symbol_operand): Likewise.
3916 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3917 for -mindirect-branch-register.
3918 (tablejump): Likewise.
3919 (*sibcall_memory): Likewise.
3920 (*sibcall_value_memory): Likewise.
3921 Disallow peepholes of indirect call and jump via memory for
3922 -mindirect-branch-register.
3923 (*call_pop): Replace m with Bw.
3924 (*call_value_pop): Likewise.
3925 (*sibcall_pop_memory): Replace m with Bs.
3926 * config/i386/i386.opt (mindirect-branch-register): New option.
3927 * doc/invoke.texi: Document -mindirect-branch-register option.
3929 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3931 * config/i386/i386-protos.h (ix86_output_function_return): New.
3932 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3933 set function_return_type.
3934 (indirect_thunk_name): Add ret_p to indicate thunk for function
3936 (output_indirect_thunk_function): Pass false to
3937 indirect_thunk_name.
3938 (ix86_output_indirect_branch_via_reg): Likewise.
3939 (ix86_output_indirect_branch_via_push): Likewise.
3940 (output_indirect_thunk_function): Create alias for function
3941 return thunk if regno < 0.
3942 (ix86_output_function_return): New function.
3943 (ix86_handle_fndecl_attribute): Handle function_return.
3944 (ix86_attribute_table): Add function_return.
3945 * config/i386/i386.h (machine_function): Add
3946 function_return_type.
3947 * config/i386/i386.md (simple_return_internal): Use
3948 ix86_output_function_return.
3949 (simple_return_internal_long): Likewise.
3950 * config/i386/i386.opt (mfunction-return=): New option.
3951 (indirect_branch): Mention -mfunction-return=.
3952 * doc/extend.texi: Document function_return function attribute.
3953 * doc/invoke.texi: Document -mfunction-return= option.
3955 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3957 * config/i386/i386-opts.h (indirect_branch): New.
3958 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3959 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3960 with local indirect jump when converting indirect call and jump.
3961 (ix86_set_indirect_branch_type): New.
3962 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3963 (indirectlabelno): New.
3964 (indirect_thunk_needed): Likewise.
3965 (indirect_thunk_bnd_needed): Likewise.
3966 (indirect_thunks_used): Likewise.
3967 (indirect_thunks_bnd_used): Likewise.
3968 (INDIRECT_LABEL): Likewise.
3969 (indirect_thunk_name): Likewise.
3970 (output_indirect_thunk): Likewise.
3971 (output_indirect_thunk_function): Likewise.
3972 (ix86_output_indirect_branch_via_reg): Likewise.
3973 (ix86_output_indirect_branch_via_push): Likewise.
3974 (ix86_output_indirect_branch): Likewise.
3975 (ix86_output_indirect_jmp): Likewise.
3976 (ix86_code_end): Call output_indirect_thunk_function if needed.
3977 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3979 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3980 (ix86_attribute_table): Add indirect_branch.
3981 * config/i386/i386.h (machine_function): Add indirect_branch_type
3982 and has_local_indirect_jump.
3983 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3985 (tablejump): Likewise.
3986 (*indirect_jump): Use ix86_output_indirect_jmp.
3987 (*tablejump_1): Likewise.
3988 (simple_return_indirect_internal): Likewise.
3989 * config/i386/i386.opt (mindirect-branch=): New option.
3990 (indirect_branch): New.
3993 (thunk-inline): Likewise.
3994 (thunk-extern): Likewise.
3995 * doc/extend.texi: Document indirect_branch function attribute.
3996 * doc/invoke.texi: Document -mindirect-branch= option.
3998 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
4001 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
4003 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
4005 * ipa-inline.c (want_inline_small_function_p): Return false if
4006 inlining has already failed with CIF_FINAL_ERROR.
4007 (update_caller_keys): Call want_inline_small_function_p before
4009 (update_callee_keys): Likewise.
4011 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
4013 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
4015 (rs6000_quadword_masked_address_p): Likewise.
4016 (quad_aligned_load_p): Likewise.
4017 (quad_aligned_store_p): Likewise.
4018 (const_load_sequence_p): Add comment to describe the outer-most loop.
4019 (mimic_memory_attributes_and_flags): New function.
4020 (rs6000_gen_stvx): Likewise.
4021 (replace_swapped_aligned_store): Likewise.
4022 (rs6000_gen_lvx): Likewise.
4023 (replace_swapped_aligned_load): Likewise.
4024 (replace_swapped_load_constant): Capitalize argument name in
4025 comment describing this function.
4026 (rs6000_analyze_swaps): Add a third pass to search for vector loads
4027 and stores that access quad-word aligned addresses and replace
4028 with stvx or lvx instructions when appropriate.
4029 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
4030 New function prototype.
4031 (rs6000_quadword_masked_address_p): Likewise.
4032 (rs6000_gen_lvx): Likewise.
4033 (rs6000_gen_stvx): Likewise.
4034 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
4035 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
4036 when memory address is aligned.
4037 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
4038 this split to select lvx instruction when memory address is aligned.
4039 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
4040 instruction when memory address is aligned.
4041 (*vsx_le_perm_load_v16qi): Likewise.
4042 (four unnamed splitters): Modify to select the stvx instruction
4043 when memory is aligned.
4045 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
4047 * predict.c (determine_unlikely_bbs): Handle correctly BBs
4048 which appears in the queue multiple times.
4050 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4051 Alan Hayward <alan.hayward@arm.com>
4052 David Sherwood <david.sherwood@arm.com>
4054 * tree-vectorizer.h (vec_lower_bound): New structure.
4055 (_loop_vec_info): Add check_nonzero and lower_bounds.
4056 (LOOP_VINFO_CHECK_NONZERO): New macro.
4057 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
4058 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
4059 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
4060 fields. Make seg_len the distance travelled, not including the
4062 (dr_direction_indicator): Declare.
4063 (dr_zero_step_indicator): Likewise.
4064 (dr_known_forward_stride_p): Likewise.
4065 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
4067 (runtime_alias_check_p): Allow runtime alias checks with
4069 (operator ==): Compare access_size and align.
4070 (prune_runtime_alias_test_list): Rework for new distinction between
4071 the access_size and seg_len.
4072 (create_intersect_range_checks_index): Likewise. Cope with polynomial
4074 (get_segment_min_max): New function.
4075 (create_intersect_range_checks): Use it.
4076 (dr_step_indicator): New function.
4077 (dr_direction_indicator): Likewise.
4078 (dr_zero_step_indicator): Likewise.
4079 (dr_known_forward_stride_p): Likewise.
4080 * tree-loop-distribution.c (data_ref_segment_size): Return
4081 DR_STEP * (niters - 1).
4082 (compute_alias_check_pairs): Update call to the dr_with_seg_len
4084 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
4085 (vect_preserves_scalar_order_p): New function, split out from...
4086 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
4087 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
4088 (vect_vfa_access_size): New function.
4089 (vect_vfa_align): Likewise.
4090 (vect_compile_time_alias): Take access_size_a and access_b arguments.
4091 (dump_lower_bound): New function.
4092 (vect_check_lower_bound): Likewise.
4093 (vect_small_gap_p): Likewise.
4094 (vectorizable_with_step_bound_p): Likewise.
4095 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
4096 depencies if the vectorization factor is 1. Convert the checks
4097 for nonzero steps into checks on the bounds of DR_STEP. Try using
4098 a bunds check for variable steps if the minimum required step is
4099 relatively small. Update calls to the dr_with_seg_len
4100 constructor and to vect_compile_time_alias.
4101 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
4103 (vect_loop_versioning): Call it.
4104 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
4106 (vect_estimate_min_profitable_iters): Account for any bounds checks.
4108 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4109 Alan Hayward <alan.hayward@arm.com>
4110 David Sherwood <david.sherwood@arm.com>
4112 * doc/sourcebuild.texi (vect_scatter_store): Document.
4113 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
4115 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
4117 * genopinit.c (main): Add supports_vec_scatter_store and
4118 supports_vec_scatter_store_cached to target_optabs.
4119 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
4120 IFN_MASK_SCATTER_STORE.
4121 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
4123 * internal-fn.h (internal_store_fn_p): Declare.
4124 (internal_fn_stored_value_index): Likewise.
4125 * internal-fn.c (scatter_store_direct): New macro.
4126 (expand_scatter_store_optab_fn): New function.
4127 (direct_scatter_store_optab_supported_p): New macro.
4128 (internal_store_fn_p): New function.
4129 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
4130 IFN_MASK_SCATTER_STORE.
4131 (internal_fn_mask_index): Likewise.
4132 (internal_fn_stored_value_index): New function.
4133 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
4135 * optabs-query.h (supports_vec_scatter_store_p): Declare.
4136 * optabs-query.c (supports_vec_scatter_store_p): New function.
4137 * tree-vectorizer.h (vect_get_store_rhs): Declare.
4138 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
4139 true for scatter stores.
4140 (vect_gather_scatter_fn_p): Handle scatter stores too.
4141 (vect_check_gather_scatter): Consider using scatter stores if
4142 supports_vec_scatter_store_p.
4143 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
4145 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
4146 internal_fn_stored_value_index.
4147 (check_load_store_masking): Handle scatter stores too.
4148 (vect_get_store_rhs): Make public.
4149 (vectorizable_call): Use internal_store_fn_p.
4150 (vectorizable_store): Handle scatter store internal functions.
4151 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
4152 when deciding whether the end of the group has been reached.
4153 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
4154 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
4155 (mask_scatter_store<mode>): New insns.
4157 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4158 Alan Hayward <alan.hayward@arm.com>
4159 David Sherwood <david.sherwood@arm.com>
4161 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
4162 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
4163 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
4165 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
4166 Use vect_truncate_gather_scatter_offset if we can't treat the
4167 operation as a normal gather load or scatter store.
4168 (get_group_load_store_type): Take the gather_scatter_info
4169 as argument. Try using a gather load or scatter store for
4170 single-element groups.
4171 (get_load_store_type): Update calls to get_group_load_store_type
4172 and vect_use_strided_gather_scatters_p.
4174 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4175 Alan Hayward <alan.hayward@arm.com>
4176 David Sherwood <david.sherwood@arm.com>
4178 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
4179 optional tree argument.
4180 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
4182 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
4183 but continue to use the current value as a fallback.
4184 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
4185 to compare the updates.
4186 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
4187 (get_load_store_type): Use it when handling a strided access.
4188 (vect_get_strided_load_store_ops): New function.
4189 (vect_get_data_ptr_increment): Likewise.
4190 (vectorizable_load): Handle strided gather loads. Always pass
4191 a step to vect_create_data_ref_ptr and bump_vector_ptr.
4193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4194 Alan Hayward <alan.hayward@arm.com>
4195 David Sherwood <david.sherwood@arm.com>
4197 * doc/md.texi (gather_load@var{m}): Document.
4198 (mask_gather_load@var{m}): Likewise.
4199 * genopinit.c (main): Add supports_vec_gather_load and
4200 supports_vec_gather_load_cached to target_optabs.
4201 * optabs-tree.c (init_tree_optimization_optabs): Use
4202 ggc_cleared_alloc to allocate target_optabs.
4203 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
4204 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
4206 * internal-fn.h (internal_load_fn_p): Declare.
4207 (internal_gather_scatter_fn_p): Likewise.
4208 (internal_fn_mask_index): Likewise.
4209 (internal_gather_scatter_fn_supported_p): Likewise.
4210 * internal-fn.c (gather_load_direct): New macro.
4211 (expand_gather_load_optab_fn): New function.
4212 (direct_gather_load_optab_supported_p): New macro.
4213 (direct_internal_fn_optab): New function.
4214 (internal_load_fn_p): Likewise.
4215 (internal_gather_scatter_fn_p): Likewise.
4216 (internal_fn_mask_index): Likewise.
4217 (internal_gather_scatter_fn_supported_p): Likewise.
4218 * optabs-query.c (supports_at_least_one_mode_p): New function.
4219 (supports_vec_gather_load_p): Likewise.
4220 * optabs-query.h (supports_vec_gather_load_p): Declare.
4221 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
4222 and memory_type field.
4223 (NUM_PATTERNS): Bump to 15.
4224 * tree-vect-data-refs.c: Include internal-fn.h.
4225 (vect_gather_scatter_fn_p): New function.
4226 (vect_describe_gather_scatter_call): Likewise.
4227 (vect_check_gather_scatter): Try using internal functions for
4228 gather loads. Recognize existing calls to a gather load function.
4229 (vect_analyze_data_refs): Consider using gather loads if
4230 supports_vec_gather_load_p.
4231 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
4232 (vect_get_gather_scatter_offset_type): Likewise.
4233 (vect_convert_mask_for_vectype): Likewise.
4234 (vect_add_conversion_to_patterm): Likewise.
4235 (vect_try_gather_scatter_pattern): Likewise.
4236 (vect_recog_gather_scatter_pattern): New pattern recognizer.
4237 (vect_vect_recog_func_ptrs): Add it.
4238 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
4239 internal_fn_mask_index and internal_gather_scatter_fn_p.
4240 (check_load_store_masking): Take the gather_scatter_info as an
4241 argument and handle gather loads.
4242 (vect_get_gather_scatter_ops): New function.
4243 (vectorizable_call): Check internal_load_fn_p.
4244 (vectorizable_load): Likewise. Handle gather load internal
4246 (vectorizable_store): Update call to check_load_store_masking.
4247 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
4248 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
4249 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
4250 (aarch64_gather_scale_operand_d): New predicates.
4251 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
4252 (mask_gather_load<mode>): New insns.
4254 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4255 Alan Hayward <alan.hayward@arm.com>
4256 David Sherwood <david.sherwood@arm.com>
4258 * optabs.def (fold_left_plus_optab): New optab.
4259 * doc/md.texi (fold_left_plus_@var{m}): Document.
4260 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
4261 * internal-fn.c (fold_left_direct): Define.
4262 (expand_fold_left_optab_fn): Likewise.
4263 (direct_fold_left_optab_supported_p): Likewise.
4264 * fold-const-call.c (fold_const_fold_left): New function.
4265 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
4266 * tree-parloops.c (valid_reduction_p): New function.
4267 (gather_scalar_reductions): Use it.
4268 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
4269 (vect_finish_replace_stmt): Declare.
4270 * tree-vect-loop.c (fold_left_reduction_fn): New function.
4271 (needs_fold_left_reduction_p): New function, split out from...
4272 (vect_is_simple_reduction): ...here. Accept reductions that
4273 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
4274 (vect_force_simple_reduction): Also store the reduction type in
4275 the assignment's STMT_VINFO_REDUC_TYPE.
4276 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
4277 (merge_with_identity): New function.
4278 (vect_expand_fold_left): Likewise.
4279 (vectorize_fold_left_reduction): Likewise.
4280 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
4281 scalar phi in place for it. Check for target support and reject
4282 cases that would reassociate the operation. Defer the transform
4283 phase to vectorize_fold_left_reduction.
4284 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
4285 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
4286 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
4288 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4290 * tree-if-conv.c (predicate_mem_writes): Remove redundant
4291 call to ifc_temp_var.
4293 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4294 Alan Hayward <alan.hayward@arm.com>
4295 David Sherwood <david.sherwood@arm.com>
4297 * target.def (legitimize_address_displacement): Take the original
4298 offset as a poly_int.
4299 * targhooks.h (default_legitimize_address_displacement): Update
4301 * targhooks.c (default_legitimize_address_displacement): Likewise.
4302 * doc/tm.texi: Regenerate.
4303 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
4304 as an argument, moving assert of ad->disp == ad->disp_term to...
4305 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
4306 Try calling targetm.legitimize_address_displacement before expanding
4307 the address rather than afterwards, and adjust for the new interface.
4308 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
4309 Match the new hook interface. Handle SVE addresses.
4310 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
4313 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4315 * Makefile.in (OBJS): Add early-remat.o.
4316 * target.def (select_early_remat_modes): New hook.
4317 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
4318 * doc/tm.texi: Regenerate.
4319 * targhooks.h (default_select_early_remat_modes): Declare.
4320 * targhooks.c (default_select_early_remat_modes): New function.
4321 * timevar.def (TV_EARLY_REMAT): New timevar.
4322 * passes.def (pass_early_remat): New pass.
4323 * tree-pass.h (make_pass_early_remat): Declare.
4324 * early-remat.c: New file.
4325 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
4327 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
4329 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4330 Alan Hayward <alan.hayward@arm.com>
4331 David Sherwood <david.sherwood@arm.com>
4333 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
4334 vfm1 with a bound_epilog parameter.
4335 (vect_do_peeling): Update calls accordingly, and move the prologue
4336 call earlier in the function. Treat the base bound_epilog as 0 for
4337 fully-masked loops and retain vf - 1 for other loops. Add 1 to
4338 this base when peeling for gaps.
4339 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
4340 with fully-masked loops.
4341 (vect_estimate_min_profitable_iters): Handle the single peeled
4342 iteration in that case.
