1 2018-03-08 David Malcolm <dmalcolm@redhat.com>
3 PR tree-optimization/84178
4 * tree-if-conv.c (release_bb_predicate): Remove the
5 the assertion that the stmts have NULL use_ops.
6 Discard the statements, asserting that they haven't
7 yet been added to a BB.
9 2018-03-08 Richard Biener <rguenther@suse.de>
11 PR tree-optimization/84746
12 * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
13 (phi_translate): Pass in destination ANTIC_OUT set.
14 (phi_translate_1): Likewise. For a simplified result lookup
15 a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
16 (phi_translate_set): Adjust.
17 (do_pre_regular_insertion): Likewise.
18 (do_pre_partial_partial_insertion): Likewise.
20 2018-03-08 Martin Liska <mliska@suse.cz>
23 * doc/gcov.texi: Document usage of profile files.
24 * gcov-io.h: Document changes in the format.
26 2018-03-08 Alexandre Oliva <aoliva@redhat.com>
30 * dwarf2out.c (struct dw_line_info_table): Update comments for
32 (FORCE_RESET_NEXT_VIEW): New.
33 (FORCE_RESETTING_VIEW_P): New.
34 (RESETTING_VIEW_P): Check for -1 too.
35 (ZERO_VIEW_P): Likewise.
36 (new_line_info_table): Force-reset next view.
37 (dwarf2out_begin_function): Likewise.
38 (dwarf2out_source_line): Simplify zero_view_p initialization.
39 Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
40 view directly. Omit view when omitting .loc at line 0.
42 2018-03-08 Jakub Jelinek <jakub@redhat.com>
44 PR tree-optimization/84740
45 * tree-switch-conversion.c (process_switch): Call build_constructors
46 only if info.phi_count is non-zero.
48 PR tree-optimization/84739
49 * tree-tailcall.c (find_tail_calls): Check call arguments against
50 DECL_ARGUMENTS (current_function_decl) rather than
51 DECL_ARGUMENTS (func) when checking for tail recursion.
53 2018-03-07 Jakub Jelinek <jakub@redhat.com>
55 * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
56 Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
57 Volker Reichelt's entry and add entries for people that perform
58 GCC fuzzy testing and report numerous bugs.
60 2018-03-07 Segher Boessenkool <segher@kernel.crashing.org>
63 * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
64 readonly data in sdata, if that is disabled.
65 * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
66 * doc/invoke.texi (RS/6000 and PowerPC Options): Document
67 -mreadonly-in-sdata option.
69 2018-03-07 Martin Sebor <msebor@redhat.com>
71 PR tree-optimization/84468
72 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
73 basic block when looking for nul assignment.
75 2018-03-07 Eric Botcazou <ebotcazou@adacore.com>
78 * except.h (output_function_exception_table): Adjust prototype.
79 * except.c (output_function_exception_table): Remove FNNAME parameter
80 and add SECTION parameter. Ouput one part of the table at a time.
81 * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
82 the first part of the exception table and emit unwind directives.
83 * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
84 (i386_pe_seh_cold_init): Likewise.
85 * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
86 (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
87 * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
88 (ix86_output_call_insn): Emit a nop in one more case for SEH.
89 * config/i386/winnt.c: Include except.h.
90 (struct seh_frame_state): Add reg_offset, after_prologue and
91 in_cold_section fields.
92 (i386_pe_seh_end_prologue): Set seh->after_prologue.
93 (i386_pe_seh_cold_init): New function.
94 (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
95 to seh->in_cold_section.
96 (seh_emit_push): Record the offset of the push.
97 (seh_emit_save): Record the offet of the save.
98 (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
99 Test seh->after_prologue to disregard the epilogue.
100 (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
101 (i386_pe_end_cold_function): New function.
103 2018-03-07 Jakub Jelinek <jakub@redhat.com>
106 * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
107 aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
110 * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
112 * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
113 don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
116 * multiple_target.c: Include tree-inline.h and intl.h.
117 (expand_target_clones): Diagnose and fail if node->definition and
118 !tree_versionable_function_p (node->decl).
120 2018-03-06 John David Anglin <danglin@gcc.gnu.org>
122 * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
124 (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
125 (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
126 * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
128 2018-03-06 Jakub Jelinek <jakub@redhat.com>
131 * combine.c (try_combine): Use reg_or_subregno instead of handling
132 just paradoxical SUBREGs and REGs.
134 2018-03-06 Claudiu Zissulescu <claziss@synopsys.com>
136 * config/arc/arc.c (arc_finalize_pic): Remove function.
137 (arc_must_save_register): We use single base PIC register, remove
138 checks to save/restore the PIC register.
139 (arc_expand_prologue): Likewise.
140 * config/arc/arc-protos.h (arc_set_default_type_attributes):
142 (arc_verify_short): Likewise.
143 (arc_attr_type): Likewise.
144 * config/arc/arc.c (arc_set_default_type_attributes): Remove.
145 (walk_stores): Likewise.
146 (arc_address_cost): Make it static.
147 (arc_verify_short): Likewise.
148 (branch_dest): Likewise.
149 (arc_attr_type): Likewise.
150 * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
151 (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
152 (arc_final_prescan_insn): Remove inserting the nops due to
153 hardware hazards. It is done in reorg step.
154 (insn_length_variant_t): Remove.
155 (insn_length_parameters_t): Likewise.
156 (arc_insn_length_parameters): Likewise.
157 (arc_get_insn_variants): Likewise.
158 * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
160 2018-03-06 Jakub Jelinek <jakub@redhat.com>
163 * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
166 PR tree-optimization/84687
167 * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
169 * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
171 2018-03-05 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
173 * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
174 Rename to ppc_speculation_barrier.
175 * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
176 __builtin_ppc_speculation_barrier.
178 2018-03-05 Jakub Jelinek <jakub@redhat.com>
181 * combine.c (combine_simplify_rtx): Don't try to simplify if
182 if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
185 2018-03-05 Segher Boessenkool <segher@kernel.crashing.org>
187 * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
188 to 32 bytes when compiling for POWER9.
190 2018-03-05 Jakub Jelinek <jakub@redhat.com>
193 * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
194 regparm >= 3 with no arg reg available also for calls with
195 flag_force_indirect_call. Pass decl to ix86_function_regparm.
198 * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
200 (*<plusminus_insn><mode>3): Likewise. Remove <mask_operand3> uses.
202 2018-03-05 Peter Bergner <bergner@vnet.ibm.com>
205 * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
207 2018-03-05 Richard Biener <rguenther@suse.de>
209 PR tree-optimization/84486
210 * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
211 When inserting a __builtin_assume_aligned call set the LHS
212 SSA name alignment info accordingly.
214 2018-03-05 Wilco Dijkstra <wdijkstr@arm.com>
216 PR tree-optimization/84114
217 * config/aarch64/aarch64.c (aarch64_reassociation_width)
218 Avoid reassociation of FLOAT_MODE addition.
220 2018-03-05 Olga Makhotina <olga.makhotina@intel.com>
222 * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
223 OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
224 OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
225 (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
226 * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
227 * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
228 * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
230 * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
231 __builtin_ia32_wbinvd): New builtins.
232 (SPECIAL_ARGS2): New.
233 * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
234 (SPECIAL_ARGS2): New.
235 * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
236 (ix86_valid_target_attribute_inner_p): Ditto.
237 (ix86_init_mmx_sse_builtins): Add special_args2.
238 * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
239 TARGET_WBNOINVD_P): New.
240 * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
241 (define_insn "wbinvd", define_insn "wbnoinvd"): New.
242 * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
243 * config/i386/immintrin.h (_wbinvd): New intrinsic.
244 * config/i386/pconfigintrin.h: New file.
245 * config/i386/wbnoinvdintrin.h: Ditto.
246 * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
247 * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
249 2018-03-05 Richard Biener <rguenther@suse.de>
251 PR tree-optimization/84670
252 * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
254 (BB_VISITED_WITH_VISITED_SUCCS): New define.
255 (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
256 (compute_antic_aux): Only assert the number of values in ANTIC_IN
257 doesn't grow if all successors (recursively) were visited at least
260 2018-03-05 Richard Biener <rguenther@suse.de>
262 PR tree-optimization/84650
263 * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
264 if executed in the loop pipeline.
266 2018-03-05 Sandra Loosemore <sandra@codesourcery.com>
268 * doc/configfiles.texi (Configuration Files): Move info about
269 conditionalizing $target-protos.h to...
270 * doc/sourcebuild.texi (Back End): Here. Explain how $target.h
271 differs from $target-protos.h.
273 2018-03-05 Kito Cheng <kito.cheng@gmail.com>
274 Chung-Ju Wu <jasonwucj@gmail.com>
276 * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
277 * config/nds32/nds32-multiple.md (setmemsi): Define.
278 * config/nds32/nds32-memory-manipulation.c
279 (nds32_gen_dup_4_byte_to_word_value): New.
280 (emit_setmem_word_loop): New.
281 (emit_setmem_byte_loop): New.
282 (nds32_expand_setmem_loop): New.
283 (nds32_expand_setmem_loop_v3m): New.
284 (nds32_expand_setmem_unroll): New.
285 (nds32_expand_setmem): New.
287 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
288 Chung-Ju Wu <jasonwucj@gmail.com>
290 * config/nds32/nds32-memory-manipulation.c
291 (nds32_emit_load_store): New.
292 (nds32_emit_post_inc_load_store): New.
293 (nds32_emit_mem_move): New.
294 (nds32_emit_mem_move_block): New.
295 (nds32_expand_movmemsi_loop_unknown_size): New.
296 (nds32_expand_movmemsi_loop_known_size): New.
297 (nds32_expand_movmemsi_loop): New.
298 (nds32_expand_movmemsi_unroll): New.
299 (nds32_expand_movmemqi): Rename ...
300 (nds32_expand_movmemsi): ... to this.
301 * config/nds32/nds32-multiple.md (movmemqi): Rename ...
302 (movmemsi): ... to this.
303 * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
304 (nds32_expand_movmemsi): ... to this.
306 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
307 Monk Chiang <sh.chiang04@gmail.com>
308 Chung-Ju Wu <jasonwucj@gmail.com>
310 * config/nds32/nds32-protos.h
311 (nds32_expand_load_multiple): New arguments.
312 (nds32_expand_store_multiple): Ditto.
313 (nds32_valid_multiple_load_store): Rename ...
314 (nds32_valid_multiple_load_store_p): ... to this.
315 * config/nds32/nds32-memory-manipulation.c
316 (nds32_expand_load_multiple): Refine implementation.
317 (nds32_expand_store_multiple): Ditto.
318 * config/nds32/nds32-multiple.md
319 (load_multiple): Update nds32_expand_load_multiple interface.
320 (store_multiple): Update nds32_expand_store_multiple interface.
321 * config/nds32/nds32-predicates.c
322 (nds32_valid_multiple_load_store): Rename ...
323 (nds32_valid_multiple_load_store_p): ... to this and refine
325 * config/nds32/predicates.md
326 (nds32_load_multiple_and_update_address_operation): New predicate.
327 (nds32_store_multiple_and_update_address_operation): New predicate.
329 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
330 Chung-Ju Wu <jasonwucj@gmail.com>
332 * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
333 (combo): New attribute.
334 * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
336 2018-03-03 Chung-Ju Wu <jasonwucj@gmail.com>
338 * config/nds32/nds32.opt: Change -mcmodel= default value.
340 2018-03-03 Kito Cheng <kito.cheng@gmail.com>
341 Monk Chiang <sh.chiang04@gmail.com>
342 Chung-Ju Wu <jasonwucj@gmail.com>
344 * config/nds32/constants.md (unspec_element): New enum.
345 * config/nds32/constraints.md (Umw): New constraint.
346 * config/nds32/nds32-intrinsic.c: Add more builtin functions.
347 * config/nds32/nds32-intrinsic.md: Likewise.
348 * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
349 (nds32_valid_smw_lwm_base_p): New.
350 (nds32_output_smw_single_word): New.
351 (nds32_output_lmw_single_word): New.
352 (nds32_expand_unaligned_load): New.
353 (nds32_expand_unaligned_store): New.
354 * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
355 (nds32_output_smw_single_word): Declare.
356 (nds32_output_lmw_single_word): Declare.
357 (nds32_expand_unaligned_load): Declare.
358 (nds32_expand_unaligned_store): Declare.
359 * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
360 NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
361 NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
362 NDS32_BUILTIN_UASTORE_DW.
363 * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
366 2018-03-03 Monk Chiang <sh.chiang04@gmail.com>
367 Kito Cheng <kito.cheng@gmail.com>
368 Chung-Ju Wu <jasonwucj@gmail.com>
370 * config/nds32/nds32-intrinsic.c
371 (nds32_expand_builtin_null_ftype_reg): Delete.
372 (nds32_expand_builtin_reg_ftype_imm): Ditto.
373 (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
374 (nds32_read_argument): New.
375 (nds32_legitimize_target): Ditto.
376 (nds32_legitimize_argument): Ditto.
377 (nds32_check_constant_argument): Ditto.
378 (nds32_expand_unop_builtin): Ditto.
379 (nds32_expand_unopimm_builtin): Ditto.
380 (nds32_expand_binop_builtin): Ditto.
381 (nds32_builtin_decl_impl): Ditto.
382 (builtin_description): Ditto.
383 (nds32_expand_builtin_impl): Rewrite with new infrastructure.
384 (nds32_init_builtins_impl): Ditto.
385 * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
386 (nds32_builtin_decl): New.
387 * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
388 * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
390 2018-03-02 Jeff Law <law@redhat.com>
392 * reorg.c (stop_search_p): Handle DEBUG_INSN.
393 (redundant_insn, fill_simple_delay_slots): Likewise.
394 (fill_slots_from_thread): Likewise.
395 * resource.c (mark_referenced_resources): Likewise.
396 (mark_set_resources, find_dead_or_set_registers): Likewise.
398 2018-03-02 Jakub Jelinek <jakub@redhat.com>
400 * substring-locations.h (format_warning_va): Formatting fix for
402 (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
404 (format_warning_n_va, format_warning_at_substring_n): New prototypes.
405 * substring-locations.c: Include intl.h.
406 (format_warning_va): Turned into small wrapper around
407 format_warning_n_va, renamed to ...
408 (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
409 rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
411 (format_warning_at_substring_n): New function.
412 * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
413 (fmtwarn): Add ATTRIBUTE_GCC_DIAG. Turn into a copy of
414 format_warning_at_substring with just a shorter name instead of
415 const function pointer.
416 (fmtwarn_n): New function.
417 (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
418 appropriate, get rid of all the fmtstr temporaries, move conditionals
419 with G_() wrapped string literals directly into fmtwarn arguments,
420 cast dir.len to (int), formatting fixes.
422 2018-03-02 Thomas Schwinge <thomas@codesourcery.com>
424 * doc/invoke.texi: Remove "Cilk Plus" references.
426 2018-03-02 Jakub Jelinek <jakub@redhat.com>
427 Richard Biener <rguenther@suse.de>
430 * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
431 for error or warning attributes if CALL_FROM_THUNK_P is set.
434 2018-03-02 Jakub Jelinek <jakub@redhat.com>
437 * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
438 __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
441 * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
442 __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
444 * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
445 instead of -1U in last predictors element's probability member.
447 2018-03-02 Eric Botcazou <ebotcazou@adacore.com>
450 * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
451 arguments if they are comparable.
453 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
455 PR tree-optimization/84634
456 * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
457 masks and masked_loop_p with a single loop_masks, making sure it's
458 null for bb vectorization.
460 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
462 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
463 (vect_analyze_data_ref_access): Use loop->safe_len rather than
464 loop->force_vectorize to check whether there is no alias.
466 2018-03-02 Jakub Jelinek <jakub@redhat.com>
469 * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
471 * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
473 (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
474 * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
475 instead of a loop around prev_real_insn.
476 * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
480 * config/i386/i386.c (ix86_print_operand): Use conditional
481 output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
484 2018-03-02 Richard Biener <rguenther@suse.de>
486 PR tree-optimization/84427
487 * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
488 (bitmap_set_subtract_values): Rewrite to handle multiple
491 (prune_clobbered_mems): Likewise.
492 (phi_translate): Take edge instead of pred/phiblock.
493 (phi_translate_1): Likewise.
494 (phi_translate_set): Likewise. Insert all translated
495 exprs for a value into the set, keeping possibly multiple
496 expressions per value.
497 (compute_antic_aux): Adjust for phi_translate changes.
498 When intersecting union the expressions and prune those
499 not in the final value set, keeping possibly multiple
500 expressions per value. Do not use value-insertion
501 for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
502 all expressions. Add verification that the value-sets
503 only shrink during iteration.
504 (compute_partial_antic_aux): Adjust for the phi_translate changes.
505 (do_pre_regular_insertion): Likewise.
506 (do_pre_partial_partial_insertion): Likewise.
508 2018-03-02 Richard Biener <rguenther@suse.de>
511 * config/darwin.c (saved_debug_info_level): New static global.
512 (darwin_asm_lto_start): Disable debug info generation for LTO out.
513 (darwin_asm_lto_end): Restore debug info generation settings.
515 2018-03-01 Martin Liska <mliska@suse.cz>
518 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
521 2018-03-01 Richard Biener <rguenther@suse.de>
524 * dwarf2out.c (gen_variable_die): Properly handle late VLA
525 type annotation with LTO when debug was disabled at compile-time.
527 2018-03-01 Matthew Fortune <mfortune@gmail.com>
529 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
531 (mips_final_postscan_insn): Likewise.
533 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
535 PR rtl-optimization/84528
536 * alias.c (init_alias_target): Add commentary.
537 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
538 a unique base value if the frame pointer is not eliminated
539 to the stack pointer.
541 2018-03-01 Tom de Vries <tom@codesourcery.com>
543 PR rtl-optimization/83327
544 * lra-int.h (hard_regs_spilled_into): Declare.
545 * lra.c (hard_regs_spilled_into): Define.
546 (init_reg_info): Init hard_regs_spilled_into.
547 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
548 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
549 (process_bb_lives): Handle hard_regs_spilled_into.
550 (lra_create_live_ranges_1): Before doing liveness propagation, clear
551 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
553 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
555 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
556 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
557 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
558 * config/rs6000/aix72.h: New file.
560 2018-02-28 Jakub Jelinek <jakub@redhat.com>
562 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
563 instead of warning_at with conditional singular and plural messages
567 * stor-layout.c (update_alignment_for_field): For
568 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
569 && !DECL_PACKED (field), do the alignment update, just use
570 only desired_align instead of MAX (type_align, desired_align)
572 (place_field): Don't do known_align < desired_align handling
573 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
574 is non-NULL, instead do it after rli->prev_field handling and
575 only if not within a bitfield word. For DECL_PACKED (field)
576 use type_align of BITS_PER_UNIT.
578 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
580 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
581 superfluous parentheses and trailing spaces.
583 2018-02-28 Richard Biener <rguenther@suse.de>
585 PR tree-optimization/84584
586 * graphite-scop-detection.c (scop_detection::add_scop): Discard
587 SCoPs with fake exit edge.
589 2018-02-28 Martin Liska <mliska@suse.cz>
592 * timevar.c (timer::print): Fix format to properly print 100%
595 2018-02-28 Richard Biener <rguenther@suse.de>
598 * genmatch.c (capture_info::walk_match): Do not mark
599 captured expressions without operands as expr_p given
600 they act more like predicates and should be subject to
601 "lost tail" side-effect preserving.
603 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
605 PR rtl-optimization/81611
606 * auto-inc-dec.c (attempt_change): Move dead note from
607 mem_insn if it's the next use of regno
608 (find_address): Take address use of reg holding
609 non-incremented value. Add parm to limit search to the named
611 (merge_in_block): Attempt to use a mem insn that is the next
612 use of the original regno.
614 2018-02-27 Martin Sebor <msebor@redhat.com>
617 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
618 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
620 2018-02-27 Martin Sebor <msebor@redhat.com>
623 * diagnostic-core.h (warning_n, error_n, inform_n): Change
624 n argument to unsigned HOST_WIDE_INT.
625 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
626 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
627 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
628 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
630 2018-02-27 Richard Biener <rguenther@suse.de>
632 PR tree-optimization/84512
633 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
634 Do not use the estimate returned from record_stmt_cost for
635 the scalar iteration cost but sum properly using add_stmt_cost.
637 2018-02-27 Richard Biener <rguenther@suse.de>
639 PR tree-optimization/84466
640 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
641 Adjust last change to less strictly validate use operands.
643 2018-02-27 Martin Liska <mliska@suse.cz>
645 PR gcov-profile/84548
646 * gcov.c (process_file): Allow partial overlap and consider it
647 also as group functions.
648 (output_lines): Properly calculate range of lines for a group.
650 2018-02-27 Martin Liska <mliska@suse.cz>
652 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
653 'ggc' suffixes. Change first column width.
654 (timer::print): Fix formatting of the column.
656 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
658 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
659 preserve inline entry blocks for the sake of debug inline
660 entry point markers alone.
661 (remove_unused_locals): Suggest in comments a better place to
662 force the preservation of inline entry blocks that are
663 otherwise unused, but do not preserve them.
665 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
667 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
669 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
672 * config/i386/constraints.md (Bs): Replace
673 ix86_indirect_branch_register with
674 TARGET_INDIRECT_BRANCH_REGISTER.
676 * config/i386/i386.md (indirect_jump): Likewise.
677 (tablejump): Likewise.
678 (*sibcall_memory): Likewise.
679 (*sibcall_value_memory): Likewise.
680 Peepholes of indirect call and jump via memory: Likewise.
681 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
682 (*sibcall_value_GOT_32): Likewise.
683 * config/i386/predicates.md (indirect_branch_operand): Likewise.
684 (GOT_memory_operand): Likewise.
685 (call_insn_operand): Likewise.
686 (sibcall_insn_operand): Likewise.
687 (GOT32_symbol_operand): Likewise.
688 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
690 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
692 PR rtl-optimization/83496
693 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
694 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
695 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
696 redundant insn, if any.
697 (relax_delay_slots): Likewise.
698 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
700 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
702 PR tree-optimization/83965
703 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
704 that grouped statements are part of a reduction chain. Return
705 true if the statement is not marked as a reduction itself but
707 (vect_recog_dot_prod_pattern): Don't check whether the statement
708 is part of a group here.
709 (vect_recog_sad_pattern): Likewise.
710 (vect_recog_widen_sum_pattern): Likewise.
712 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
715 * final.c (rest_of_clean_state): Also look for calls inside sequences.
717 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
720 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
722 (ix86_output_indirect_function_return): New prototype.
723 (ix86_split_simple_return_pop_internal): Likewise.
724 * config/i386/i386.c (indirect_return_via_cx): New.
725 (indirect_return_via_cx_bnd): Likewise.
726 (indirect_thunk_name): Handle return va CX_REG.
727 (output_indirect_thunk_function): Create alias for
728 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
729 (ix86_output_indirect_jmp): Remove the bool argument.
730 (ix86_output_indirect_function_return): New function.
731 (ix86_split_simple_return_pop_internal): Likewise.
732 * config/i386/i386.md (*indirect_jump): Don't pass false
733 to ix86_output_indirect_jmp.
734 (*tablejump_1): Likewise.
735 (simple_return_pop_internal): Change it to define_insn_and_split.
736 Call ix86_split_simple_return_pop_internal to split it for
738 (simple_return_indirect_internal): Call
739 ix86_output_indirect_function_return instead of
740 ix86_output_indirect_jmp.
742 2018-02-26 Jakub Jelinek <jakub@redhat.com>
745 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
746 memset and value initialization afterwards.
748 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
750 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
752 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
755 * common/config/aarch64/aarch64-common.c
756 (aarch_option_optimization_table[]): Switch
757 off fomit-frame-pointer
759 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
760 Chung-Ju Wu <jasonwucj@gmail.com>
762 * config/nds32/nds32-multiple.md (load_multiple): Disallow
764 (store_multiple): Ditto.
766 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
768 * config.gcc: Add --with-cpu support for nds32 target.
769 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
770 * config/nds32/nds32.opt: Add -mcpu= option.
772 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
774 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
775 isel=yes): Warn for these deprecated options.
777 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
779 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
780 ISA_2_5_MASKS_EMBEDDED.
782 2018-02-23 Jakub Jelinek <jakub@redhat.com>
784 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
785 p->max as pointers rather than using iterative_hash_expr.
787 2018-02-23 Carl Love <cel@us.ibm.com>
789 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
790 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
792 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
793 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
794 P8V_BUILTIN_VEC_VUNSIGNED2.
796 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
799 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
800 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
802 * lra-constraints.c (curr_insn_transform): Set up
803 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
805 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
806 * lra-eliminations.c (spill_pseudos): Ditto.
807 (process_insn_for_elimination): Ditto.
808 * lra-lives.c (reg_early_clobber_p): Use the new macros.
809 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
810 LRA_NON_CLOBBERED_ALT.
812 2018-02-22 Martin Sebor <msebor@redhat.com>
814 PR tree-optimization/84480
815 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
816 to maybe_diag_stxncpy_trunc. Call it.
817 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
818 from gimple_fold_builtin_strcpy. Print inlining stack.
819 (handle_builtin_stxncpy): Print inlining stack.
820 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
822 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
825 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
826 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
827 and -fcheck-pointer-bounds are used together.
828 (indirect_thunk_prefix): New enum.
829 (indirect_thunk_need_prefix): New function.
830 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
831 "_nt" instead of "_bnd" for NOTRACK prefix.
832 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
833 (output_indirect_thunk_function): Likewise.
835 (ix86_code_end): Update output_indirect_thunk_function calls.
836 (ix86_output_indirect_branch_via_reg): Replace
837 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
838 (ix86_output_indirect_branch_via_push): Likewise.
839 (ix86_output_function_return): Likewise.
840 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
841 incompatible with -fcf-protection=branch and
842 -fcheck-pointer-bounds.
844 2018-02-22 Steve Ellcey <sellcey@cavium.com>
847 * config/aarch64/aarch64.c (aarch64_print_address_internal):
848 Change gcc_assert call to output_operand_lossage.
850 2018-02-22 Steve Ellcey <sellcey@cavium.com>
852 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
854 2018-02-22 DJ Delorie <dj@redhat.com>
855 Sebastian Perta <sebastian.perta@renesas.com>
856 Oleg Endo <olegendo@gcc.gnu.org>
858 * config/rx/rx.c (rx_rtx_costs): New function.
859 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
861 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
863 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
865 2018-02-22 Martin Liska <mliska@suse.cz>
868 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
869 Add "native" as a possible value.
871 2018-02-22 Martin Liska <mliska@suse.cz>
874 * config/i386/i386.c (ix86_option_override_internal):
875 Add "native" as a possible value for -march and -mtune.
877 2018-02-22 Jakub Jelinek <jakub@redhat.com>
880 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
881 to all type variants.