4344 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4345 Alan Hayward <alan.hayward@arm.com>
4346 David Sherwood <david.sherwood@arm.com>
4348 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
4349 single-element interleaving even if the size is not a power of 2.
4350 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
4351 accesses for single-element interleaving if the group size is
4354 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4355 Alan Hayward <alan.hayward@arm.com>
4356 David Sherwood <david.sherwood@arm.com>
4358 * doc/md.texi (fold_extract_last_@var{m}): Document.
4359 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
4360 * optabs.def (fold_extract_last_optab): New optab.
4361 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
4362 * internal-fn.c (fold_extract_direct): New macro.
4363 (expand_fold_extract_optab_fn): Likewise.
4364 (direct_fold_extract_optab_supported_p): Likewise.
4365 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
4366 * tree-vect-loop.c (vect_model_reduction_cost): Handle
4367 EXTRACT_LAST_REDUCTION.
4368 (get_initial_def_for_reduction): Do not create an initial vector
4369 for EXTRACT_LAST_REDUCTION reductions.
4370 (vectorizable_reduction): Leave the scalar phi in place for
4371 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
4372 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
4373 epilogue code for EXTRACT_LAST_REDUCTION and defer the
4374 transform phase to vectorizable_condition.
4375 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
4377 (vect_finish_stmt_generation): ...here.
4378 (vect_finish_replace_stmt): New function.
4379 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
4380 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
4382 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
4384 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4385 Alan Hayward <alan.hayward@arm.com>
4386 David Sherwood <david.sherwood@arm.com>
4388 * doc/md.texi (extract_last_@var{m}): Document.
4389 * optabs.def (extract_last_optab): New optab.
4390 * internal-fn.def (EXTRACT_LAST): New internal function.
4391 * internal-fn.c (cond_unary_direct): New macro.
4392 (expand_cond_unary_optab_fn): Likewise.
4393 (direct_cond_unary_optab_supported_p): Likewise.
4394 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
4395 loops using EXTRACT_LAST.
4396 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
4397 (extract_last_<mode>): ...this optab.
4398 (vec_extract<mode><Vel>): Update accordingly.
4400 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4401 Alan Hayward <alan.hayward@arm.com>
4402 David Sherwood <david.sherwood@arm.com>
4404 * target.def (empty_mask_is_expensive): New hook.
4405 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
4406 * doc/tm.texi: Regenerate.
4407 * targhooks.h (default_empty_mask_is_expensive): Declare.
4408 * targhooks.c (default_empty_mask_is_expensive): New function.
4409 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
4410 if the target says that empty masks are expensive.
4411 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4413 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4416 Alan Hayward <alan.hayward@arm.com>
4417 David Sherwood <david.sherwood@arm.com>
4419 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4420 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4421 (vect_use_loop_mask_for_alignment_p): New function.
4422 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4423 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4424 niters_skip argument. Make sure that the first niters_skip elements
4425 of the first iteration are inactive.
4426 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4427 Update call to vect_set_loop_masks_directly.
4428 (get_misalign_in_elems): New function, split out from...
4429 (vect_gen_prolog_loop_niters): ...here.
4430 (vect_update_init_of_dr): Take a code argument that specifies whether
4431 the adjustment should be added or subtracted.
4432 (vect_update_init_of_drs): Likewise.
4433 (vect_prepare_for_masked_peels): New function.
4434 (vect_do_peeling): Skip prologue peeling if we're using a mask
4435 instead. Update call to vect_update_inits_of_drs.
4436 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4438 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4439 alignment. Do not include the number of peeled iterations in
4440 the minimum threshold in that case.
4441 (vectorizable_induction): Adjust the start value down by
4442 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4443 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4444 Take the number of skipped iterations into account when calculating
4446 * tree-vect-stmts.c (vect_gen_while_not): New function.
4448 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4449 Alan Hayward <alan.hayward@arm.com>
4450 David Sherwood <david.sherwood@arm.com>
4452 * doc/sourcebuild.texi (vect_fully_masked): Document.
4453 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4455 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4457 (vect_analyze_loop_2): ...here. Don't check the vectorization
4458 factor against the number of loop iterations if the loop is
4461 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4462 Alan Hayward <alan.hayward@arm.com>
4463 David Sherwood <david.sherwood@arm.com>
4465 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4466 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4467 (dump_groups): Update accordingly.
4468 (iv_use::mem_type): New member variable.
4469 (address_p): New function.
4470 (record_use): Add a mem_type argument and initialize the new
4472 (record_group_use): Add a mem_type argument. Use address_p.
4473 Remove obsolete null checks of base_object. Update call to record_use.
4474 (find_interesting_uses_op): Update call to record_group_use.
4475 (find_interesting_uses_cond): Likewise.
4476 (find_interesting_uses_address): Likewise.
4477 (get_mem_type_for_internal_fn): New function.
4478 (find_address_like_use): Likewise.
4479 (find_interesting_uses_stmt): Try find_address_like_use before
4480 calling find_interesting_uses_op.
4481 (addr_offset_valid_p): Use the iv mem_type field as the type
4482 of the addressed memory.
4483 (add_autoinc_candidates): Likewise.
4484 (get_address_cost): Likewise.
4485 (split_small_address_groups_p): Use address_p.
4486 (split_address_groups): Likewise.
4487 (add_iv_candidate_for_use): Likewise.
4488 (autoinc_possible_for_pair): Likewise.
4489 (rewrite_groups): Likewise.
4490 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4491 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4492 (get_alias_ptr_type_for_ptr_address): New function.
4493 (rewrite_use_address): Rewrite address uses in calls that were
4494 identified by find_address_like_use.
4496 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4497 Alan Hayward <alan.hayward@arm.com>
4498 David Sherwood <david.sherwood@arm.com>
4500 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4502 * gimple-expr.h (is_gimple_addressable: Likewise.
4503 * gimple-expr.c (is_gimple_address): Likewise.
4504 * internal-fn.c (expand_call_mem_ref): New function.
4505 (expand_mask_load_optab_fn): Use it.
4506 (expand_mask_store_optab_fn): Likewise.
4508 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4509 Alan Hayward <alan.hayward@arm.com>
4510 David Sherwood <david.sherwood@arm.com>
4512 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4513 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4514 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4515 (cond_umax@var{mode}): Document.
4516 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4517 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4518 (cond_umin_optab, cond_umax_optab): New optabs.
4519 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4520 (COND_IOR, COND_XOR): New internal functions.
4521 * internal-fn.h (get_conditional_internal_fn): Declare.
4522 * internal-fn.c (cond_binary_direct): New macro.
4523 (expand_cond_binary_optab_fn): Likewise.
4524 (direct_cond_binary_optab_supported_p): Likewise.
4525 (get_conditional_internal_fn): New function.
4526 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4527 Cope with reduction statements that are vectorized as calls rather
4529 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4530 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4531 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4532 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4533 (UNSPEC_COND_EOR): New unspecs.
4534 (optab): Add mappings for them.
4535 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4536 (sve_int_op, sve_fp_op): New int attributes.
4538 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4539 Alan Hayward <alan.hayward@arm.com>
4540 David Sherwood <david.sherwood@arm.com>
4542 * optabs.def (while_ult_optab): New optab.
4543 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4544 * internal-fn.def (WHILE_ULT): New internal function.
4545 * internal-fn.h (direct_internal_fn_supported_p): New override
4546 that takes two types as argument.
4547 * internal-fn.c (while_direct): New macro.
4548 (expand_while_optab_fn): New function.
4549 (convert_optab_supported_p): Likewise.
4550 (direct_while_optab_supported_p): New macro.
4551 * wide-int.h (wi::udiv_ceil): New function.
4552 * tree-vectorizer.h (rgroup_masks): New structure.
4553 (vec_loop_masks): New typedef.
4554 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4556 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4557 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4558 (vect_max_vf): New function.
4559 (slpeel_make_loop_iterate_ntimes): Delete.
4560 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4561 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4562 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4563 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4564 internal-fn.h, stor-layout.h and optabs-query.h.
4565 (vect_set_loop_mask): New function.
4566 (add_preheader_seq): Likewise.
4567 (add_header_seq): Likewise.
4568 (interleave_supported_p): Likewise.
4569 (vect_maybe_permute_loop_masks): Likewise.
4570 (vect_set_loop_masks_directly): Likewise.
4571 (vect_set_loop_condition_masked): Likewise.
4572 (vect_set_loop_condition_unmasked): New function, split out from
4573 slpeel_make_loop_iterate_ntimes.
4574 (slpeel_make_loop_iterate_ntimes): Rename to..
4575 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4576 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4577 (vect_do_peeling): Update call accordingly.
4578 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4580 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4581 mask_compare_type, can_fully_mask_p and fully_masked_p.
4582 (release_vec_loop_masks): New function.
4583 (_loop_vec_info): Use it to free the loop masks.
4584 (can_produce_all_loop_masks_p): New function.
4585 (vect_get_max_nscalars_per_iter): Likewise.
4586 (vect_verify_full_masking): Likewise.
4587 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4588 retries, and free the mask rgroups before retrying. Check loop-wide
4589 reasons for disallowing fully-masked loops. Make the final decision
4590 about whether use a fully-masked loop or not.
4591 (vect_estimate_min_profitable_iters): Do not assume that peeling
4592 for the number of iterations will be needed for fully-masked loops.
4593 (vectorizable_reduction): Disable fully-masked loops.
4594 (vectorizable_live_operation): Likewise.
4595 (vect_halve_mask_nunits): New function.
4596 (vect_double_mask_nunits): Likewise.
4597 (vect_record_loop_mask): Likewise.
4598 (vect_get_loop_mask): Likewise.
4599 (vect_transform_loop): Handle the case in which the final loop
4600 iteration might handle a partial vector. Call vect_set_loop_condition
4601 instead of slpeel_make_loop_iterate_ntimes.
4602 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4603 (check_load_store_masking): New function.
4604 (prepare_load_store_mask): Likewise.
4605 (vectorizable_store): Handle fully-masked loops.
4606 (vectorizable_load): Likewise.
4607 (supportable_widening_operation): Use vect_halve_mask_nunits for
4609 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4610 (vect_gen_while): New function.
4611 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4612 (aarch64_uqdec<mode>): New insn.
4614 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4615 Alan Hayward <alan.hayward@arm.com>
4616 David Sherwood <david.sherwood@arm.com>
4618 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4619 (reduc_xor_scal_optab): New optabs.
4620 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4621 (reduc_xor_scal_@var{m}): Document.
4622 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4623 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4625 * fold-const-call.c (fold_const_call): Handle them.
4626 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4627 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4628 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4629 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4630 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4631 (UNSPEC_XORV): New unspecs.
4632 (optab): Add entries for them.
4633 (BITWISEV): New int iterator.
4634 (bit_reduc_op): New int attributes.
4636 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4637 Alan Hayward <alan.hayward@arm.com>
4638 David Sherwood <david.sherwood@arm.com>
4640 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4641 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4642 * optabs.def (vec_shl_insert_optab): New optab.
4643 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4644 (duplicate_and_interleave): Likewise.
4645 * tree-vect-loop.c: Include internal-fn.h.
4646 (neutral_op_for_slp_reduction): New function, split out from
4647 get_initial_defs_for_reduction.
4648 (get_initial_def_for_reduction): Handle option 2 for variable-length
4649 vectors by loading the neutral value into a vector and then shifting
4650 the initial value into element 0.
4651 (get_initial_defs_for_reduction): Replace the code argument with
4652 the neutral value calculated by neutral_op_for_slp_reduction.
4653 Use gimple_build_vector for constant-length vectors.
4654 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4655 but the first group_size elements have a neutral value.
4656 Use duplicate_and_interleave otherwise.
4657 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4658 Update call to get_initial_defs_for_reduction. Handle SLP
4659 reductions for variable-length vectors by creating one vector
4660 result for each scalar result, with the elements associated
4661 with other scalar results stubbed out with the neutral value.
4662 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4663 Require IFN_VEC_SHL_INSERT for double reductions on
4664 variable-length vectors, or SLP reductions that have
4665 a neutral value. Require can_duplicate_and_interleave_p
4666 support for variable-length unchained SLP reductions if there
4667 is no neutral value, such as for MIN/MAX reductions. Also require
4668 the number of vector elements to be a multiple of the number of
4669 SLP statements when doing variable-length unchained SLP reductions.
4670 Update call to vect_create_epilog_for_reduction.
4671 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4672 and remove initial values.
4673 (duplicate_and_interleave): Make public.
4674 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4675 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4677 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4678 Alan Hayward <alan.hayward@arm.com>
4679 David Sherwood <david.sherwood@arm.com>
4681 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4682 (can_duplicate_and_interleave_p): New function.
4683 (vect_get_and_check_slp_defs): Take the vector of statements
4684 rather than just the current one. Remove excess parentheses.
4685 Restriction rejectinon of vect_constant_def and vect_external_def
4686 for variable-length vectors to boolean types, or types for which
4687 can_duplicate_and_interleave_p is false.
4688 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4689 (duplicate_and_interleave): New function.
4690 (vect_get_constant_vectors): Use gimple_build_vector for
4691 constant-length vectors and suitable variable-length constant
4692 vectors. Use duplicate_and_interleave for other variable-length
4693 vectors. Don't defer the update when inserting new statements.
4695 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4696 Alan Hayward <alan.hayward@arm.com>
4697 David Sherwood <david.sherwood@arm.com>
4699 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4700 min_profitable_iters doesn't go negative.
4702 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4703 Alan Hayward <alan.hayward@arm.com>
4704 David Sherwood <david.sherwood@arm.com>
4706 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4707 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4708 * optabs.def (vec_mask_load_lanes_optab): New optab.
4709 (vec_mask_store_lanes_optab): Likewise.
4710 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4711 (MASK_STORE_LANES): Likewise.
4712 * internal-fn.c (mask_load_lanes_direct): New macro.
4713 (mask_store_lanes_direct): Likewise.
4714 (expand_mask_load_optab_fn): Handle masked operations.
4715 (expand_mask_load_lanes_optab_fn): New macro.
4716 (expand_mask_store_optab_fn): Handle masked operations.
4717 (expand_mask_store_lanes_optab_fn): New macro.
4718 (direct_mask_load_lanes_optab_supported_p): Likewise.
4719 (direct_mask_store_lanes_optab_supported_p): Likewise.
4720 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4722 (vect_load_lanes_supported): Likewise.
4723 * tree-vect-data-refs.c (strip_conversion): New function.
4724 (can_group_stmts_p): Likewise.
4725 (vect_analyze_data_ref_accesses): Use it instead of checking
4726 for a pair of assignments.
4727 (vect_store_lanes_supported): Take a masked_p parameter.
4728 (vect_load_lanes_supported): Likewise.
4729 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4730 vect_store_lanes_supported and vect_load_lanes_supported.
4731 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4732 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4733 parameter. Don't allow gaps for masked accesses.
4734 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4735 and vect_load_lanes_supported.
4736 (get_load_store_type): Take a masked_p parameter and update
4737 call to get_group_load_store_type.
4738 (vectorizable_store): Update call to get_load_store_type.
4739 Handle IFN_MASK_STORE_LANES.
4740 (vectorizable_load): Update call to get_load_store_type.
4741 Handle IFN_MASK_LOAD_LANES.
4743 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4744 Alan Hayward <alan.hayward@arm.com>
4745 David Sherwood <david.sherwood@arm.com>
4747 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4749 * config/aarch64/aarch64-protos.h
4750 (aarch64_sve_struct_memory_operand_p): Declare.
4751 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4752 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4753 (VPRED, vpred): Handle SVE structure modes.
4754 * config/aarch64/constraints.md (Utx): New constraint.
4755 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4756 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4757 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4758 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4759 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4760 structure modes. Split into pieces after RA.
4761 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4762 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4764 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4765 SVE structure modes.
4766 (aarch64_classify_address): Likewise.
4767 (sizetochar): Move earlier in file.
4768 (aarch64_print_operand): Handle SVE register lists.
4769 (aarch64_array_mode): New function.
4770 (aarch64_sve_struct_memory_operand_p): Likewise.
4771 (TARGET_ARRAY_MODE): Redefine.
4773 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4774 Alan Hayward <alan.hayward@arm.com>
4775 David Sherwood <david.sherwood@arm.com>
4777 * target.def (array_mode): New target hook.
4778 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4779 * doc/tm.texi: Regenerate.
4780 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4781 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4782 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4784 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4787 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4788 Alan Hayward <alan.hayward@arm.com>
4789 David Sherwood <david.sherwood@arm.com>
4791 * fold-const.c (fold_binary_loc): Check the argument types
4792 rather than the result type when testing for a vector operation.
4794 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4796 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4797 * doc/tm.texi: Regenerate.
4799 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4800 Alan Hayward <alan.hayward@arm.com>
4801 David Sherwood <david.sherwood@arm.com>
4803 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4804 (sve): Document new AArch64 extension.
4805 * doc/md.texi (w): Extend the description of the AArch64
4806 constraint to include SVE vectors.
4807 (Upl, Upa): Document new AArch64 predicate constraints.
4808 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4810 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4811 (msve-vector-bits=): New option.
4812 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4813 SVE when these are disabled.
4814 (sve): New extension.
4815 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4816 modes. Adjust their number of units based on aarch64_sve_vg.
4817 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4818 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4819 aarch64_addr_query_type.
4820 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4821 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4822 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4823 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4824 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4825 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4826 (aarch64_simd_imm_zero_p): Delete.
4827 (aarch64_check_zero_based_sve_index_immediate): Declare.
4828 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4829 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4830 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4831 (aarch64_sve_float_mul_immediate_p): Likewise.
4832 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4834 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4835 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4836 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4837 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4838 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4839 (aarch64_regmode_natural_size): Likewise.
4840 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4841 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4843 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4844 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4845 for VG and the SVE predicate registers.
4846 (V_ALIASES): Add a "z"-prefixed alias.
4847 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4848 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4849 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4850 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4851 (REG_CLASS_NAMES): Add entries for them.
4852 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4853 and the predicate registers.
4854 (aarch64_sve_vg): Declare.
4855 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4856 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4857 (REGMODE_NATURAL_SIZE): Define.
4858 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4860 * config/aarch64/aarch64.c: Include cfgrtl.h.
4861 (simd_immediate_info): Add a constructor for series vectors,
4862 and an associated step field.
4863 (aarch64_sve_vg): New variable.
4864 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4865 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4866 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4867 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4868 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4869 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4870 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4871 (aarch64_get_mask_mode): New functions.
4872 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4873 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4874 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4875 predicate modes and predicate registers. Explicitly restrict
4876 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4877 to store a vector mode if it is recognized by
4878 aarch64_classify_vector_mode.
4879 (aarch64_regmode_natural_size): New function.
4880 (aarch64_hard_regno_caller_save_mode): Return the original mode
4882 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4883 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4884 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4885 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4887 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4888 does not overlap dest if the function is frame-related. Handle
4890 (aarch64_split_add_offset): New function.
4891 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4892 them aarch64_add_offset.
4893 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4894 and update call to aarch64_sub_sp.
4895 (aarch64_add_cfa_expression): New function.
4896 (aarch64_expand_prologue): Pass extra temporary registers to the
4897 functions above. Handle the case in which we need to emit new
4898 DW_CFA_expressions for registers that were originally saved
4899 relative to the stack pointer, but now have to be expressed
4900 relative to the frame pointer.
4901 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4903 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4904 IP0 and IP1 values for SVE frames.
4905 (aarch64_expand_vec_series): New function.
4906 (aarch64_expand_sve_widened_duplicate): Likewise.
4907 (aarch64_expand_sve_const_vector): Likewise.
4908 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4909 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4910 into the register, rather than emitting a SET directly.
4911 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4912 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4913 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4914 (offset_9bit_signed_scaled_p): New functions.
4915 (aarch64_replicate_bitmask_imm): New function.
4916 (aarch64_bitmask_imm): Use it.
4917 (aarch64_cannot_force_const_mem): Reject expressions involving
4918 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4919 (aarch64_classify_index): Handle SVE indices, by requiring
4920 a plain register index with a scale that matches the element size.
4921 (aarch64_classify_address): Handle SVE addresses. Assert that
4922 the mode of the address is VOIDmode or an integer mode.
4923 Update call to aarch64_classify_symbol.
4924 (aarch64_classify_symbolic_expression): Update call to
4925 aarch64_classify_symbol.
4926 (aarch64_const_vec_all_in_range_p): New function.
4927 (aarch64_print_vector_float_operand): Likewise.
4928 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4929 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4930 and the FP immediates 1.0 and 0.5.
4931 (aarch64_print_address_internal): Handle SVE addresses.
4932 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4933 (aarch64_regno_regclass): Handle predicate registers.
4934 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4936 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4937 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4938 (aarch64_convert_sve_vector_bits): New function.
4939 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4940 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4942 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4943 Handle SVE vector and predicate modes. Accept VL-based constants
4944 that need only one temporary register, and VL offsets that require
4945 no temporary registers.
4946 (aarch64_conditional_register_usage): Mark the predicate registers
4947 as fixed if SVE isn't available.
4948 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4949 Return true for SVE vector and predicate modes.
4950 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4951 rather than an unsigned int. Handle SVE modes.
4952 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4954 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4956 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4957 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4958 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4959 (aarch64_sve_float_mul_immediate_p): New functions.
4960 (aarch64_sve_valid_immediate): New function.
4961 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4962 Explicitly reject structure modes. Check for INDEX constants.
4963 Handle PTRUE and PFALSE constants.
4964 (aarch64_check_zero_based_sve_index_immediate): New function.
4965 (aarch64_simd_imm_zero_p): Delete.
4966 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4967 vector modes. Accept constants in the range of CNT[BHWD].
4968 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4969 ask for an Advanced SIMD mode.
4970 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4971 (aarch64_simd_vector_alignment): Handle SVE predicates.
4972 (aarch64_vectorize_preferred_vector_alignment): New function.
4973 (aarch64_simd_vector_alignment_reachable): Use it instead of
4975 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4976 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4978 (MAX_VECT_LEN): Delete.
4979 (expand_vec_perm_d): Add a vec_flags field.
4980 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4981 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4982 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4984 (aarch64_evpc_rev): Rename to...
4985 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4986 (aarch64_evpc_rev_global): New function.
4987 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4988 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4990 (aarch64_evpc_sve_tbl): New function.
4991 (aarch64_expand_vec_perm_const_1): Update after rename of
4992 aarch64_evpc_rev. Handle SVE permutes too, trying
4993 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4994 than aarch64_evpc_tbl.
4995 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4996 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4997 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4998 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4999 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
5000 (aarch64_expand_sve_vcond): New functions.
5001 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
5002 of aarch64_vector_mode_p.
5003 (aarch64_dwarf_poly_indeterminate_value): New function.
5004 (aarch64_compute_pressure_classes): Likewise.
5005 (aarch64_can_change_mode_class): Likewise.
5006 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
5007 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
5008 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
5009 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
5010 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
5011 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
5012 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
5013 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
5015 (Dn, Dl, Dr): Accept const as well as const_vector.
5016 (Dz): Likewise. Compare against CONST0_RTX.
5017 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
5018 of "vector" where appropriate.
5019 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
5020 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
5021 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
5022 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
5023 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
5024 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
5025 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
5026 (v_int_equiv): Extend to SVE modes.
5027 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
5029 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
5030 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
5031 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
5032 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
5033 (SVE_COND_FP_CMP): New int iterators.
5034 (perm_hilo): Handle the new unpack unspecs.
5035 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
5037 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
5038 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
5039 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
5040 (aarch64_equality_operator, aarch64_constant_vector_operand)
5041 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
5042 (aarch64_sve_nonimmediate_operand): Likewise.
5043 (aarch64_sve_general_operand): Likewise.
5044 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
5045 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
5046 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
5047 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
5048 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
5049 (aarch64_sve_float_arith_immediate): Likewise.
5050 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
5051 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
5052 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
5053 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
5054 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
5055 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
5056 (aarch64_sve_float_arith_operand): Likewise.
5057 (aarch64_sve_float_arith_with_sub_operand): Likewise.
5058 (aarch64_sve_float_mul_operand): Likewise.
5059 (aarch64_sve_vec_perm_operand): Likewise.
5060 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
5061 (aarch64_mov_operand): Accept const_poly_int and const_vector.
5062 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
5063 as well as const_vector.
5064 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
5065 in file. Use CONST0_RTX and CONSTM1_RTX.
5066 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
5067 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
5068 Use aarch64_simd_imm_zero.
5069 * config/aarch64/aarch64-sve.md: New file.
5070 * config/aarch64/aarch64.md: Include it.
5071 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
5072 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
5073 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
5074 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
5075 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
5076 (sve): New attribute.
5077 (enabled): Disable instructions with the sve attribute unless
5079 (movqi, movhi): Pass CONST_POLY_INT operaneds through
5080 aarch64_expand_mov_immediate.
5081 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
5082 CNT[BHSD] immediates.
5083 (movti): Split CONST_POLY_INT moves into two halves.
5084 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
5085 Split additions that need a temporary here if the destination
5086 is the stack pointer.
5087 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
5088 (*add<mode>3_poly_1): New instruction.
5089 (set_clobber_cc): New expander.
5091 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
5093 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
5094 parameter and use it instead of GET_MODE_SIZE (innermode). Use
5095 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
5096 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
5097 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
5098 Change innermode from fixed_mode_size to machine_mode.
5099 (simplify_subreg): Update call accordingly. Handle a constant-sized
5100 subreg of a variable-length CONST_VECTOR.
5102 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
5103 Alan Hayward <alan.hayward@arm.com>
5104 David Sherwood <david.sherwood@arm.com>
5106 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
5107 (add_offset_to_base): New function, split out from...
5108 (create_mem_ref): ...here. When handling a scale other than 1,
5109 check first whether the address is valid without the offset.
5110 Add it into the base if so, leaving the index and scale as-is.
5112 2018-01-12 Jakub Jelinek <jakub@redhat.com>
5115 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
5116 fold_for_warn before checking if arg2 is INTEGER_CST.
5118 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
5120 * config/rs6000/predicates.md (load_multiple_operation): Delete.
5121 (store_multiple_operation): Delete.
5122 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
5123 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
5124 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
5125 guarded by TARGET_STRING.
5126 (rs6000_output_load_multiple): Delete.
5127 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
5128 OPTION_MASK_STRING / TARGET_STRING handling.
5129 (print_operand) <'N', 'O'>: Add comment that these are unused now.
5130 (const rs6000_opt_masks) <"string">: Change mask to 0.
5131 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
5132 (MASK_STRING): Delete.
5133 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
5135 (load_multiple): Delete.
5142 (store_multiple): Delete.
5149 (movmemsi_8reg): Delete.
5150 (corresponding unnamed define_insn): Delete.
5151 (movmemsi_6reg): Delete.
5152 (corresponding unnamed define_insn): Delete.
5153 (movmemsi_4reg): Delete.
5154 (corresponding unnamed define_insn): Delete.
5155 (movmemsi_2reg): Delete.
5156 (corresponding unnamed define_insn): Delete.
5157 (movmemsi_1reg): Delete.
5158 (corresponding unnamed define_insn): Delete.
5159 * config/rs6000/rs6000.opt (mno-string): New.
5160 (mstring): Replace by deprecation warning stub.
5161 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
5163 2018-01-12 Jakub Jelinek <jakub@redhat.com>
5165 * regrename.c (regrename_do_replace): If replacing the same
5166 reg multiple times, try to reuse last created gen_raw_REG.
5169 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
5170 main to workaround a bug in GDB.
5172 2018-01-12 Tom de Vries <tom@codesourcery.com>
5175 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
5177 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
5179 PR rtl-optimization/80481
5180 * ira-color.c (get_cap_member): New function.
5181 (allocnos_conflict_by_live_ranges_p): Use it.
5182 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
5183 (setup_slot_coalesced_allocno_live_ranges): Ditto.
5185 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
5188 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
5189 (*saddl_se_1): Ditto.
5191 (*ssubl_se_1): Ditto.
5193 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5195 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
5196 rather than wi::to_widest for DR_INITs.
5197 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
5198 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
5199 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
5201 (vect_analyze_group_access_1): Note that here.
5203 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5205 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
5206 polynomial type sizes.
5208 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5210 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
5211 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
5212 (gimple_add_tmp_var): Likewise.
5214 2018-01-12 Martin Liska <mliska@suse.cz>
5216 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
5217 (gimple_alloc_sizes): Likewise.
5218 (dump_gimple_statistics): Use PRIu64 in printf format.
5219 * gimple.h: Change uint64_t to int.
5221 2018-01-12 Martin Liska <mliska@suse.cz>
5223 * tree-core.h: Use uint64_t instead of int.
5224 * tree.c (tree_node_counts): Likewise.
5225 (tree_node_sizes): Likewise.
5226 (dump_tree_statistics): Use PRIu64 in printf format.
5228 2018-01-12 Martin Liska <mliska@suse.cz>
5230 * Makefile.in: As qsort_chk is implemented in vec.c, add
5231 vec.o to linkage of gencfn-macros.
5232 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
5233 passing the info to record_node_allocation_statistics.
5234 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
5236 * ggc-common.c (struct ggc_usage): Add operator== and use
5237 it in operator< and compare function.
5238 * mem-stats.h (struct mem_usage): Likewise.
5239 * vec.c (struct vec_usage): Remove operator< and compare
5240 function. Can be simply inherited.
5242 2018-01-12 Martin Jambor <mjambor@suse.cz>
5245 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
5246 * tree-ssa-math-opts.c: Include domwalk.h.
5247 (convert_mult_to_fma_1): New function.
5248 (fma_transformation_info): New type.
5249 (fma_deferring_state): Likewise.
5250 (cancel_fma_deferring): New function.
5251 (result_of_phi): Likewise.
5252 (last_fma_candidate_feeds_initial_phi): Likewise.
5253 (convert_mult_to_fma): Added deferring logic, split actual
5254 transformation to convert_mult_to_fma_1.
5255 (math_opts_dom_walker): New type.
5256 (math_opts_dom_walker::after_dom_children): New method, body moved
5257 here from pass_optimize_widening_mul::execute, added deferring logic
5259 (pass_optimize_widening_mul::execute): Moved most of code to
5260 math_opts_dom_walker::after_dom_children.
5261 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
5262 * config/i386/i386.c (ix86_option_override_internal): Added
5263 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
5265 2018-01-12 Richard Biener <rguenther@suse.de>
5268 * dwarf2out.c (gen_variable_die): Do not reset old_die for
5269 inline instance vars.
5271 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
5274 * config/rx/rx.c (rx_is_restricted_memory_address):
5277 2018-01-12 Richard Biener <rguenther@suse.de>
5279 PR tree-optimization/80846
5280 * target.def (split_reduction): New target hook.
5281 * targhooks.c (default_split_reduction): New function.
5282 * targhooks.h (default_split_reduction): Declare.
5283 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
5284 target requests first reduce vectors by combining low and high
5286 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
5287 (get_vectype_for_scalar_type_and_size): Export.
5288 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
5289 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
5290 * doc/tm.texi: Regenerate.
5291 * config/i386/i386.c (ix86_split_reduction): Implement
5292 TARGET_VECTORIZE_SPLIT_REDUCTION.
5294 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5297 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
5298 in PIC mode except for TARGET_VXWORKS_RTP.
5299 * config/sparc/sparc.c: Include cfgrtl.h.
5300 (TARGET_INIT_PIC_REG): Define.
5301 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
5302 (sparc_pic_register_p): New predicate.
5303 (sparc_legitimate_address_p): Use it.
5304 (sparc_legitimize_pic_address): Likewise.
5305 (sparc_delegitimize_address): Likewise.
5306 (sparc_mode_dependent_address_p): Likewise.
5307 (gen_load_pcrel_sym): Remove 4th parameter.
5308 (load_got_register): Adjust call to above. Remove obsolete stuff.
5309 (sparc_expand_prologue): Do not call load_got_register here.
5310 (sparc_flat_expand_prologue): Likewise.
5311 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
5312 (sparc_use_pseudo_pic_reg): New function.
5313 (sparc_init_pic_reg): Likewise.
5314 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5315 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
5317 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
5319 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
5320 Add item for branch_cost.
5322 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5324 PR rtl-optimization/83565
5325 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
5326 not extend the result to a larger mode for rotate operations.
5327 (num_sign_bit_copies1): Likewise.
5329 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5332 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
5334 Use values-Xc.o for -pedantic.
5335 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
5337 2018-01-12 Martin Liska <mliska@suse.cz>
5340 * ipa-devirt.c (final_warning_record::grow_type_warnings):
5342 (possible_polymorphic_call_targets): Use it.
5343 (ipa_devirt): Likewise.
5345 2018-01-12 Martin Liska <mliska@suse.cz>
5347 * profile-count.h (enum profile_quality): Use 0 as invalid
5348 enum value of profile_quality.
5350 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
5352 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
5353 -mext-string options.
5355 2018-01-12 Richard Biener <rguenther@suse.de>
5357 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
5358 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
5359 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
5361 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
5363 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
5365 * configure.ac (--with-long-double-format): Add support for the
5366 configuration option to change the default long double format on
5368 * config.gcc (powerpc*-linux*-*): Likewise.
5369 * configure: Regenerate.
5370 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
5371 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
5372 used without modification.
5374 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5376 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
5377 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
5378 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
5379 MISC_BUILTIN_SPEC_BARRIER.
5380 (rs6000_init_builtins): Likewise.
5381 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
5383 (speculation_barrier): New define_insn.
5384 * doc/extend.texi: Document __builtin_speculation_barrier.
5386 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5389 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
5390 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
5391 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
5393 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
5394 integral modes instead of "ss" and "sd".