883 PR tree-optimization/84503
884 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
885 width as info->bitpos + info->bitsize - start.
886 (merged_store_group::merge_overlapping): Simplify width computation.
887 (check_no_overlap): New function.
888 (imm_store_chain_info::try_coalesce_bswap): Compute expected
889 start + width and last_order of the group, fail if check_no_overlap
891 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
892 to group if check_no_overlap fails.
894 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
896 * config/rs6000/altivec.md: Delete contraint arguments to
897 define_expand, define_split, and define_peephole2, and in
898 define_insn_and_split if always unused.
899 * config/rs6000/darwin.md: Ditto.
900 * config/rs6000/dfp.md: Ditto.
901 * config/rs6000/rs6000.md: Ditto.
902 * config/rs6000/sync.md: Ditto.
903 * config/rs6000/vector.md: Ditto.
904 * config/rs6000/vsx.md: Ditto.
906 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
908 * config/rs6000/altivec.md: Write output control strings as braced
909 blocks instead of double-quoted strings.
910 * config/rs6000/darwin.md: Ditto.
911 * config/rs6000/rs6000.md: Ditto.
912 * config/rs6000/vector.md: Ditto.
913 * config/rs6000/vsx.md: Ditto.
915 2018-02-21 Jason Merrill <jason@redhat.com>
917 PR c++/84314 - ICE with templates and fastcall attribute.
918 * attribs.c (build_type_attribute_qual_variant): Remove assert.
920 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
922 * ipa-cp.c (determine_versionability): Fix comment typos.
924 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
927 * ipa-cp.c (determine_versionability): Do not version functions caling
930 2018-02-21 Martin Liska <mliska@suse.cz>
933 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
934 Add "native" as a possible value.
935 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
936 the macro when native cpu detection is available.
938 2018-02-21 Martin Liska <mliska@suse.cz>
941 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
942 Add "native" as a possible value.
943 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
944 when native cpu detection is available.
946 2018-02-21 Jakub Jelinek <jakub@redhat.com>
947 Martin Sebor <msebor@redhat.com>
949 PR tree-optimization/84478
950 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
952 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
953 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
954 support which is conservatively correct, for 2 only stay conservative
955 for maxlen. Formatting and comment capitalization fixes. Add STRICT
956 argument to the 2 argument get_range_strlen, adjust 6 arg
957 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
959 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
960 (gimple_fold_builtin_strlen): Pass true as last argument to
963 2018-02-20 Martin Sebor <msebor@redhat.com>
966 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
967 (builtin_memref::set_base_and_offset): Same. Handle inner references.
968 (builtin_memref::builtin_memref): Factor out parts into
969 set_base_and_offset and call it.
971 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
974 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
975 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
976 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
977 search at the associated MODE_INT.
979 2018-02-20 Jeff Law <law@redhat.com>
982 PR tree-optimization/81592
984 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
985 for range data rather than using global data.
986 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
987 range data rather than using global data.
988 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
989 pass it to children as needed.
990 (struct directive::fmtresult): Similarly.
991 (struct directive::set_width): Similarly.
992 (struct directive::set_precision): Similarly.
993 (format_integer, format_directive, parse_directive): Similarly.
994 (format_none): Accept unnamed vr_values parameter.
995 (format_percent, format_floating, format_character): Similarly.
996 (format_string, format_plain): Similarly.
997 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
998 the EVRP range analyzer for range data rather than using global data.
999 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
1000 gimple-ssa-evrp-analyze.h
1001 (class sprintf_dom_walker): Add after_dom_children member function.
1002 Add evrp_range_analyzer member.
1003 (sprintf_dom_walker::before_dom_children): Call into the EVRP
1004 range analyzer as needed.
1005 (sprintf_dom_walker::after_dom_children): New member function.
1006 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
1008 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
1009 (evrp_range_analyzer::pop_to_marker): Likewise.
1011 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
1013 PR tree-optimization/84419
1014 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
1015 with the required type if its current type is compatible but
1018 2018-02-20 Jakub Jelinek <jakub@redhat.com>
1021 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
1022 after vectorization.
1024 2018-02-20 Martin Liska <mliska@suse.cz>
1027 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
1028 possible values if we don't have a hint.
1030 2018-02-20 Martin Liska <mliska@suse.cz>
1034 * final.c (shorten_branches): Build align_tab array with one
1036 * opts.c (finish_options): Add alignment option limit check.
1037 (MAX_CODE_ALIGN): Likewise.
1038 (MAX_CODE_ALIGN_VALUE): Likewise.
1039 * doc/invoke.texi: Document maximum allowed option value for
1040 all -falign-* options.
1042 2018-02-19 Jakub Jelinek <jakub@redhat.com>
1045 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
1046 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
1047 * var-tracking.c (emit_note_insn_var_location): Remove all references
1048 to NOTE_INSN_CALL_ARG_LOCATION.
1049 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
1050 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
1051 Use copy_rtx_if_shared.
1052 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
1053 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
1054 (dwarf2out_var_location): Remove handling of
1055 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
1057 * final.c (final_scan_insn): Remove all references to
1058 NOTE_INSN_CALL_ARG_LOCATION.
1059 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
1060 before dumping final insns.
1061 * except.c (emit_note_eh_region_end): Remove all references to
1062 NOTE_INSN_CALL_ARG_LOCATION.
1063 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
1064 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
1065 * config/arc/arc.c (hwloop_optimize): Likewise.
1066 * config/arm/arm.c (create_fix_barrier): Likewise.
1067 * config/s390/s390.c (s390_chunkify_start): Likewise.
1068 * config/sh/sh.c (find_barrier): Likewise.
1069 * config/i386/i386.c (rest_of_insert_endbranch,
1070 ix86_seh_fixup_eh_fallthru): Likewise.
1071 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
1072 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
1073 * config/frv/frv.c (frv_function_prologue): Likewise.
1074 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
1076 (note_outside_basic_block_p): Remove all references to
1077 NOTE_INSN_CALL_ARG_LOCATION.
1078 * gengtype.c (adjust_field_rtx_def): Likewise.
1079 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
1081 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
1082 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
1085 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
1088 PR tree-optimization/84452
1089 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
1090 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
1093 2018-02-19 Martin Liska <mliska@suse.cz>
1096 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
1098 2018-02-19 Martin Liska <mliska@suse.cz>
1099 Richard Sandiford <richard.sandiford@linaro.org>
1101 PR tree-optimization/82491
1102 * gimple-fold.c (get_base_constructor): Make earlier bail out
1105 2018-02-19 Carl Love <cel@us.ibm.com>
1107 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
1108 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
1110 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
1111 P8V_BUILTIN_VEC_NEG.
1113 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
1115 * config/rl78/rl78.md (movdf): New define expand.
1117 2018-02-19 Martin Liska <mliska@suse.cz>
1120 * doc/invoke.texi: Fix typo.
1121 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
1123 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
1125 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
1126 handle rs6000_single_float and rs6000_double_float specially for
1129 2018-02-16 Jeff Law <law@redhat.com>
1131 * config/rx/rx.c (add_pop_cfi_notes): New function.;
1134 2018-02-16 Jakub Jelinek <jakub@redhat.com>
1137 * ipa-inline.c (inline_small_functions): Fix a typo.
1139 2018-02-16 Nathan Sidwell <nathan@acm.org>
1141 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
1143 2018-02-16 Carl Love <cel@us.ibm.com>
1145 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
1146 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
1147 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
1148 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
1149 expansion to P8V_BUILTIN_VEC_FLOAT2.
1151 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
1153 PR rtl-optimization/70023
1154 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
1155 src_regno into account.
1157 2018-02-16 Carl Love <cel@us.ibm.com>
1159 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
1160 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
1161 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
1162 * config/rs6000/rs6000.c: Remove case statements for
1163 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
1164 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
1165 and P9V_BUILTIN_VEC_VINSERT4B.
1166 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
1167 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
1168 * config/rs6000/vsx.md:
1169 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
1172 2018-02-16 Carl Love <cel@us.ibm.com>
1174 * config/rs6000/altivec.h: Add builtin names vec_extract4b
1176 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
1178 * config/rs6000/rs6000-c.c: Add the definitions for
1179 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
1180 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
1181 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
1182 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
1183 definition for insert4b and define insn *insert3b_internal.
1184 * doc/extend.texi: Add documentation for vec_extract4b.
1186 2018-02-16 Nathan Sidwell <nathan@acm.org>
1188 * doc/extend.texi (Backwards Compatibility): Mention friend
1189 injection. Note for-scope is deprecated.
1190 * doc/invoke.texi (-ffriend-injection): Deprecate.
1192 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
1194 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
1195 that moved to I2, also allow destinations that are a paradoxical
1196 subreg (instead of a normal reg).
1198 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
1201 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
1204 2018-02-16 Richard Biener <rguenther@suse.de>
1206 PR tree-optimization/84037
1207 PR tree-optimization/84016
1209 * config/i386/i386.c (ix86_builtin_vectorization_cost):
1210 Adjust vec_construct for the fact we need additional higher latency
1211 128bit inserts for AVX256 and AVX512 vector builds.
1212 (ix86_add_stmt_cost): Scale vector construction cost for
1215 2018-02-16 Richard Biener <rguenther@suse.de>
1217 PR tree-optimization/84417
1218 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
1219 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
1220 (non_rewritable_lvalue_p): Likewise, use poly-ints.
1222 2018-02-16 Martin Liska <mliska@suse.cz>
1225 * internal-fn.def (ASAN_CHECK): Set proper flags.
1226 (ASAN_MARK): Likewise.
1228 2018-02-16 Julia Koval <julia.koval@intel.com>
1230 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
1231 from PTA_CANNONLAKE.
1233 2018-02-16 Jakub Jelinek <jakub@redhat.com>
1236 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
1237 Use ++iter rather than iter++ for std::list iterators.
1238 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
1239 defer deleting them until all nodes in the forest are processed. Do
1240 free even leaf nodes. Change to_process into auto_vec.
1243 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
1244 * vec.h (vec_default_construct): Use memset instead of placement new
1245 if BROKEN_VALUE_INITIALIZATION is defined.
1246 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
1247 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
1250 PR rtl-optimization/83723
1251 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
1252 * lra.c (lra_substitute_pseudo): Likewise. If true, use
1253 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
1255 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
1257 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
1259 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
1261 PR rtl-optimization/81443
1262 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
1263 from inner REGs to paradoxical SUBREGs.
1265 2018-02-16 Richard Biener <rguenther@suse.de>
1267 PR tree-optimization/84399
1268 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
1269 For operands we can analyze at their definition make sure we can
1270 analyze them at each use as well.
1272 2018-02-16 Richard Biener <rguenther@suse.de>
1274 PR tree-optimization/84190
1275 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
1276 volatile accesses if the decl isn't volatile.
1278 2018-02-15 Jason Merrill <jason@redhat.com>
1280 PR c++/84314 - ICE with templates and fastcall attribute.
1281 * attribs.c (build_type_attribute_qual_variant): Don't clobber
1282 TYPE_CANONICAL on an existing type.
1284 2018-02-15 Jakub Jelinek <jakub@redhat.com>
1286 PR tree-optimization/84383
1287 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
1288 dstoff nor call operand_equal_p if dstbase is NULL.
1290 PR tree-optimization/84334
1291 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
1292 also a CONSTANT_CLASS_P, punt.
1294 2018-02-14 Jim Wilson <jimw@sifive.com>
1296 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
1297 first SMALL_OPERAND check. New local min_second_step. Move assert
1298 to where locals are set. Add TARGET_RVC support.
1299 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
1301 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
1303 * doc/invoke.texi: Correct -Wformat-overflow code sample.
1305 2018-02-14 Martin Sebor <msebor@redhat.com>
1307 PR tree-optimization/83698
1308 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
1309 arrays constrain the offset range to their bounds.
1310 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
1311 (builtin_access::overlap): Avoid setting the size of overlap if it's
1313 (maybe_diag_overlap): Also consider arrays when deciding what values
1314 of offsets to include in diagnostics.
1316 2018-02-14 Martin Sebor <msebor@redhat.com>
1319 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
1320 that correspond to the kind of a declaration.
1322 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
1325 * config/pa/pa.md: Load address of PIC label using the linkage table
1326 if the label is nonlocal.
1328 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
1330 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
1331 warning message if user requests -maltivec=be.
1332 * doc/invoke.texi: Document deprecation of -maltivec=be.
1334 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
1337 * config/rs6000/rs6000-c.c: Update definitions for
1338 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
1339 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
1341 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1344 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
1345 add _get_ssp intrinsics. Remove argument from
1346 __builtin_ia32_rdssp[d|q].
1347 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
1348 * config/i386/i386-builtin.def: Remove argument from
1349 __builtin_ia32_rdssp[d|q].
1350 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
1351 ix86_expand_special_args_builtin for _rdssp[d|q].
1352 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
1353 Clear register before usage.
1354 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
1355 Add documentation for new _get_ssp and _inc_ssp intrinsics.
1357 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
1359 PR tree-optimization/84357
1360 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
1361 operand 1 of an ARRAY_REF too.
1363 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
1366 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
1367 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
1369 (set_of_reg): New struct.
1370 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
1371 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
1372 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
1374 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
1375 Split into bitclr, bitset, bitinvert patterns if appropriate.
1376 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
1377 use rx_fuse_in_memory_bitop.
1378 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
1379 to named insn, correct maximum insn length.
1381 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
1384 * machmode.def: Define a complex mode for PARTIAL_INT.
1385 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
1387 * doc/rtl.texi: Document CSPImode.
1388 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
1390 (msp430_hard_regno_nregs_with_padding): Likewise.
1392 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
1395 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
1397 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
1399 PR rtl-optimization/84169
1400 * combine.c (try_combine): New variable split_i2i3. Set it to true if
1401 we generated a parallel as new i3 and we split that to new i2 and i3
1402 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
1403 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
1404 those to i2, not i1. Partially rewrite this scan code.
1406 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1409 * stor-layout.c (place_field): For variable length fields, adjust
1410 offset_align afterwards not just based on the field's alignment,
1411 but also on the size.
1414 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
1415 of exps and logs in the use_exp2 case.
1417 2018-02-13 Jeff Law <law@redhat.com>
1419 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
1422 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
1425 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
1429 * common.opt (gas-loc-support, gas-locview-support): New.
1430 (ginline-points, ginternal-reset-location-views): New.
1431 * doc/invoke.texi: Document them. Use @itemx where intended.
1432 (gvariable-location-views): Adjust.
1433 * target.def (reset_location_view): New.
1434 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
1435 (TARGET_RESET_LOCATION_VIEW): New.
1436 * doc/tm.texi: Rebuilt.
1437 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
1438 (dwarf2out_default_as_locview_support): New.
1439 (output_asm_line_debug_info): Use option variables.
1440 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
1441 (output_loc_list): Likewise.
1442 (add_high_low_attributes): Check option variables.
1443 Don't output entry view attribute in strict mode.
1444 (gen_inlined_subroutine_die): Check option variables.
1445 (dwarf2out_inline_entry): Likewise.
1446 (init_sections_and_labels): Likewise.
1447 (dwarf2out_early_finish): Likewise.
1448 (maybe_reset_location_view): New, from...
1449 (dwarf2out_var_location): ... here. Call it.
1450 * debug.h (dwarf2out_default_as_loc_support): Declare.
1451 (dwarf2out_default_as_locview_support): Declare.
1452 * hooks.c (hook_int_rtx_insn_0): New.
1453 * hooks.h (hook_int_rtx_insn_0): Declare.
1454 * toplev.c (process_options): Take -gas-loc-support and
1455 -gas-locview-support from dwarf2out. Enable
1456 -gvariable-location-views by default only with locview
1457 assembler support. Enable -ginternal-reset-location-views by
1458 default only if the target defines the corresponding hook.
1459 Enable -ginline-points by default if location views are
1460 enabled; force it disabled if statement frontiers are
1462 * tree-inline.c (expand_call_inline): Check option variables.
1463 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
1465 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
1467 PR tree-optimization/84321
1468 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
1469 handling. Also check whether the anti-range contains any values
1470 that satisfy the mask; switch to a VR_RANGE if not.
1472 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
1475 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
1477 2018-02-13 Martin Jambor <mjambor@suse.cz>
1480 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
1481 of call statements, also set location of a load to a temporary.
1483 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
1485 * config/rl78/rl78.c (add_vector_labels): New function.
1486 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
1487 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
1488 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
1489 which checks that no arguments are passed.
1490 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
1491 * doc/extend.texi: Documentation for the new attribute.
1493 2018-02-13 Andreas Schwab <schwab@suse.de>
1495 * config/riscv/linux.h (CPP_SPEC): Define.
1497 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1500 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
1501 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
1502 OPTION_MASK_ISA_AES as first argument to def_builtin_const
1503 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
1504 instead of OPTION_MASK_ISA_PCLMUL as first argument to
1505 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
1506 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
1507 temporarily for AES and PCLMUL builtins.
1509 PR tree-optimization/84339
1510 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
1511 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1515 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1516 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1517 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1519 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1521 * omp-simd-clone.h: New file.
1522 * omp-simd-clone.c: Include omp-simd-clone.h.
1523 (expand_simd_clones): No longer static.
1524 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1525 cgraph.h and omp-simd-clone.h.
1526 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1527 (vect_recog_widen_shift_pattern): Formatting fix.
1528 (vect_pattern_recog_1): Don't check optab for calls.
1531 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1532 operands[2] into a REG before using gen_lowpart on it.
1534 2018-02-12 Jeff Law <law@redhat.com>
1537 * config/sh/sh.c (find_barrier): Consider a sibling call
1540 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1541 successfully back substituting a reg.
1543 2018-02-12 Richard Biener <rguenther@suse.de>
1545 PR tree-optimization/84037
1546 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1547 parameter, move visited init to caller.
1548 (vect_slp_analyze_operations): Separate cost from validity
1549 check, initialize visited once for all instances.
1550 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1552 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1554 (vect_model_promotion_demotion_cost): Likewise.
1555 (vectorizable_bswap): Guard cost modeling with !slp_node
1556 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1558 (vectorizable_call): Likewise.
1559 (vectorizable_conversion): Likewise.
1560 (vectorizable_assignment): Likewise.
1561 (vectorizable_shift): Likewise.
1562 (vectorizable_operation): Likewise.
1563 (vectorizable_store): Likewise.
1564 (vectorizable_load): Likewise.
1565 (vectorizable_condition): Likewise.
1566 (vectorizable_comparison): Likewise.
1568 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1571 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1572 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1573 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1575 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1578 * params.def (inline-min-speedup): Increase from 8 to 15.
1579 (max-inline-insns-auto): Decrease from 40 to 30.
1580 * ipa-split.c (consider_split): Add some buffer for function to
1581 be considered inlining candidate.
1582 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1585 2018-02-12 Richard Biener <rguenther@suse.de>
1587 PR tree-optimization/84037
1588 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1589 matched stmts if we cannot swap the non-matched ones.
1591 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1593 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1594 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1595 _mm_maskz_scalef_round_ss): New intrinsics.
1596 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1597 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1598 __builtin_ia32_scalefss_round): Remove.
1599 (__builtin_ia32_scalefsd_mask_round,
1600 __builtin_ia32_scalefss_mask_round): New intrinsics.
1601 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1602 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1603 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1604 "<round_constraint>")): Changed to ...
1605 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1606 "<round_scalar_constraint>")): ... this.
1607 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1608 %0, %1, %2<round_op3>}"): Changed to ...
1609 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1610 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1611 %2<round_scalar_mask_op3>}"): ... this.
1612 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1614 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1616 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1617 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1618 (_mm_maskz_sqrt_round_ss): New intrinsics.
1619 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1620 (__builtin_ia32_sqrtsd_mask_round)
1621 (__builtin_ia32_sqrtss_mask_round): New builtins.
1622 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1623 (__builtin_ia32_sqrtss_round): Remove.
1624 (__builtin_ia32_sqrtsd_mask_round)
1625 (__builtin_ia32_sqrtss_mask_round): New builtins.
1626 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1627 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1628 ((match_operand:VF_128 1 "vector_operand"
1629 "xBm,<round_constraint>")): Changed to ...
1630 ((match_operand:VF_128 1 "vector_operand"
1631 "xBm,<round_scalar_constraint>")): ... this.
1632 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1633 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1634 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1635 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1636 %<iptr>1<round_scalar_mask_op3>}): ... this.
1637 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1638 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1640 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1643 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1644 Cast vec_cmpeq result to correct type.
1645 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1646 Cast vec_cmpgt result to correct type.
1648 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1650 * final.c (final_scan_insn_1): Renamed from...
1651 (final_scan_insn): ... this. New wrapper, to recover
1652 seen from the outermost call in recursive ones.
1653 * config/sparc/sparc.c (output_return): Drop seen from call.
1654 (output_sibcall): Likewise.
1655 * config/visium/visium.c (output_branch): Likewise.
1657 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1659 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1662 2018-02-10 Alan Modra <amodra@gmail.com>
1665 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1666 Specify LR as an input.
1668 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1671 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1672 remove_member_access_dummy_vars): New functions.
1673 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1674 lower_omp_1, execute_lower_omp): Use them.
1676 PR rtl-optimization/84308
1677 * shrink-wrap.c (spread_components): Release todo vector.
1679 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1681 PR rtl-optimization/57193
1682 * ira-color.c (struct allocno_color_data): Add member
1683 conflict_allocno_hard_prefs.
1684 (update_conflict_allocno_hard_prefs): New.
1685 (bucket_allocno_compare_func): Add a preference based on
1686 conflict_allocno_hard_prefs.
1687 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1688 (color_allocnos): Remove a dead code. Initiate
1689 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1691 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1694 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1695 constraint from =wa to wa. Avoid a subreg on the output operand,
1696 instead use a pseudo and subreg it in a move.
1697 (p9_xxbrd_<mode>): Changed to ...
1698 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1699 (p9_xxbrd_v2df): New expander.
1700 (p9_xxbrw_<mode>): Changed to ...
1701 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1702 (p9_xxbrw_v4sf): New expander.
1704 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1706 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1708 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1711 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1712 multiply in 32-bit mode.
1713 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1714 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1717 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1719 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1720 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1721 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1722 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1724 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1727 * dwarf2out.c (is_trivial_indirect_ref): New function.
1728 (dwarf2out_late_global_decl): Do not generate a location
1729 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1730 and that are not defined in the current unit.
1732 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1734 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1735 instead of a libcall for UNORDERED.
1737 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1740 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1741 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1743 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1746 * config/s390/s390.c (s390_set_current_function): Invoke
1747 s390_indirect_branch_settings also if fndecl didn't change.
1749 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1751 * config/rs6000/rs6000.md (blockage): Set length to zero.
1753 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1755 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1757 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1760 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1761 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1765 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1766 PARALLEL incoming that failed vt_get_decl_and_offset check.
1769 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1770 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1772 (get_variable_section): For decls in named .bss* sections pass true as
1773 second argument to bss_initializer_p.
1775 2018-02-09 Marek Polacek <polacek@redhat.com>
1776 Jakub Jelinek <jakub@redhat.com>
1779 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1780 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1781 Sync some changes from cxx_fold_indirect_ref.
1783 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1785 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1787 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1788 (BLOCK_INLINE_ENTRY_LABEL): New.
1789 (dwarf2out_var_location): Disregard inline entry markers.
1790 (inline_entry_data): New struct.
1791 (inline_entry_data_hasher): New hashtable type.
1792 (inline_entry_data_hasher::hash): New.
1793 (inline_entry_data_hasher::equal): New.
1794 (inline_entry_data_table): New variable.
1795 (add_high_low_attributes): Add DW_AT_entry_pc and
1796 DW_AT_GNU_entry_view attributes if a pending entry is found
1797 in inline_entry_data_table. Add old entry_pc attribute only
1798 if debug nonbinding markers are disabled.
1799 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1800 markers are enabled.
1801 (block_within_block_p, dwarf2out_inline_entry): New.
1802 (dwarf2out_finish): Check that no entries remained in
1803 inline_entry_data_table.
1804 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1805 (final_scan_insn, notice_source_line): Likewise.
1806 (rest_of_clean_state): Skip inline entry markers.
1807 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1809 * gimple.c (gimple_build_debug_inline_entry): New.
1810 * gimple.h (enum gimple_debug_subcode): Add
1811 GIMPLE_DEBUG_INLINE_ENTRY.
1812 (gimple_build_debug_inline_entry): Declare.
1813 (gimple_debug_inline_entry_p): New.
1814 (gimple_debug_nonbind_marker_p): Adjust.
1815 * insn-notes.def (INLINE_ENTRY): New.
1816 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1817 inline entry marker notes.
1818 (print_insn): Likewise.
1819 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1820 (INSN_DEBUG_MARKER_KIND): Likewise.
1821 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1822 * tree-inline.c (expand_call_inline): Build and insert
1823 debug_inline_entry stmt.
1824 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1825 inline entry blocks early, if nonbind markers are enabled.
1826 (dump_scope_block): Dump fragment info.
1827 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1828 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1829 (gimple_build_debug_inline_entry): New.
1830 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1831 Enable/disable inline entry points too.
1832 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1833 (DEBUG_INSN): Describe inline entry markers.
1835 * common.opt (gvariable-location-views): New.
1836 (gvariable-location-views=incompat5): New.
1837 * config.in: Rebuilt.
1838 * configure: Rebuilt.
1839 * configure.ac: Test assembler for view support.
1840 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1841 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1842 * dwarf2out.c (var_loc_view): New typedef.
1843 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1844 (dwarf2out_locviews_in_attribute): New.
1845 (dwarf2out_locviews_in_loclist): New.
1846 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1847 (enum dw_line_info_opcode): Add LI_adv_address.
1848 (struct dw_line_info_table): Add view.
1849 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1850 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1851 (zero_view_p): New variable.
1852 (ZERO_VIEW_P): New macro.
1853 (output_asm_line_debug_info): New.
1854 (struct var_loc_node): Add view.
1855 (add_AT_view_list, AT_loc_list): New.
1856 (add_var_loc_to_decl): Add view param. Test it against last.
1857 (new_loc_list): Add view params. Record them.
1858 (AT_loc_list_ptr): Handle loc and view lists.
1859 (view_list_to_loc_list_val_node): New.
1860 (print_dw_val): Handle dw_val_class_view_list.
1861 (size_of_die): Likewise.
1862 (value_format): Likewise.
1863 (loc_list_has_views): New.
1864 (gen_llsym): Set vl_symbol too.
1865 (maybe_gen_llsym, skip_loc_list_entry): New.
1866 (dwarf2out_maybe_output_loclist_view_pair): New.
1867 (output_loc_list): Output view list or entries too.