5395 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
5396 vectors with 32-bit and 64-bit elements.
5397 (vecdupssescalarmodesuffix): New mode attribute.
5398 (vec_dup<mode>): Use it.
5400 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
5403 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
5404 frame if argument is passed on stack.
5406 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5409 * ree.c (combine_reaching_defs): Optimize also
5410 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
5411 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5413 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5416 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5418 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5421 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5422 after they are computed.
5424 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5426 PR tree-optimization/83695
5427 * gimple-loop-linterchange.cc
5428 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5429 reset cached scev information after interchange.
5430 (pass_linterchange::execute): Remove call to scev_reset_htab.
5432 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5434 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5435 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5436 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5437 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5438 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5439 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5440 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5441 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5442 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5443 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5444 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5445 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5446 (V_lane_reg): Likewise.
5447 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5449 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5450 (vfmal_lane_low<mode>_intrinsic,
5451 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5452 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5453 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5454 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5455 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5456 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5458 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5460 * config/arm/arm-cpus.in (fp16fml): New feature.
5461 (ALL_SIMD): Add fp16fml.
5462 (armv8.2-a): Add fp16fml as an option.
5463 (armv8.3-a): Likewise.
5464 (armv8.4-a): Add fp16fml as part of fp16.
5465 * config/arm/arm.h (TARGET_FP16FML): Define.
5466 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5468 * config/arm/arm-modes.def (V2HF): Define.
5469 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5470 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5471 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5472 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5473 vfmsl_low, vfmsl_high): New set of builtins.
5474 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5475 (vfml_op): New code attribute.
5476 (VFMLHALVES): New int iterator.
5477 (VFML, VFMLSEL): New mode attributes.
5478 (V_reg): Define mapping for V2HF.
5479 (V_hi, V_lo): New mode attributes.
5480 (VF_constraint): Likewise.
5481 (vfml_half, vfml_half_selector): New int attributes.
5482 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5484 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5485 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5487 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5488 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5489 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5490 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5492 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5493 Document new effective target and option set.
5495 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5497 * config/arm/arm-cpus.in (armv8_4): New feature.
5498 (ARMv8_4a): New fgroup.
5499 (armv8.4-a): New arch.
5500 * config/arm/arm-tables.opt: Regenerate.
5501 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5502 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5503 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5504 Add matching rules for -march=armv8.4-a and extensions.
5505 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5507 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5510 * config/rx/rx.md (BW): New mode attribute.
5511 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5513 2018-01-11 Richard Biener <rguenther@suse.de>
5515 PR tree-optimization/83435
5516 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5517 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5518 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5520 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5521 Alan Hayward <alan.hayward@arm.com>
5522 David Sherwood <david.sherwood@arm.com>
5524 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5526 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5527 (aarch64_print_address_internal): Use it to check for a zero offset.
5529 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5530 Alan Hayward <alan.hayward@arm.com>
5531 David Sherwood <david.sherwood@arm.com>
5533 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5534 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5535 Return a poly_int64 rather than a HOST_WIDE_INT.
5536 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5537 rather than a HOST_WIDE_INT.
5538 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5539 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5540 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5541 final_offset from HOST_WIDE_INT to poly_int64.
5542 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5543 to_constant when getting the number of units in an Advanced SIMD
5545 (aarch64_builtin_vectorized_function): Check for a constant number
5547 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5549 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5550 attribute instead of GET_MODE_NUNITS.
5551 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5552 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5553 GET_MODE_SIZE for fixed-size registers.
5554 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5555 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5556 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5557 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5558 (aarch64_print_operand, aarch64_print_address_internal)
5559 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5560 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5561 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5562 Handle polynomial GET_MODE_SIZE.
5563 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5564 wider than SImode without modification.
5565 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5566 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5567 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5568 passing and returning SVE modes.
5569 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5570 rather than GEN_INT.
5571 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5572 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5573 (aarch64_allocate_and_probe_stack_space): Likewise.
5574 (aarch64_layout_frame): Cope with polynomial offsets.
5575 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5576 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5578 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5579 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5580 poly_int64 rather than a HOST_WIDE_INT.
5581 (aarch64_get_separate_components, aarch64_process_components)
5582 (aarch64_expand_prologue, aarch64_expand_epilogue)
5583 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5584 (aarch64_anchor_offset): New function, split out from...
5585 (aarch64_legitimize_address): ...here.
5586 (aarch64_builtin_vectorization_cost): Handle polynomial
5587 TYPE_VECTOR_SUBPARTS.
5588 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5590 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5591 number of elements from the PARALLEL rather than the mode.
5592 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5593 rather than GET_MODE_BITSIZE.
5594 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5595 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5596 (aarch64_expand_vec_perm_const_1): Handle polynomial
5597 d->perm.length () and d->perm elements.
5598 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5599 Apply to_constant to d->perm elements.
5600 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5601 polynomial CONST_VECTOR_NUNITS.
5602 (aarch64_move_pointer): Take amount as a poly_int64 rather
5604 (aarch64_progress_pointer): Avoid temporary variable.
5605 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5606 the mode attribute instead of GET_MODE.
5608 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5609 Alan Hayward <alan.hayward@arm.com>
5610 David Sherwood <david.sherwood@arm.com>
5612 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5613 x exists before using it.
5614 (aarch64_add_constant_internal): Rename to...
5615 (aarch64_add_offset_1): ...this. Replace regnum with separate
5616 src and dest rtxes. Handle the case in which they're different,
5617 including when the offset is zero. Replace scratchreg with an rtx.
5618 Use 2 additions if there is no spare register into which we can
5619 move a 16-bit constant.
5620 (aarch64_add_constant): Delete.
5621 (aarch64_add_offset): Replace reg with separate src and dest
5622 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5623 Use aarch64_add_offset_1.
5624 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5625 an rtx rather than an int. Take the delta as a poly_int64
5626 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5627 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5628 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5629 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5630 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5632 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5633 aarch64_add_constant.
5635 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5637 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5638 Use scalar_float_mode.
5640 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5642 * config/aarch64/aarch64-simd.md
5643 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5644 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5645 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5646 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5647 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5648 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5649 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5650 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5651 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5652 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5654 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5657 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5658 targ_options->x_arm_arch_string is non NULL.
5660 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5662 * config/aarch64/aarch64.h
5663 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5665 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5668 * expmed.c (emit_store_flag_force): Swap if const op0
5669 and change VOIDmode to mode of op0.
5671 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5673 PR rtl-optimization/83761
5674 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5675 than bytes to mode_for_size.
5677 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5680 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5681 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5684 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5687 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5688 when in layout mode.
5689 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5690 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5693 2018-01-10 Michael Collison <michael.collison@arm.com>
5695 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5696 * config/aarch64/aarch64-option-extension.def: Add
5697 AARCH64_OPT_EXTENSION of 'fp16fml'.
5698 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5699 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5700 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5701 * config/aarch64/constraints.md (Ui7): New constraint.
5702 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5703 (VFMLA_SEL_W): Ditto.
5706 (VFMLA16_LOW): New int iterator.
5707 (VFMLA16_HIGH): Ditto.
5708 (UNSPEC_FMLAL): New unspec.
5709 (UNSPEC_FMLSL): Ditto.
5710 (UNSPEC_FMLAL2): Ditto.
5711 (UNSPEC_FMLSL2): Ditto.
5712 (f16mac): New code attribute.
5713 * config/aarch64/aarch64-simd-builtins.def
5714 (aarch64_fmlal_lowv2sf): Ditto.
5715 (aarch64_fmlsl_lowv2sf): Ditto.
5716 (aarch64_fmlalq_lowv4sf): Ditto.
5717 (aarch64_fmlslq_lowv4sf): Ditto.
5718 (aarch64_fmlal_highv2sf): Ditto.
5719 (aarch64_fmlsl_highv2sf): Ditto.
5720 (aarch64_fmlalq_highv4sf): Ditto.
5721 (aarch64_fmlslq_highv4sf): Ditto.
5722 (aarch64_fmlal_lane_lowv2sf): Ditto.
5723 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5724 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5725 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5726 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5727 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5728 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5729 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5730 (aarch64_fmlal_lane_highv2sf): Ditto.
5731 (aarch64_fmlsl_lane_highv2sf): Ditto.
5732 (aarch64_fmlal_laneq_highv2sf): Ditto.
5733 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5734 (aarch64_fmlalq_lane_highv4sf): Ditto.
5735 (aarch64_fmlsl_lane_highv4sf): Ditto.
5736 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5737 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5738 * config/aarch64/aarch64-simd.md:
5739 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5740 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5741 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5742 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5743 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5744 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5745 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5746 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5747 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5748 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5749 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5750 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5751 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5752 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5753 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5754 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5755 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5756 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5757 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5758 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5759 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5760 (vfmlsl_low_u32): Ditto.
5761 (vfmlalq_low_u32): Ditto.
5762 (vfmlslq_low_u32): Ditto.
5763 (vfmlal_high_u32): Ditto.
5764 (vfmlsl_high_u32): Ditto.
5765 (vfmlalq_high_u32): Ditto.
5766 (vfmlslq_high_u32): Ditto.
5767 (vfmlal_lane_low_u32): Ditto.
5768 (vfmlsl_lane_low_u32): Ditto.
5769 (vfmlal_laneq_low_u32): Ditto.
5770 (vfmlsl_laneq_low_u32): Ditto.
5771 (vfmlalq_lane_low_u32): Ditto.
5772 (vfmlslq_lane_low_u32): Ditto.
5773 (vfmlalq_laneq_low_u32): Ditto.
5774 (vfmlslq_laneq_low_u32): Ditto.
5775 (vfmlal_lane_high_u32): Ditto.
5776 (vfmlsl_lane_high_u32): Ditto.
5777 (vfmlal_laneq_high_u32): Ditto.
5778 (vfmlsl_laneq_high_u32): Ditto.
5779 (vfmlalq_lane_high_u32): Ditto.
5780 (vfmlslq_lane_high_u32): Ditto.
5781 (vfmlalq_laneq_high_u32): Ditto.
5782 (vfmlslq_laneq_high_u32): Ditto.
5783 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5784 (AARCH64_FL_FOR_ARCH8_4): New.
5785 (AARCH64_ISA_F16FML): New ISA flag.
5786 (TARGET_F16FML): New feature flag for fp16fml.
5787 (doc/invoke.texi): Document new fp16fml option.
5789 2018-01-10 Michael Collison <michael.collison@arm.com>
5791 * config/aarch64/aarch64-builtins.c:
5792 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5793 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5794 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5795 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5796 (AARCH64_ISA_SHA3): New ISA flag.
5797 (TARGET_SHA3): New feature flag for sha3.
5798 * config/aarch64/iterators.md (sha512_op): New int attribute.
5799 (CRYPTO_SHA512): New int iterator.
5800 (UNSPEC_SHA512H): New unspec.
5801 (UNSPEC_SHA512H2): Ditto.
5802 (UNSPEC_SHA512SU0): Ditto.
5803 (UNSPEC_SHA512SU1): Ditto.
5804 * config/aarch64/aarch64-simd-builtins.def
5805 (aarch64_crypto_sha512hqv2di): New builtin.
5806 (aarch64_crypto_sha512h2qv2di): Ditto.
5807 (aarch64_crypto_sha512su0qv2di): Ditto.
5808 (aarch64_crypto_sha512su1qv2di): Ditto.
5809 (aarch64_eor3qv8hi): Ditto.
5810 (aarch64_rax1qv2di): Ditto.
5811 (aarch64_xarqv2di): Ditto.
5812 (aarch64_bcaxqv8hi): Ditto.
5813 * config/aarch64/aarch64-simd.md:
5814 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5815 (aarch64_crypto_sha512su0qv2di): Ditto.
5816 (aarch64_crypto_sha512su1qv2di): Ditto.
5817 (aarch64_eor3qv8hi): Ditto.
5818 (aarch64_rax1qv2di): Ditto.
5819 (aarch64_xarqv2di): Ditto.
5820 (aarch64_bcaxqv8hi): Ditto.
5821 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5822 (vsha512h2q_u64): Ditto.
5823 (vsha512su0q_u64): Ditto.
5824 (vsha512su1q_u64): Ditto.
5825 (veor3q_u16): Ditto.
5826 (vrax1q_u64): Ditto.
5828 (vbcaxq_u16): Ditto.
5829 * config/arm/types.md (crypto_sha512): New type attribute.
5830 (crypto_sha3): Ditto.
5831 (doc/invoke.texi): Document new sha3 option.
5833 2018-01-10 Michael Collison <michael.collison@arm.com>
5835 * config/aarch64/aarch64-builtins.c:
5836 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5837 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5838 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5839 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5840 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5841 (AARCH64_ISA_SM4): New ISA flag.
5842 (TARGET_SM4): New feature flag for sm4.
5843 * config/aarch64/aarch64-simd-builtins.def
5844 (aarch64_sm3ss1qv4si): Ditto.
5845 (aarch64_sm3tt1aq4si): Ditto.
5846 (aarch64_sm3tt1bq4si): Ditto.
5847 (aarch64_sm3tt2aq4si): Ditto.
5848 (aarch64_sm3tt2bq4si): Ditto.
5849 (aarch64_sm3partw1qv4si): Ditto.
5850 (aarch64_sm3partw2qv4si): Ditto.
5851 (aarch64_sm4eqv4si): Ditto.
5852 (aarch64_sm4ekeyqv4si): Ditto.
5853 * config/aarch64/aarch64-simd.md:
5854 (aarch64_sm3ss1qv4si): Ditto.
5855 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5856 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5857 (aarch64_sm4eqv4si): Ditto.
5858 (aarch64_sm4ekeyqv4si): Ditto.
5859 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5860 (sm3part_op): Ditto.
5861 (CRYPTO_SM3TT): Ditto.
5862 (CRYPTO_SM3PART): Ditto.
5863 (UNSPEC_SM3SS1): New unspec.
5864 (UNSPEC_SM3TT1A): Ditto.
5865 (UNSPEC_SM3TT1B): Ditto.
5866 (UNSPEC_SM3TT2A): Ditto.
5867 (UNSPEC_SM3TT2B): Ditto.
5868 (UNSPEC_SM3PARTW1): Ditto.
5869 (UNSPEC_SM3PARTW2): Ditto.
5870 (UNSPEC_SM4E): Ditto.
5871 (UNSPEC_SM4EKEY): Ditto.
5872 * config/aarch64/constraints.md (Ui2): New constraint.
5873 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5874 * config/arm/types.md (crypto_sm3): New type attribute.
5875 (crypto_sm4): Ditto.
5876 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5877 (vsm3tt1aq_u32): Ditto.
5878 (vsm3tt1bq_u32): Ditto.
5879 (vsm3tt2aq_u32): Ditto.
5880 (vsm3tt2bq_u32): Ditto.
5881 (vsm3partw1q_u32): Ditto.
5882 (vsm3partw2q_u32): Ditto.
5883 (vsm4eq_u32): Ditto.
5884 (vsm4ekeyq_u32): Ditto.
5885 (doc/invoke.texi): Document new sm4 option.
5887 2018-01-10 Michael Collison <michael.collison@arm.com>
5889 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5890 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5891 (AARCH64_FL_FOR_ARCH8_4): New.
5892 (AARCH64_FL_V8_4): New flag.
5893 (doc/invoke.texi): Document new armv8.4-a option.
5895 2018-01-10 Michael Collison <michael.collison@arm.com>
5897 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5898 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5899 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5900 * config/aarch64/aarch64-option-extension.def: Add
5901 AARCH64_OPT_EXTENSION of 'sha2'.
5902 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5903 (crypto): Disable sha2 and aes if crypto disabled.
5904 (crypto): Enable aes and sha2 if enabled.
5905 (simd): Disable sha2 and aes if simd disabled.
5906 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5908 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5909 (TARGET_SHA2): New feature flag for sha2.
5910 (TARGET_AES): New feature flag for aes.
5911 * config/aarch64/aarch64-simd.md:
5912 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5913 conditional on TARGET_AES.
5914 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5915 (aarch64_crypto_sha1hsi): Make pattern conditional
5917 (aarch64_crypto_sha1hv4si): Ditto.
5918 (aarch64_be_crypto_sha1hv4si): Ditto.
5919 (aarch64_crypto_sha1su1v4si): Ditto.
5920 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5921 (aarch64_crypto_sha1su0v4si): Ditto.
5922 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5923 (aarch64_crypto_sha256su0v4si): Ditto.
5924 (aarch64_crypto_sha256su1v4si): Ditto.
5925 (doc/invoke.texi): Document new aes and sha2 options.
5927 2018-01-10 Martin Sebor <msebor@redhat.com>
5929 PR tree-optimization/83781
5930 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5933 2018-01-11 Martin Sebor <msebor@gmail.com>
5934 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5936 PR tree-optimization/83501
5937 PR tree-optimization/81703
5939 * tree-ssa-strlen.c (get_string_cst): Rename...
5940 (get_string_len): ...to this. Handle global constants.