1868 (output_view_list_offset): New.
1869 (output_die): Handle dw_val_class_view_list.
1870 (output_dwarf_version): New.
1871 (output_compilation_unit_header): Use it.
1872 (output_skeleton_debug_sections): Likewise.
1873 (output_rnglists, output_line_info): Likewise.
1874 (output_pubnames, output_aranges): Update version comments.
1875 (output_one_line_info_table): Output view numbers in asm comments.
1876 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1877 Call maybe_gen_llsym.
1878 (loc_list_from_tree_1): Adjust.
1879 (add_AT_location_description): Create view list attribute if
1880 needed, check it's absent otherwise.
1881 (convert_cfa_to_fb_loc_list): Adjust.
1882 (maybe_emit_file): Call output_asm_line_debug_info for test.
1883 (dwarf2out_var_location): Reset views as needed. Precompute
1884 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1885 attribute. Set view.
1886 (new_line_info_table): Reset next view.
1887 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1888 (dwarf2out_source_line): Likewise. Output view resets and labels to
1889 the assembler, or select appropriate line info opcodes.
1890 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1891 (optimize_string_length): Catch it. Adjust.
1892 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1893 dw_val_class_view_list, and remove it if no longer needed.
1894 (hash_loc_list): Hash view numbers.
1895 (loc_list_hasher::equal): Compare them.
1896 (optimize_location_lists): Check whether a view list symbol is
1897 needed, and whether the locview attribute is present, and
1898 whether they match. Remove the locview attribute if no longer
1900 (index_location_lists): Call skip_loc_list_entry for test.
1901 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1902 Use output_dwarf_version.
1903 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1904 (struct dw_val_node): Add val_view_list.
1905 * final.c (SEEN_NEXT_VIEW): New.
1906 (set_next_view_needed): New.
1907 (clear_next_view_needed): New.
1908 (maybe_output_next_view): New.
1909 (final_start_function): Rename to...
1910 (final_start_function_1): ... this. Take pointer to FIRST,
1911 add SEEN parameter. Emit param bindings in the initial view.
1912 (final_start_function): Reintroduce SEEN-less interface.
1913 (final): Rename to...
1914 (final_1): ... this. Take SEEN parameter. Output final pending
1915 next view at the end.
1916 (final): Reintroduce seen-less interface.
1917 (final_scan_insn): Output pending next view before switching
1918 sections or ending a block. Mark the next view as needed when
1919 outputting variable locations. Notify debug backend of section
1920 changes, and of location view changes.
1921 (rest_of_handle_final): Adjust.
1922 * toplev.c (process_options): Autodetect value for debug variable
1923 location views option. Warn on incompat5 without -gdwarf-5.
1924 * doc/invoke.texi (gvariable-location-views): New.
1925 (gvariable-location-views=incompat5): New.
1926 (gno-variable-location-views): New.
1928 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1930 PR tree-optimization/84136
1931 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1932 that the result of find_edge is non-NULL.
1934 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1937 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1938 storing integer register in SImode. Fix cost of 256 and 512
1939 byte aligned SSE register store.
1941 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1943 * config/i386/i386.c (ix86_multiplication_cost): Fix
1944 multiplication cost for TARGET_AVX512DQ.
1946 2018-02-08 Marek Polacek <polacek@redhat.com>
1948 PR tree-optimization/84238
1949 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1952 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1954 PR tree-optimization/84265
1955 * tree-vect-stmts.c (vectorizable_store): Don't treat
1956 VMAT_CONTIGUOUS accesses as grouped.
1957 (vectorizable_load): Likewise.
1959 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1961 PR tree-optimization/81635
1962 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1963 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1964 (test_round_for_mask): New functions.
1965 (wide_int_cc_tests): Call test_round_for_mask.
1966 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1967 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1968 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1969 range returned by get_range_info.
1971 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1974 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1975 * symtab.c: Include builtins.h
1976 (symtab_node::output_to_lto_symbol_table_p): Move here
1977 from lto-streamer-out.c:output_symbol_p.
1978 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1979 (output_symbol_p): Move all logic to symtab.c
1980 (produce_symtab): Update.
1982 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1984 * config/s390/s390-opts.h (enum indirect_branch): Define.
1985 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1986 (s390_indirect_branch_via_thunk)
1987 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1988 (enum s390_indirect_branch_type): Define.
1989 * config/s390/s390.c (struct s390_frame_layout, struct
1990 machine_function): Remove.
1991 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1992 (indirect_branch_table_label_no, indirect_branch_table_name):
1994 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1995 (enum s390_indirect_branch_option): Define.
1996 (s390_return_addr_from_memory): New function.
1997 (s390_handle_string_attribute): New function.
1998 (s390_attribute_table): Add new attribute handler.
1999 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
2000 (s390_indirect_branch_via_thunk): New function.
2001 (s390_indirect_branch_via_inline_thunk): New function.
2002 (s390_function_ok_for_sibcall): When jumping via thunk disallow
2003 sibling call optimization for non z10 compiles.
2004 (s390_emit_call): Force indirect branch target to be a single
2005 register. Add r1 clobber for non-z10 compiles.
2006 (s390_emit_epilogue): Emit return jump via return_use expander.
2007 (s390_reorg): Handle JUMP_INSNs as execute targets.
2008 (s390_option_override_internal): Perform validity checks for the
2009 new command line options.
2010 (s390_indirect_branch_attrvalue): New function.
2011 (s390_indirect_branch_settings): New function.
2012 (s390_set_current_function): Invoke s390_indirect_branch_settings.
2013 (s390_output_indirect_thunk_function): New function.
2014 (s390_code_end): Implement target hook.
2015 (s390_case_values_threshold): Implement target hook.
2016 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
2018 * config/s390/s390.h (struct s390_frame_layout)
2019 (struct machine_function): Move here from s390.c.
2020 (TARGET_INDIRECT_BRANCH_NOBP_RET)
2021 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
2022 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
2023 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
2024 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
2025 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
2026 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
2027 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
2028 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
2029 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
2030 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
2031 (mnemonic attribute): Add values which aren't recognized
2033 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
2034 pattern for branch conversion. Fix mnemonic attribute.
2035 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
2036 indirect branch via thunk if requested.
2037 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
2038 ("*indirect_jump"): Disable for branch conversion using out of
2040 ("indirect_jump_via_thunk<mode>_z10")
2041 ("indirect_jump_via_thunk<mode>")
2042 ("indirect_jump_via_inlinethunk<mode>_z10")
2043 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
2044 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
2045 ("casesi_jump_via_inlinethunk<mode>_z10")
2046 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
2047 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
2048 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
2049 ("*indirect2_jump"): Disable for branch conversion.
2050 ("casesi_jump"): Turn into expander and expand patterns for branch
2052 ("return_use"): New expander.
2053 ("*return"): Emit return via thunk and rename it to ...
2054 ("*return<mode>"): ... this one.
2055 * config/s390/s390.opt: Add new options and and enum for the
2058 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
2060 * lra-constraints.c (match_reload): Unconditionally use
2061 gen_lowpart_SUBREG, rather than selecting between that
2062 and equivalent gen_rtx_SUBREG code.
2064 2018-02-08 Richard Biener <rguenther@suse.de>
2066 PR tree-optimization/84233
2067 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
2068 changed flag instead of boguously re-using phi_inserted.
2070 2018-02-08 Martin Jambor <mjambor@suse.cz>
2072 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
2073 static local variables.
2075 2018-02-08 Richard Biener <rguenther@suse.de>
2077 PR tree-optimization/84278
2078 * tree-vect-stmts.c (vectorizable_store): When looking for
2079 smaller vector types to perform grouped strided loads/stores
2080 make sure the mode is supported by the target.
2081 (vectorizable_load): Likewise.
2083 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
2085 * config/aarch64/aarch64.c (aarch64_components_for_bb):
2086 Increase LDP/STP opportunities by adding adjacent callee-saves.
2088 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
2090 PR rtl-optimization/84068
2091 PR rtl-optimization/83459
2092 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
2094 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
2096 PR tree-optimization/84224
2097 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
2098 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
2101 2018-02-07 Iain Sandoe <iain@codesourcery.com>
2104 * config/rs6000/altivec.md (*restore_world): Remove LR use.
2105 * config/rs6000/predicates.md (restore_world_operation): Adjust op
2106 count, remove one USE.
2108 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
2110 * doc/install.texi (Configuration): Document the
2111 --with-long-double-format={ibm,ieee} PowerPC configuration
2115 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
2116 Convert from define_expand to be define_insn_and_split. Rework
2117 float/double/_Float128 conversions to QI/HI/SImode to work with
2118 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
2119 conversions to QI/HImode types did a store and then a load to
2120 truncate the value. For conversions to VSX registers, don't split
2121 the insn, instead emit the code directly. Use the code iterator
2122 any_fix to combine signed and unsigned conversions.
2123 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
2124 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
2125 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
2126 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
2127 (fix_<mode>di2_hw): Likewise.
2128 (fixuns_<mode>di2_hw): Likewise.
2129 (fix_<mode>si2_hw): Likewise.
2130 (fixuns_<mode>si2_hw): Likewise.
2131 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
2132 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
2133 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
2134 fix<uns>_trunc<SFDF:mode>si2_p8.
2135 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
2137 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
2138 (fix<uns>_<mode>_mem): Likewise.
2139 (fctiw<u>z_<mode>_mem): Likewise.
2140 (fix<uns>_<mode>_mem): Likewise.
2141 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
2142 the register allocator from doing a direct move to the GPRs to do
2143 a store, and instead use the ISA 3.0 store byte/half-word from
2144 vector register instruction. For IEEE 128-bit floating point,
2145 also optimize stores of 32-bit ints.
2146 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
2148 2018-02-07 Alan Hayward <alan.hayward@arm.com>
2150 * genextract.c (push_pathstr_operand): New function to support
2152 (walk_rtx): Call push_pathstr_operand.
2153 (print_path): Support [a-zA-Z].
2155 2018-02-07 Richard Biener <rguenther@suse.de>
2157 PR tree-optimization/84037
2158 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
2159 (cse_and_gimplify_to_preheader): Declare.
2160 (vect_get_place_in_interleaving_chain): Likewise.
2161 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2163 (_loop_vec_info::~_loop_vec_info): Delete it.
2164 (cse_and_gimplify_to_preheader): New function.
2165 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
2166 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
2167 (vectorizable_load): Likewise. For grouped stores always base
2168 the IV on the first element.
2169 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
2170 condition before gimplifying.
2172 2018-02-07 Jakub Jelinek <jakub@redhat.com>
2174 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
2175 *DIV_EXPR and *MOD_EXPR.
2177 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
2180 * config/i386/i386.c (ix86_option_override_internal): Mask out
2181 the CF_SET bit when checking -fcf-protection.
2183 2018-02-07 Tom de Vries <tom@codesourcery.com>
2186 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
2189 2018-02-07 Richard Biener <rguenther@suse.de>
2191 PR tree-optimization/84204
2192 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
2195 PR tree-optimization/84205
2196 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
2197 special-case isl_ast_op_zdiv_r.
2199 PR tree-optimization/84223
2200 * graphite-scop-detection.c (gather_bbs::before_dom_children):
2201 Only add conditions from within the region.
2202 (gather_bbs::after_dom_children): Adjust.
2204 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
2207 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
2208 * config/avr/avr.md: Only post-reload split REG-REG moves if
2209 either register is GENERAL_REG_P.
2211 2018-02-07 Jakub Jelinek <jakub@redhat.com>
2213 PR tree-optimization/84235
2214 * tree-ssa-scopedtables.c
2215 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
2216 if the subtraction is performed in floating point type where NaNs are
2217 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
2218 build 1. Formatting fix.
2220 2018-02-06 Jakub Jelinek <jakub@redhat.com>
2223 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
2224 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
2225 and skip it regardless of bb boundaries. Use CALL_P macro,
2226 don't test INSN_P (insn) together with CALL_P or JUMP_P check
2227 unnecessarily, formatting fix.
2229 2018-02-06 Michael Collison <michael.collison@arm.com>
2231 * config/arm/thumb2.md:
2232 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
2233 (*thumb_mov_notscc): Ditto.
2235 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
2238 * config/rs6000/rs6000.md (su code attribute): Use "u" for
2239 unsigned_fix, not "s".
2241 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2243 * configure.ac (gcc_fn_eh_frame_ro): New function.
2244 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
2245 correct .eh_frame permissions.
2246 * configure: Regenerate.
2248 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
2250 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
2253 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2255 * config/rs6000/rs6000.c (rs6000_option_override_internal):
2256 Display warning message for -mno-speculate-indirect-jumps.
2258 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
2260 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
2262 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
2264 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
2266 PR tree-optimization/84225
2267 * tree-eh.c (find_trapping_overflow): Only call
2268 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
2270 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
2273 * config/i386/i386.c: Reimplement the check of possible options
2274 -mibt/-mshstk conbination. Change error messages.
2275 * doc/invoke.texi: Fix a typo: remove extra '='.
2277 2018-02-06 Marek Polacek <polacek@redhat.com>
2279 PR tree-optimization/84228
2280 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
2282 2018-02-06 Tamar Christina <tamar.christina@arm.com>
2285 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
2286 emitted arch directives.
2287 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
2288 __ARM_FEATURE_COPROC before changing architectures.
2290 2018-02-06 Richard Biener <rguenther@suse.de>
2292 * config/i386/i386.c (print_reg): Fix typo.
2293 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
2295 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
2297 * configure: Regenerate.
2299 2018-02-05 Martin Sebor <msebor@redhat.com>
2301 PR tree-optimization/83369
2302 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
2305 2018-02-05 Martin Liska <mliska@suse.cz>
2307 * doc/invoke.texi: Cherry-pick upstream r323995.
2309 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
2311 * ira.c (ira_init_register_move_cost): Adjust comment.
2313 2018-02-05 Martin Liska <mliska@suse.cz>
2315 PR gcov-profile/84137
2316 * doc/gcov.texi: Fix typo in documentation.
2318 2018-02-05 Martin Liska <mliska@suse.cz>
2320 PR gcov-profile/83879
2321 * doc/gcov.texi: Document necessity of --dynamic-list-data when
2322 using dlopen functionality.
2324 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
2326 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
2327 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
2328 _mm_maskz_range_ss, _mm_mask_range_round_ss,
2329 _mm_maskz_range_round_ss): New intrinsics.
2330 (__builtin_ia32_rangesd128_round)
2331 (__builtin_ia32_rangess128_round): Remove.
2332 (__builtin_ia32_rangesd128_mask_round,
2333 __builtin_ia32_rangess128_mask_round): New builtins.
2334 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
2335 __builtin_ia32_rangess128_round): Remove.
2336 (__builtin_ia32_rangesd128_mask_round,
2337 __builtin_ia32_rangess128_mask_round): New builtins.
2338 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
2339 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
2340 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
2341 "<round_saeonly_constraint>")): Changed to ...
2342 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
2343 "<round_saeonly_scalar_constraint>")): ... this.
2344 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
2345 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
2346 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
2347 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
2348 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
2350 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
2352 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
2354 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
2355 Remove all values except native, 8540 and 8548.
2357 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
2359 * config/i386/i386.c (ix86_output_function_return): Pass
2360 INVALID_REGNUM, instead of -1, as invalid register number to
2361 indirect_thunk_name and output_indirect_thunk.
2363 2018-02-02 Julia Koval <julia.koval@intel.com>
2365 * config.gcc: Add -march=icelake.
2366 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
2367 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
2368 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
2369 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
2370 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
2371 (processor_target_table): Add icelake.
2372 (ix86_option_override_internal): Handle new PTAs.
2373 (get_builtin_code_for_version): Handle icelake.
2374 (M_INTEL_COREI7_ICELAKE): New.
2375 (fold_builtin_cpu): Handle icelake.
2376 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
2377 * doc/invoke.texi: Add -march=icelake.
2379 2018-02-02 Julia Koval <julia.koval@intel.com>
2381 * config/i386/i386.c (ix86_option_override_internal): Change flags type
2382 to wide_int_bitmask.
2383 * wide-int-bitmask.h: New.
2385 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
2388 * config/i386/i386.md: Replace Pmode with word_mode in
2389 builtin_setjmp_setup and builtin_longjmp to support x32.
2391 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
2395 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
2397 (rs6000_supported_cpu_names): New static variable.
2398 (linux_cpu_translation_table): Likewise.
2399 (elf_platform) <cpu>: Define new static variable and use it.
2400 Translate kernel AT_PLATFORM name to canonical name if needed.
2401 Error if platform name is unknown.
2403 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
2406 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
2408 2018-02-01 Jeff Law <law@redhat.com>
2411 * config/i386/i386.c (release_scratch_register_on_entry): Add new
2412 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
2413 the scratch if RELEASE_VIA_POP is false.
2414 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
2415 If we have to save a temporary register, decrement SIZE appropriately.
2416 Pass new arguments to release_scratch_register_on_entry.
2417 (ix86_adjust_stack_and_probe): Likewise.
2418 (ix86_emit_probe_stack_range): Pass new arguments to
2419 release_scratch_register_on_entry.
2421 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
2423 PR rtl-optimization/84157
2424 * combine.c (change_zero_ext): Use REG_P predicate in
2425 front of HARD_REGISTER_P predicate.
2427 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
2429 * config/avr/avr.c (avr_option_override): Move disabling of
2430 -fdelete-null-pointer-checks to...
2431 * common/config/avr/avr-common.c (avr_option_optimization_table):
2434 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2436 PR tree-optimization/81635
2437 * tree-data-ref.c (split_constant_offset_1): For types that
2438 wrap on overflow, try to use range info to prove that wrapping
2441 2018-02-01 Renlin Li <renlin.li@arm.com>
2444 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
2446 (aarch64_register_move_cost): Likewise.
2447 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
2449 (REG_CLASS_NAMES): Likewise.
2450 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
2451 TAILCALL_ADDR_REGS. Remove IP registers.
2452 * config/aarch64/aarch64.md (Ucs): Update register constraint.
2454 2018-02-01 Richard Biener <rguenther@suse.de>
2456 * domwalk.h (dom_walker::dom_walker): Add additional constructor
2457 for specifying RPO order and allow NULL for that.
2458 * domwalk.c (dom_walker::dom_walker): Likewise.
2459 (dom_walker::walk): Handle NULL RPO order.
2460 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
2462 (rewrite_update_dom_walker): Likewise.
2463 (mark_def_dom_walker): Likewise.
2465 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2467 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
2468 (aarch64_maybe_expand_sve_subreg_move): Declare.
2469 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
2470 * config/aarch64/predicates.md (aarch64_any_register_operand): New
2472 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
2473 that are semantically a reverse operation.
2474 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
2475 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
2476 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
2478 (aarch64_can_change_mode_class): For big-endian, forbid changes
2479 between two SVE modes if they have different element sizes.
2481 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2483 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
2484 the TImode handling for big-endian targets.
2486 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2488 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
2489 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
2491 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
2492 Remove BSWAP handing for big-endian targets and use the form of
2493 LD1RQ appropariate for the mode.
2495 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2497 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
2498 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
2501 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2504 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
2505 check for operands that need to go through aarch64_sve_reload_be.
2507 2018-02-01 Jakub Jelinek <jakub@redhat.com>
2509 PR tree-optimization/81661
2510 PR tree-optimization/84117
2511 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
2512 * tree-eh.c: Include gimplify.h.
2513 (find_trapping_overflow, replace_trapping_overflow,
2514 rewrite_to_non_trapping_overflow): New functions.
2515 * tree-vect-loop.c: Include tree-eh.h.
2516 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2517 * tree-data-ref.c: Include tree-eh.h.
2518 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2520 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2522 PR rtl-optimization/84123
2523 * combine.c (change_zero_ext): Check if hard register satisfies
2524 can_change_dest_mode before calling gen_lowpart_SUBREG.
2526 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2529 * ira.c (ira_init_register_move_cost): Remove assert.
2531 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2533 PR rtl-optimization/84071
2534 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2535 * doc/tm.texi: Regenerate.
2537 2018-01-31 Richard Biener <rguenther@suse.de>
2539 PR tree-optimization/84132
2540 * tree-data-ref.c (analyze_miv_subscript): Properly
2541 check whether evolution_function_is_affine_multivariate_p
2542 before calling gcd_of_steps_may_divide_p.
2544 2018-01-31 Julia Koval <julia.koval@intel.com>
2547 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2548 * config/i386/i386.md (rdpid_rex64) New.
2549 (rdpid): Make 32bit only.
2551 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2554 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2555 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2557 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2560 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2562 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2564 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2566 PR rtl-optimization/84071
2567 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2568 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2570 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2572 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2573 (arc_attribute_table): Add 'aux' attribute.
2574 (arc_in_small_data_p): Consider aux like variables.
2575 (arc_is_aux_reg_p): New function.
2576 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2577 (arc_get_aux_arg): New function.
2578 (prepare_move_operands): Handle aux-register access.
2579 (arc_handle_aux_attribute): New function.
2580 * doc/extend.texi (ARC Variable attributes): Add subsection.
2582 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2584 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2585 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2586 (arc_attribute_table): Add 'uncached' attribute.
2587 (arc_print_operand): Print '.di' flag for uncached memory
2589 (arc_in_small_data_p): Do not consider for small data the uncached
2591 (arc_is_uncached_mem_p): New function.
2592 * config/arc/predicates.md (compact_store_memory_operand): Check
2593 for uncached memory accesses.
2594 (nonvol_nonimm_operand): Likewise.
2595 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2597 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2600 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2601 falign-loops=): Add Optimization flag.
2603 2018-01-30 Jeff Law <law@redhat.com>
2606 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2607 INT_REGISTERS_SAVED. Check it prior to calling
2608 get_scratch_register_on_entry.
2609 (ix86_adjust_stack_and_probe): Similarly.
2610 (ix86_emit_probe_stack_range): Similarly.
2611 (ix86_expand_prologue): Corresponding changes.
2613 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2616 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2617 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2619 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2622 * lra-constraints.c (curr_insn_transform): Process AND in the
2625 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2627 PR rtl-optimization/83986
2628 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2629 dependence against last_pending_memory_flush in addition to
2632 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2634 PR tree-optimization/81611
2635 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2638 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2641 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2644 2018-01-30 Richard Biener <rguenther@suse.de>
2645 Jakub Jelinek <jakub@redhat.com>
2647 PR tree-optimization/84111
2648 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2649 inner loops added during recursion, as they don't have up-to-date
2652 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2655 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2656 (can_inline_edge_by_limits_p): ... here.
2657 (can_early_inline_edge_p, check_callers,
2658 update_caller_keys, update_callee_keys, recursive_inlining,
2659 add_new_edges_to_heap, speculation_useful_p,
2660 inline_small_functions,
2661 inline_small_functions, flatten_function,
2662 inline_to_all_callers_1): Update.
2664 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2666 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2669 2018-01-30 Richard Biener <rguenther@suse.de>
2671 PR tree-optimization/83008
2672 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2673 invariant and constant vector uses in stmts when they need
2676 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2679 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2680 * configure: Regenerate.
2682 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2684 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2686 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2687 Use gen_rtx_REG rather than gen_lowpart.
2689 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2691 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2692 rather than 0 when creating partial subregs.
2694 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2696 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2699 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2702 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2703 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2704 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2705 flags. This restores the settings used before the 2017-07-24.
2706 Turning off pre increment/decrement/modify allows IVOPTS to
2707 optimize DF/SF loops where the index is an int.
2709 2018-01-29 Richard Biener <rguenther@suse.de>
2710 Kelvin Nilsen <kelvin@gcc.gnu.org>
2713 * tree-vect-stmts.c (vectorizable_call): Don't call
2714 targetm.vectorize_builtin_md_vectorized_function if callee is
2717 2018-01-22 Carl Love <cel@us.ibm.com>
2719 * doc/extend.tex: Fix typo in second arg in
2720 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2722 2018-01-29 Richard Biener <rguenther@suse.de>
2724 PR tree-optimization/84086
2725 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2726 (flush_ssaname_freelist): When SSA names were released reset
2727 the SCEV hash table.
2729 2018-01-29 Richard Biener <rguenther@suse.de>
2731 PR tree-optimization/84057
2732 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2733 removed paths when removing edges.
2735 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2737 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2738 -mfunction-return=@var{choice}.
2740 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2743 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2745 (layout::print_source_line): Likewise.
2746 (test_get_line_width_without_trailing_whitespace): Add test cases.
2748 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2751 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2754 2018-01-26 Jim Wilson <jimw@sifive.com>
2756 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2758 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2761 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2763 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2764 and CMP + SUB-immediate -> SUBS.
2766 2018-01-26 Martin Sebor <msebor@redhat.com>
2768 PR tree-optimization/83896
2769 * tree-ssa-strlen.c (get_string_len): Rename...
2770 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2771 Avoid assuming length is constant.
2772 (handle_char_store): Use HOST_WIDE_INT for string length.
2774 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2777 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2778 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2780 2018-01-26 Richard Biener <rguenther@suse.de>
2782 PR rtl-optimization/84003
2783 * dse.c (record_store): Only record redundant stores when
2784 the earlier store aliases at least all accesses the later one does.
2786 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2788 PR rtl-optimization/83985
2789 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2790 REG_CFA_RESTORE insns.
2791 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2792 REG_CFA_RESTORE insns here.
2795 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2796 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2798 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2800 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2801 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2802 (arc_init): Likewise.
2803 (arc_override_options): Likewise.
2804 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2806 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2808 * config/arc/arc.h (TARGET_DBNZ): Define.
2809 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2810 properly set the tune attribute.
2811 (dbnz): Use TARGET_DBNZ guard.
2812 * config/arc/arc.opt (mtune): Add core3 option.
2814 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2816 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2817 recognize new pic like addresses.
2818 (arc_delegitimize_address): Clean up.
2820 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2822 * config/arc/arc-arches.def: Option mrf16 valid for all
2824 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2825 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2826 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2827 * config/arc/arc-tables.opt: Regenerate.
2828 * config/arc/arc.c (arc_conditional_register_usage): Handle
2829 reduced register file case.
2830 (arc_file_start): Set must have build attributes.
2831 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2833 * config/arc/arc.opt (mrf16): Add new option.
2834 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2835 * config/arc/genmultilib.awk: Handle new mrf16 option.
2836 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2837 * config/arc/t-multilib: Regenerate.
2838 * doc/invoke.texi (ARC Options): Document mrf16 option.
2840 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2842 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2843 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2844 (arc_attribute_table): Add 'secure_call' attribute.
2845 (arc_print_operand): Print secure call operand.
2846 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2848 (arc_is_secure_call_p): New function. * config/arc/arc.md
2849 (call_i): Add support for sjli instruction.
2850 (call_value_i): Likewise.
2851 * config/arc/constraints.md (Csc): New constraint.
2853 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2854 John Eric Martin <John.Martin@emmicro-us.com>
2856 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2857 * config/arc/arc.c (_arc_jli_section): New struct.
2858 (arc_jli_section): New type.
2859 (rc_jli_sections): New static variable.
2860 (arc_handle_jli_attribute): New function.
2861 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2862 (arc_file_end): New function.
2863 (TARGET_ASM_FILE_END): Define.
2864 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2865 (arc_add_jli_section): New function.
2866 (jli_call_scan): Likewise.
2867 (arc_reorg): Call jli_call_scan.
2868 (arc_output_addsi): Remove 'S' from printing asm operand.
2869 (arc_is_jli_call_p): New function.
2870 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2872 (movhi_insn): Likewise.
2873 (movsi_insn): Likewise.
2874 (movsi_set_cc_insn): Likewise.
2875 (loadqi_update): Likewise.
2876 (load_zeroextendqisi_update): Likewise.
2877 (load_signextendqisi_update): Likewise.
2878 (loadhi_update): Likewise.
2879 (load_zeroextendhisi_update): Likewise.
2880 (load_signextendhisi_update): Likewise.
2881 (loadsi_update): Likewise.
2882 (loadsf_update): Likewise.
2883 (movsicc_insn): Likewise.
2884 (bset_insn): Likewise.
2885 (bxor_insn): Likewise.
2886 (bclr_insn): Likewise.
2887 (bmsk_insn): Likewise.
2888 (bicsi3_insn): Likewise.
2889 (cmpsi_cc_c_insn): Likewise.
2890 (movsi_ne): Likewise.
2891 (movsi_cond_exec): Likewise.
2892 (clrsbsi2): Likewise.
2903 (call_i): Remove 'S' asm letter, add jli instruction.
2904 (call_value_i): Likewise.
2905 * config/arc/arc.op (mjli-always): New option.
2906 * config/arc/constraints.md (Cji): New constraint.
2907 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2909 (subsf3_fpx): Likewise.
2910 (mulsf3_fpx): Likewise.
2911 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2913 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2915 * doc/invoke.texi (ARC): Document mjli-always option.
2917 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2919 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2920 avoid addition with 0 and use incw and decw where possible.
2922 2018-01-26 Richard Biener <rguenther@suse.de>
2924 PR tree-optimization/81082
2925 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2926 association if it requires casting to unsigned.
2927 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2928 from fold_plusminus_mult_expr to catch important cases late when
2929 range info is available.
2931 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2933 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2934 * configure.ac (hidden_linkonce): New test.
2935 * configure: Regenerate.
2936 * config.in: Regenerate.
2938 2018-01-26 Julia Koval <julia.koval@intel.com>
2940 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2941 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2942 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2943 _mm_mask_bitshuffle_epi64_mask): Fix type.
2944 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2945 USI_FTYPE_V4DI_V4DI_USI): Remove.
2946 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2947 __builtin_ia32_vpshufbitqmb256_mask,
2948 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2949 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2950 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2952 2018-01-26 Alan Modra <amodra@gmail.com>
2955 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2956 UNSPEC_VBPERMQ. Sort other unspecs.
2958 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2960 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2962 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2965 * predict.c (drop_profile): Do not push/pop cfun; update also
2967 (handle_missing_profiles): Fix logic looking for zero profiles.
2969 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2972 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2973 on functions with #pragma omp declare simd or functions with simd
2975 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2976 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2977 Remove trailing \n from warning_at calls.
2979 2018-01-25 Tom de Vries <tom@codesourcery.com>
2982 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2983 for neutered workers.
2985 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2988 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2989 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2991 2018-01-24 Jeff Law <law@redhat.com>
2994 * i386.c (get_probe_interval): Move to earlier point.
2995 (ix86_compute_frame_layout): If -fstack-clash-protection and
2996 the frame is larger than the probe interval, then use pushes
2997 to save registers rather than reg->mem moves.
2998 (ix86_expand_prologue): Remove conditional for int_registers_saved
3001 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
3004 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
3005 min/max for never referenced object.
3007 2018-01-24 Jakub Jelinek <jakub@redhat.com>
3010 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
3012 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
3013 attributes from DECL_ATTRIBUTES (decl) without affecting
3014 DECL_ATTRIBUTES (current_function_decl).
3015 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
3016 functions with non-NULL DECL_ABSTRACT_ORIGIN.
3018 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
3020 PR tree-optimization/83979
3021 * fold-const.c (fold_comparison): Use constant_boolean_node
3022 instead of boolean_{true,false}_node.
3024 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
3026 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
3029 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3031 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3032 Simplify the clause that sets the length attribute.
3033 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3034 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
3035 clause that sets the length attribute.
3036 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3038 2018-01-24 Tom de Vries <tom@codesourcery.com>
3041 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
3042 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
3043 Add strict parameter.
3044 (prevent_branch_around_nothing): Insert dummy insn between branch to
3045 label and label with no ptx insn inbetween.
3046 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
3048 2018-01-24 Tom de Vries <tom@codesourcery.com>
3051 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
3052 for neutered threads in warp.
3053 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
3055 2018-01-24 Richard Biener <rguenther@suse.de>
3057 PR tree-optimization/83176
3058 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
3061 2018-01-24 Richard Biener <rguenther@suse.de>
3063 PR tree-optimization/82819
3064 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
3065 code generating pluses that are no-ops in the target precision.
3067 2018-01-24 Richard Biener <rguenther@suse.de>
3070 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
3072 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3074 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
3075 to merge probabilities.
3076 * predict.c (probably_never_executed): Also mark as cold functions
3077 with global 0 profile and guessed local profile.
3078 * profile-count.c (profile_probability::combine_with_count): New
3080 * profile-count.h (profile_probability::operator*,
3081 profile_probability::operator*=, profile_probability::operator/,
3082 profile_probability::operator/=): Reduce precision to adjusted
3083 and set value to guessed on contradictory divisions.
3084 (profile_probability::combine_with_freq): Remove.
3085 (profile_probability::combine_wiht_count): Declare.
3086 (profile_count::force_nonzero):: Set to adjusted.
3087 (profile_count::probability_in):: Set quality to adjusted.
3088 * tree-ssa-tail-merge.c (replace_block_by): Use
3091 2018-01-23 Andrew Waterman <andrew@sifive.com>
3092 Jim Wilson <jimw@sifive.com>
3094 * config/riscv/riscv.c (riscv_stack_boundary): New.
3095 (riscv_option_override): Set riscv_stack_boundary. Handle
3096 riscv_preferred_stack_boundary_arg.
3097 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
3098 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
3099 (STACK_BOUNDARY): Set to riscv_stack_boundary.
3100 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
3101 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
3102 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
3104 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
3107 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
3108 of struct ix86_frame.
3109 (ix86_expand_epilogue): Likewise. Add a local variable for
3110 the reg_save_offset field in struct ix86_frame.
3112 2018-01-23 Bin Cheng <bin.cheng@arm.com>
3114 PR tree-optimization/82604
3115 * tree-loop-distribution.c (enum partition_kind): New enum item
3116 PKIND_PARTIAL_MEMSET.
3117 (partition_builtin_p): Support above new enum item.
3118 (generate_code_for_partition): Ditto.
3119 (compute_access_range): Differentiate cases that equality can be
3120 proven at all loops, the innermost loops or no loops.
3121 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
3122 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
3123 (finalize_partitions, distribute_loop): Don't fuse partition of
3124 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
3125 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
3128 2018-01-23 Martin Liska <mliska@suse.cz>
3130 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
3131 order to ignore the predictor.
3132 (PRED_POLYMORPHIC_CALL): Likewise.
3133 (PRED_RECURSIVE_CALL): Likewise.
3135 2018-01-23 Martin Liska <mliska@suse.cz>
3137 * tree-profile.c (tree_profiling): Print function header to
3138 aware reader which function we are working on.
3139 * value-prof.c (gimple_find_values_to_profile): Do not print
3140 not interesting value histograms.
3142 2018-01-23 Martin Liska <mliska@suse.cz>
3144 * profile-count.h (enum profile_quality): Add
3145 profile_uninitialized as the first value. Do not number values
3146 as they are zero based.
3147 (profile_count::verify): Update sanity check.
3148 (profile_probability::verify): Likewise.
3150 2018-01-23 Nathan Sidwell <nathan@acm.org>
3152 * doc/invoke.texi (ffor-scope): Deprecate.
3154 2018-01-23 David Malcolm <dmalcolm@redhat.com>
3156 PR tree-optimization/83510
3157 * domwalk.c (set_all_edges_as_executable): New function.
3158 (dom_walker::dom_walker): Convert bool param
3159 "skip_unreachable_blocks" to enum reachability. Move setup of
3160 edge flags to set_all_edges_as_executable and only do it when
3161 reachability is REACHABLE_BLOCKS.
3162 * domwalk.h (enum dom_walker::reachability): New enum.
3163 (dom_walker::dom_walker): Convert bool param
3164 "skip_unreachable_blocks" to enum reachability.
3165 (set_all_edges_as_executable): New decl.
3166 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
3167 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
3169 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
3170 but converting true to REACHABLE_BLOCKS.
3171 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
3173 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
3174 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
3175 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
3177 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
3178 if check_all_array_refs will be called.
3180 2018-01-23 David Malcolm <dmalcolm@redhat.com>
3182 * tree.c (selftest::test_location_wrappers): Add more test
3185 2018-01-23 David Malcolm <dmalcolm@redhat.com>
3187 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
3188 (selftest::test_bit_in_range): Likewise.
3190 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
3193 * doc/sourcebuild.texi (vect_float): Say that the selector
3194 only describes the situation when -funsafe-math-optimizations is on.
3195 (vect_float_strict): Document.
3197 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
3199 PR tree-optimization/83965
3200 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
3201 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
3202 instead of checking only for a reduction.
3203 (vect_recog_widen_sum_pattern): Likewise.
3205 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3207 * predict.c (probably_never_executed): Only use precise profile info.
3208 (compute_function_frequency): Skip after inlining hack since we now
3209 have quality checking.
3211 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3213 * profile-count.h (profile_probability::very_unlikely,
3214 profile_probability::unlikely, profile_probability::even): Set
3215 precision to guessed.
3217 2018-01-23 Richard Biener <rguenther@suse.de>
3219 PR tree-optimization/83963
3220 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
3221 Properly terminate dominator walk when crossing the exit edge not
3222 when visiting its source block.
3224 2018-01-23 Jakub Jelinek <jakub@redhat.com>
3227 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
3228 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
3230 2018-01-22 Jakub Jelinek <jakub@redhat.com>
3232 PR tree-optimization/83957
3233 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
3234 semicolon after for body surrounded by braces.
3236 PR tree-optimization/83081
3237 * profile-count.h (profile_probability::split): New method.
3238 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
3239 Use profile_probability::split.
3240 (do_compare_rtx_and_jump): Fix adjustment of probabilities
3241 when splitting a single conditional jump into 2.
3243 2018-01-22 David Malcolm <dmalcolm@redhat.com>
3245 PR tree-optimization/69452
3246 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
3249 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3251 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
3252 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
3253 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
3255 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3257 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
3258 * config/rl78/rl78.md (movdi): New define_expand.
3259 * config/rl78/rl78.c (rl78_split_movdi): New function.
3261 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
3264 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
3266 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
3267 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
3268 128-bit to produce an UNSPEC move to get the double word with the
3269 signbit and then a shift directly to do signbit.
3270 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
3271 implementation with a new version that just does either a direct
3272 move or a regular move. Move memory interface to separate insns.
3273 Move insns so they are next to the expander.
3274 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
3275 with signbit move. Split big and little endian case.
3276 (signbit<mode>2_dm_mem_le): Likewise.
3277 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
3278 (signbit<mode>2_dm2): Likewise.
3280 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3282 * config/rl78/rl78.md (anddi3): New define_expand.
3284 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3286 * config/rl78/rl78.md (umindi3): New define_expand.
3288 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3290 * config/rl78/rl78.md (smindi3): New define_expand.
3292 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3294 * config/rl78/rl78.md (smaxdi3): New define_expand.
3296 2018-01-22 Carl Love <cel@us.ibm.com>
3298 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
3299 LVX_V1TI): Add macro expansion.
3300 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
3301 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
3302 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
3303 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
3304 Change check to determine if the instruction is a byte reversing
3305 entry. Fix typo in comment.
3306 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
3307 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
3308 Add def_builtin calls for new builtins.
3309 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
3310 Add define_insn expansion.
3312 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3314 * config/rl78/rl78.md (umaxdi3): New define_expand.
3316 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3318 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
3319 for non-QImode registers.
3321 2018-01-22 Richard Biener <rguenther@suse.de>
3323 PR tree-optimization/83963
3324 * graphite-scop-detection.c (scop_detection::get_sese): Delay
3325 including the loop exit block.
3326 (scop_detection::merge_sese): Likewise.
3327 (scop_detection::add_scop): Do it here instead.
3329 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3331 * doc/sourcebuild.texi (arm_softfloat): Document.
3333 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
3336 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
3337 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
3338 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
3340 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3341 David Edelsohn <dje.gcc@gmail.com>
3344 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3345 Change "crset eq" to "crset 2".
3346 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3347 (*call_indirect_aix<mode>_nospec): Likewise.
3348 (*call_value_indirect_aix<mode>_nospec): Likewise.
3349 (*call_indirect_elfv2<mode>_nospec): Likewise.
3350 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
3351 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
3352 change assembly output from . to $.
3353 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3354 (indirect_jump<mode>_nospec): Change assembly output from . to $.
3355 (*tablejump<mode>_internal1_nospec): Likewise.
3357 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
3360 * config/sh/sh_optimize_sett_clrt.cc:
3361 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
3363 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
3365 PR tree-optimization/83940
3366 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
3367 offset_dt to vect_constant_def rather than vect_unknown_def_type.
3368 (vect_check_load_store_mask): Add a mask_dt_out parameter and
3369 use it to pass back the definition type.
3370 (vect_check_store_rhs): Likewise rhs_dt_out.
3371 (vect_build_gather_load_calls): Add a mask_dt argument and use
3372 it instead of a call to vect_is_simple_use.
3373 (vectorizable_store): Update calls to vect_check_load_store_mask
3374 and vect_check_store_rhs. Use the dt returned by the latter instead
3375 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
3376 instead of calls to vect_is_simple_use. Pass the scalar rather
3377 than the vector operand to vect_is_simple_use when handling
3378 second and subsequent copies of an rhs value.
3379 (vectorizable_load): Update calls to vect_check_load_store_mask
3380 and vect_build_gather_load_calls. Use the cached mask_dt and
3381 gs_info.offset_dt instead of calls to vect_is_simple_use.
3383 2018-01-20 Jakub Jelinek <jakub@redhat.com>
3386 * tree-emutls.c: Include gimplify.h.
3387 (lower_emutls_2): New function.
3388 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
3389 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
3390 it before further processing.
3393 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
3394 UINTVAL (trueop1) instead of INTVAL (op1).
3396 2018-01-19 Jakub Jelinek <jakub@redhat.com>
3400 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
3401 INCOMING_FRAME_SP_OFFSET if not defined.
3402 (scan_trace): Add ENTRY argument. If true and
3403 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
3404 emit a note to adjust the CFA offset.
3405 (create_cfi_notes): Adjust scan_trace callers.
3406 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
3407 INCOMING_FRAME_SP_OFFSET in the CIE.
3408 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
3409 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
3411 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
3412 * doc/tm.texi: Regenerated.
3414 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
3416 PR rtl-optimization/83147
3417 * lra-constraints.c (remove_inheritance_pseudos): Use
3418 lra_substitute_pseudo_within_insn.
3420 2018-01-19 Tom de Vries <tom@codesourcery.com>
3421 Cesar Philippidis <cesar@codesourcery.com>
3424 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
3426 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
3429 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
3430 spaces for function labels.
3432 2018-01-19 Martin Liska <mliska@suse.cz>
3434 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
3435 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
3436 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
3437 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
3438 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
3439 (PRED_CONST_RETURN): Change from 69 to 65.
3440 (PRED_NULL_RETURN): Change from 91 to 71.
3441 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
3442 (PRED_LOOP_GUARD): Change from 66 to 73.
3444 2018-01-19 Martin Liska <mliska@suse.cz>
3446 * predict.c (predict_insn_def): Add new assert.
3447 (struct branch_predictor): Change type to signed integer.
3448 (test_prediction_value_range): Amend test to cover
3450 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
3451 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
3452 (PRED_LOOP_ITERATIONS_MAX): Likewise.
3453 (PRED_LOOP_IV_COMPARE): Likewise.
3454 * predict.h (PROB_UNINITIALIZED): Define new constant.
3456 2018-01-19 Martin Liska <mliska@suse.cz>
3458 * predict.c (dump_prediction): Add new format for
3459 analyze_brprob.py script which is enabled with -details
3461 * profile-count.h (precise_p): New function.
3463 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3465 PR tree-optimization/83922
3466 * tree-vect-loop.c (vect_verify_full_masking): Return false if
3467 there are no statements that need masking.
3468 (vect_active_double_reduction_p): New function.
3469 (vect_analyze_loop_operations): Use it when handling phis that
3470 are not in the loop header.
3472 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3474 PR tree-optimization/83914
3475 * tree-vect-loop.c (vectorizable_induction): Don't convert
3476 init_expr or apply the peeling adjustment for inductions
3477 that are nested within the vectorized loop.
3479 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3481 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
3484 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3488 * function.h (gimplify_parameters): Add gimple_seq * argument.
3489 * function.c: Include gimple.h and options.h.
3490 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
3491 for the added local temporaries if needed.
3492 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
3493 if there are any parameter cleanups, wrap whole body into a
3494 try/finally with the cleanups.
3496 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
3499 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
3500 Use GET_MODE_CLASS for scalar floating point.
3502 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3506 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
3507 Fix call of call_cgraph_insertion_hooks.
3509 2018-01-18 Martin Sebor <msebor@redhat.com>
3511 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
3513 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3516 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3519 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3522 * common.opt: (-ffile-prefix-map): New option.
3523 * opts.c (common_handle_option): Defer it.
3524 * opts-global.c (handle_common_deferred_options): Handle it.
3525 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3526 * file-prefix-map.h: New file.
3527 (remap_debug_filename, add_debug_prefix_map): ...here.
3528 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3529 * final.c (debug_prefix_map, add_debug_prefix_map
3530 remap_debug_filename): Move to...
3531 * file-prefix-map.c: New file.
3532 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3533 generalize, get rid of alloca(), use strrchr() instead of strchr().
3534 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3535 Implement in terms of add_prefix_map().
3536 (remap_macro_filename, remap_debug_filename): Implement in term of
3538 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3539 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3540 * dbxout.c: Include file-prefix-map.h.
3541 * varasm.c: Likewise.
3542 * vmsdbgout.c: Likewise.
3543 * xcoffout.c: Likewise.
3544 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3545 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3546 * doc/invoke.texi (-ffile-prefix-map): Document.
3547 (-fdebug-prefix-map): Update description.
3549 2018-01-18 Martin Liska <mliska@suse.cz>
3551 * config/i386/i386.c (indirect_thunk_name): Document that also
3553 (output_indirect_thunk): Document why both instructions
3554 (pause and lfence) are generated.
3556 2018-01-18 Richard Biener <rguenther@suse.de>
3558 PR tree-optimization/83887
3559 * graphite-scop-detection.c
3560 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3561 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3562 (scop_detection::merge_sese): Re-implement with a flood-fill
3563 algorithm that properly finds a SESE region if it exists.
3565 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3568 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3569 pointer_diff optimizations use view_convert instead of convert.
3571 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3573 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3574 Generate different code for -mno-speculate-indirect-jumps.
3575 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3576 (*call_indirect_aix<mode>): Disable for
3577 -mno-speculate-indirect-jumps.
3578 (*call_indirect_aix<mode>_nospec): New define_insn.
3579 (*call_value_indirect_aix<mode>): Disable for
3580 -mno-speculate-indirect-jumps.
3581 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3582 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3583 -mno-speculate-indirect-jumps.
3584 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3586 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3588 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3589 long double type, set the flags for noting the default long double
3590 type, even if we don't pass or return a long double type.
3592 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3595 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3598 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3600 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3601 support for merge[hl].
3602 (fold_mergehl_helper): New helper function.
3603 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3604 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3605 (altivec_vmrglw_direct): Add xxmrglw insn.
3607 2018-01-17 Andrew Waterman <andrew@sifive.com>
3609 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3610 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3612 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3615 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3616 call the lto_location_cache before reading the
3617 DECL_SOURCE_LOCATION of the types.
3619 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3620 Richard Sandiford <richard.sandiford@linaro.org>
3622 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3623 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3624 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3625 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3626 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3628 * config/aarch64/constraints.md (aarch64_movti_operand):
3630 * config/aarch64/predicates.md (Uti): Add new constraint.
3632 2018-01-17 Carl Love <cel@us.ibm.com>
3634 * config/rs6000/vsx.md (define_expand xl_len_r,
3635 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3636 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3638 (define_expand, define_insn): Move the shift left from the
3639 define_insn to the define_expand for lxvl and stxvl instructions.
3640 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3641 and XL_LEN_R definitions to PURE.
3643 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3645 * config/i386/i386.c (indirect_thunk_name): Declare regno
3646 as unsigned int. Compare regno with INVALID_REGNUM.
3647 (output_indirect_thunk): Ditto.
3648 (output_indirect_thunk_function): Ditto.
3649 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3650 in the call to output_indirect_thunk_function.
3652 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3655 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3656 rather than the size of inner_type to determine the stack slot size
3657 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3659 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3662 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3665 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3667 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3668 endian Linux systems to optionally enable multilibs for selecting
3669 the long double type if the user configured an explicit type.
3670 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3671 have no long double multilibs if not defined.
3672 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3673 warn if the user used -mabi={ieee,ibm}longdouble and we built
3674 multilibs for long double.
3675 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3676 appropriate multilib option.
3677 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3679 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3680 for building long double multilibs.
3681 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3683 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3685 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3688 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3690 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3693 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3696 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3699 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3701 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3702 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3704 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3706 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3707 different rtl trees depending on TARGET_64BIT.
3708 (rs6000_gen_lvx): Likewise.
3710 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3712 * config/visium/visium.md (nop): Tweak comment.
3713 (hazard_nop): Likewise.
3715 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3717 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3718 -mspeculate-indirect-jumps.
3719 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3720 for -mno-speculate-indirect-jumps.
3721 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3722 (*call_value_indirect_elfv2<mode>): Disable for
3723 -mno-speculate-indirect-jumps.
3724 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3725 (indirect_jump): Emit different RTL for
3726 -mno-speculate-indirect-jumps.
3727 (*indirect_jump<mode>): Disable for
3728 -mno-speculate-indirect-jumps.
3729 (*indirect_jump<mode>_nospec): New define_insn.
3730 (tablejump): Emit different RTL for
3731 -mno-speculate-indirect-jumps.
3732 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3733 (tablejumpsi_nospec): New define_expand.
3734 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3735 (tablejumpdi_nospec): New define_expand.
3736 (*tablejump<mode>_internal1): Disable for
3737 -mno-speculate-indirect-jumps.
3738 (*tablejump<mode>_internal1_nospec): New define_insn.
3739 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3742 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3744 * caller-save.c (insert_save): Drop unnecessary parameter. All
3747 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3748 Richard Biener <rguenth@suse.de>
3751 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3752 return early, inline manually is_gimple_sizepos. Make sure if we
3753 call gimplify_expr we don't end up with a gimple constant.
3754 * tree.c (variably_modified_type_p): Don't return true for
3755 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3756 * gimplify.h (is_gimple_sizepos): Remove.
3758 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3760 PR tree-optimization/83857
3761 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3762 vectorizable_live_operation for pure SLP statements.
3763 (vectorizable_live_operation): Handle PHIs.
3765 2018-01-16 Richard Biener <rguenther@suse.de>
3767 PR tree-optimization/83867
3768 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3769 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3771 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3774 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3775 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3776 If off is not INTEGER_CST, issue a may not be aligned warning
3777 rather than isn't aligned. Use isn%'t rather than isn't.
3778 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3780 <case MULT_EXPR>: Improve the case when bottom and one of the
3781 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3782 operand, in that case check if the other operand is multiple of
3783 bottom divided by the INTEGER_CST operand.
3785 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3788 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3789 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3790 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3791 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3792 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3793 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3794 (pa_function_arg_size): New function.
3796 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3798 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3799 in a separate statement.
3801 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3803 PR tree-optimization/83847
3804 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3805 group gathers and scatters.
3807 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3809 PR rtl-optimization/86620
3810 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3812 PR rtl-optimization/83213
3813 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3814 to last if both are JUMP_INSNs.
3816 PR tree-optimization/83843
3817 * gimple-ssa-store-merging.c
3818 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3819 store_immediate_info for bswap/nop orig_stores.
3821 2018-01-15 Andrew Waterman <andrew@sifive.com>
3823 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3825 <UDIV>: Increase cost if !TARGET_DIV.
3827 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3829 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3830 (define_attr "cr_logical_3op"): New.
3831 (cceq_ior_compare): Adjust.
3832 (cceq_ior_compare_complement): Adjust.
3833 (*cceq_rev_compare): Adjust.
3834 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3835 (is_cracked_insn): Adjust.
3836 (insn_must_be_first_in_group): Adjust.
3837 * config/rs6000/40x.md: Adjust.
3838 * config/rs6000/440.md: Adjust.
3839 * config/rs6000/476.md: Adjust.
3840 * config/rs6000/601.md: Adjust.
3841 * config/rs6000/603.md: Adjust.
3842 * config/rs6000/6xx.md: Adjust.
3843 * config/rs6000/7450.md: Adjust.
3844 * config/rs6000/7xx.md: Adjust.
3845 * config/rs6000/8540.md: Adjust.
3846 * config/rs6000/cell.md: Adjust.
3847 * config/rs6000/e300c2c3.md: Adjust.
3848 * config/rs6000/e500mc.md: Adjust.
3849 * config/rs6000/e500mc64.md: Adjust.
3850 * config/rs6000/e5500.md: Adjust.
3851 * config/rs6000/e6500.md: Adjust.
3852 * config/rs6000/mpc.md: Adjust.
3853 * config/rs6000/power4.md: Adjust.
3854 * config/rs6000/power5.md: Adjust.
3855 * config/rs6000/power6.md: Adjust.
3856 * config/rs6000/power7.md: Adjust.
3857 * config/rs6000/power8.md: Adjust.
3858 * config/rs6000/power9.md: Adjust.
3859 * config/rs6000/rs64.md: Adjust.
3860 * config/rs6000/titan.md: Adjust.
3862 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3864 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3865 ix86_indirect_branch_register logic.