5941 (handle_char_store): Adjust.
5943 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5944 Jim Wilson <jimw@sifive.com>
5946 * config/riscv/riscv-protos.h (riscv_output_return): New.
5947 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5948 (riscv_attribute_table, riscv_output_return),
5949 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5950 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5951 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5953 (riscv_expand_prologue): Add early return for naked function.
5954 (riscv_expand_epilogue): Likewise.
5955 (riscv_function_ok_for_sibcall): Return false for naked function.
5956 (riscv_set_current_function): New.
5957 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5958 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5959 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5960 * doc/extend.texi (RISC-V Function Attributes): New.
5962 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5964 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5965 check for 128-bit long double before checking TCmode.
5966 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5967 128-bit long doubles before checking TFmode or TCmode.
5968 (FLOAT128_IBM_P): Likewise.
5970 2018-01-10 Martin Sebor <msebor@redhat.com>
5972 PR tree-optimization/83671
5973 * builtins.c (c_strlen): Unconditionally return zero for the empty
5975 Use -Warray-bounds for warnings.
5976 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5977 for non-constant array indices with COMPONENT_REF, arrays of
5978 arrays, and pointers to arrays.
5979 (gimple_fold_builtin_strlen): Determine and set length range for
5980 non-constant character arrays.
5982 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5985 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5988 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5990 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5992 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5995 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5996 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5997 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5998 indexed_or_indirect_operand predicate.
5999 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
6000 (*vsx_le_perm_load_v8hi): Likewise.
6001 (*vsx_le_perm_load_v16qi): Likewise.
6002 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
6003 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
6004 (*vsx_le_perm_store_v8hi): Likewise.
6005 (*vsx_le_perm_store_v16qi): Likewise.
6006 (eight unnamed splitters): Likewise.
6008 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
6010 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
6011 * config/rs6000/emmintrin.h: Likewise.
6012 * config/rs6000/mmintrin.h: Likewise.
6013 * config/rs6000/xmmintrin.h: Likewise.
6015 2018-01-10 David Malcolm <dmalcolm@redhat.com>
6018 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
6020 * tree.c (tree_nop_conversion): Return true for location wrapper
6022 (maybe_wrap_with_location): New function.
6023 (selftest::check_strip_nops): New function.
6024 (selftest::test_location_wrappers): New function.
6025 (selftest::tree_c_tests): Call it.
6026 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
6027 (maybe_wrap_with_location): New decl.
6028 (EXPR_LOCATION_WRAPPER_P): New macro.
6029 (location_wrapper_p): New inline function.
6030 (tree_strip_any_location_wrapper): New inline function.
6032 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
6035 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
6036 stack_realign_offset for the largest alignment of stack slot
6038 (ix86_find_max_used_stack_alignment): New function.
6039 (ix86_finalize_stack_frame_flags): Use it. Set
6040 max_used_stack_alignment if we don't realign stack.
6041 * config/i386/i386.h (machine_function): Add
6042 max_used_stack_alignment.
6044 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
6046 * config/arm/arm.opt (-mbranch-cost): New option.
6047 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
6050 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
6053 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
6054 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
6056 2018-01-10 Richard Biener <rguenther@suse.de>
6059 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
6060 early out so it also covers the case where we have a non-NULL
6063 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
6065 PR tree-optimization/83753
6066 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
6067 for non-strided grouped accesses if the number of elements is 1.
6069 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
6072 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
6073 * i386.h (TARGET_USE_GATHER): Define.
6074 * x86-tune.def (X86_TUNE_USE_GATHER): New.
6076 2018-01-10 Martin Liska <mliska@suse.cz>
6079 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
6080 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
6082 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
6083 CLEANUP_NO_PARTITIONING is not set.
6085 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
6087 * doc/rtl.texi: Remove documentation of (const ...) wrappers
6088 for vectors, as a partial revert of r254296.
6089 * rtl.h (const_vec_p): Delete.
6090 (const_vec_duplicate_p): Don't test for vector CONSTs.
6091 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
6092 * expmed.c (make_tree): Likewise.
6095 * common.md (E, F): Use CONSTANT_P instead of checking for
6097 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
6098 checking for CONST_VECTOR.
6100 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6103 * predict.c (force_edge_cold): Handle in more sane way edges
6106 2018-01-09 Carl Love <cel@us.ibm.com>
6108 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
6110 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
6111 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
6112 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
6113 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
6114 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
6115 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
6116 * config/rs6000/rs6000-protos.h: Add extern defition for
6117 rs6000_generate_float2_double_code.
6118 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
6120 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
6121 (float2_v2df): Add define_expand.
6123 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
6126 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
6127 op_mode in the force_to_mode call.
6129 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6131 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
6132 instead of checking each element individually.
6133 (aarch64_evpc_uzp): Likewise.
6134 (aarch64_evpc_zip): Likewise.
6135 (aarch64_evpc_ext): Likewise.
6136 (aarch64_evpc_rev): Likewise.
6137 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
6138 instead of checking each element individually. Return true without
6140 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
6141 whether all selected elements come from the same input, instead of
6142 checking each element individually. Remove calls to gen_rtx_REG,
6143 start_sequence and end_sequence and instead assert that no rtl is
6146 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6148 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
6149 order of HIGH and CONST checks.
6151 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6153 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
6154 if the destination isn't an SSA_NAME.
6156 2018-01-09 Richard Biener <rguenther@suse.de>
6158 PR tree-optimization/83668
6159 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
6161 (canonicalize_loop_form): ... here, renamed from ...
6162 (canonicalize_loop_closed_ssa_form): ... this and amended to
6163 swap successor edges for loop exit blocks to make us use
6164 the RPO order we need for initial schedule generation.
6166 2018-01-09 Joseph Myers <joseph@codesourcery.com>
6168 PR tree-optimization/64811
6169 * match.pd: When optimizing comparisons with Inf, avoid
6170 introducing or losing exceptions from comparisons with NaN.
6172 2018-01-09 Martin Liska <mliska@suse.cz>
6175 * asan.c (shadow_mem_size): Add gcc_assert.
6177 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
6179 Don't save registers in main().
6182 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
6183 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
6184 * config/avr/avr.c (avr_set_current_function): Don't error if
6185 naked, OS_task or OS_main are specified at the same time.
6186 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
6188 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
6190 * common/config/avr/avr-common.c (avr_option_optimization_table):
6191 Switch on -mmain-is-OS_task for optimizing compilations.
6193 2018-01-09 Richard Biener <rguenther@suse.de>
6195 PR tree-optimization/83572
6196 * graphite.c: Include cfganal.h.
6197 (graphite_transform_loops): Connect infinite loops to exit
6198 and remove fake edges at the end.
6200 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6202 * ipa-inline.c (edge_badness): Revert accidental checkin.
6204 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6207 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
6208 symbols; not inline clones.
6210 2018-01-09 Jakub Jelinek <jakub@redhat.com>
6213 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
6214 hard registers. Formatting fixes.
6216 PR preprocessor/83722
6217 * gcc.c (try_generate_repro): Pass
6218 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
6219 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
6222 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
6223 Kito Cheng <kito.cheng@gmail.com>
6225 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
6226 (riscv_leaf_function_p): Delete.
6227 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
6229 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6231 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
6233 (do_ifelse): New function.
6234 (do_isel): New function.
6235 (do_sub3): New function.
6236 (do_add3): New function.
6237 (do_load_mask_compare): New function.
6238 (do_overlap_load_compare): New function.
6239 (expand_compare_loop): New function.
6240 (expand_block_compare): Call expand_compare_loop() when appropriate.
6241 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
6243 (-mblock-compare-inline-loop-limit): New option.
6245 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6248 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
6249 Reverse order of second and third operands in first alternative.
6250 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
6251 of first and second elements in UNSPEC_VPERMR vector.
6252 (altivec_expand_vec_perm_le): Likewise.
6254 2018-01-08 Jeff Law <law@redhat.com>
6256 PR rtl-optimizatin/81308
6257 * tree-switch-conversion.c (cfg_altered): New file scoped static.
6258 (process_switch): If group_case_labels makes a change, then set
6260 (pass_convert_switch::execute): If a switch is converted, then
6261 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
6263 PR rtl-optimization/81308
6264 * recog.c (split_all_insns): Conditionally cleanup the CFG after
6267 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
6269 PR target/83663 - Revert r255946
6270 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
6271 generation for cases where splatting a value is not useful.
6272 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
6273 across a vec_duplicate and a paradoxical subreg forming a vector
6274 mode to a vec_concat.
6276 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
6278 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
6279 -march=armv8.3-a variants.
6280 * config/arm/t-multilib: Likewise.
6281 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
6283 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6285 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
6287 (cceq_ior_compare_complement): Give it a name so I can use it, and
6288 change boolean_or_operator predicate to boolean_operator so it can
6289 be used to generate a crand.
6290 (eqne): New code iterator.
6291 (bd/bd_neg): New code_attrs.
6292 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
6293 a single define_insn.
6294 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
6295 decrement (bdnzt/bdnzf/bdzt/bdzf).
6296 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
6297 with the new names of the branch decrement patterns, and added the
6298 names of the branch decrement conditional patterns.
6300 2018-01-08 Richard Biener <rguenther@suse.de>
6302 PR tree-optimization/83563
6303 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
6306 2018-01-08 Richard Biener <rguenther@suse.de>
6309 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
6311 2018-01-08 Richard Biener <rguenther@suse.de>
6313 PR tree-optimization/83685
6314 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
6315 references to abnormals.
6317 2018-01-08 Richard Biener <rguenther@suse.de>
6320 * dwarf2out.c (output_indirect_strings): Handle empty
6321 skeleton_debug_str_hash.
6322 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
6324 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6326 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
6327 (emit_store_direct): Likewise.
6328 (arc_trampoline_adjust_address): Likewise.
6329 (arc_asm_trampoline_template): New function.
6330 (arc_initialize_trampoline): Use asm_trampoline_template.
6331 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
6332 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
6333 * config/arc/arc.md (flush_icache): Delete pattern.
6335 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6337 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
6338 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
6341 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6344 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
6345 by not USED_FOR_TARGET.
6346 (make_pass_resolve_sw_modes): Likewise.
6348 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6350 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
6353 2018-01-08 Richard Biener <rguenther@suse.de>
6356 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
6358 2018-01-08 Richard Biener <rguenther@suse.de>
6361 * match.pd ((t * 2) / 2) -> t): Add missing :c.
6363 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
6366 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
6367 basic blocks with a small number of successors.
6368 (convert_control_dep_chain_into_preds): Improve handling of
6370 (dump_predicates): Split apart into...
6371 (dump_pred_chain): ...here...
6372 (dump_pred_info): ...and here.
6373 (can_one_predicate_be_invalidated_p): Add debugging printfs.
6374 (can_chain_union_be_invalidated_p): Improve check for invalidation
6376 (uninit_uses_cannot_happen): Avoid unnecessary if
6377 convert_control_dep_chain_into_preds yielded nothing.
6379 2018-01-06 Martin Sebor <msebor@redhat.com>
6381 PR tree-optimization/83640
6382 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
6383 subtracting negative offset from size.
6384 (builtin_access::overlap): Adjust offset bounds of the access to fall
6385 within the size of the object if possible.
6387 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
6389 PR rtl-optimization/83699
6390 * expmed.c (extract_bit_field_1): Restrict the vector usage of
6391 extract_bit_field_as_subreg to cases in which the extracted
6392 value is also a vector.
6394 * lra-constraints.c (process_alt_operands): Test for the equivalence
6395 substitutions when detecting a possible reload cycle.
6397 2018-01-06 Jakub Jelinek <jakub@redhat.com>
6400 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
6401 by default if flag_selective_schedling{,2}. Formatting fixes.
6403 PR rtl-optimization/83682
6404 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
6405 if it has non-VECTOR_MODE element mode.
6406 (vec_duplicate_p): Likewise.
6409 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
6410 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6412 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6415 * config/i386/i386-builtin.def
6416 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6417 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6418 Require also OPTION_MASK_ISA_AVX512F in addition to
6419 OPTION_MASK_ISA_GFNI.
6420 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6421 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6422 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6423 to OPTION_MASK_ISA_GFNI.
6424 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6425 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6426 OPTION_MASK_ISA_AVX512BW.
6427 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6428 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6429 addition to OPTION_MASK_ISA_GFNI.
6430 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6431 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6432 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6433 to OPTION_MASK_ISA_GFNI.
6434 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6435 a requirement for all ISAs rather than any of them with a few
6437 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6439 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6440 bitmasks to be enabled with 3 exceptions, instead of requiring any
6441 enabled ISA with lots of exceptions.
6442 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6443 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6444 Change avx512bw in isa attribute to avx512f.
6445 * config/i386/sgxintrin.h: Add license boilerplate.
6446 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6447 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6448 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6449 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6451 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6452 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6453 temporarily sse2 rather than sse if not enabled already.
6456 * config/i386/sse.md (VI248_VLBW): Rename to ...
6457 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6458 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6459 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6460 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6461 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6462 mode iterator instead of VI248_VLBW.
6464 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6466 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6467 (record_modified): Skip clobbers; add debug output.
6468 (param_change_prob): Use sreal frequencies.
6470 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6472 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6473 punt for user-aligned variables.
6475 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6477 * tree-chrec.c (chrec_contains_symbols): Return true for
6480 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6483 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6484 of (x|y) == x for BICS pattern.
6486 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6488 PR tree-optimization/83605
6489 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6490 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6493 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6495 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6496 * config/epiphany/rtems.h: New file.
6498 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6499 Uros Bizjak <ubizjak@gmail.com>
6502 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6503 QIreg_operand instead of register_operand predicate.
6504 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6505 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6506 comments instead of -fmitigate[-_]rop.
6508 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6511 * cgraphunit.c (symbol_table::compile): Switch to text_section
6512 before calling assembly_start debug hook.
6513 * run-rtl-passes.c (run_rtl_passes): Likewise.
6516 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6518 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6519 range_int_cst_p rather than !symbolic_range_p before calling
6520 extract_range_from_multiplicative_op_1.
6522 2018-01-04 Jeff Law <law@redhat.com>
6524 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6525 redundant test in assertion.
6527 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6529 * doc/rtl.texi: Document machine_mode wrapper classes.
6531 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6533 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6536 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6538 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6539 the VEC_PERM_EXPR fold to fail.
6541 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6544 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6545 to switched_sections.
6547 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6550 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6553 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6556 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6557 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6559 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6562 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6563 is BLKmode and bitpos not zero or mode change is needed.
6565 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6568 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6571 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6574 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6575 instead of MULT rtx. Update all corresponding splitters.
6577 (*ssub<modesuffix>): Ditto.
6579 (*cmp_sadd_di): Update split patterns.
6580 (*cmp_sadd_si): Ditto.
6581 (*cmp_sadd_sidi): Ditto.
6582 (*cmp_ssub_di): Ditto.
6583 (*cmp_ssub_si): Ditto.
6584 (*cmp_ssub_sidi): Ditto.
6585 * config/alpha/predicates.md (const23_operand): New predicate.
6586 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6587 Look for ASHIFT, not MULT inner operand.
6588 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6590 2018-01-04 Martin Liska <mliska@suse.cz>
6592 PR gcov-profile/83669
6593 * gcov.c (output_intermediate_file): Add version to intermediate
6595 * doc/gcov.texi: Document new field 'version' in intermediate
6596 file format. Fix location of '-k' option of gcov command.
6598 2018-01-04 Martin Liska <mliska@suse.cz>
6601 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6603 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6605 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6607 2018-01-03 Martin Sebor <msebor@redhat.com>
6609 PR tree-optimization/83655
6610 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6611 checking calls with invalid arguments.
6613 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6615 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6616 (vectorizable_mask_load_store): Delete.
6617 (vectorizable_call): Return false for masked loads and stores.
6618 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6619 instead of gimple_assign_rhs1.
6620 (vectorizable_load): Handle IFN_MASK_LOAD.
6621 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6623 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6625 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6627 (vectorizable_mask_load_store): ...here.
6628 (vectorizable_load): ...and here.
6630 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6632 * tree-vect-stmts.c (vect_build_all_ones_mask)
6633 (vect_build_zero_merge_argument): New functions, split out from...
6634 (vectorizable_load): ...here.
6636 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6638 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6640 (vectorizable_mask_load_store): ...here.
6641 (vectorizable_store): ...and here.
6643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6645 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6647 (vectorizable_mask_load_store): ...here.
6649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6651 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6652 (vect_model_store_cost): Take a vec_load_store_type instead of a
6654 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6655 (vect_model_store_cost): Take a vec_load_store_type instead of a
6657 (vectorizable_mask_load_store): Update accordingly.
6658 (vectorizable_store): Likewise.
6659 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6661 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6663 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6664 IFN_MASK_LOAD calls here rather than...