3867 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3869 * config/i386/constraints.md (Bs): Update
3870 ix86_indirect_branch_register check. Don't check
3871 ix86_indirect_branch_register with GOT_memory_operand.
3873 * config/i386/predicates.md (GOT_memory_operand): Don't check
3874 ix86_indirect_branch_register here.
3875 (GOT32_symbol_operand): Likewise.
3877 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3879 * config/i386/predicates.md (constant_call_address_operand):
3880 Rewrite ix86_indirect_branch_register logic.
3881 (sibcall_insn_operand): Likewise.
3883 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3885 * config/i386/constraints.md (Bs): Replace
3886 ix86_indirect_branch_thunk_register with
3887 ix86_indirect_branch_register.
3889 * config/i386/i386.md (indirect_jump): Likewise.
3890 (tablejump): Likewise.
3891 (*sibcall_memory): Likewise.
3892 (*sibcall_value_memory): Likewise.
3893 Peepholes of indirect call and jump via memory: Likewise.
3894 * config/i386/i386.opt: Likewise.
3895 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3896 (GOT_memory_operand): Likewise.
3897 (call_insn_operand): Likewise.
3898 (sibcall_insn_operand): Likewise.
3899 (GOT32_symbol_operand): Likewise.
3901 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3904 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3905 type rather than type addr's type points to.
3906 (expand_omp_atomic_mutex): Likewise.
3907 (expand_omp_atomic): Likewise.
3909 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3912 * config/i386/i386.c (output_indirect_thunk_function): Use
3913 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3914 for __x86_return_thunk.
3916 2018-01-15 Richard Biener <rguenther@suse.de>
3919 * expmed.c (extract_bit_field_1): Fix typo.
3921 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3924 * config/arm/iterators.md (VF): New mode iterator.
3925 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3926 Remove integer-related logic from pattern.
3927 (neon_vabd<mode>_3): Likewise.
3929 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3932 * common.opt (fstrict-overflow): No longer an alias.
3933 (fwrapv-pointer): New option.
3934 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3935 also for pointer types based on flag_wrapv_pointer.
3936 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3937 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3938 opts->x_flag_wrapv got set.
3939 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3940 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3941 POINTER_TYPE_OVERFLOW_UNDEFINED.
3942 * match.pd: Likewise in address comparison pattern.
3943 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3945 2018-01-15 Richard Biener <rguenther@suse.de>
3948 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3949 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3950 Reset type names to their identifier if their TYPE_DECL doesn't
3951 have linkage (and thus is used for ODR and devirt).
3952 (save_debug_info_for_decl): Remove.
3953 (save_debug_info_for_type): Likewise.
3954 (add_tree_to_fld_list): Adjust.
3955 * tree-pretty-print.c (dump_generic_node): Make dumping of
3956 type names more robust.
3958 2018-01-15 Richard Biener <rguenther@suse.de>
3960 * BASE-VER: Bump to 8.0.1.
3962 2018-01-14 Martin Sebor <msebor@redhat.com>
3965 * builtins.c (check_access): Avoid warning when the no-warning bit
3968 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3970 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3971 * ira-color (allocno_hard_regs_compare): Likewise.
3973 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3976 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3977 Use .pushsection/.popsection.
3979 2018-01-14 Martin Sebor <msebor@redhat.com>
3982 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3984 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3986 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3987 entry from extra_headers.
3988 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3989 extra_headers, make the list bitwise identical to the i?86-*-* one.
3991 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3993 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3994 -mcmodel=large with -mindirect-branch=thunk,
3995 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3996 -mfunction-return=thunk-extern.
3997 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3998 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3999 -mfunction-return=thunk and -mfunction-return=thunk-extern.
4001 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
4003 * config/i386/i386.c (print_reg): Print the name of the full
4004 integer register without '%'.
4005 (ix86_print_operand): Handle 'V'.
4006 * doc/extend.texi: Document 'V' modifier.
4008 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
4010 * config/i386/constraints.md (Bs): Disallow memory operand for
4011 -mindirect-branch-register.
4013 * config/i386/predicates.md (indirect_branch_operand): Likewise.
4014 (GOT_memory_operand): Likewise.
4015 (call_insn_operand): Likewise.
4016 (sibcall_insn_operand): Likewise.
4017 (GOT32_symbol_operand): Likewise.
4018 * config/i386/i386.md (indirect_jump): Call convert_memory_address
4019 for -mindirect-branch-register.
4020 (tablejump): Likewise.
4021 (*sibcall_memory): Likewise.
4022 (*sibcall_value_memory): Likewise.
4023 Disallow peepholes of indirect call and jump via memory for
4024 -mindirect-branch-register.
4025 (*call_pop): Replace m with Bw.
4026 (*call_value_pop): Likewise.
4027 (*sibcall_pop_memory): Replace m with Bs.
4028 * config/i386/i386.opt (mindirect-branch-register): New option.
4029 * doc/invoke.texi: Document -mindirect-branch-register option.
4031 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
4033 * config/i386/i386-protos.h (ix86_output_function_return): New.
4034 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
4035 set function_return_type.
4036 (indirect_thunk_name): Add ret_p to indicate thunk for function
4038 (output_indirect_thunk_function): Pass false to
4039 indirect_thunk_name.
4040 (ix86_output_indirect_branch_via_reg): Likewise.
4041 (ix86_output_indirect_branch_via_push): Likewise.
4042 (output_indirect_thunk_function): Create alias for function
4043 return thunk if regno < 0.
4044 (ix86_output_function_return): New function.
4045 (ix86_handle_fndecl_attribute): Handle function_return.
4046 (ix86_attribute_table): Add function_return.
4047 * config/i386/i386.h (machine_function): Add
4048 function_return_type.
4049 * config/i386/i386.md (simple_return_internal): Use
4050 ix86_output_function_return.
4051 (simple_return_internal_long): Likewise.
4052 * config/i386/i386.opt (mfunction-return=): New option.
4053 (indirect_branch): Mention -mfunction-return=.
4054 * doc/extend.texi: Document function_return function attribute.
4055 * doc/invoke.texi: Document -mfunction-return= option.
4057 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
4059 * config/i386/i386-opts.h (indirect_branch): New.
4060 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
4061 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
4062 with local indirect jump when converting indirect call and jump.
4063 (ix86_set_indirect_branch_type): New.
4064 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
4065 (indirectlabelno): New.
4066 (indirect_thunk_needed): Likewise.
4067 (indirect_thunk_bnd_needed): Likewise.
4068 (indirect_thunks_used): Likewise.
4069 (indirect_thunks_bnd_used): Likewise.
4070 (INDIRECT_LABEL): Likewise.
4071 (indirect_thunk_name): Likewise.
4072 (output_indirect_thunk): Likewise.
4073 (output_indirect_thunk_function): Likewise.
4074 (ix86_output_indirect_branch_via_reg): Likewise.
4075 (ix86_output_indirect_branch_via_push): Likewise.
4076 (ix86_output_indirect_branch): Likewise.
4077 (ix86_output_indirect_jmp): Likewise.
4078 (ix86_code_end): Call output_indirect_thunk_function if needed.
4079 (ix86_output_call_insn): Call ix86_output_indirect_branch if
4081 (ix86_handle_fndecl_attribute): Handle indirect_branch.
4082 (ix86_attribute_table): Add indirect_branch.
4083 * config/i386/i386.h (machine_function): Add indirect_branch_type
4084 and has_local_indirect_jump.
4085 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
4087 (tablejump): Likewise.
4088 (*indirect_jump): Use ix86_output_indirect_jmp.
4089 (*tablejump_1): Likewise.
4090 (simple_return_indirect_internal): Likewise.
4091 * config/i386/i386.opt (mindirect-branch=): New option.
4092 (indirect_branch): New.
4095 (thunk-inline): Likewise.
4096 (thunk-extern): Likewise.
4097 * doc/extend.texi: Document indirect_branch function attribute.
4098 * doc/invoke.texi: Document -mindirect-branch= option.
4100 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
4103 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
4105 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
4107 * ipa-inline.c (want_inline_small_function_p): Return false if
4108 inlining has already failed with CIF_FINAL_ERROR.
4109 (update_caller_keys): Call want_inline_small_function_p before
4111 (update_callee_keys): Likewise.
4113 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
4115 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
4117 (rs6000_quadword_masked_address_p): Likewise.
4118 (quad_aligned_load_p): Likewise.
4119 (quad_aligned_store_p): Likewise.
4120 (const_load_sequence_p): Add comment to describe the outer-most loop.
4121 (mimic_memory_attributes_and_flags): New function.
4122 (rs6000_gen_stvx): Likewise.
4123 (replace_swapped_aligned_store): Likewise.
4124 (rs6000_gen_lvx): Likewise.
4125 (replace_swapped_aligned_load): Likewise.
4126 (replace_swapped_load_constant): Capitalize argument name in
4127 comment describing this function.
4128 (rs6000_analyze_swaps): Add a third pass to search for vector loads
4129 and stores that access quad-word aligned addresses and replace
4130 with stvx or lvx instructions when appropriate.
4131 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
4132 New function prototype.
4133 (rs6000_quadword_masked_address_p): Likewise.
4134 (rs6000_gen_lvx): Likewise.
4135 (rs6000_gen_stvx): Likewise.
4136 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
4137 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
4138 when memory address is aligned.
4139 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
4140 this split to select lvx instruction when memory address is aligned.
4141 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
4142 instruction when memory address is aligned.
4143 (*vsx_le_perm_load_v16qi): Likewise.
4144 (four unnamed splitters): Modify to select the stvx instruction
4145 when memory is aligned.
4147 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
4149 * predict.c (determine_unlikely_bbs): Handle correctly BBs
4150 which appears in the queue multiple times.
4152 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4153 Alan Hayward <alan.hayward@arm.com>
4154 David Sherwood <david.sherwood@arm.com>
4156 * tree-vectorizer.h (vec_lower_bound): New structure.
4157 (_loop_vec_info): Add check_nonzero and lower_bounds.
4158 (LOOP_VINFO_CHECK_NONZERO): New macro.
4159 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
4160 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
4161 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
4162 fields. Make seg_len the distance travelled, not including the
4164 (dr_direction_indicator): Declare.
4165 (dr_zero_step_indicator): Likewise.
4166 (dr_known_forward_stride_p): Likewise.
4167 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
4169 (runtime_alias_check_p): Allow runtime alias checks with
4171 (operator ==): Compare access_size and align.
4172 (prune_runtime_alias_test_list): Rework for new distinction between
4173 the access_size and seg_len.
4174 (create_intersect_range_checks_index): Likewise. Cope with polynomial
4176 (get_segment_min_max): New function.
4177 (create_intersect_range_checks): Use it.
4178 (dr_step_indicator): New function.
4179 (dr_direction_indicator): Likewise.
4180 (dr_zero_step_indicator): Likewise.
4181 (dr_known_forward_stride_p): Likewise.
4182 * tree-loop-distribution.c (data_ref_segment_size): Return
4183 DR_STEP * (niters - 1).
4184 (compute_alias_check_pairs): Update call to the dr_with_seg_len
4186 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
4187 (vect_preserves_scalar_order_p): New function, split out from...
4188 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
4189 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
4190 (vect_vfa_access_size): New function.
4191 (vect_vfa_align): Likewise.
4192 (vect_compile_time_alias): Take access_size_a and access_b arguments.
4193 (dump_lower_bound): New function.
4194 (vect_check_lower_bound): Likewise.
4195 (vect_small_gap_p): Likewise.
4196 (vectorizable_with_step_bound_p): Likewise.
4197 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
4198 depencies if the vectorization factor is 1. Convert the checks
4199 for nonzero steps into checks on the bounds of DR_STEP. Try using
4200 a bunds check for variable steps if the minimum required step is
4201 relatively small. Update calls to the dr_with_seg_len
4202 constructor and to vect_compile_time_alias.
4203 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
4205 (vect_loop_versioning): Call it.
4206 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
4208 (vect_estimate_min_profitable_iters): Account for any bounds checks.
4210 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4211 Alan Hayward <alan.hayward@arm.com>
4212 David Sherwood <david.sherwood@arm.com>
4214 * doc/sourcebuild.texi (vect_scatter_store): Document.
4215 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
4217 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
4219 * genopinit.c (main): Add supports_vec_scatter_store and
4220 supports_vec_scatter_store_cached to target_optabs.
4221 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
4222 IFN_MASK_SCATTER_STORE.
4223 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
4225 * internal-fn.h (internal_store_fn_p): Declare.
4226 (internal_fn_stored_value_index): Likewise.
4227 * internal-fn.c (scatter_store_direct): New macro.
4228 (expand_scatter_store_optab_fn): New function.
4229 (direct_scatter_store_optab_supported_p): New macro.
4230 (internal_store_fn_p): New function.
4231 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
4232 IFN_MASK_SCATTER_STORE.
4233 (internal_fn_mask_index): Likewise.
4234 (internal_fn_stored_value_index): New function.
4235 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
4237 * optabs-query.h (supports_vec_scatter_store_p): Declare.
4238 * optabs-query.c (supports_vec_scatter_store_p): New function.
4239 * tree-vectorizer.h (vect_get_store_rhs): Declare.
4240 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
4241 true for scatter stores.
4242 (vect_gather_scatter_fn_p): Handle scatter stores too.
4243 (vect_check_gather_scatter): Consider using scatter stores if
4244 supports_vec_scatter_store_p.
4245 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
4247 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
4248 internal_fn_stored_value_index.
4249 (check_load_store_masking): Handle scatter stores too.
4250 (vect_get_store_rhs): Make public.
4251 (vectorizable_call): Use internal_store_fn_p.
4252 (vectorizable_store): Handle scatter store internal functions.
4253 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
4254 when deciding whether the end of the group has been reached.
4255 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
4256 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
4257 (mask_scatter_store<mode>): New insns.
4259 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4260 Alan Hayward <alan.hayward@arm.com>
4261 David Sherwood <david.sherwood@arm.com>
4263 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
4264 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
4265 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
4267 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
4268 Use vect_truncate_gather_scatter_offset if we can't treat the
4269 operation as a normal gather load or scatter store.
4270 (get_group_load_store_type): Take the gather_scatter_info
4271 as argument. Try using a gather load or scatter store for
4272 single-element groups.
4273 (get_load_store_type): Update calls to get_group_load_store_type
4274 and vect_use_strided_gather_scatters_p.
4276 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4277 Alan Hayward <alan.hayward@arm.com>
4278 David Sherwood <david.sherwood@arm.com>
4280 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
4281 optional tree argument.
4282 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
4284 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
4285 but continue to use the current value as a fallback.
4286 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
4287 to compare the updates.
4288 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
4289 (get_load_store_type): Use it when handling a strided access.
4290 (vect_get_strided_load_store_ops): New function.
4291 (vect_get_data_ptr_increment): Likewise.
4292 (vectorizable_load): Handle strided gather loads. Always pass
4293 a step to vect_create_data_ref_ptr and bump_vector_ptr.
4295 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4296 Alan Hayward <alan.hayward@arm.com>
4297 David Sherwood <david.sherwood@arm.com>
4299 * doc/md.texi (gather_load@var{m}): Document.
4300 (mask_gather_load@var{m}): Likewise.
4301 * genopinit.c (main): Add supports_vec_gather_load and
4302 supports_vec_gather_load_cached to target_optabs.
4303 * optabs-tree.c (init_tree_optimization_optabs): Use
4304 ggc_cleared_alloc to allocate target_optabs.
4305 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
4306 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
4308 * internal-fn.h (internal_load_fn_p): Declare.
4309 (internal_gather_scatter_fn_p): Likewise.
4310 (internal_fn_mask_index): Likewise.
4311 (internal_gather_scatter_fn_supported_p): Likewise.
4312 * internal-fn.c (gather_load_direct): New macro.
4313 (expand_gather_load_optab_fn): New function.
4314 (direct_gather_load_optab_supported_p): New macro.
4315 (direct_internal_fn_optab): New function.
4316 (internal_load_fn_p): Likewise.
4317 (internal_gather_scatter_fn_p): Likewise.
4318 (internal_fn_mask_index): Likewise.
4319 (internal_gather_scatter_fn_supported_p): Likewise.
4320 * optabs-query.c (supports_at_least_one_mode_p): New function.
4321 (supports_vec_gather_load_p): Likewise.
4322 * optabs-query.h (supports_vec_gather_load_p): Declare.
4323 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
4324 and memory_type field.
4325 (NUM_PATTERNS): Bump to 15.
4326 * tree-vect-data-refs.c: Include internal-fn.h.
4327 (vect_gather_scatter_fn_p): New function.
4328 (vect_describe_gather_scatter_call): Likewise.
4329 (vect_check_gather_scatter): Try using internal functions for
4330 gather loads. Recognize existing calls to a gather load function.
4331 (vect_analyze_data_refs): Consider using gather loads if
4332 supports_vec_gather_load_p.
4333 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
4334 (vect_get_gather_scatter_offset_type): Likewise.
4335 (vect_convert_mask_for_vectype): Likewise.
4336 (vect_add_conversion_to_patterm): Likewise.
4337 (vect_try_gather_scatter_pattern): Likewise.
4338 (vect_recog_gather_scatter_pattern): New pattern recognizer.
4339 (vect_vect_recog_func_ptrs): Add it.
4340 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
4341 internal_fn_mask_index and internal_gather_scatter_fn_p.
4342 (check_load_store_masking): Take the gather_scatter_info as an
4343 argument and handle gather loads.
4344 (vect_get_gather_scatter_ops): New function.
4345 (vectorizable_call): Check internal_load_fn_p.
4346 (vectorizable_load): Likewise. Handle gather load internal
4348 (vectorizable_store): Update call to check_load_store_masking.
4349 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
4350 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
4351 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
4352 (aarch64_gather_scale_operand_d): New predicates.
4353 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
4354 (mask_gather_load<mode>): New insns.
4356 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4357 Alan Hayward <alan.hayward@arm.com>
4358 David Sherwood <david.sherwood@arm.com>
4360 * optabs.def (fold_left_plus_optab): New optab.
4361 * doc/md.texi (fold_left_plus_@var{m}): Document.
4362 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
4363 * internal-fn.c (fold_left_direct): Define.
4364 (expand_fold_left_optab_fn): Likewise.
4365 (direct_fold_left_optab_supported_p): Likewise.
4366 * fold-const-call.c (fold_const_fold_left): New function.
4367 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
4368 * tree-parloops.c (valid_reduction_p): New function.
4369 (gather_scalar_reductions): Use it.
4370 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
4371 (vect_finish_replace_stmt): Declare.
4372 * tree-vect-loop.c (fold_left_reduction_fn): New function.
4373 (needs_fold_left_reduction_p): New function, split out from...
4374 (vect_is_simple_reduction): ...here. Accept reductions that
4375 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
4376 (vect_force_simple_reduction): Also store the reduction type in
4377 the assignment's STMT_VINFO_REDUC_TYPE.
4378 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
4379 (merge_with_identity): New function.
4380 (vect_expand_fold_left): Likewise.
4381 (vectorize_fold_left_reduction): Likewise.
4382 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
4383 scalar phi in place for it. Check for target support and reject
4384 cases that would reassociate the operation. Defer the transform
4385 phase to vectorize_fold_left_reduction.
4386 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
4387 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
4388 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
4390 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4392 * tree-if-conv.c (predicate_mem_writes): Remove redundant
4393 call to ifc_temp_var.
4395 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4396 Alan Hayward <alan.hayward@arm.com>
4397 David Sherwood <david.sherwood@arm.com>
4399 * target.def (legitimize_address_displacement): Take the original
4400 offset as a poly_int.
4401 * targhooks.h (default_legitimize_address_displacement): Update
4403 * targhooks.c (default_legitimize_address_displacement): Likewise.
4404 * doc/tm.texi: Regenerate.
4405 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
4406 as an argument, moving assert of ad->disp == ad->disp_term to...
4407 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
4408 Try calling targetm.legitimize_address_displacement before expanding
4409 the address rather than afterwards, and adjust for the new interface.
4410 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
4411 Match the new hook interface. Handle SVE addresses.
4412 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
4415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4417 * Makefile.in (OBJS): Add early-remat.o.
4418 * target.def (select_early_remat_modes): New hook.
4419 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
4420 * doc/tm.texi: Regenerate.
4421 * targhooks.h (default_select_early_remat_modes): Declare.
4422 * targhooks.c (default_select_early_remat_modes): New function.
4423 * timevar.def (TV_EARLY_REMAT): New timevar.
4424 * passes.def (pass_early_remat): New pass.
4425 * tree-pass.h (make_pass_early_remat): Declare.
4426 * early-remat.c: New file.
4427 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
4429 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
4431 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4432 Alan Hayward <alan.hayward@arm.com>
4433 David Sherwood <david.sherwood@arm.com>
4435 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
4436 vfm1 with a bound_epilog parameter.
4437 (vect_do_peeling): Update calls accordingly, and move the prologue
4438 call earlier in the function. Treat the base bound_epilog as 0 for
4439 fully-masked loops and retain vf - 1 for other loops. Add 1 to
4440 this base when peeling for gaps.
4441 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
4442 with fully-masked loops.
4443 (vect_estimate_min_profitable_iters): Handle the single peeled
4444 iteration in that case.
4446 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4447 Alan Hayward <alan.hayward@arm.com>
4448 David Sherwood <david.sherwood@arm.com>
4450 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
4451 single-element interleaving even if the size is not a power of 2.
4452 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
4453 accesses for single-element interleaving if the group size is
4456 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4457 Alan Hayward <alan.hayward@arm.com>
4458 David Sherwood <david.sherwood@arm.com>
4460 * doc/md.texi (fold_extract_last_@var{m}): Document.
4461 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
4462 * optabs.def (fold_extract_last_optab): New optab.
4463 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
4464 * internal-fn.c (fold_extract_direct): New macro.
4465 (expand_fold_extract_optab_fn): Likewise.
4466 (direct_fold_extract_optab_supported_p): Likewise.
4467 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
4468 * tree-vect-loop.c (vect_model_reduction_cost): Handle
4469 EXTRACT_LAST_REDUCTION.
4470 (get_initial_def_for_reduction): Do not create an initial vector
4471 for EXTRACT_LAST_REDUCTION reductions.
4472 (vectorizable_reduction): Leave the scalar phi in place for
4473 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
4474 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
4475 epilogue code for EXTRACT_LAST_REDUCTION and defer the
4476 transform phase to vectorizable_condition.
4477 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
4479 (vect_finish_stmt_generation): ...here.
4480 (vect_finish_replace_stmt): New function.
4481 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
4482 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
4484 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
4486 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4487 Alan Hayward <alan.hayward@arm.com>
4488 David Sherwood <david.sherwood@arm.com>
4490 * doc/md.texi (extract_last_@var{m}): Document.
4491 * optabs.def (extract_last_optab): New optab.
4492 * internal-fn.def (EXTRACT_LAST): New internal function.
4493 * internal-fn.c (cond_unary_direct): New macro.
4494 (expand_cond_unary_optab_fn): Likewise.
4495 (direct_cond_unary_optab_supported_p): Likewise.
4496 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
4497 loops using EXTRACT_LAST.
4498 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
4499 (extract_last_<mode>): ...this optab.
4500 (vec_extract<mode><Vel>): Update accordingly.
4502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4503 Alan Hayward <alan.hayward@arm.com>
4504 David Sherwood <david.sherwood@arm.com>
4506 * target.def (empty_mask_is_expensive): New hook.
4507 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
4508 * doc/tm.texi: Regenerate.
4509 * targhooks.h (default_empty_mask_is_expensive): Declare.
4510 * targhooks.c (default_empty_mask_is_expensive): New function.
4511 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
4512 if the target says that empty masks are expensive.
4513 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4515 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4517 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4518 Alan Hayward <alan.hayward@arm.com>
4519 David Sherwood <david.sherwood@arm.com>
4521 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4522 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4523 (vect_use_loop_mask_for_alignment_p): New function.
4524 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4525 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4526 niters_skip argument. Make sure that the first niters_skip elements
4527 of the first iteration are inactive.
4528 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4529 Update call to vect_set_loop_masks_directly.
4530 (get_misalign_in_elems): New function, split out from...
4531 (vect_gen_prolog_loop_niters): ...here.
4532 (vect_update_init_of_dr): Take a code argument that specifies whether
4533 the adjustment should be added or subtracted.
4534 (vect_update_init_of_drs): Likewise.
4535 (vect_prepare_for_masked_peels): New function.
4536 (vect_do_peeling): Skip prologue peeling if we're using a mask
4537 instead. Update call to vect_update_inits_of_drs.
4538 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4540 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4541 alignment. Do not include the number of peeled iterations in
4542 the minimum threshold in that case.
4543 (vectorizable_induction): Adjust the start value down by
4544 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4545 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4546 Take the number of skipped iterations into account when calculating
4548 * tree-vect-stmts.c (vect_gen_while_not): New function.
4550 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4551 Alan Hayward <alan.hayward@arm.com>
4552 David Sherwood <david.sherwood@arm.com>
4554 * doc/sourcebuild.texi (vect_fully_masked): Document.
4555 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4557 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4559 (vect_analyze_loop_2): ...here. Don't check the vectorization
4560 factor against the number of loop iterations if the loop is
4563 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4564 Alan Hayward <alan.hayward@arm.com>
4565 David Sherwood <david.sherwood@arm.com>
4567 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4568 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4569 (dump_groups): Update accordingly.
4570 (iv_use::mem_type): New member variable.
4571 (address_p): New function.
4572 (record_use): Add a mem_type argument and initialize the new
4574 (record_group_use): Add a mem_type argument. Use address_p.
4575 Remove obsolete null checks of base_object. Update call to record_use.
4576 (find_interesting_uses_op): Update call to record_group_use.
4577 (find_interesting_uses_cond): Likewise.
4578 (find_interesting_uses_address): Likewise.
4579 (get_mem_type_for_internal_fn): New function.
4580 (find_address_like_use): Likewise.
4581 (find_interesting_uses_stmt): Try find_address_like_use before
4582 calling find_interesting_uses_op.
4583 (addr_offset_valid_p): Use the iv mem_type field as the type
4584 of the addressed memory.
4585 (add_autoinc_candidates): Likewise.
4586 (get_address_cost): Likewise.
4587 (split_small_address_groups_p): Use address_p.
4588 (split_address_groups): Likewise.
4589 (add_iv_candidate_for_use): Likewise.
4590 (autoinc_possible_for_pair): Likewise.
4591 (rewrite_groups): Likewise.
4592 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4593 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4594 (get_alias_ptr_type_for_ptr_address): New function.
4595 (rewrite_use_address): Rewrite address uses in calls that were
4596 identified by find_address_like_use.
4598 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4599 Alan Hayward <alan.hayward@arm.com>
4600 David Sherwood <david.sherwood@arm.com>
4602 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4604 * gimple-expr.h (is_gimple_addressable: Likewise.
4605 * gimple-expr.c (is_gimple_address): Likewise.
4606 * internal-fn.c (expand_call_mem_ref): New function.
4607 (expand_mask_load_optab_fn): Use it.
4608 (expand_mask_store_optab_fn): Likewise.
4610 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4611 Alan Hayward <alan.hayward@arm.com>
4612 David Sherwood <david.sherwood@arm.com>
4614 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4615 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4616 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4617 (cond_umax@var{mode}): Document.
4618 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4619 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4620 (cond_umin_optab, cond_umax_optab): New optabs.
4621 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4622 (COND_IOR, COND_XOR): New internal functions.
4623 * internal-fn.h (get_conditional_internal_fn): Declare.
4624 * internal-fn.c (cond_binary_direct): New macro.
4625 (expand_cond_binary_optab_fn): Likewise.
4626 (direct_cond_binary_optab_supported_p): Likewise.
4627 (get_conditional_internal_fn): New function.
4628 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4629 Cope with reduction statements that are vectorized as calls rather
4631 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4632 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4633 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4634 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4635 (UNSPEC_COND_EOR): New unspecs.
4636 (optab): Add mappings for them.
4637 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4638 (sve_int_op, sve_fp_op): New int attributes.
4640 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4641 Alan Hayward <alan.hayward@arm.com>
4642 David Sherwood <david.sherwood@arm.com>
4644 * optabs.def (while_ult_optab): New optab.
4645 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4646 * internal-fn.def (WHILE_ULT): New internal function.
4647 * internal-fn.h (direct_internal_fn_supported_p): New override
4648 that takes two types as argument.
4649 * internal-fn.c (while_direct): New macro.
4650 (expand_while_optab_fn): New function.
4651 (convert_optab_supported_p): Likewise.
4652 (direct_while_optab_supported_p): New macro.
4653 * wide-int.h (wi::udiv_ceil): New function.
4654 * tree-vectorizer.h (rgroup_masks): New structure.
4655 (vec_loop_masks): New typedef.
4656 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4658 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4659 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4660 (vect_max_vf): New function.
4661 (slpeel_make_loop_iterate_ntimes): Delete.
4662 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4663 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4664 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4665 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4666 internal-fn.h, stor-layout.h and optabs-query.h.
4667 (vect_set_loop_mask): New function.
4668 (add_preheader_seq): Likewise.
4669 (add_header_seq): Likewise.
4670 (interleave_supported_p): Likewise.
4671 (vect_maybe_permute_loop_masks): Likewise.
4672 (vect_set_loop_masks_directly): Likewise.
4673 (vect_set_loop_condition_masked): Likewise.
4674 (vect_set_loop_condition_unmasked): New function, split out from
4675 slpeel_make_loop_iterate_ntimes.
4676 (slpeel_make_loop_iterate_ntimes): Rename to..
4677 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4678 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4679 (vect_do_peeling): Update call accordingly.
4680 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4682 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4683 mask_compare_type, can_fully_mask_p and fully_masked_p.
4684 (release_vec_loop_masks): New function.
4685 (_loop_vec_info): Use it to free the loop masks.
4686 (can_produce_all_loop_masks_p): New function.
4687 (vect_get_max_nscalars_per_iter): Likewise.
4688 (vect_verify_full_masking): Likewise.
4689 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4690 retries, and free the mask rgroups before retrying. Check loop-wide
4691 reasons for disallowing fully-masked loops. Make the final decision
4692 about whether use a fully-masked loop or not.
4693 (vect_estimate_min_profitable_iters): Do not assume that peeling
4694 for the number of iterations will be needed for fully-masked loops.
4695 (vectorizable_reduction): Disable fully-masked loops.
4696 (vectorizable_live_operation): Likewise.
4697 (vect_halve_mask_nunits): New function.
4698 (vect_double_mask_nunits): Likewise.
4699 (vect_record_loop_mask): Likewise.
4700 (vect_get_loop_mask): Likewise.
4701 (vect_transform_loop): Handle the case in which the final loop
4702 iteration might handle a partial vector. Call vect_set_loop_condition
4703 instead of slpeel_make_loop_iterate_ntimes.
4704 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4705 (check_load_store_masking): New function.
4706 (prepare_load_store_mask): Likewise.
4707 (vectorizable_store): Handle fully-masked loops.
4708 (vectorizable_load): Likewise.
4709 (supportable_widening_operation): Use vect_halve_mask_nunits for
4711 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4712 (vect_gen_while): New function.
4713 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4714 (aarch64_uqdec<mode>): New insn.
4716 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4717 Alan Hayward <alan.hayward@arm.com>
4718 David Sherwood <david.sherwood@arm.com>
4720 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4721 (reduc_xor_scal_optab): New optabs.
4722 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4723 (reduc_xor_scal_@var{m}): Document.
4724 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4725 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4727 * fold-const-call.c (fold_const_call): Handle them.
4728 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4729 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4730 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4731 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4732 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4733 (UNSPEC_XORV): New unspecs.
4734 (optab): Add entries for them.
4735 (BITWISEV): New int iterator.
4736 (bit_reduc_op): New int attributes.
4738 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4739 Alan Hayward <alan.hayward@arm.com>
4740 David Sherwood <david.sherwood@arm.com>
4742 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4743 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4744 * optabs.def (vec_shl_insert_optab): New optab.
4745 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4746 (duplicate_and_interleave): Likewise.
4747 * tree-vect-loop.c: Include internal-fn.h.
4748 (neutral_op_for_slp_reduction): New function, split out from
4749 get_initial_defs_for_reduction.
4750 (get_initial_def_for_reduction): Handle option 2 for variable-length
4751 vectors by loading the neutral value into a vector and then shifting
4752 the initial value into element 0.
4753 (get_initial_defs_for_reduction): Replace the code argument with
4754 the neutral value calculated by neutral_op_for_slp_reduction.
4755 Use gimple_build_vector for constant-length vectors.
4756 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4757 but the first group_size elements have a neutral value.
4758 Use duplicate_and_interleave otherwise.
4759 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4760 Update call to get_initial_defs_for_reduction. Handle SLP
4761 reductions for variable-length vectors by creating one vector
4762 result for each scalar result, with the elements associated
4763 with other scalar results stubbed out with the neutral value.
4764 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4765 Require IFN_VEC_SHL_INSERT for double reductions on
4766 variable-length vectors, or SLP reductions that have
4767 a neutral value. Require can_duplicate_and_interleave_p
4768 support for variable-length unchained SLP reductions if there
4769 is no neutral value, such as for MIN/MAX reductions. Also require
4770 the number of vector elements to be a multiple of the number of
4771 SLP statements when doing variable-length unchained SLP reductions.
4772 Update call to vect_create_epilog_for_reduction.
4773 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4774 and remove initial values.
4775 (duplicate_and_interleave): Make public.
4776 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4777 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4779 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4780 Alan Hayward <alan.hayward@arm.com>
4781 David Sherwood <david.sherwood@arm.com>
4783 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4784 (can_duplicate_and_interleave_p): New function.
4785 (vect_get_and_check_slp_defs): Take the vector of statements
4786 rather than just the current one. Remove excess parentheses.
4787 Restriction rejectinon of vect_constant_def and vect_external_def
4788 for variable-length vectors to boolean types, or types for which
4789 can_duplicate_and_interleave_p is false.
4790 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4791 (duplicate_and_interleave): New function.
4792 (vect_get_constant_vectors): Use gimple_build_vector for
4793 constant-length vectors and suitable variable-length constant
4794 vectors. Use duplicate_and_interleave for other variable-length
4795 vectors. Don't defer the update when inserting new statements.
4797 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4798 Alan Hayward <alan.hayward@arm.com>
4799 David Sherwood <david.sherwood@arm.com>
4801 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4802 min_profitable_iters doesn't go negative.
4804 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4805 Alan Hayward <alan.hayward@arm.com>
4806 David Sherwood <david.sherwood@arm.com>
4808 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4809 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4810 * optabs.def (vec_mask_load_lanes_optab): New optab.
4811 (vec_mask_store_lanes_optab): Likewise.
4812 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4813 (MASK_STORE_LANES): Likewise.
4814 * internal-fn.c (mask_load_lanes_direct): New macro.
4815 (mask_store_lanes_direct): Likewise.
4816 (expand_mask_load_optab_fn): Handle masked operations.
4817 (expand_mask_load_lanes_optab_fn): New macro.
4818 (expand_mask_store_optab_fn): Handle masked operations.
4819 (expand_mask_store_lanes_optab_fn): New macro.
4820 (direct_mask_load_lanes_optab_supported_p): Likewise.
4821 (direct_mask_store_lanes_optab_supported_p): Likewise.
4822 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4824 (vect_load_lanes_supported): Likewise.
4825 * tree-vect-data-refs.c (strip_conversion): New function.
4826 (can_group_stmts_p): Likewise.
4827 (vect_analyze_data_ref_accesses): Use it instead of checking
4828 for a pair of assignments.
4829 (vect_store_lanes_supported): Take a masked_p parameter.
4830 (vect_load_lanes_supported): Likewise.
4831 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4832 vect_store_lanes_supported and vect_load_lanes_supported.
4833 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4834 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4835 parameter. Don't allow gaps for masked accesses.
4836 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4837 and vect_load_lanes_supported.
4838 (get_load_store_type): Take a masked_p parameter and update
4839 call to get_group_load_store_type.
4840 (vectorizable_store): Update call to get_load_store_type.
4841 Handle IFN_MASK_STORE_LANES.
4842 (vectorizable_load): Update call to get_load_store_type.
4843 Handle IFN_MASK_LOAD_LANES.
4845 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4846 Alan Hayward <alan.hayward@arm.com>
4847 David Sherwood <david.sherwood@arm.com>
4849 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4851 * config/aarch64/aarch64-protos.h
4852 (aarch64_sve_struct_memory_operand_p): Declare.
4853 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4854 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4855 (VPRED, vpred): Handle SVE structure modes.
4856 * config/aarch64/constraints.md (Utx): New constraint.
4857 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4858 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4859 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4860 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4861 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4862 structure modes. Split into pieces after RA.
4863 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4864 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4866 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4867 SVE structure modes.
4868 (aarch64_classify_address): Likewise.
4869 (sizetochar): Move earlier in file.
4870 (aarch64_print_operand): Handle SVE register lists.
4871 (aarch64_array_mode): New function.
4872 (aarch64_sve_struct_memory_operand_p): Likewise.
4873 (TARGET_ARRAY_MODE): Redefine.
4875 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4876 Alan Hayward <alan.hayward@arm.com>
4877 David Sherwood <david.sherwood@arm.com>
4879 * target.def (array_mode): New target hook.
4880 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4881 * doc/tm.texi: Regenerate.
4882 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4883 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4884 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4886 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4889 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4890 Alan Hayward <alan.hayward@arm.com>
4891 David Sherwood <david.sherwood@arm.com>
4893 * fold-const.c (fold_binary_loc): Check the argument types
4894 rather than the result type when testing for a vector operation.
4896 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4898 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4899 * doc/tm.texi: Regenerate.
4901 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4902 Alan Hayward <alan.hayward@arm.com>
4903 David Sherwood <david.sherwood@arm.com>
4905 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4906 (sve): Document new AArch64 extension.
4907 * doc/md.texi (w): Extend the description of the AArch64
4908 constraint to include SVE vectors.
4909 (Upl, Upa): Document new AArch64 predicate constraints.
4910 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4912 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4913 (msve-vector-bits=): New option.
4914 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4915 SVE when these are disabled.
4916 (sve): New extension.
4917 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4918 modes. Adjust their number of units based on aarch64_sve_vg.
4919 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4920 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4921 aarch64_addr_query_type.
4922 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4923 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4924 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4925 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4926 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4927 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4928 (aarch64_simd_imm_zero_p): Delete.
4929 (aarch64_check_zero_based_sve_index_immediate): Declare.
4930 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4931 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4932 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4933 (aarch64_sve_float_mul_immediate_p): Likewise.
4934 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4936 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4937 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4938 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4939 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4940 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4941 (aarch64_regmode_natural_size): Likewise.
4942 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4943 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4945 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4946 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4947 for VG and the SVE predicate registers.
4948 (V_ALIASES): Add a "z"-prefixed alias.
4949 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4950 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4951 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4952 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4953 (REG_CLASS_NAMES): Add entries for them.
4954 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4955 and the predicate registers.
4956 (aarch64_sve_vg): Declare.
4957 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4958 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4959 (REGMODE_NATURAL_SIZE): Define.
4960 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4962 * config/aarch64/aarch64.c: Include cfgrtl.h.
4963 (simd_immediate_info): Add a constructor for series vectors,
4964 and an associated step field.
4965 (aarch64_sve_vg): New variable.
4966 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4967 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4968 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4969 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4970 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4971 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4972 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4973 (aarch64_get_mask_mode): New functions.
4974 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4975 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4976 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4977 predicate modes and predicate registers. Explicitly restrict
4978 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4979 to store a vector mode if it is recognized by
4980 aarch64_classify_vector_mode.
4981 (aarch64_regmode_natural_size): New function.
4982 (aarch64_hard_regno_caller_save_mode): Return the original mode
4984 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4985 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4986 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4987 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4989 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4990 does not overlap dest if the function is frame-related. Handle
4992 (aarch64_split_add_offset): New function.
4993 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4994 them aarch64_add_offset.
4995 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4996 and update call to aarch64_sub_sp.
4997 (aarch64_add_cfa_expression): New function.
4998 (aarch64_expand_prologue): Pass extra temporary registers to the
4999 functions above. Handle the case in which we need to emit new
5000 DW_CFA_expressions for registers that were originally saved
5001 relative to the stack pointer, but now have to be expressed
5002 relative to the frame pointer.
5003 (aarch64_output_mi_thunk): Pass extra temporary registers to the
5005 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
5006 IP0 and IP1 values for SVE frames.
5007 (aarch64_expand_vec_series): New function.
5008 (aarch64_expand_sve_widened_duplicate): Likewise.
5009 (aarch64_expand_sve_const_vector): Likewise.
5010 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
5011 Handle SVE constants. Use emit_move_insn to move a force_const_mem
5012 into the register, rather than emitting a SET directly.
5013 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
5014 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
5015 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
5016 (offset_9bit_signed_scaled_p): New functions.
5017 (aarch64_replicate_bitmask_imm): New function.
5018 (aarch64_bitmask_imm): Use it.
5019 (aarch64_cannot_force_const_mem): Reject expressions involving
5020 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
5021 (aarch64_classify_index): Handle SVE indices, by requiring
5022 a plain register index with a scale that matches the element size.
5023 (aarch64_classify_address): Handle SVE addresses. Assert that
5024 the mode of the address is VOIDmode or an integer mode.
5025 Update call to aarch64_classify_symbol.
5026 (aarch64_classify_symbolic_expression): Update call to
5027 aarch64_classify_symbol.
5028 (aarch64_const_vec_all_in_range_p): New function.
5029 (aarch64_print_vector_float_operand): Likewise.
5030 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
5031 "vN" for FP registers with SVE modes. Handle (const ...) vectors
5032 and the FP immediates 1.0 and 0.5.
5033 (aarch64_print_address_internal): Handle SVE addresses.
5034 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
5035 (aarch64_regno_regclass): Handle predicate registers.
5036 (aarch64_secondary_reload): Handle big-endian reloads of SVE
5038 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
5039 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
5040 (aarch64_convert_sve_vector_bits): New function.
5041 (aarch64_override_options): Use it to handle -msve-vector-bits=.
5042 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
5044 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
5045 Handle SVE vector and predicate modes. Accept VL-based constants
5046 that need only one temporary register, and VL offsets that require
5047 no temporary registers.
5048 (aarch64_conditional_register_usage): Mark the predicate registers
5049 as fixed if SVE isn't available.
5050 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
5051 Return true for SVE vector and predicate modes.
5052 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
5053 rather than an unsigned int. Handle SVE modes.
5054 (aarch64_preferred_simd_mode): Update call accordingly. Handle
5056 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
5058 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
5059 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
5060 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
5061 (aarch64_sve_float_mul_immediate_p): New functions.
5062 (aarch64_sve_valid_immediate): New function.
5063 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
5064 Explicitly reject structure modes. Check for INDEX constants.
5065 Handle PTRUE and PFALSE constants.
5066 (aarch64_check_zero_based_sve_index_immediate): New function.
5067 (aarch64_simd_imm_zero_p): Delete.
5068 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
5069 vector modes. Accept constants in the range of CNT[BHWD].
5070 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
5071 ask for an Advanced SIMD mode.
5072 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
5073 (aarch64_simd_vector_alignment): Handle SVE predicates.
5074 (aarch64_vectorize_preferred_vector_alignment): New function.
5075 (aarch64_simd_vector_alignment_reachable): Use it instead of
5077 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
5078 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
5080 (MAX_VECT_LEN): Delete.
5081 (expand_vec_perm_d): Add a vec_flags field.
5082 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
5083 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
5084 (aarch64_evpc_ext): Don't apply a big-endian lane correction
5086 (aarch64_evpc_rev): Rename to...
5087 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
5088 (aarch64_evpc_rev_global): New function.
5089 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
5090 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
5092 (aarch64_evpc_sve_tbl): New function.
5093 (aarch64_expand_vec_perm_const_1): Update after rename of
5094 aarch64_evpc_rev. Handle SVE permutes too, trying
5095 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
5096 than aarch64_evpc_tbl.
5097 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
5098 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
5099 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
5100 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
5101 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
5102 (aarch64_expand_sve_vcond): New functions.
5103 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
5104 of aarch64_vector_mode_p.
5105 (aarch64_dwarf_poly_indeterminate_value): New function.
5106 (aarch64_compute_pressure_classes): Likewise.
5107 (aarch64_can_change_mode_class): Likewise.
5108 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
5109 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
5110 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
5111 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
5112 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
5113 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
5114 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
5115 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
5117 (Dn, Dl, Dr): Accept const as well as const_vector.
5118 (Dz): Likewise. Compare against CONST0_RTX.
5119 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
5120 of "vector" where appropriate.
5121 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
5122 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
5123 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
5124 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
5125 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
5126 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
5127 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
5128 (v_int_equiv): Extend to SVE modes.
5129 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
5131 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
5132 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
5133 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
5134 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
5135 (SVE_COND_FP_CMP): New int iterators.
5136 (perm_hilo): Handle the new unpack unspecs.
5137 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
5139 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
5140 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
5141 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
5142 (aarch64_equality_operator, aarch64_constant_vector_operand)
5143 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
5144 (aarch64_sve_nonimmediate_operand): Likewise.
5145 (aarch64_sve_general_operand): Likewise.
5146 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
5147 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
5148 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
5149 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
5150 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
5151 (aarch64_sve_float_arith_immediate): Likewise.
5152 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
5153 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
5154 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
5155 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
5156 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
5157 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
5158 (aarch64_sve_float_arith_operand): Likewise.
5159 (aarch64_sve_float_arith_with_sub_operand): Likewise.
5160 (aarch64_sve_float_mul_operand): Likewise.
5161 (aarch64_sve_vec_perm_operand): Likewise.
5162 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
5163 (aarch64_mov_operand): Accept const_poly_int and const_vector.
5164 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
5165 as well as const_vector.
5166 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
5167 in file. Use CONST0_RTX and CONSTM1_RTX.
5168 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
5169 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
5170 Use aarch64_simd_imm_zero.
5171 * config/aarch64/aarch64-sve.md: New file.
5172 * config/aarch64/aarch64.md: Include it.
5173 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
5174 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
5175 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
5176 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
5177 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
5178 (sve): New attribute.
5179 (enabled): Disable instructions with the sve attribute unless
5181 (movqi, movhi): Pass CONST_POLY_INT operaneds through
5182 aarch64_expand_mov_immediate.
5183 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
5184 CNT[BHSD] immediates.
5185 (movti): Split CONST_POLY_INT moves into two halves.
5186 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
5187 Split additions that need a temporary here if the destination
5188 is the stack pointer.
5189 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
5190 (*add<mode>3_poly_1): New instruction.
5191 (set_clobber_cc): New expander.
5193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
5195 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
5196 parameter and use it instead of GET_MODE_SIZE (innermode). Use
5197 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
5198 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
5199 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
5200 Change innermode from fixed_mode_size to machine_mode.
5201 (simplify_subreg): Update call accordingly. Handle a constant-sized
5202 subreg of a variable-length CONST_VECTOR.
5204 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
5205 Alan Hayward <alan.hayward@arm.com>
5206 David Sherwood <david.sherwood@arm.com>
5208 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
5209 (add_offset_to_base): New function, split out from...
5210 (create_mem_ref): ...here. When handling a scale other than 1,
5211 check first whether the address is valid without the offset.
5212 Add it into the base if so, leaving the index and scale as-is.
5214 2018-01-12 Jakub Jelinek <jakub@redhat.com>
5217 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
5218 fold_for_warn before checking if arg2 is INTEGER_CST.
5220 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
5222 * config/rs6000/predicates.md (load_multiple_operation): Delete.
5223 (store_multiple_operation): Delete.
5224 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
5225 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
5226 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
5227 guarded by TARGET_STRING.
5228 (rs6000_output_load_multiple): Delete.
5229 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
5230 OPTION_MASK_STRING / TARGET_STRING handling.
5231 (print_operand) <'N', 'O'>: Add comment that these are unused now.
5232 (const rs6000_opt_masks) <"string">: Change mask to 0.
5233 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
5234 (MASK_STRING): Delete.
5235 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
5237 (load_multiple): Delete.
5244 (store_multiple): Delete.
5251 (movmemsi_8reg): Delete.
5252 (corresponding unnamed define_insn): Delete.
5253 (movmemsi_6reg): Delete.
5254 (corresponding unnamed define_insn): Delete.
5255 (movmemsi_4reg): Delete.
5256 (corresponding unnamed define_insn): Delete.
5257 (movmemsi_2reg): Delete.
5258 (corresponding unnamed define_insn): Delete.
5259 (movmemsi_1reg): Delete.
5260 (corresponding unnamed define_insn): Delete.
5261 * config/rs6000/rs6000.opt (mno-string): New.
5262 (mstring): Replace by deprecation warning stub.
5263 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
5265 2018-01-12 Jakub Jelinek <jakub@redhat.com>
5267 * regrename.c (regrename_do_replace): If replacing the same
5268 reg multiple times, try to reuse last created gen_raw_REG.
5271 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
5272 main to workaround a bug in GDB.
5274 2018-01-12 Tom de Vries <tom@codesourcery.com>
5277 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
5279 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
5281 PR rtl-optimization/80481
5282 * ira-color.c (get_cap_member): New function.
5283 (allocnos_conflict_by_live_ranges_p): Use it.
5284 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
5285 (setup_slot_coalesced_allocno_live_ranges): Ditto.
5287 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
5290 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
5291 (*saddl_se_1): Ditto.
5293 (*ssubl_se_1): Ditto.
5295 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5297 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
5298 rather than wi::to_widest for DR_INITs.
5299 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
5300 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
5301 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
5303 (vect_analyze_group_access_1): Note that here.
5305 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5307 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
5308 polynomial type sizes.
5310 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5312 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
5313 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
5314 (gimple_add_tmp_var): Likewise.
5316 2018-01-12 Martin Liska <mliska@suse.cz>
5318 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
5319 (gimple_alloc_sizes): Likewise.
5320 (dump_gimple_statistics): Use PRIu64 in printf format.
5321 * gimple.h: Change uint64_t to int.
5323 2018-01-12 Martin Liska <mliska@suse.cz>
5325 * tree-core.h: Use uint64_t instead of int.
5326 * tree.c (tree_node_counts): Likewise.
5327 (tree_node_sizes): Likewise.
5328 (dump_tree_statistics): Use PRIu64 in printf format.
5330 2018-01-12 Martin Liska <mliska@suse.cz>
5332 * Makefile.in: As qsort_chk is implemented in vec.c, add
5333 vec.o to linkage of gencfn-macros.
5334 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
5335 passing the info to record_node_allocation_statistics.
5336 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
5338 * ggc-common.c (struct ggc_usage): Add operator== and use
5339 it in operator< and compare function.
5340 * mem-stats.h (struct mem_usage): Likewise.
5341 * vec.c (struct vec_usage): Remove operator< and compare
5342 function. Can be simply inherited.
5344 2018-01-12 Martin Jambor <mjambor@suse.cz>
5347 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
5348 * tree-ssa-math-opts.c: Include domwalk.h.
5349 (convert_mult_to_fma_1): New function.
5350 (fma_transformation_info): New type.
5351 (fma_deferring_state): Likewise.
5352 (cancel_fma_deferring): New function.
5353 (result_of_phi): Likewise.
5354 (last_fma_candidate_feeds_initial_phi): Likewise.
5355 (convert_mult_to_fma): Added deferring logic, split actual
5356 transformation to convert_mult_to_fma_1.
5357 (math_opts_dom_walker): New type.
5358 (math_opts_dom_walker::after_dom_children): New method, body moved
5359 here from pass_optimize_widening_mul::execute, added deferring logic
5361 (pass_optimize_widening_mul::execute): Moved most of code to
5362 math_opts_dom_walker::after_dom_children.
5363 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
5364 * config/i386/i386.c (ix86_option_override_internal): Added
5365 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
5367 2018-01-12 Richard Biener <rguenther@suse.de>
5370 * dwarf2out.c (gen_variable_die): Do not reset old_die for
5371 inline instance vars.
5373 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
5376 * config/rx/rx.c (rx_is_restricted_memory_address):
5379 2018-01-12 Richard Biener <rguenther@suse.de>
5381 PR tree-optimization/80846
5382 * target.def (split_reduction): New target hook.
5383 * targhooks.c (default_split_reduction): New function.
5384 * targhooks.h (default_split_reduction): Declare.
5385 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
5386 target requests first reduce vectors by combining low and high
5388 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
5389 (get_vectype_for_scalar_type_and_size): Export.
5390 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
5391 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
5392 * doc/tm.texi: Regenerate.
5393 * config/i386/i386.c (ix86_split_reduction): Implement
5394 TARGET_VECTORIZE_SPLIT_REDUCTION.
5396 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5399 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
5400 in PIC mode except for TARGET_VXWORKS_RTP.
5401 * config/sparc/sparc.c: Include cfgrtl.h.
5402 (TARGET_INIT_PIC_REG): Define.
5403 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
5404 (sparc_pic_register_p): New predicate.
5405 (sparc_legitimate_address_p): Use it.
5406 (sparc_legitimize_pic_address): Likewise.
5407 (sparc_delegitimize_address): Likewise.
5408 (sparc_mode_dependent_address_p): Likewise.