6665 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6667 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6668 Alan Hayward <alan.hayward@arm.com>
6669 David Sherwood <david.sherwood@arm.com>
6671 * expmed.c (extract_bit_field_1): For vector extracts,
6672 fall back to extract_bit_field_as_subreg if vec_extract
6675 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6676 Alan Hayward <alan.hayward@arm.com>
6677 David Sherwood <david.sherwood@arm.com>
6679 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6680 they are variable or constant sized.
6681 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6682 slots for constant-sized data.
6684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6685 Alan Hayward <alan.hayward@arm.com>
6686 David Sherwood <david.sherwood@arm.com>
6688 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6689 handling COND_EXPRs with boolean comparisons, try to find a better
6690 basis for the mask type than the boolean itself.
6692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6694 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6695 is calculated and how it can be overridden.
6696 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6697 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6699 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6703 Alan Hayward <alan.hayward@arm.com>
6704 David Sherwood <david.sherwood@arm.com>
6706 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6707 Remove the mode argument.
6708 (aarch64_simd_valid_immediate): Remove the mode and inverse
6710 * config/aarch64/iterators.md (bitsize): New iterator.
6711 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6712 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6713 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6714 aarch64_simd_valid_immediate.
6715 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6716 (aarch64_reg_or_bic_imm): Likewise.
6717 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6718 with an insn_type enum and msl with a modifier_type enum.
6719 Replace element_width with a scalar_mode. Change the shift
6720 to unsigned int. Add constructors for scalar_float_mode and
6721 scalar_int_mode elements.
6722 (aarch64_vect_float_const_representable_p): Delete.
6723 (aarch64_can_const_movi_rtx_p)
6724 (aarch64_simd_scalar_immediate_valid_for_move)
6725 (aarch64_simd_make_constant): Update call to
6726 aarch64_simd_valid_immediate.
6727 (aarch64_advsimd_valid_immediate_hs): New function.
6728 (aarch64_advsimd_valid_immediate): Likewise.
6729 (aarch64_simd_valid_immediate): Remove mode and inverse
6730 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6731 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6732 and aarch64_float_const_representable_p on the result.
6733 (aarch64_output_simd_mov_immediate): Remove mode argument.
6734 Update call to aarch64_simd_valid_immediate and use of
6735 simd_immediate_info.
6736 (aarch64_output_scalar_simd_mov_immediate): Update call
6739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6740 Alan Hayward <alan.hayward@arm.com>
6741 David Sherwood <david.sherwood@arm.com>
6743 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6744 (mode_nunits): Likewise CONST_MODE_NUNITS.
6745 * machmode.def (ADJUST_NUNITS): Document.
6746 * genmodes.c (mode_data::need_nunits_adj): New field.
6747 (blank_mode): Update accordingly.
6748 (adj_nunits): New variable.
6749 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6751 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6752 listed in adj_nunits.
6753 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6754 listed in adj_nunits. Don't emit case statements for such modes.
6755 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6756 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6757 nothing if adj_nunits is nonnull.
6758 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6759 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6760 (emit_mode_fbit): Update use of print_maybe_const_decl.
6761 (emit_move_size): Likewise. Treat the array as non-const
6763 (emit_mode_adjustments): Handle adj_nunits.
6765 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6767 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6768 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6769 (VECTOR_MODES): Use it.
6770 (make_vector_modes): Take the prefix as an argument.
6772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6773 Alan Hayward <alan.hayward@arm.com>
6774 David Sherwood <david.sherwood@arm.com>
6776 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6777 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6778 for MODE_VECTOR_BOOL.
6779 * machmode.def (VECTOR_BOOL_MODE): Document.
6780 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6781 (make_vector_bool_mode): New function.
6782 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6784 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6785 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6787 * stor-layout.c (int_mode_for_mode): Likewise.
6788 * tree.c (build_vector_type_for_mode): Likewise.
6789 * varasm.c (output_constant_pool_2): Likewise.
6790 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6791 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6792 for MODE_VECTOR_BOOL.
6793 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6794 of mode class checks.
6795 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6796 instead of a list of mode class checks.
6797 (expand_vector_scalar_condition): Likewise.
6798 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6800 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6801 Alan Hayward <alan.hayward@arm.com>
6802 David Sherwood <david.sherwood@arm.com>
6804 * machmode.h (mode_size): Change from unsigned short to
6806 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6807 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6808 or if measurement_type is not polynomial.
6809 (fixed_size_mode::includes_p): Check for constant-sized modes.
6810 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6811 return a poly_uint16 rather than an unsigned short.
6812 (emit_mode_size): Change the type of mode_size from unsigned short
6813 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6814 (emit_mode_adjustments): Cope with polynomial vector sizes.
6815 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6817 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6819 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6820 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6821 * caller-save.c (setup_save_areas): Likewise.
6822 (replace_reg_with_saved_mem): Likewise.
6823 * calls.c (emit_library_call_value_1): Likewise.
6824 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6825 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6826 (gen_lowpart_for_combine): Likewise.
6827 * convert.c (convert_to_integer_1): Likewise.
6828 * cse.c (equiv_constant, cse_insn): Likewise.
6829 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6830 (cselib_subst_to_values): Likewise.
6831 * dce.c (word_dce_process_block): Likewise.
6832 * df-problems.c (df_word_lr_mark_ref): Likewise.
6833 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6834 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6835 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6836 (rtl_for_decl_location): Likewise.
6837 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6838 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6839 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6840 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6841 (expand_expr_real_1): Likewise.
6842 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6843 (pad_below): Likewise.
6844 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6845 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6846 * ira.c (get_subreg_tracking_sizes): Likewise.
6847 * ira-build.c (ira_create_allocno_objects): Likewise.
6848 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6849 (ira_sort_regnos_for_alter_reg): Likewise.
6850 * ira-costs.c (record_operand_costs): Likewise.
6851 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6852 (resolve_simple_move): Likewise.
6853 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6854 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6855 (lra_constraints): Likewise.
6856 (CONST_POOL_OK_P): Reject variable-sized modes.
6857 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6858 (add_pseudo_to_slot, lra_spill): Likewise.
6859 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6860 * optabs-query.c (get_best_extraction_insn): Likewise.
6861 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6862 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6863 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6864 * recog.c (offsettable_address_addr_space_p): Likewise.
6865 * regcprop.c (maybe_mode_change): Likewise.
6866 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6867 * regrename.c (build_def_use): Likewise.
6868 * regstat.c (dump_reg_info): Likewise.
6869 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6870 (find_reloads, find_reloads_subreg_address): Likewise.
6871 * reload1.c (eliminate_regs_1): Likewise.
6872 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6873 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6874 (simplify_binary_operation_1, simplify_subreg): Likewise.
6875 * targhooks.c (default_function_arg_padding): Likewise.
6876 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6877 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6878 (verify_gimple_assign_ternary): Likewise.
6879 * tree-inline.c (estimate_move_cost): Likewise.
6880 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6881 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6882 (get_address_cost_ainc): Likewise.
6883 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6884 (vect_supportable_dr_alignment): Likewise.
6885 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6886 (vectorizable_reduction): Likewise.
6887 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6888 (vectorizable_operation, vectorizable_load): Likewise.
6889 * tree.c (build_same_sized_truth_vector_type): Likewise.
6890 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6891 * var-tracking.c (emit_note_insn_var_location): Likewise.
6892 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6893 (ADDR_VEC_ALIGN): Likewise.
6895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6896 Alan Hayward <alan.hayward@arm.com>
6897 David Sherwood <david.sherwood@arm.com>
6899 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6901 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6902 or if measurement_type is polynomial.
6903 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6904 * combine.c (make_extraction): Likewise.
6905 * dse.c (find_shift_sequence): Likewise.
6906 * dwarf2out.c (mem_loc_descriptor): Likewise.
6907 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6908 (extract_bit_field, extract_low_bits): Likewise.
6909 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6910 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6911 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6912 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6913 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6914 * reload.c (find_reloads): Likewise.
6915 * reload1.c (alter_reg): Likewise.
6916 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6917 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6918 * tree-if-conv.c (predicate_mem_writes): Likewise.
6919 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6920 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6921 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6922 * valtrack.c (dead_debug_insert_temp): Likewise.
6923 * varasm.c (mergeable_constant_section): Likewise.
6924 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6926 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6927 Alan Hayward <alan.hayward@arm.com>
6928 David Sherwood <david.sherwood@arm.com>
6930 * expr.c (expand_assignment): Cope with polynomial mode sizes
6931 when assigning to a CONCAT.
6933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6934 Alan Hayward <alan.hayward@arm.com>
6935 David Sherwood <david.sherwood@arm.com>
6937 * machmode.h (mode_precision): Change from unsigned short to
6939 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6941 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6942 or if measurement_type is not polynomial.
6943 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6944 in which the mode is already known to be a scalar_int_mode.
6945 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6946 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6948 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6949 for GET_MODE_PRECISION.
6950 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6951 for GET_MODE_PRECISION.
6952 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6954 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6955 (expand_field_assignment, make_extraction): Likewise.
6956 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6957 (get_last_value): Likewise.
6958 * convert.c (convert_to_integer_1): Likewise.
6959 * cse.c (cse_insn): Likewise.
6960 * expr.c (expand_expr_real_1): Likewise.
6961 * lra-constraints.c (simplify_operand_subreg): Likewise.
6962 * optabs-query.c (can_atomic_load_p): Likewise.
6963 * optabs.c (expand_atomic_load): Likewise.
6964 (expand_atomic_store): Likewise.
6965 * ree.c (combine_reaching_defs): Likewise.
6966 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6967 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6968 * tree.h (type_has_mode_precision_p): Likewise.
6969 * ubsan.c (instrument_si_overflow): Likewise.
6971 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6972 Alan Hayward <alan.hayward@arm.com>
6973 David Sherwood <david.sherwood@arm.com>
6975 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6976 polynomial numbers of units.
6977 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6978 (valid_vector_subparts_p): New function.
6979 (build_vector_type): Remove temporary shim and take the number
6980 of units as a poly_uint64 rather than an int.
6981 (build_opaque_vector_type): Take the number of units as a
6982 poly_uint64 rather than an int.
6983 * tree.c (build_vector_from_ctor): Handle polynomial
6984 TYPE_VECTOR_SUBPARTS.
6985 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6986 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6987 (build_vector_from_val): If the number of units is variable,
6988 use build_vec_duplicate_cst for constant operands and
6989 VEC_DUPLICATE_EXPR otherwise.
6990 (make_vector_type): Remove temporary is_constant ().
6991 (build_vector_type, build_opaque_vector_type): Take the number of
6992 units as a poly_uint64 rather than an int.
6993 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6995 * cfgexpand.c (expand_debug_expr): Likewise.
6996 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6997 (store_constructor, expand_expr_real_1): Likewise.
6998 (const_scalar_mask_from_tree): Likewise.
6999 * fold-const-call.c (fold_const_reduction): Likewise.
7000 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
7001 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
7002 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
7003 (fold_relational_const): Likewise.
7004 (native_interpret_vector): Likewise. Change the size from an
7005 int to an unsigned int.
7006 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
7007 TYPE_VECTOR_SUBPARTS.
7008 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
7009 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
7010 duplicating a non-constant operand into a variable-length vector.
7011 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
7012 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
7013 * ipa-icf.c (sem_variable::equals): Likewise.
7014 * match.pd: Likewise.
7015 * omp-simd-clone.c (simd_clone_subparts): Likewise.
7016 * print-tree.c (print_node): Likewise.
7017 * stor-layout.c (layout_type): Likewise.
7018 * targhooks.c (default_builtin_vectorization_cost): Likewise.
7019 * tree-cfg.c (verify_gimple_comparison): Likewise.
7020 (verify_gimple_assign_binary): Likewise.
7021 (verify_gimple_assign_ternary): Likewise.
7022 (verify_gimple_assign_single): Likewise.
7023 * tree-pretty-print.c (dump_generic_node): Likewise.
7024 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7025 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
7026 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
7027 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
7028 (vect_shift_permute_load_chain): Likewise.
7029 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
7030 (expand_vector_condition, optimize_vector_constructor): Likewise.
7031 (lower_vec_perm, get_compute_type): Likewise.
7032 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
7033 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
7034 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
7035 (vect_recog_mask_conversion_pattern): Likewise.
7036 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
7037 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
7038 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7039 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
7040 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
7041 (vectorizable_shift, vectorizable_operation, vectorizable_store)
7042 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
7043 (supportable_widening_operation): Likewise.
7044 (supportable_narrowing_operation): Likewise.
7045 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
7047 * varasm.c (output_constant): Likewise.
7049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7050 Alan Hayward <alan.hayward@arm.com>
7051 David Sherwood <david.sherwood@arm.com>
7053 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
7054 so that both the length == 3 and length != 3 cases set up their
7055 own permute vectors. Add comments explaining why we know the
7056 number of elements is constant.
7057 (vect_permute_load_chain): Likewise.
7059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7060 Alan Hayward <alan.hayward@arm.com>
7061 David Sherwood <david.sherwood@arm.com>
7063 * machmode.h (mode_nunits): Change from unsigned char to
7065 (ONLY_FIXED_SIZE_MODES): New macro.
7066 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
7067 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
7068 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
7070 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
7071 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
7072 or if measurement_type is not polynomial.
7073 * genmodes.c (ZERO_COEFFS): New macro.
7074 (emit_mode_nunits_inline): Make mode_nunits_inline return a
7076 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
7077 Use ZERO_COEFFS when emitting initializers.
7078 * data-streamer.h (bp_pack_poly_value): New function.
7079 (bp_unpack_poly_value): Likewise.
7080 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
7081 for GET_MODE_NUNITS.
7082 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
7083 for GET_MODE_NUNITS.
7084 * tree.c (make_vector_type): Remove temporary shim and make
7085 the real function take the number of units as a poly_uint64
7087 (build_vector_type_for_mode): Handle polynomial nunits.
7088 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
7089 * emit-rtl.c (const_vec_series_p_1): Likewise.
7090 (gen_rtx_CONST_VECTOR): Likewise.
7091 * fold-const.c (test_vec_duplicate_folding): Likewise.
7092 * genrecog.c (validate_pattern): Likewise.
7093 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
7094 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
7095 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
7096 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
7097 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
7098 * rtlanal.c (subreg_get_info): Likewise.
7099 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7100 (vect_grouped_load_supported): Likewise.
7101 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
7102 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
7103 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
7104 (simplify_const_unary_operation, simplify_binary_operation_1)
7105 (simplify_const_binary_operation, simplify_ternary_operation)
7106 (test_vector_ops_duplicate, test_vector_ops): Likewise.
7107 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
7108 instead of CONST_VECTOR_NUNITS.
7109 * varasm.c (output_constant_pool_2): Likewise.
7110 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
7111 explicit-encoded elements in the XVEC for variable-length vectors.
7113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7115 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
7117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7118 Alan Hayward <alan.hayward@arm.com>
7119 David Sherwood <david.sherwood@arm.com>
7121 * coretypes.h (fixed_size_mode): Declare.
7122 (fixed_size_mode_pod): New typedef.
7123 * builtins.h (target_builtins::x_apply_args_mode)
7124 (target_builtins::x_apply_result_mode): Change type to
7125 fixed_size_mode_pod.
7126 * builtins.c (apply_args_size, apply_result_size, result_vector)
7127 (expand_builtin_apply_args_1, expand_builtin_apply)
7128 (expand_builtin_return): Update accordingly.
7130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7132 * cse.c (hash_rtx_cb): Hash only the encoded elements.
7133 * cselib.c (cselib_hash_rtx): Likewise.
7134 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
7135 CONST_VECTOR encoding.
7137 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7138 Jeff Law <law@redhat.com>
7141 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
7142 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
7143 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
7144 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
7147 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
7148 explicitly probe *sp in a noreturn function if there were any callee
7149 register saves or frame pointer is needed.
7151 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7154 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
7155 BLKmode for ternary, binary or unary expressions.
7158 * var-tracking.c (delete_vta_debug_insn): New inline function.
7159 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
7160 insns from get_insns () to NULL instead of each bb separately.
7161 Use delete_vta_debug_insn. No longer static.
7162 (vt_debug_insns_local, variable_tracking_main_1): Adjust
7163 delete_vta_debug_insns callers.
7164 * rtl.h (delete_vta_debug_insns): Declare.
7165 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
7166 instead of variable_tracking_main.
7168 2018-01-03 Martin Sebor <msebor@redhat.com>
7170 PR tree-optimization/83603
7171 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
7172 arguments past the endof the argument list in functions declared
7173 without a prototype.
7174 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
7175 Avoid checking when arguments are null.
7177 2018-01-03 Martin Sebor <msebor@redhat.com>
7180 * doc/extend.texi (attribute const): Fix a typo.
7181 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
7182 issuing -Wsuggest-attribute for void functions.
7184 2018-01-03 Martin Sebor <msebor@redhat.com>
7186 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
7187 offset_int::from instead of wide_int::to_shwi.
7188 (maybe_diag_overlap): Remove assertion.
7189 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
7190 * gimple-ssa-sprintf.c (format_directive): Same.