5409 (gen_load_pcrel_sym): Remove 4th parameter.
5410 (load_got_register): Adjust call to above. Remove obsolete stuff.
5411 (sparc_expand_prologue): Do not call load_got_register here.
5412 (sparc_flat_expand_prologue): Likewise.
5413 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
5414 (sparc_use_pseudo_pic_reg): New function.
5415 (sparc_init_pic_reg): Likewise.
5416 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5417 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
5419 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
5421 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
5422 Add item for branch_cost.
5424 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5426 PR rtl-optimization/83565
5427 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
5428 not extend the result to a larger mode for rotate operations.
5429 (num_sign_bit_copies1): Likewise.
5431 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5434 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
5436 Use values-Xc.o for -pedantic.
5437 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
5439 2018-01-12 Martin Liska <mliska@suse.cz>
5442 * ipa-devirt.c (final_warning_record::grow_type_warnings):
5444 (possible_polymorphic_call_targets): Use it.
5445 (ipa_devirt): Likewise.
5447 2018-01-12 Martin Liska <mliska@suse.cz>
5449 * profile-count.h (enum profile_quality): Use 0 as invalid
5450 enum value of profile_quality.
5452 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
5454 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
5455 -mext-string options.
5457 2018-01-12 Richard Biener <rguenther@suse.de>
5459 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
5460 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
5461 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
5463 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
5465 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
5467 * configure.ac (--with-long-double-format): Add support for the
5468 configuration option to change the default long double format on
5470 * config.gcc (powerpc*-linux*-*): Likewise.
5471 * configure: Regenerate.
5472 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
5473 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
5474 used without modification.
5476 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5478 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
5479 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
5480 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
5481 MISC_BUILTIN_SPEC_BARRIER.
5482 (rs6000_init_builtins): Likewise.
5483 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
5485 (speculation_barrier): New define_insn.
5486 * doc/extend.texi: Document __builtin_speculation_barrier.
5488 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5491 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
5492 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
5493 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
5495 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
5496 integral modes instead of "ss" and "sd".
5497 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
5498 vectors with 32-bit and 64-bit elements.
5499 (vecdupssescalarmodesuffix): New mode attribute.
5500 (vec_dup<mode>): Use it.
5502 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
5505 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
5506 frame if argument is passed on stack.
5508 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5511 * ree.c (combine_reaching_defs): Optimize also
5512 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
5513 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5515 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5518 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5520 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5523 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5524 after they are computed.
5526 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5528 PR tree-optimization/83695
5529 * gimple-loop-linterchange.cc
5530 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5531 reset cached scev information after interchange.
5532 (pass_linterchange::execute): Remove call to scev_reset_htab.
5534 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5536 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5537 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5538 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5539 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5540 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5541 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5542 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5543 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5544 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5545 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5546 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5547 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5548 (V_lane_reg): Likewise.
5549 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5551 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5552 (vfmal_lane_low<mode>_intrinsic,
5553 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5554 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5555 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5556 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5557 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5558 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5560 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5562 * config/arm/arm-cpus.in (fp16fml): New feature.
5563 (ALL_SIMD): Add fp16fml.
5564 (armv8.2-a): Add fp16fml as an option.
5565 (armv8.3-a): Likewise.
5566 (armv8.4-a): Add fp16fml as part of fp16.
5567 * config/arm/arm.h (TARGET_FP16FML): Define.
5568 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5570 * config/arm/arm-modes.def (V2HF): Define.
5571 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5572 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5573 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5574 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5575 vfmsl_low, vfmsl_high): New set of builtins.
5576 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5577 (vfml_op): New code attribute.
5578 (VFMLHALVES): New int iterator.
5579 (VFML, VFMLSEL): New mode attributes.
5580 (V_reg): Define mapping for V2HF.
5581 (V_hi, V_lo): New mode attributes.
5582 (VF_constraint): Likewise.
5583 (vfml_half, vfml_half_selector): New int attributes.
5584 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5586 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5587 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5589 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5590 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5591 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5592 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5594 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5595 Document new effective target and option set.
5597 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5599 * config/arm/arm-cpus.in (armv8_4): New feature.
5600 (ARMv8_4a): New fgroup.
5601 (armv8.4-a): New arch.
5602 * config/arm/arm-tables.opt: Regenerate.
5603 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5604 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5605 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5606 Add matching rules for -march=armv8.4-a and extensions.
5607 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5609 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5612 * config/rx/rx.md (BW): New mode attribute.
5613 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5615 2018-01-11 Richard Biener <rguenther@suse.de>
5617 PR tree-optimization/83435
5618 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5619 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5620 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5622 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5623 Alan Hayward <alan.hayward@arm.com>
5624 David Sherwood <david.sherwood@arm.com>
5626 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5628 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5629 (aarch64_print_address_internal): Use it to check for a zero offset.
5631 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5632 Alan Hayward <alan.hayward@arm.com>
5633 David Sherwood <david.sherwood@arm.com>
5635 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5636 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5637 Return a poly_int64 rather than a HOST_WIDE_INT.
5638 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5639 rather than a HOST_WIDE_INT.
5640 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5641 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5642 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5643 final_offset from HOST_WIDE_INT to poly_int64.
5644 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5645 to_constant when getting the number of units in an Advanced SIMD
5647 (aarch64_builtin_vectorized_function): Check for a constant number
5649 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5651 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5652 attribute instead of GET_MODE_NUNITS.
5653 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5654 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5655 GET_MODE_SIZE for fixed-size registers.
5656 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5657 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5658 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5659 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5660 (aarch64_print_operand, aarch64_print_address_internal)
5661 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5662 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5663 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5664 Handle polynomial GET_MODE_SIZE.
5665 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5666 wider than SImode without modification.
5667 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5668 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5669 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5670 passing and returning SVE modes.
5671 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5672 rather than GEN_INT.
5673 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5674 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5675 (aarch64_allocate_and_probe_stack_space): Likewise.
5676 (aarch64_layout_frame): Cope with polynomial offsets.
5677 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5678 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5680 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5681 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5682 poly_int64 rather than a HOST_WIDE_INT.
5683 (aarch64_get_separate_components, aarch64_process_components)
5684 (aarch64_expand_prologue, aarch64_expand_epilogue)
5685 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5686 (aarch64_anchor_offset): New function, split out from...
5687 (aarch64_legitimize_address): ...here.
5688 (aarch64_builtin_vectorization_cost): Handle polynomial
5689 TYPE_VECTOR_SUBPARTS.
5690 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5692 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5693 number of elements from the PARALLEL rather than the mode.
5694 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5695 rather than GET_MODE_BITSIZE.
5696 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5697 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5698 (aarch64_expand_vec_perm_const_1): Handle polynomial
5699 d->perm.length () and d->perm elements.
5700 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5701 Apply to_constant to d->perm elements.
5702 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5703 polynomial CONST_VECTOR_NUNITS.
5704 (aarch64_move_pointer): Take amount as a poly_int64 rather
5706 (aarch64_progress_pointer): Avoid temporary variable.
5707 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5708 the mode attribute instead of GET_MODE.
5710 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5711 Alan Hayward <alan.hayward@arm.com>
5712 David Sherwood <david.sherwood@arm.com>
5714 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5715 x exists before using it.
5716 (aarch64_add_constant_internal): Rename to...
5717 (aarch64_add_offset_1): ...this. Replace regnum with separate
5718 src and dest rtxes. Handle the case in which they're different,
5719 including when the offset is zero. Replace scratchreg with an rtx.
5720 Use 2 additions if there is no spare register into which we can
5721 move a 16-bit constant.
5722 (aarch64_add_constant): Delete.
5723 (aarch64_add_offset): Replace reg with separate src and dest
5724 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5725 Use aarch64_add_offset_1.
5726 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5727 an rtx rather than an int. Take the delta as a poly_int64
5728 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5729 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5730 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5731 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5732 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5734 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5735 aarch64_add_constant.
5737 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5739 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5740 Use scalar_float_mode.
5742 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5744 * config/aarch64/aarch64-simd.md
5745 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5746 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5747 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5748 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5749 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5750 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5751 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5752 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5753 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5754 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5756 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5759 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5760 targ_options->x_arm_arch_string is non NULL.
5762 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5764 * config/aarch64/aarch64.h
5765 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5767 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5770 * expmed.c (emit_store_flag_force): Swap if const op0
5771 and change VOIDmode to mode of op0.
5773 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5775 PR rtl-optimization/83761
5776 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5777 than bytes to mode_for_size.
5779 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5782 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5783 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5786 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5789 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5790 when in layout mode.
5791 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5792 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5795 2018-01-10 Michael Collison <michael.collison@arm.com>
5797 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5798 * config/aarch64/aarch64-option-extension.def: Add
5799 AARCH64_OPT_EXTENSION of 'fp16fml'.
5800 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5801 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5802 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5803 * config/aarch64/constraints.md (Ui7): New constraint.
5804 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5805 (VFMLA_SEL_W): Ditto.
5808 (VFMLA16_LOW): New int iterator.
5809 (VFMLA16_HIGH): Ditto.
5810 (UNSPEC_FMLAL): New unspec.
5811 (UNSPEC_FMLSL): Ditto.
5812 (UNSPEC_FMLAL2): Ditto.
5813 (UNSPEC_FMLSL2): Ditto.
5814 (f16mac): New code attribute.
5815 * config/aarch64/aarch64-simd-builtins.def
5816 (aarch64_fmlal_lowv2sf): Ditto.
5817 (aarch64_fmlsl_lowv2sf): Ditto.
5818 (aarch64_fmlalq_lowv4sf): Ditto.
5819 (aarch64_fmlslq_lowv4sf): Ditto.
5820 (aarch64_fmlal_highv2sf): Ditto.
5821 (aarch64_fmlsl_highv2sf): Ditto.
5822 (aarch64_fmlalq_highv4sf): Ditto.
5823 (aarch64_fmlslq_highv4sf): Ditto.
5824 (aarch64_fmlal_lane_lowv2sf): Ditto.
5825 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5826 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5827 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5828 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5829 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5830 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5831 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5832 (aarch64_fmlal_lane_highv2sf): Ditto.
5833 (aarch64_fmlsl_lane_highv2sf): Ditto.
5834 (aarch64_fmlal_laneq_highv2sf): Ditto.
5835 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5836 (aarch64_fmlalq_lane_highv4sf): Ditto.
5837 (aarch64_fmlsl_lane_highv4sf): Ditto.
5838 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5839 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5840 * config/aarch64/aarch64-simd.md:
5841 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5842 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5843 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5844 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5845 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5846 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5847 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5848 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5849 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5850 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5851 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5852 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5853 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5854 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5855 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5856 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5857 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5858 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5859 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5860 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5861 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5862 (vfmlsl_low_u32): Ditto.
5863 (vfmlalq_low_u32): Ditto.
5864 (vfmlslq_low_u32): Ditto.
5865 (vfmlal_high_u32): Ditto.
5866 (vfmlsl_high_u32): Ditto.
5867 (vfmlalq_high_u32): Ditto.
5868 (vfmlslq_high_u32): Ditto.
5869 (vfmlal_lane_low_u32): Ditto.
5870 (vfmlsl_lane_low_u32): Ditto.
5871 (vfmlal_laneq_low_u32): Ditto.
5872 (vfmlsl_laneq_low_u32): Ditto.
5873 (vfmlalq_lane_low_u32): Ditto.
5874 (vfmlslq_lane_low_u32): Ditto.
5875 (vfmlalq_laneq_low_u32): Ditto.
5876 (vfmlslq_laneq_low_u32): Ditto.
5877 (vfmlal_lane_high_u32): Ditto.
5878 (vfmlsl_lane_high_u32): Ditto.
5879 (vfmlal_laneq_high_u32): Ditto.
5880 (vfmlsl_laneq_high_u32): Ditto.
5881 (vfmlalq_lane_high_u32): Ditto.
5882 (vfmlslq_lane_high_u32): Ditto.
5883 (vfmlalq_laneq_high_u32): Ditto.
5884 (vfmlslq_laneq_high_u32): Ditto.
5885 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5886 (AARCH64_FL_FOR_ARCH8_4): New.
5887 (AARCH64_ISA_F16FML): New ISA flag.
5888 (TARGET_F16FML): New feature flag for fp16fml.
5889 (doc/invoke.texi): Document new fp16fml option.
5891 2018-01-10 Michael Collison <michael.collison@arm.com>
5893 * config/aarch64/aarch64-builtins.c:
5894 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5895 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5896 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5897 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5898 (AARCH64_ISA_SHA3): New ISA flag.
5899 (TARGET_SHA3): New feature flag for sha3.
5900 * config/aarch64/iterators.md (sha512_op): New int attribute.
5901 (CRYPTO_SHA512): New int iterator.
5902 (UNSPEC_SHA512H): New unspec.
5903 (UNSPEC_SHA512H2): Ditto.
5904 (UNSPEC_SHA512SU0): Ditto.
5905 (UNSPEC_SHA512SU1): Ditto.
5906 * config/aarch64/aarch64-simd-builtins.def
5907 (aarch64_crypto_sha512hqv2di): New builtin.
5908 (aarch64_crypto_sha512h2qv2di): Ditto.
5909 (aarch64_crypto_sha512su0qv2di): Ditto.
5910 (aarch64_crypto_sha512su1qv2di): Ditto.
5911 (aarch64_eor3qv8hi): Ditto.
5912 (aarch64_rax1qv2di): Ditto.
5913 (aarch64_xarqv2di): Ditto.
5914 (aarch64_bcaxqv8hi): Ditto.
5915 * config/aarch64/aarch64-simd.md:
5916 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5917 (aarch64_crypto_sha512su0qv2di): Ditto.
5918 (aarch64_crypto_sha512su1qv2di): Ditto.
5919 (aarch64_eor3qv8hi): Ditto.
5920 (aarch64_rax1qv2di): Ditto.
5921 (aarch64_xarqv2di): Ditto.
5922 (aarch64_bcaxqv8hi): Ditto.
5923 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5924 (vsha512h2q_u64): Ditto.
5925 (vsha512su0q_u64): Ditto.
5926 (vsha512su1q_u64): Ditto.
5927 (veor3q_u16): Ditto.
5928 (vrax1q_u64): Ditto.
5930 (vbcaxq_u16): Ditto.
5931 * config/arm/types.md (crypto_sha512): New type attribute.
5932 (crypto_sha3): Ditto.
5933 (doc/invoke.texi): Document new sha3 option.
5935 2018-01-10 Michael Collison <michael.collison@arm.com>
5937 * config/aarch64/aarch64-builtins.c:
5938 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5939 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5940 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5941 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5942 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5943 (AARCH64_ISA_SM4): New ISA flag.
5944 (TARGET_SM4): New feature flag for sm4.
5945 * config/aarch64/aarch64-simd-builtins.def
5946 (aarch64_sm3ss1qv4si): Ditto.
5947 (aarch64_sm3tt1aq4si): Ditto.
5948 (aarch64_sm3tt1bq4si): Ditto.
5949 (aarch64_sm3tt2aq4si): Ditto.
5950 (aarch64_sm3tt2bq4si): Ditto.
5951 (aarch64_sm3partw1qv4si): Ditto.
5952 (aarch64_sm3partw2qv4si): Ditto.
5953 (aarch64_sm4eqv4si): Ditto.
5954 (aarch64_sm4ekeyqv4si): Ditto.
5955 * config/aarch64/aarch64-simd.md:
5956 (aarch64_sm3ss1qv4si): Ditto.
5957 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5958 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5959 (aarch64_sm4eqv4si): Ditto.
5960 (aarch64_sm4ekeyqv4si): Ditto.
5961 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5962 (sm3part_op): Ditto.
5963 (CRYPTO_SM3TT): Ditto.
5964 (CRYPTO_SM3PART): Ditto.
5965 (UNSPEC_SM3SS1): New unspec.
5966 (UNSPEC_SM3TT1A): Ditto.
5967 (UNSPEC_SM3TT1B): Ditto.
5968 (UNSPEC_SM3TT2A): Ditto.
5969 (UNSPEC_SM3TT2B): Ditto.
5970 (UNSPEC_SM3PARTW1): Ditto.
5971 (UNSPEC_SM3PARTW2): Ditto.
5972 (UNSPEC_SM4E): Ditto.
5973 (UNSPEC_SM4EKEY): Ditto.
5974 * config/aarch64/constraints.md (Ui2): New constraint.
5975 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5976 * config/arm/types.md (crypto_sm3): New type attribute.
5977 (crypto_sm4): Ditto.
5978 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5979 (vsm3tt1aq_u32): Ditto.
5980 (vsm3tt1bq_u32): Ditto.
5981 (vsm3tt2aq_u32): Ditto.
5982 (vsm3tt2bq_u32): Ditto.
5983 (vsm3partw1q_u32): Ditto.
5984 (vsm3partw2q_u32): Ditto.
5985 (vsm4eq_u32): Ditto.
5986 (vsm4ekeyq_u32): Ditto.
5987 (doc/invoke.texi): Document new sm4 option.
5989 2018-01-10 Michael Collison <michael.collison@arm.com>
5991 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5992 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5993 (AARCH64_FL_FOR_ARCH8_4): New.
5994 (AARCH64_FL_V8_4): New flag.
5995 (doc/invoke.texi): Document new armv8.4-a option.
5997 2018-01-10 Michael Collison <michael.collison@arm.com>
5999 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
6000 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
6001 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
6002 * config/aarch64/aarch64-option-extension.def: Add
6003 AARCH64_OPT_EXTENSION of 'sha2'.
6004 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
6005 (crypto): Disable sha2 and aes if crypto disabled.
6006 (crypto): Enable aes and sha2 if enabled.
6007 (simd): Disable sha2 and aes if simd disabled.
6008 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
6010 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
6011 (TARGET_SHA2): New feature flag for sha2.
6012 (TARGET_AES): New feature flag for aes.
6013 * config/aarch64/aarch64-simd.md:
6014 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
6015 conditional on TARGET_AES.
6016 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
6017 (aarch64_crypto_sha1hsi): Make pattern conditional
6019 (aarch64_crypto_sha1hv4si): Ditto.
6020 (aarch64_be_crypto_sha1hv4si): Ditto.
6021 (aarch64_crypto_sha1su1v4si): Ditto.
6022 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
6023 (aarch64_crypto_sha1su0v4si): Ditto.
6024 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
6025 (aarch64_crypto_sha256su0v4si): Ditto.
6026 (aarch64_crypto_sha256su1v4si): Ditto.
6027 (doc/invoke.texi): Document new aes and sha2 options.
6029 2018-01-10 Martin Sebor <msebor@redhat.com>
6031 PR tree-optimization/83781
6032 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
6035 2018-01-11 Martin Sebor <msebor@gmail.com>
6036 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6038 PR tree-optimization/83501
6039 PR tree-optimization/81703
6041 * tree-ssa-strlen.c (get_string_cst): Rename...
6042 (get_string_len): ...to this. Handle global constants.
6043 (handle_char_store): Adjust.
6045 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
6046 Jim Wilson <jimw@sifive.com>
6048 * config/riscv/riscv-protos.h (riscv_output_return): New.
6049 * config/riscv/riscv.c (struct machine_function): New naked_p field.
6050 (riscv_attribute_table, riscv_output_return),
6051 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
6052 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
6053 (riscv_compute_frame_info): Only compute frame->mask if not a naked
6055 (riscv_expand_prologue): Add early return for naked function.
6056 (riscv_expand_epilogue): Likewise.
6057 (riscv_function_ok_for_sibcall): Return false for naked function.
6058 (riscv_set_current_function): New.
6059 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
6060 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
6061 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
6062 * doc/extend.texi (RISC-V Function Attributes): New.
6064 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
6066 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
6067 check for 128-bit long double before checking TCmode.
6068 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
6069 128-bit long doubles before checking TFmode or TCmode.
6070 (FLOAT128_IBM_P): Likewise.
6072 2018-01-10 Martin Sebor <msebor@redhat.com>
6074 PR tree-optimization/83671
6075 * builtins.c (c_strlen): Unconditionally return zero for the empty
6077 Use -Warray-bounds for warnings.
6078 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
6079 for non-constant array indices with COMPONENT_REF, arrays of
6080 arrays, and pointers to arrays.
6081 (gimple_fold_builtin_strlen): Determine and set length range for
6082 non-constant character arrays.
6084 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
6087 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
6090 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
6092 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
6094 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
6097 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
6098 VECTOR_MEM_ALTIVEC_OR_VSX_P.
6099 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
6100 indexed_or_indirect_operand predicate.
6101 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
6102 (*vsx_le_perm_load_v8hi): Likewise.
6103 (*vsx_le_perm_load_v16qi): Likewise.
6104 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
6105 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
6106 (*vsx_le_perm_store_v8hi): Likewise.
6107 (*vsx_le_perm_store_v16qi): Likewise.
6108 (eight unnamed splitters): Likewise.
6110 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
6112 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
6113 * config/rs6000/emmintrin.h: Likewise.
6114 * config/rs6000/mmintrin.h: Likewise.
6115 * config/rs6000/xmmintrin.h: Likewise.
6117 2018-01-10 David Malcolm <dmalcolm@redhat.com>
6120 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
6122 * tree.c (tree_nop_conversion): Return true for location wrapper
6124 (maybe_wrap_with_location): New function.
6125 (selftest::check_strip_nops): New function.
6126 (selftest::test_location_wrappers): New function.
6127 (selftest::tree_c_tests): Call it.
6128 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
6129 (maybe_wrap_with_location): New decl.
6130 (EXPR_LOCATION_WRAPPER_P): New macro.
6131 (location_wrapper_p): New inline function.
6132 (tree_strip_any_location_wrapper): New inline function.
6134 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
6137 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
6138 stack_realign_offset for the largest alignment of stack slot
6140 (ix86_find_max_used_stack_alignment): New function.
6141 (ix86_finalize_stack_frame_flags): Use it. Set
6142 max_used_stack_alignment if we don't realign stack.
6143 * config/i386/i386.h (machine_function): Add
6144 max_used_stack_alignment.
6146 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
6148 * config/arm/arm.opt (-mbranch-cost): New option.
6149 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
6152 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
6155 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
6156 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
6158 2018-01-10 Richard Biener <rguenther@suse.de>
6161 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
6162 early out so it also covers the case where we have a non-NULL
6165 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
6167 PR tree-optimization/83753
6168 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
6169 for non-strided grouped accesses if the number of elements is 1.
6171 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
6174 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
6175 * i386.h (TARGET_USE_GATHER): Define.
6176 * x86-tune.def (X86_TUNE_USE_GATHER): New.
6178 2018-01-10 Martin Liska <mliska@suse.cz>
6181 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
6182 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
6184 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
6185 CLEANUP_NO_PARTITIONING is not set.
6187 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
6189 * doc/rtl.texi: Remove documentation of (const ...) wrappers
6190 for vectors, as a partial revert of r254296.
6191 * rtl.h (const_vec_p): Delete.
6192 (const_vec_duplicate_p): Don't test for vector CONSTs.
6193 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
6194 * expmed.c (make_tree): Likewise.
6197 * common.md (E, F): Use CONSTANT_P instead of checking for
6199 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
6200 checking for CONST_VECTOR.
6202 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6205 * predict.c (force_edge_cold): Handle in more sane way edges
6208 2018-01-09 Carl Love <cel@us.ibm.com>
6210 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
6212 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
6213 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
6214 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
6215 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
6216 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
6217 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
6218 * config/rs6000/rs6000-protos.h: Add extern defition for
6219 rs6000_generate_float2_double_code.
6220 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
6222 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
6223 (float2_v2df): Add define_expand.
6225 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
6228 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
6229 op_mode in the force_to_mode call.
6231 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6233 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
6234 instead of checking each element individually.
6235 (aarch64_evpc_uzp): Likewise.
6236 (aarch64_evpc_zip): Likewise.
6237 (aarch64_evpc_ext): Likewise.
6238 (aarch64_evpc_rev): Likewise.
6239 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
6240 instead of checking each element individually. Return true without
6242 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
6243 whether all selected elements come from the same input, instead of
6244 checking each element individually. Remove calls to gen_rtx_REG,
6245 start_sequence and end_sequence and instead assert that no rtl is
6248 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6250 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
6251 order of HIGH and CONST checks.
6253 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6255 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
6256 if the destination isn't an SSA_NAME.
6258 2018-01-09 Richard Biener <rguenther@suse.de>
6260 PR tree-optimization/83668
6261 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
6263 (canonicalize_loop_form): ... here, renamed from ...
6264 (canonicalize_loop_closed_ssa_form): ... this and amended to
6265 swap successor edges for loop exit blocks to make us use
6266 the RPO order we need for initial schedule generation.
6268 2018-01-09 Joseph Myers <joseph@codesourcery.com>
6270 PR tree-optimization/64811
6271 * match.pd: When optimizing comparisons with Inf, avoid
6272 introducing or losing exceptions from comparisons with NaN.
6274 2018-01-09 Martin Liska <mliska@suse.cz>
6277 * asan.c (shadow_mem_size): Add gcc_assert.
6279 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
6281 Don't save registers in main().
6284 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
6285 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
6286 * config/avr/avr.c (avr_set_current_function): Don't error if
6287 naked, OS_task or OS_main are specified at the same time.
6288 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
6290 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
6292 * common/config/avr/avr-common.c (avr_option_optimization_table):
6293 Switch on -mmain-is-OS_task for optimizing compilations.
6295 2018-01-09 Richard Biener <rguenther@suse.de>
6297 PR tree-optimization/83572
6298 * graphite.c: Include cfganal.h.
6299 (graphite_transform_loops): Connect infinite loops to exit
6300 and remove fake edges at the end.
6302 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6304 * ipa-inline.c (edge_badness): Revert accidental checkin.
6306 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6309 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
6310 symbols; not inline clones.
6312 2018-01-09 Jakub Jelinek <jakub@redhat.com>
6315 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
6316 hard registers. Formatting fixes.
6318 PR preprocessor/83722
6319 * gcc.c (try_generate_repro): Pass
6320 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
6321 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
6324 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
6325 Kito Cheng <kito.cheng@gmail.com>
6327 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
6328 (riscv_leaf_function_p): Delete.
6329 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
6331 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6333 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
6335 (do_ifelse): New function.
6336 (do_isel): New function.
6337 (do_sub3): New function.
6338 (do_add3): New function.
6339 (do_load_mask_compare): New function.
6340 (do_overlap_load_compare): New function.
6341 (expand_compare_loop): New function.
6342 (expand_block_compare): Call expand_compare_loop() when appropriate.
6343 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
6345 (-mblock-compare-inline-loop-limit): New option.
6347 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6350 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
6351 Reverse order of second and third operands in first alternative.
6352 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
6353 of first and second elements in UNSPEC_VPERMR vector.