7191 (parse_directive): Same.
7192 (sprintf_dom_walker::compute_format_length): Same.
7193 (try_substitute_return_value): Same.
7195 2018-01-03 Jeff Law <law@redhat.com>
7198 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
7199 non-constant residual for zero at runtime and avoid probing in
7200 that case. Reorganize code for trailing problem to mirror handling
7203 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
7205 PR tree-optimization/83501
7206 * tree-ssa-strlen.c (get_string_cst): New.
7207 (handle_char_store): Call get_string_cst.
7209 2018-01-03 Martin Liska <mliska@suse.cz>
7211 PR tree-optimization/83593
7212 * tree-ssa-strlen.c: Include tree-cfg.h.
7213 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
7214 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
7215 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
7217 (strlen_dom_walker::before_dom_children): Call
7218 gimple_purge_dead_eh_edges. Dump tranformation with details
7220 (strlen_dom_walker::before_dom_children): Update call by adding
7221 new argument cleanup_eh.
7222 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
7224 2018-01-03 Martin Liska <mliska@suse.cz>
7227 * cif-code.def (VARIADIC_THUNK): New enum value.
7228 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
7231 2018-01-03 Jan Beulich <jbeulich@suse.com>
7233 * sse.md (mov<mode>_internal): Tighten condition for when to use
7234 vmovdqu<ssescalarsize> for TI and OI modes.
7236 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7238 Update copyright years.
7240 2018-01-03 Martin Liska <mliska@suse.cz>
7243 * ipa-visibility.c (function_and_variable_visibility): Skip
7244 functions with noipa attribure.
7246 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7248 * gcc.c (process_command): Update copyright notice dates.
7249 * gcov-dump.c (print_version): Ditto.
7250 * gcov.c (print_version): Ditto.
7251 * gcov-tool.c (print_version): Ditto.
7252 * gengtype.c (create_file): Ditto.
7253 * doc/cpp.texi: Bump @copying's copyright year.
7254 * doc/cppinternals.texi: Ditto.
7255 * doc/gcc.texi: Ditto.
7256 * doc/gccint.texi: Ditto.
7257 * doc/gcov.texi: Ditto.
7258 * doc/install.texi: Ditto.
7259 * doc/invoke.texi: Ditto.
7261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7263 * vector-builder.h (vector_builder::m_full_nelts): Change from
7264 unsigned int to poly_uint64.
7265 (vector_builder::full_nelts): Update prototype accordingly.
7266 (vector_builder::new_vector): Likewise.
7267 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
7268 (vector_builder::operator ==): Likewise.
7269 (vector_builder::finalize): Likewise.
7270 * int-vector-builder.h (int_vector_builder::int_vector_builder):
7271 Take the number of elements as a poly_uint64 rather than an
7273 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
7274 from unsigned int to poly_uint64.
7275 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
7276 (vec_perm_indices::new_vector): Likewise.
7277 (vec_perm_indices::length): Likewise.
7278 (vec_perm_indices::nelts_per_input): Likewise.
7279 (vec_perm_indices::input_nelts): Likewise.
7280 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
7281 number of elements per input as a poly_uint64 rather than an
7282 unsigned int. Use the original encoding for variable-length
7283 vectors, rather than clamping each individual element.
7284 For the second and subsequent elements in each pattern,
7285 clamp the step and base before clamping their sum.
7286 (vec_perm_indices::series_p): Handle polynomial element counts.
7287 (vec_perm_indices::all_in_range_p): Likewise.
7288 (vec_perm_indices_to_tree): Likewise.
7289 (vec_perm_indices_to_rtx): Likewise.
7290 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
7291 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
7292 (tree_vector_builder::new_binary_operation): Handle polynomial
7293 element counts. Return false if we need to know the number
7294 of elements at compile time.
7295 * fold-const.c (fold_vec_perm): Punt if the number of elements
7296 isn't known at compile time.
7298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7300 * vec-perm-indices.h (vec_perm_builder): Change element type
7301 from HOST_WIDE_INT to poly_int64.
7302 (vec_perm_indices::element_type): Update accordingly.
7303 (vec_perm_indices::clamp): Handle polynomial element_types.
7304 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7305 (vec_perm_indices::all_in_range_p): Likewise.
7306 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
7308 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
7309 polynomial vec_perm_indices element types.
7310 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
7311 * fold-const.c (fold_vec_perm): Likewise.
7312 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
7313 * tree-vect-generic.c (lower_vec_perm): Likewise.
7314 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7315 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
7316 element type to HOST_WIDE_INT.
7318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7319 Alan Hayward <alan.hayward@arm.com>
7320 David Sherwood <david.sherwood@arm.com>
7322 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
7323 rather than an int. Use plus_constant.
7324 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
7325 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
7327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7328 Alan Hayward <alan.hayward@arm.com>
7329 David Sherwood <david.sherwood@arm.com>
7331 * calls.c (emit_call_1, expand_call): Change struct_value_size from
7332 a HOST_WIDE_INT to a poly_int64.
7334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7335 Alan Hayward <alan.hayward@arm.com>
7336 David Sherwood <david.sherwood@arm.com>
7338 * calls.c (load_register_parameters): Cope with polynomial
7339 mode sizes. Require a constant size for BLKmode parameters
7340 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
7341 forces a parameter to be padded at the lsb end in order to
7342 fill a complete number of words, require the parameter size
7343 to be ordered wrt UNITS_PER_WORD.
7345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7346 Alan Hayward <alan.hayward@arm.com>
7347 David Sherwood <david.sherwood@arm.com>
7349 * reload1.c (spill_stack_slot_width): Change element type
7350 from unsigned int to poly_uint64_pod.
7351 (alter_reg): Treat mode sizes as polynomial.
7353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7354 Alan Hayward <alan.hayward@arm.com>
7355 David Sherwood <david.sherwood@arm.com>
7357 * reload.c (complex_word_subreg_p): New function.
7358 (reload_inner_reg_of_subreg, push_reload): Use it.
7360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7361 Alan Hayward <alan.hayward@arm.com>
7362 David Sherwood <david.sherwood@arm.com>
7364 * lra-constraints.c (process_alt_operands): Reject matched
7365 operands whose sizes aren't ordered.
7366 (match_reload): Refer to this check here.
7368 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7369 Alan Hayward <alan.hayward@arm.com>
7370 David Sherwood <david.sherwood@arm.com>
7372 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
7373 that the mode size is in the set {1, 2, 4, 8, 16}.
7375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7376 Alan Hayward <alan.hayward@arm.com>
7377 David Sherwood <david.sherwood@arm.com>
7379 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
7380 Use plus_constant instead of gen_rtx_PLUS.
7382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7383 Alan Hayward <alan.hayward@arm.com>
7384 David Sherwood <david.sherwood@arm.com>
7386 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
7387 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
7388 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
7389 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
7390 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
7391 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
7392 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
7393 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
7394 * config/i386/i386.c (ix86_push_rounding): ...this new function.
7395 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
7397 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
7398 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
7399 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
7400 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
7401 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
7402 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
7403 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
7404 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
7405 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
7406 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
7408 * expr.c (emit_move_resolve_push): Treat the input and result
7409 of PUSH_ROUNDING as a poly_int64.
7410 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
7411 (emit_push_insn): Likewise.
7412 * lra-eliminations.c (mark_not_eliminable): Likewise.
7413 * recog.c (push_operand): Likewise.
7414 * reload1.c (elimination_effects): Likewise.
7415 * rtlanal.c (nonzero_bits1): Likewise.
7416 * calls.c (store_one_arg): Likewise. Require the padding to be
7417 known at compile time.
7419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7420 Alan Hayward <alan.hayward@arm.com>
7421 David Sherwood <david.sherwood@arm.com>
7423 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7424 Use plus_constant instead of gen_rtx_PLUS.
7426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7427 Alan Hayward <alan.hayward@arm.com>
7428 David Sherwood <david.sherwood@arm.com>
7430 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7433 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7434 Alan Hayward <alan.hayward@arm.com>
7435 David Sherwood <david.sherwood@arm.com>
7437 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7438 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7439 via stack temporaries. Treat the mode size as polynomial too.
7441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7442 Alan Hayward <alan.hayward@arm.com>
7443 David Sherwood <david.sherwood@arm.com>
7445 * expr.c (expand_expr_real_2): When handling conversions involving
7446 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7447 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7448 as a poly_uint64 too.
7450 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7451 Alan Hayward <alan.hayward@arm.com>
7452 David Sherwood <david.sherwood@arm.com>
7454 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
7456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7457 Alan Hayward <alan.hayward@arm.com>
7458 David Sherwood <david.sherwood@arm.com>
7460 * combine.c (can_change_dest_mode): Handle polynomial
7461 REGMODE_NATURAL_SIZE.
7462 * expmed.c (store_bit_field_1): Likewise.
7463 * expr.c (store_constructor): Likewise.
7464 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7465 and polynomial REGMODE_NATURAL_SIZE.
7466 (gen_lowpart_common): Likewise.
7467 * reginfo.c (record_subregs_of_mode): Likewise.
7468 * rtlanal.c (read_modify_subreg_p): Likewise.
7470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7471 Alan Hayward <alan.hayward@arm.com>
7472 David Sherwood <david.sherwood@arm.com>
7474 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7475 numbers of elements.
7477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7478 Alan Hayward <alan.hayward@arm.com>
7479 David Sherwood <david.sherwood@arm.com>
7481 * match.pd: Cope with polynomial numbers of vector elements.
7483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7484 Alan Hayward <alan.hayward@arm.com>
7485 David Sherwood <david.sherwood@arm.com>
7487 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7488 in a POINTER_PLUS_EXPR.
7490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7491 Alan Hayward <alan.hayward@arm.com>
7492 David Sherwood <david.sherwood@arm.com>
7494 * omp-simd-clone.c (simd_clone_subparts): New function.
7495 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7496 (ipa_simd_modify_function_body): Likewise.
7498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7499 Alan Hayward <alan.hayward@arm.com>
7500 David Sherwood <david.sherwood@arm.com>
7502 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7503 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7504 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7505 (expand_vector_condition, vector_element): Likewise.
7506 (subparts_gt): New function.
7507 (get_compute_type): Use subparts_gt.
7508 (count_type_subparts): Delete.
7509 (expand_vector_operations_1): Use subparts_gt instead of
7510 count_type_subparts.
7512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7513 Alan Hayward <alan.hayward@arm.com>
7514 David Sherwood <david.sherwood@arm.com>
7516 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7517 (vect_compile_time_alias): ...this new function. Do the calculation
7518 on poly_ints rather than trees.
7519 (vect_prune_runtime_alias_test_list): Update call accordingly.
7521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7522 Alan Hayward <alan.hayward@arm.com>
7523 David Sherwood <david.sherwood@arm.com>
7525 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7527 (vect_schedule_slp_instance): Likewise.
7529 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7530 Alan Hayward <alan.hayward@arm.com>
7531 David Sherwood <david.sherwood@arm.com>
7533 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7534 constant and extern definitions for variable-length vectors.
7535 (vect_get_constant_vectors): Note that the number of units
7536 is known to be constant.
7538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7539 Alan Hayward <alan.hayward@arm.com>
7540 David Sherwood <david.sherwood@arm.com>
7542 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7543 of units as polynomial. Choose between WIDE and NARROW based
7546 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7547 Alan Hayward <alan.hayward@arm.com>
7548 David Sherwood <david.sherwood@arm.com>
7550 * tree-vect-stmts.c (simd_clone_subparts): New function.
7551 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7553 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7554 Alan Hayward <alan.hayward@arm.com>
7555 David Sherwood <david.sherwood@arm.com>
7557 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7558 vectors as polynomial. Use build_index_vector for
7561 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7562 Alan Hayward <alan.hayward@arm.com>
7563 David Sherwood <david.sherwood@arm.com>
7565 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7566 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7567 for variable-length vectors.
7568 (vectorizable_mask_load_store): Treat the number of units as
7569 polynomial, asserting that it is constant if the condition has
7570 already been enforced.
7571 (vectorizable_store, vectorizable_load): Likewise.
7573 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7574 Alan Hayward <alan.hayward@arm.com>
7575 David Sherwood <david.sherwood@arm.com>
7577 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7578 of units as polynomial. Punt if we can't tell at compile time
7579 which vector contains the final result.
7581 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7582 Alan Hayward <alan.hayward@arm.com>
7583 David Sherwood <david.sherwood@arm.com>
7585 * tree-vect-loop.c (vectorizable_induction): Treat the number
7586 of units as polynomial. Punt on SLP inductions. Use an integer
7587 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7588 cast of such a series for variable-length floating-point
7591 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7592 Alan Hayward <alan.hayward@arm.com>
7593 David Sherwood <david.sherwood@arm.com>
7595 * tree.h (build_index_vector): Declare.
7596 * tree.c (build_index_vector): New function.
7597 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7598 of units as polynomial, forcibly converting it to a constant if
7599 vectorizable_reduction has already enforced the condition.
7600 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7601 to create a {1,2,3,...} vector.
7602 (vectorizable_reduction): Treat the number of units as polynomial.
7603 Choose vectype_in based on the largest scalar element size rather
7604 than the smallest number of units. Enforce the restrictions
7607 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7608 Alan Hayward <alan.hayward@arm.com>
7609 David Sherwood <david.sherwood@arm.com>
7611 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7612 number of units as polynomial.
7614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7615 Alan Hayward <alan.hayward@arm.com>
7616 David Sherwood <david.sherwood@arm.com>
7618 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7619 * target.def (autovectorize_vector_sizes): Return the vector sizes
7620 by pointer, using vector_sizes rather than a bitmask.
7621 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7622 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7623 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7625 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7626 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7627 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7628 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7629 * omp-general.c (omp_max_vf): Likewise.
7630 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7631 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7632 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7633 * tree-vect-slp.c (vect_slp_bb): Likewise.
7634 * doc/tm.texi: Regenerate.
7635 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7637 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7638 the vector size as a poly_uint64 rather than an unsigned int.
7639 (current_vector_size): Change from an unsigned int to a poly_uint64.
7640 (get_vectype_for_scalar_type): Update accordingly.
7641 * tree.h (build_truth_vector_type): Take the size and number of
7642 units as a poly_uint64 rather than an unsigned int.
7643 (build_vector_type): Add a temporary overload that takes
7644 the number of units as a poly_uint64 rather than an unsigned int.
7645 * tree.c (make_vector_type): Likewise.
7646 (build_truth_vector_type): Take the number of units as a poly_uint64
7647 rather than an unsigned int.
7649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7650 Alan Hayward <alan.hayward@arm.com>
7651 David Sherwood <david.sherwood@arm.com>
7653 * target.def (get_mask_mode): Take the number of units and length
7654 as poly_uint64s rather than unsigned ints.
7655 * targhooks.h (default_get_mask_mode): Update accordingly.
7656 * targhooks.c (default_get_mask_mode): Likewise.
7657 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7658 * doc/tm.texi: Regenerate.
7660 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7661 Alan Hayward <alan.hayward@arm.com>
7662 David Sherwood <david.sherwood@arm.com>
7664 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7665 * omp-general.c (omp_max_vf): Likewise.
7666 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7667 (expand_omp_simd): Handle polynomial safelen.
7668 * omp-low.c (omplow_simd_context): Add a default constructor.
7669 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7670 (lower_rec_simd_input_clauses): Update accordingly.
7671 (lower_rec_input_clauses): Likewise.
7673 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7674 Alan Hayward <alan.hayward@arm.com>
7675 David Sherwood <david.sherwood@arm.com>
7677 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7678 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7679 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7680 (vect_analyze_slp_cost): Likewise.
7681 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7682 (vect_model_load_cost): Likewise.
7684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7685 Alan Hayward <alan.hayward@arm.com>
7686 David Sherwood <david.sherwood@arm.com>
7688 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7689 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7690 from an unsigned int * to a poly_uint64_pod *.
7691 (calculate_unrolling_factor): New function.
7692 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7695 Alan Hayward <alan.hayward@arm.com>
7696 David Sherwood <david.sherwood@arm.com>
7698 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7699 from an unsigned int to a poly_uint64.
7700 (_loop_vec_info::slp_unrolling_factor): Likewise.
7701 (_loop_vec_info::vectorization_factor): Change from an int
7703 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7704 (vect_get_num_vectors): New function.
7705 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7706 (vect_get_num_copies): Use vect_get_num_vectors.
7707 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7708 to an unsigned int *.
7709 (vect_analyze_data_refs): Change min_vf from an int * to a
7711 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7712 than an unsigned HOST_WIDE_INT.
7713 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7714 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7715 to an unsigned int *.
7716 (vect_analyze_data_ref_dependences): Likewise.
7717 (vect_compute_data_ref_alignment): Handle polynomial vf.
7718 (vect_enhance_data_refs_alignment): Likewise.
7719 (vect_prune_runtime_alias_test_list): Likewise.
7720 (vect_shift_permute_load_chain): Likewise.
7721 (vect_supportable_dr_alignment): Likewise.
7722 (dependence_distance_ge_vf): Take the vectorization factor as a
7723 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7724 (vect_analyze_data_refs): Change min_vf from an int * to a
7726 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7727 vfm1 as a poly_uint64 rather than an int. Make the same change
7728 for the returned bound_scalar.
7729 (vect_gen_vector_loop_niters): Handle polynomial vf.
7730 (vect_do_peeling): Likewise. Update call to
7731 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7732 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7734 * tree-vect-loop.c (vect_determine_vectorization_factor)
7735 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7736 (vect_get_known_peeling_cost): Likewise.
7737 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7738 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7739 (vect_transform_loop): Likewise. Use the lowest possible VF when
7740 updating the upper bounds of the loop.
7741 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7743 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7744 polynomial unroll factors.
7745 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7746 (vect_make_slp_decision): Likewise.
7747 (vect_supported_load_permutation_p): Likewise, and polynomial
7749 (vect_analyze_slp_cost): Handle polynomial vf.
7750 (vect_slp_analyze_node_operations): Likewise.
7751 (vect_slp_analyze_bb_1): Likewise.
7752 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7753 than an unsigned HOST_WIDE_INT.
7754 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7755 (vectorizable_load): Handle polynomial vf.
7756 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7758 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7761 Alan Hayward <alan.hayward@arm.com>
7762 David Sherwood <david.sherwood@arm.com>
7764 * match.pd: Handle bit operations involving three constants
7765 and try to fold one pair.
7767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7769 * tree-vect-loop-manip.c: Include gimple-fold.h.
7770 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7771 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7772 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7773 Add a path that uses a step of VF instead of 1, but disable it
7775 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7776 and niters_no_overflow parameters. Update calls to
7777 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7778 Create a new SSA name if the latter choses to use a ste other
7779 than zero, and return it via niters_vector_mult_vf_var.
7780 * tree-vect-loop.c (vect_transform_loop): Update calls to
7781 vect_do_peeling, vect_gen_vector_loop_niters and
7782 slpeel_make_loop_iterate_ntimes.
7783 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7784 (vect_gen_vector_loop_niters): Update declarations after above changes.
7786 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7788 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7789 128-bit round to integer instructions.
7790 (ceil<mode>2): Likewise.
7791 (btrunc<mode>2): Likewise.
7792 (round<mode>2): Likewise.
7794 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7796 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7797 unaligned VSX load/store on P8/P9.
7798 (expand_block_clear): Allow the use of unaligned VSX
7799 load/store on P8/P9.
7801 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7803 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7805 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7806 swap associated with both a load and a store.
7808 2018-01-02 Andrew Waterman <andrew@sifive.com>
7810 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7811 * config/riscv/riscv.md (clear_cache): Use it.
7813 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7815 * web.c: Remove out-of-date comment.
7817 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7819 * expr.c (fixup_args_size_notes): Check that any existing
7820 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7821 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7822 (emit_single_push_insn): ...here.
7824 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7826 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7827 (const_vector_encoded_nelts): New function.
7828 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7829 (const_vector_int_elt, const_vector_elt): Declare.
7830 * emit-rtl.c (const_vector_int_elt_1): New function.
7831 (const_vector_elt): Likewise.
7832 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7833 of CONST_VECTOR_ELT.
7835 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7837 * expr.c: Include rtx-vector-builder.h.
7838 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7839 directly on the tree encoding.
7840 (const_vector_from_tree): Likewise.
7841 * optabs.c: Include rtx-vector-builder.h.
7842 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7843 sequence of "u" values.
7844 * vec-perm-indices.c: Include rtx-vector-builder.h.
7845 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7846 directly on the vec_perm_indices encoding.
7848 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7850 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7851 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7852 * rtx-vector-builder.h: New file.
7853 * rtx-vector-builder.c: Likewise.
7854 * rtl.h (rtx_def::u2): Add a const_vector field.
7855 (CONST_VECTOR_NPATTERNS): New macro.
7856 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7857 (CONST_VECTOR_DUPLICATE_P): Likewise.
7858 (CONST_VECTOR_STEPPED_P): Likewise.
7859 (CONST_VECTOR_ENCODED_ELT): Likewise.
7860 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7861 (unwrap_const_vec_duplicate): Likewise.
7862 (const_vec_series_p): Check for a non-duplicated vector encoding.
7863 Say that the function only returns true for integer vectors.
7864 * emit-rtl.c: Include rtx-vector-builder.h.
7865 (gen_const_vec_duplicate_1): Delete.
7866 (gen_const_vector): Call gen_const_vec_duplicate instead of
7867 gen_const_vec_duplicate_1.
7868 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7869 (gen_const_vec_duplicate): Use rtx_vector_builder.
7870 (gen_const_vec_series): Likewise.
7871 (gen_rtx_CONST_VECTOR): Likewise.
7872 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7873 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7874 Build a new vector rather than modifying a CONST_VECTOR in-place.
7875 (handle_special_swappables): Update call accordingly.
7876 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7877 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7878 Build a new vector rather than modifying a CONST_VECTOR in-place.
7879 (handle_special_swappables): Update call accordingly.
7881 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7883 * simplify-rtx.c (simplify_const_binary_operation): Use
7884 CONST_VECTOR_ELT instead of XVECEXP.
7886 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7888 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7889 the selector elements to be different from the data elements
7890 if the selector is a VECTOR_CST.
7891 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7892 ssizetype for the selector.
7894 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7896 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7897 before testing each element individually.
7898 * tree-vect-generic.c (lower_vec_perm): Likewise.
7900 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7902 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7903 * selftest-run-tests.c (selftest::run_tests): Call it.
7904 * vector-builder.h (vector_builder::operator ==): New function.
7905 (vector_builder::operator !=): Likewise.
7906 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7907 (vec_perm_indices::all_from_input_p): New function.
7908 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7909 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7910 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7911 instead of reading the VECTOR_CST directly. Detect whether both
7912 vector inputs are the same before constructing the vec_perm_indices,
7913 and update the number of inputs argument accordingly. Use the
7914 utility functions added above. Only construct sel2 if we need to.
7916 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7918 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7919 the broadcast of the low byte.
7920 (expand_mult_highpart): Use an explicit encoding for the permutes.
7921 * optabs-query.c (can_mult_highpart_p): Likewise.
7922 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7923 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7924 (vectorizable_bswap): Likewise.
7925 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7926 explicit encoding for the power-of-2 permutes.
7927 (vect_permute_store_chain): Likewise.
7928 (vect_grouped_load_supported): Likewise.
7929 (vect_permute_load_chain): Likewise.
7931 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7933 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7934 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7935 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7936 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7937 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7938 (vect_gen_perm_mask_any): Likewise.
7940 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7942 * int-vector-builder.h: New file.
7943 * vec-perm-indices.h: Include int-vector-builder.h.
7944 (vec_perm_indices): Redefine as an int_vector_builder.
7945 (auto_vec_perm_indices): Delete.
7946 (vec_perm_builder): Redefine as a stand-alone class.
7947 (vec_perm_indices::vec_perm_indices): New function.
7948 (vec_perm_indices::clamp): Likewise.
7949 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7950 (vec_perm_indices::new_vector): New function.
7951 (vec_perm_indices::new_expanded_vector): Update for new
7952 vec_perm_indices class.
7953 (vec_perm_indices::rotate_inputs): New function.
7954 (vec_perm_indices::all_in_range_p): Operate directly on the
7955 encoded form, without computing elided elements.
7956 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7957 encoding. Update for new vec_perm_indices class.
7958 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7959 the given vec_perm_builder.
7960 (expand_vec_perm_var): Update vec_perm_builder constructor.
7961 (expand_mult_highpart): Use vec_perm_builder instead of
7962 auto_vec_perm_indices.
7963 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7964 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7965 or double series encoding as appropriate.
7966 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7967 vec_perm_indices instead of auto_vec_perm_indices.
7968 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7969 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7970 (vect_permute_store_chain): Likewise.
7971 (vect_grouped_load_supported): Likewise.
7972 (vect_permute_load_chain): Likewise.
7973 (vect_shift_permute_load_chain): Likewise.
7974 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7975 (vect_transform_slp_perm_load): Likewise.
7976 (vect_schedule_slp_instance): Likewise.
7977 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7978 (vectorizable_mask_load_store): Likewise.
7979 (vectorizable_bswap): Likewise.
7980 (vectorizable_store): Likewise.
7981 (vectorizable_load): Likewise.
7982 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7983 vec_perm_indices instead of auto_vec_perm_indices. Use
7984 tree_to_vec_perm_builder to read the vector from a tree.
7985 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7986 vec_perm_builder instead of a vec_perm_indices.
7987 (have_whole_vector_shift): Use vec_perm_builder and
7988 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7989 truncation to calc_vec_perm_mask_for_shift.
7990 (vect_create_epilog_for_reduction): Likewise.
7991 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7992 from auto_vec_perm_indices to vec_perm_indices.
7993 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7994 instead of changing individual elements.
7995 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7996 the vector in d.perm.
7997 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7998 from auto_vec_perm_indices to vec_perm_indices.
7999 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
8000 instead of changing individual elements.
8001 (arm_vectorize_vec_perm_const): Use new_vector to install
8002 the vector in d.perm.
8003 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
8004 Update vec_perm_builder constructor.
8005 (rs6000_expand_interleave): Likewise.
8006 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
8007 (rs6000_expand_interleave): Likewise.
8009 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8011 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
8012 to qimode could truncate the indices.
8013 * optabs.c (expand_vec_perm_var): Likewise.
8015 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8017 * Makefile.in (OBJS): Add vec-perm-indices.o.
8018 * vec-perm-indices.h: New file.
8019 * vec-perm-indices.c: Likewise.
8020 * target.h (vec_perm_indices): Replace with a forward class
8022 (auto_vec_perm_indices): Move to vec-perm-indices.h.
8023 * optabs.h: Include vec-perm-indices.h.
8024 (expand_vec_perm): Delete.
8025 (selector_fits_mode_p, expand_vec_perm_var): Declare.
8026 (expand_vec_perm_const): Declare.
8027 * target.def (vec_perm_const_ok): Replace with...
8028 (vec_perm_const): ...this new hook.
8029 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
8030 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
8031 * doc/tm.texi: Regenerate.
8032 * optabs.def (vec_perm_const): Delete.
8033 * doc/md.texi (vec_perm_const): Likewise.
8034 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
8035 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
8036 expand_vec_perm for constant permutation vectors. Assert that
8037 the mode of variable permutation vectors is the integer equivalent
8038 of the mode that is being permuted.
8039 * optabs-query.h (selector_fits_mode_p): Declare.
8040 * optabs-query.c: Include vec-perm-indices.h.
8041 (selector_fits_mode_p): New function.
8042 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
8043 is defined, instead of checking whether the vec_perm_const_optab
8044 exists. Use targetm.vectorize.vec_perm_const instead of
8045 targetm.vectorize.vec_perm_const_ok. Check whether the indices
8046 fit in the vector mode before using a variable permute.
8047 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
8048 vec_perm_indices instead of an rtx.
8049 (expand_vec_perm): Replace with...
8050 (expand_vec_perm_const): ...this new function. Take the selector
8051 as a vec_perm_indices rather than an rtx. Also take the mode of
8052 the selector. Update call to shift_amt_for_vec_perm_mask.
8053 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
8054 Use vec_perm_indices::new_expanded_vector to expand the original
8055 selector into bytes. Check whether the indices fit in the vector
8056 mode before using a variable permute.
8057 (expand_vec_perm_var): Make global.
8058 (expand_mult_highpart): Use expand_vec_perm_const.
8059 * fold-const.c: Includes vec-perm-indices.h.
8060 * tree-ssa-forwprop.c: Likewise.
8061 * tree-vect-data-refs.c: Likewise.
8062 * tree-vect-generic.c: Likewise.
8063 * tree-vect-loop.c: Likewise.
8064 * tree-vect-slp.c: Likewise.
8065 * tree-vect-stmts.c: Likewise.
8066 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
8068 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
8069 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
8070 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
8071 (aarch64_vectorize_vec_perm_const): ...this new function.
8072 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8073 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8074 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
8075 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
8076 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8077 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8078 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
8080 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
8081 check for NEON modes.
8082 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
8083 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
8084 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
8085 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
8087 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
8088 the old VEC_PERM_CONST conditions.
8089 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
8090 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
8091 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
8092 (ia64_vectorize_vec_perm_const_ok): Merge into...
8093 (ia64_vectorize_vec_perm_const): ...this new function.
8094 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
8095 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
8096 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
8097 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
8098 * config/mips/mips.c (mips_expand_vec_perm_const)
8099 (mips_vectorize_vec_perm_const_ok): Merge into...
8100 (mips_vectorize_vec_perm_const): ...this new function.
8101 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
8102 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
8103 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
8104 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
8105 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
8106 (rs6000_expand_vec_perm_const): Delete.
8107 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
8109 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8110 (altivec_expand_vec_perm_const_le): Take each operand individually.
8111 Operate on constant selectors rather than rtxes.
8112 (altivec_expand_vec_perm_const): Likewise. Update call to
8113 altivec_expand_vec_perm_const_le.
8114 (rs6000_expand_vec_perm_const): Delete.
8115 (rs6000_vectorize_vec_perm_const_ok): Delete.
8116 (rs6000_vectorize_vec_perm_const): New function.
8117 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
8118 an element count and rtx array.
8119 (rs6000_expand_extract_even): Update call accordingly.
8120 (rs6000_expand_interleave): Likewise.
8121 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
8122 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
8123 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
8124 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
8125 (rs6000_expand_vec_perm_const): Delete.
8126 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8127 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8128 (altivec_expand_vec_perm_const_le): Take each operand individually.
8129 Operate on constant selectors rather than rtxes.
8130 (altivec_expand_vec_perm_const): Likewise. Update call to
8131 altivec_expand_vec_perm_const_le.
8132 (rs6000_expand_vec_perm_const): Delete.
8133 (rs6000_vectorize_vec_perm_const_ok): Delete.
8134 (rs6000_vectorize_vec_perm_const): New function. Remove stray
8135 reference to the SPE evmerge intructions.
8136 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
8137 an element count and rtx array.
8138 (rs6000_expand_extract_even): Update call accordingly.
8139 (rs6000_expand_interleave): Likewise.
8140 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
8141 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
8143 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8145 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8147 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
8148 vector mode and that that mode matches the mode of the data
8150 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
8151 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
8152 directly using expand_vec_perm_1 when forcing selectors into
8154 (expand_vec_perm_var): New function, split out from expand_vec_perm.
8156 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8158 * optabs-query.h (can_vec_perm_p): Delete.
8159 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
8160 * optabs-query.c (can_vec_perm_p): Split into...
8161 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
8162 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
8163 particular selector is valid.
8164 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
8165 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
8166 (vect_grouped_load_supported): Likewise.
8167 (vect_shift_permute_load_chain): Likewise.
8168 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
8169 (vect_transform_slp_perm_load): Likewise.
8170 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
8171 (vectorizable_bswap): Likewise.
8172 (vect_gen_perm_mask_checked): Likewise.
8173 * fold-const.c (fold_ternary_loc): Likewise. Don't take
8174 implementations of variable permutation vectors into account
8175 when deciding which selector to use.
8176 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
8177 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
8178 with a false third argument.
8179 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
8180 to test whether the constant selector is valid and can_vec_perm_var_p
8181 to test whether a variable selector is valid.
8183 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8185 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
8186 * optabs-query.c (can_vec_perm_p): Likewise.
8187 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
8188 instead of vec_perm_indices.
8189 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
8190 (vect_gen_perm_mask_checked): Likewise,
8191 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
8192 (vect_gen_perm_mask_checked): Likewise,
8194 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8196 * optabs-query.h (qimode_for_vec_perm): Declare.
8197 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
8198 (qimode_for_vec_perm): ...this new function.
8199 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
8201 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
8203 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
8204 does not have a conditional at the top.
8206 2018-01-02 Richard Biener <rguenther@suse.de>
8208 * ipa-inline.c (big_speedup_p): Fix expression.
8210 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
8213 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
8216 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
8220 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
8221 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
8222 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
8223 cond_taken_branch_cost 3->4.
8225 2018-01-01 Jakub Jelinek <jakub@redhat.com>
8227 PR tree-optimization/83581
8228 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
8229 TODO_cleanup_cfg if any changes have been made.
8232 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
8233 convert_modes if target mode has the right side, but different mode
8237 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
8238 last argument when extracting from CONCAT. If either from_real or
8239 from_imag is NULL, use expansion through memory. If result is not
8240 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
8241 the parts directly to inner mode, if even that fails, use expansion
8245 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
8246 check for bswap in mode rather than HImode and use that in expand_unop
8249 Copyright (C) 2018 Free Software Foundation, Inc.
8251 Copying and distribution of this file, with or without modification,
8252 are permitted in any medium without royalty provided the copyright
8253 notice and this notice are preserved.