6354 (altivec_expand_vec_perm_le): Likewise.
6356 2018-01-08 Jeff Law <law@redhat.com>
6358 PR rtl-optimizatin/81308
6359 * tree-switch-conversion.c (cfg_altered): New file scoped static.
6360 (process_switch): If group_case_labels makes a change, then set
6362 (pass_convert_switch::execute): If a switch is converted, then
6363 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
6365 PR rtl-optimization/81308
6366 * recog.c (split_all_insns): Conditionally cleanup the CFG after
6369 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
6371 PR target/83663 - Revert r255946
6372 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
6373 generation for cases where splatting a value is not useful.
6374 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
6375 across a vec_duplicate and a paradoxical subreg forming a vector
6376 mode to a vec_concat.
6378 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
6380 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
6381 -march=armv8.3-a variants.
6382 * config/arm/t-multilib: Likewise.
6383 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
6385 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6387 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
6389 (cceq_ior_compare_complement): Give it a name so I can use it, and
6390 change boolean_or_operator predicate to boolean_operator so it can
6391 be used to generate a crand.
6392 (eqne): New code iterator.
6393 (bd/bd_neg): New code_attrs.
6394 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
6395 a single define_insn.
6396 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
6397 decrement (bdnzt/bdnzf/bdzt/bdzf).
6398 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
6399 with the new names of the branch decrement patterns, and added the
6400 names of the branch decrement conditional patterns.
6402 2018-01-08 Richard Biener <rguenther@suse.de>
6404 PR tree-optimization/83563
6405 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
6408 2018-01-08 Richard Biener <rguenther@suse.de>
6411 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
6413 2018-01-08 Richard Biener <rguenther@suse.de>
6415 PR tree-optimization/83685
6416 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
6417 references to abnormals.
6419 2018-01-08 Richard Biener <rguenther@suse.de>
6422 * dwarf2out.c (output_indirect_strings): Handle empty
6423 skeleton_debug_str_hash.
6424 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
6426 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6428 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
6429 (emit_store_direct): Likewise.
6430 (arc_trampoline_adjust_address): Likewise.
6431 (arc_asm_trampoline_template): New function.
6432 (arc_initialize_trampoline): Use asm_trampoline_template.
6433 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
6434 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
6435 * config/arc/arc.md (flush_icache): Delete pattern.
6437 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6439 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
6440 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
6443 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6446 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
6447 by not USED_FOR_TARGET.
6448 (make_pass_resolve_sw_modes): Likewise.
6450 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6452 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
6455 2018-01-08 Richard Biener <rguenther@suse.de>
6458 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
6460 2018-01-08 Richard Biener <rguenther@suse.de>
6463 * match.pd ((t * 2) / 2) -> t): Add missing :c.
6465 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
6468 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
6469 basic blocks with a small number of successors.
6470 (convert_control_dep_chain_into_preds): Improve handling of
6472 (dump_predicates): Split apart into...
6473 (dump_pred_chain): ...here...
6474 (dump_pred_info): ...and here.
6475 (can_one_predicate_be_invalidated_p): Add debugging printfs.
6476 (can_chain_union_be_invalidated_p): Improve check for invalidation
6478 (uninit_uses_cannot_happen): Avoid unnecessary if
6479 convert_control_dep_chain_into_preds yielded nothing.
6481 2018-01-06 Martin Sebor <msebor@redhat.com>
6483 PR tree-optimization/83640
6484 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
6485 subtracting negative offset from size.
6486 (builtin_access::overlap): Adjust offset bounds of the access to fall
6487 within the size of the object if possible.
6489 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
6491 PR rtl-optimization/83699
6492 * expmed.c (extract_bit_field_1): Restrict the vector usage of
6493 extract_bit_field_as_subreg to cases in which the extracted
6494 value is also a vector.
6496 * lra-constraints.c (process_alt_operands): Test for the equivalence
6497 substitutions when detecting a possible reload cycle.
6499 2018-01-06 Jakub Jelinek <jakub@redhat.com>
6502 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
6503 by default if flag_selective_schedling{,2}. Formatting fixes.
6505 PR rtl-optimization/83682
6506 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
6507 if it has non-VECTOR_MODE element mode.
6508 (vec_duplicate_p): Likewise.
6511 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
6512 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6514 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6517 * config/i386/i386-builtin.def
6518 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6519 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6520 Require also OPTION_MASK_ISA_AVX512F in addition to
6521 OPTION_MASK_ISA_GFNI.
6522 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6523 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6524 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6525 to OPTION_MASK_ISA_GFNI.
6526 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6527 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6528 OPTION_MASK_ISA_AVX512BW.
6529 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6530 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6531 addition to OPTION_MASK_ISA_GFNI.
6532 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6533 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6534 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6535 to OPTION_MASK_ISA_GFNI.
6536 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6537 a requirement for all ISAs rather than any of them with a few
6539 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6541 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6542 bitmasks to be enabled with 3 exceptions, instead of requiring any
6543 enabled ISA with lots of exceptions.
6544 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6545 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6546 Change avx512bw in isa attribute to avx512f.
6547 * config/i386/sgxintrin.h: Add license boilerplate.
6548 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6549 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6550 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6551 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6553 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6554 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6555 temporarily sse2 rather than sse if not enabled already.
6558 * config/i386/sse.md (VI248_VLBW): Rename to ...
6559 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6560 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6561 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6562 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6563 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6564 mode iterator instead of VI248_VLBW.
6566 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6568 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6569 (record_modified): Skip clobbers; add debug output.
6570 (param_change_prob): Use sreal frequencies.
6572 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6574 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6575 punt for user-aligned variables.
6577 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6579 * tree-chrec.c (chrec_contains_symbols): Return true for
6582 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6585 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6586 of (x|y) == x for BICS pattern.
6588 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6590 PR tree-optimization/83605
6591 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6592 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6595 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6597 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6598 * config/epiphany/rtems.h: New file.
6600 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6601 Uros Bizjak <ubizjak@gmail.com>
6604 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6605 QIreg_operand instead of register_operand predicate.
6606 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6607 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6608 comments instead of -fmitigate[-_]rop.
6610 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6613 * cgraphunit.c (symbol_table::compile): Switch to text_section
6614 before calling assembly_start debug hook.
6615 * run-rtl-passes.c (run_rtl_passes): Likewise.
6618 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6620 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6621 range_int_cst_p rather than !symbolic_range_p before calling
6622 extract_range_from_multiplicative_op_1.
6624 2018-01-04 Jeff Law <law@redhat.com>
6626 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6627 redundant test in assertion.
6629 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6631 * doc/rtl.texi: Document machine_mode wrapper classes.
6633 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6635 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6638 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6640 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6641 the VEC_PERM_EXPR fold to fail.
6643 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6646 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6647 to switched_sections.
6649 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6652 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6655 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6658 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6659 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6661 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6664 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6665 is BLKmode and bitpos not zero or mode change is needed.
6667 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6670 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6673 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6676 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6677 instead of MULT rtx. Update all corresponding splitters.
6679 (*ssub<modesuffix>): Ditto.
6681 (*cmp_sadd_di): Update split patterns.
6682 (*cmp_sadd_si): Ditto.
6683 (*cmp_sadd_sidi): Ditto.
6684 (*cmp_ssub_di): Ditto.
6685 (*cmp_ssub_si): Ditto.
6686 (*cmp_ssub_sidi): Ditto.
6687 * config/alpha/predicates.md (const23_operand): New predicate.
6688 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6689 Look for ASHIFT, not MULT inner operand.
6690 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6692 2018-01-04 Martin Liska <mliska@suse.cz>
6694 PR gcov-profile/83669
6695 * gcov.c (output_intermediate_file): Add version to intermediate
6697 * doc/gcov.texi: Document new field 'version' in intermediate
6698 file format. Fix location of '-k' option of gcov command.
6700 2018-01-04 Martin Liska <mliska@suse.cz>
6703 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6705 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6707 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6709 2018-01-03 Martin Sebor <msebor@redhat.com>
6711 PR tree-optimization/83655
6712 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6713 checking calls with invalid arguments.
6715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6717 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6718 (vectorizable_mask_load_store): Delete.
6719 (vectorizable_call): Return false for masked loads and stores.
6720 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6721 instead of gimple_assign_rhs1.
6722 (vectorizable_load): Handle IFN_MASK_LOAD.
6723 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6727 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6729 (vectorizable_mask_load_store): ...here.
6730 (vectorizable_load): ...and here.
6732 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6734 * tree-vect-stmts.c (vect_build_all_ones_mask)
6735 (vect_build_zero_merge_argument): New functions, split out from...
6736 (vectorizable_load): ...here.
6738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6740 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6742 (vectorizable_mask_load_store): ...here.
6743 (vectorizable_store): ...and here.
6745 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6747 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6749 (vectorizable_mask_load_store): ...here.
6751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6753 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6754 (vect_model_store_cost): Take a vec_load_store_type instead of a
6756 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6757 (vect_model_store_cost): Take a vec_load_store_type instead of a
6759 (vectorizable_mask_load_store): Update accordingly.
6760 (vectorizable_store): Likewise.
6761 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6763 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6765 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6766 IFN_MASK_LOAD calls here rather than...
6767 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6769 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6770 Alan Hayward <alan.hayward@arm.com>
6771 David Sherwood <david.sherwood@arm.com>
6773 * expmed.c (extract_bit_field_1): For vector extracts,
6774 fall back to extract_bit_field_as_subreg if vec_extract
6777 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6778 Alan Hayward <alan.hayward@arm.com>
6779 David Sherwood <david.sherwood@arm.com>
6781 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6782 they are variable or constant sized.
6783 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6784 slots for constant-sized data.
6786 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6787 Alan Hayward <alan.hayward@arm.com>
6788 David Sherwood <david.sherwood@arm.com>
6790 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6791 handling COND_EXPRs with boolean comparisons, try to find a better
6792 basis for the mask type than the boolean itself.
6794 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6796 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6797 is calculated and how it can be overridden.
6798 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6799 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6801 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6804 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6805 Alan Hayward <alan.hayward@arm.com>
6806 David Sherwood <david.sherwood@arm.com>
6808 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6809 Remove the mode argument.
6810 (aarch64_simd_valid_immediate): Remove the mode and inverse
6812 * config/aarch64/iterators.md (bitsize): New iterator.
6813 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6814 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6815 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6816 aarch64_simd_valid_immediate.
6817 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6818 (aarch64_reg_or_bic_imm): Likewise.
6819 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6820 with an insn_type enum and msl with a modifier_type enum.
6821 Replace element_width with a scalar_mode. Change the shift
6822 to unsigned int. Add constructors for scalar_float_mode and
6823 scalar_int_mode elements.
6824 (aarch64_vect_float_const_representable_p): Delete.
6825 (aarch64_can_const_movi_rtx_p)
6826 (aarch64_simd_scalar_immediate_valid_for_move)
6827 (aarch64_simd_make_constant): Update call to
6828 aarch64_simd_valid_immediate.
6829 (aarch64_advsimd_valid_immediate_hs): New function.
6830 (aarch64_advsimd_valid_immediate): Likewise.
6831 (aarch64_simd_valid_immediate): Remove mode and inverse
6832 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6833 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6834 and aarch64_float_const_representable_p on the result.
6835 (aarch64_output_simd_mov_immediate): Remove mode argument.
6836 Update call to aarch64_simd_valid_immediate and use of
6837 simd_immediate_info.
6838 (aarch64_output_scalar_simd_mov_immediate): Update call
6841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6842 Alan Hayward <alan.hayward@arm.com>
6843 David Sherwood <david.sherwood@arm.com>
6845 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6846 (mode_nunits): Likewise CONST_MODE_NUNITS.
6847 * machmode.def (ADJUST_NUNITS): Document.
6848 * genmodes.c (mode_data::need_nunits_adj): New field.
6849 (blank_mode): Update accordingly.
6850 (adj_nunits): New variable.
6851 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6853 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6854 listed in adj_nunits.
6855 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6856 listed in adj_nunits. Don't emit case statements for such modes.
6857 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6858 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6859 nothing if adj_nunits is nonnull.
6860 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6861 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6862 (emit_mode_fbit): Update use of print_maybe_const_decl.
6863 (emit_move_size): Likewise. Treat the array as non-const
6865 (emit_mode_adjustments): Handle adj_nunits.
6867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6869 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6870 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6871 (VECTOR_MODES): Use it.
6872 (make_vector_modes): Take the prefix as an argument.
6874 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6875 Alan Hayward <alan.hayward@arm.com>
6876 David Sherwood <david.sherwood@arm.com>
6878 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6879 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6880 for MODE_VECTOR_BOOL.
6881 * machmode.def (VECTOR_BOOL_MODE): Document.
6882 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6883 (make_vector_bool_mode): New function.
6884 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6886 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6887 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6889 * stor-layout.c (int_mode_for_mode): Likewise.
6890 * tree.c (build_vector_type_for_mode): Likewise.
6891 * varasm.c (output_constant_pool_2): Likewise.
6892 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6893 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6894 for MODE_VECTOR_BOOL.
6895 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6896 of mode class checks.
6897 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6898 instead of a list of mode class checks.
6899 (expand_vector_scalar_condition): Likewise.
6900 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6903 Alan Hayward <alan.hayward@arm.com>
6904 David Sherwood <david.sherwood@arm.com>
6906 * machmode.h (mode_size): Change from unsigned short to
6908 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6909 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6910 or if measurement_type is not polynomial.
6911 (fixed_size_mode::includes_p): Check for constant-sized modes.
6912 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6913 return a poly_uint16 rather than an unsigned short.
6914 (emit_mode_size): Change the type of mode_size from unsigned short
6915 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6916 (emit_mode_adjustments): Cope with polynomial vector sizes.
6917 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6919 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6921 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6922 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6923 * caller-save.c (setup_save_areas): Likewise.
6924 (replace_reg_with_saved_mem): Likewise.
6925 * calls.c (emit_library_call_value_1): Likewise.
6926 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6927 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6928 (gen_lowpart_for_combine): Likewise.
6929 * convert.c (convert_to_integer_1): Likewise.
6930 * cse.c (equiv_constant, cse_insn): Likewise.
6931 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6932 (cselib_subst_to_values): Likewise.
6933 * dce.c (word_dce_process_block): Likewise.
6934 * df-problems.c (df_word_lr_mark_ref): Likewise.
6935 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6936 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6937 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6938 (rtl_for_decl_location): Likewise.
6939 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6940 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6941 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6942 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6943 (expand_expr_real_1): Likewise.
6944 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6945 (pad_below): Likewise.
6946 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6947 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6948 * ira.c (get_subreg_tracking_sizes): Likewise.
6949 * ira-build.c (ira_create_allocno_objects): Likewise.
6950 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6951 (ira_sort_regnos_for_alter_reg): Likewise.
6952 * ira-costs.c (record_operand_costs): Likewise.
6953 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6954 (resolve_simple_move): Likewise.
6955 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6956 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6957 (lra_constraints): Likewise.
6958 (CONST_POOL_OK_P): Reject variable-sized modes.
6959 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6960 (add_pseudo_to_slot, lra_spill): Likewise.
6961 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6962 * optabs-query.c (get_best_extraction_insn): Likewise.
6963 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6964 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6965 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6966 * recog.c (offsettable_address_addr_space_p): Likewise.
6967 * regcprop.c (maybe_mode_change): Likewise.
6968 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6969 * regrename.c (build_def_use): Likewise.
6970 * regstat.c (dump_reg_info): Likewise.
6971 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6972 (find_reloads, find_reloads_subreg_address): Likewise.
6973 * reload1.c (eliminate_regs_1): Likewise.
6974 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6975 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6976 (simplify_binary_operation_1, simplify_subreg): Likewise.
6977 * targhooks.c (default_function_arg_padding): Likewise.
6978 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6979 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6980 (verify_gimple_assign_ternary): Likewise.
6981 * tree-inline.c (estimate_move_cost): Likewise.
6982 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6983 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6984 (get_address_cost_ainc): Likewise.
6985 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6986 (vect_supportable_dr_alignment): Likewise.
6987 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6988 (vectorizable_reduction): Likewise.
6989 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6990 (vectorizable_operation, vectorizable_load): Likewise.
6991 * tree.c (build_same_sized_truth_vector_type): Likewise.
6992 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6993 * var-tracking.c (emit_note_insn_var_location): Likewise.
6994 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6995 (ADDR_VEC_ALIGN): Likewise.
6997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6998 Alan Hayward <alan.hayward@arm.com>
6999 David Sherwood <david.sherwood@arm.com>
7001 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
7003 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
7004 or if measurement_type is polynomial.
7005 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
7006 * combine.c (make_extraction): Likewise.
7007 * dse.c (find_shift_sequence): Likewise.
7008 * dwarf2out.c (mem_loc_descriptor): Likewise.
7009 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
7010 (extract_bit_field, extract_low_bits): Likewise.
7011 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
7012 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
7013 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
7014 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
7015 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
7016 * reload.c (find_reloads): Likewise.
7017 * reload1.c (alter_reg): Likewise.
7018 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
7019 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
7020 * tree-if-conv.c (predicate_mem_writes): Likewise.
7021 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
7022 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
7023 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
7024 * valtrack.c (dead_debug_insert_temp): Likewise.
7025 * varasm.c (mergeable_constant_section): Likewise.
7026 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
7028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7029 Alan Hayward <alan.hayward@arm.com>
7030 David Sherwood <david.sherwood@arm.com>
7032 * expr.c (expand_assignment): Cope with polynomial mode sizes
7033 when assigning to a CONCAT.
7035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7036 Alan Hayward <alan.hayward@arm.com>
7037 David Sherwood <david.sherwood@arm.com>
7039 * machmode.h (mode_precision): Change from unsigned short to
7041 (mode_to_precision): Return a poly_uint16 rather than an unsigned
7043 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
7044 or if measurement_type is not polynomial.
7045 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
7046 in which the mode is already known to be a scalar_int_mode.
7047 * genmodes.c (emit_mode_precision): Change the type of mode_precision
7048 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
7050 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
7051 for GET_MODE_PRECISION.
7052 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
7053 for GET_MODE_PRECISION.
7054 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
7056 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
7057 (expand_field_assignment, make_extraction): Likewise.
7058 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
7059 (get_last_value): Likewise.
7060 * convert.c (convert_to_integer_1): Likewise.
7061 * cse.c (cse_insn): Likewise.
7062 * expr.c (expand_expr_real_1): Likewise.
7063 * lra-constraints.c (simplify_operand_subreg): Likewise.
7064 * optabs-query.c (can_atomic_load_p): Likewise.
7065 * optabs.c (expand_atomic_load): Likewise.
7066 (expand_atomic_store): Likewise.
7067 * ree.c (combine_reaching_defs): Likewise.
7068 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
7069 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
7070 * tree.h (type_has_mode_precision_p): Likewise.
7071 * ubsan.c (instrument_si_overflow): Likewise.
7073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7074 Alan Hayward <alan.hayward@arm.com>
7075 David Sherwood <david.sherwood@arm.com>
7077 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
7078 polynomial numbers of units.
7079 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
7080 (valid_vector_subparts_p): New function.
7081 (build_vector_type): Remove temporary shim and take the number
7082 of units as a poly_uint64 rather than an int.
7083 (build_opaque_vector_type): Take the number of units as a
7084 poly_uint64 rather than an int.
7085 * tree.c (build_vector_from_ctor): Handle polynomial
7086 TYPE_VECTOR_SUBPARTS.
7087 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
7088 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
7089 (build_vector_from_val): If the number of units is variable,
7090 use build_vec_duplicate_cst for constant operands and
7091 VEC_DUPLICATE_EXPR otherwise.
7092 (make_vector_type): Remove temporary is_constant ().
7093 (build_vector_type, build_opaque_vector_type): Take the number of
7094 units as a poly_uint64 rather than an int.
7095 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
7097 * cfgexpand.c (expand_debug_expr): Likewise.
7098 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
7099 (store_constructor, expand_expr_real_1): Likewise.
7100 (const_scalar_mask_from_tree): Likewise.
7101 * fold-const-call.c (fold_const_reduction): Likewise.
7102 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
7103 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
7104 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
7105 (fold_relational_const): Likewise.
7106 (native_interpret_vector): Likewise. Change the size from an
7107 int to an unsigned int.
7108 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
7109 TYPE_VECTOR_SUBPARTS.
7110 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
7111 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
7112 duplicating a non-constant operand into a variable-length vector.
7113 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
7114 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
7115 * ipa-icf.c (sem_variable::equals): Likewise.
7116 * match.pd: Likewise.
7117 * omp-simd-clone.c (simd_clone_subparts): Likewise.
7118 * print-tree.c (print_node): Likewise.
7119 * stor-layout.c (layout_type): Likewise.
7120 * targhooks.c (default_builtin_vectorization_cost): Likewise.
7121 * tree-cfg.c (verify_gimple_comparison): Likewise.
7122 (verify_gimple_assign_binary): Likewise.
7123 (verify_gimple_assign_ternary): Likewise.
7124 (verify_gimple_assign_single): Likewise.
7125 * tree-pretty-print.c (dump_generic_node): Likewise.
7126 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7127 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
7128 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
7129 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
7130 (vect_shift_permute_load_chain): Likewise.
7131 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
7132 (expand_vector_condition, optimize_vector_constructor): Likewise.
7133 (lower_vec_perm, get_compute_type): Likewise.
7134 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
7135 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
7136 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
7137 (vect_recog_mask_conversion_pattern): Likewise.
7138 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
7139 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
7140 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7141 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
7142 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
7143 (vectorizable_shift, vectorizable_operation, vectorizable_store)
7144 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
7145 (supportable_widening_operation): Likewise.
7146 (supportable_narrowing_operation): Likewise.
7147 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
7149 * varasm.c (output_constant): Likewise.
7151 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7152 Alan Hayward <alan.hayward@arm.com>
7153 David Sherwood <david.sherwood@arm.com>
7155 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
7156 so that both the length == 3 and length != 3 cases set up their
7157 own permute vectors. Add comments explaining why we know the
7158 number of elements is constant.
7159 (vect_permute_load_chain): Likewise.
7161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7162 Alan Hayward <alan.hayward@arm.com>
7163 David Sherwood <david.sherwood@arm.com>
7165 * machmode.h (mode_nunits): Change from unsigned char to
7167 (ONLY_FIXED_SIZE_MODES): New macro.
7168 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
7169 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
7170 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
7172 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
7173 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
7174 or if measurement_type is not polynomial.
7175 * genmodes.c (ZERO_COEFFS): New macro.
7176 (emit_mode_nunits_inline): Make mode_nunits_inline return a
7178 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
7179 Use ZERO_COEFFS when emitting initializers.
7180 * data-streamer.h (bp_pack_poly_value): New function.
7181 (bp_unpack_poly_value): Likewise.
7182 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
7183 for GET_MODE_NUNITS.
7184 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
7185 for GET_MODE_NUNITS.
7186 * tree.c (make_vector_type): Remove temporary shim and make
7187 the real function take the number of units as a poly_uint64
7189 (build_vector_type_for_mode): Handle polynomial nunits.
7190 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
7191 * emit-rtl.c (const_vec_series_p_1): Likewise.
7192 (gen_rtx_CONST_VECTOR): Likewise.
7193 * fold-const.c (test_vec_duplicate_folding): Likewise.
7194 * genrecog.c (validate_pattern): Likewise.
7195 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
7196 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
7197 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
7198 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
7199 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
7200 * rtlanal.c (subreg_get_info): Likewise.
7201 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7202 (vect_grouped_load_supported): Likewise.
7203 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
7204 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
7205 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
7206 (simplify_const_unary_operation, simplify_binary_operation_1)
7207 (simplify_const_binary_operation, simplify_ternary_operation)
7208 (test_vector_ops_duplicate, test_vector_ops): Likewise.
7209 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
7210 instead of CONST_VECTOR_NUNITS.
7211 * varasm.c (output_constant_pool_2): Likewise.
7212 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
7213 explicit-encoded elements in the XVEC for variable-length vectors.
7215 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7217 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
7219 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7220 Alan Hayward <alan.hayward@arm.com>
7221 David Sherwood <david.sherwood@arm.com>
7223 * coretypes.h (fixed_size_mode): Declare.
7224 (fixed_size_mode_pod): New typedef.
7225 * builtins.h (target_builtins::x_apply_args_mode)
7226 (target_builtins::x_apply_result_mode): Change type to
7227 fixed_size_mode_pod.
7228 * builtins.c (apply_args_size, apply_result_size, result_vector)
7229 (expand_builtin_apply_args_1, expand_builtin_apply)
7230 (expand_builtin_return): Update accordingly.
7232 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7234 * cse.c (hash_rtx_cb): Hash only the encoded elements.
7235 * cselib.c (cselib_hash_rtx): Likewise.
7236 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
7237 CONST_VECTOR encoding.
7239 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7240 Jeff Law <law@redhat.com>
7243 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
7244 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
7245 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
7246 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
7249 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
7250 explicitly probe *sp in a noreturn function if there were any callee
7251 register saves or frame pointer is needed.
7253 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7256 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
7257 BLKmode for ternary, binary or unary expressions.
7260 * var-tracking.c (delete_vta_debug_insn): New inline function.
7261 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
7262 insns from get_insns () to NULL instead of each bb separately.
7263 Use delete_vta_debug_insn. No longer static.
7264 (vt_debug_insns_local, variable_tracking_main_1): Adjust
7265 delete_vta_debug_insns callers.
7266 * rtl.h (delete_vta_debug_insns): Declare.
7267 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
7268 instead of variable_tracking_main.
7270 2018-01-03 Martin Sebor <msebor@redhat.com>
7272 PR tree-optimization/83603
7273 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
7274 arguments past the endof the argument list in functions declared
7275 without a prototype.
7276 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
7277 Avoid checking when arguments are null.
7279 2018-01-03 Martin Sebor <msebor@redhat.com>
7282 * doc/extend.texi (attribute const): Fix a typo.
7283 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
7284 issuing -Wsuggest-attribute for void functions.
7286 2018-01-03 Martin Sebor <msebor@redhat.com>
7288 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
7289 offset_int::from instead of wide_int::to_shwi.
7290 (maybe_diag_overlap): Remove assertion.
7291 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
7292 * gimple-ssa-sprintf.c (format_directive): Same.
7293 (parse_directive): Same.
7294 (sprintf_dom_walker::compute_format_length): Same.
7295 (try_substitute_return_value): Same.
7297 2018-01-03 Jeff Law <law@redhat.com>
7300 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
7301 non-constant residual for zero at runtime and avoid probing in
7302 that case. Reorganize code for trailing problem to mirror handling
7305 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
7307 PR tree-optimization/83501
7308 * tree-ssa-strlen.c (get_string_cst): New.
7309 (handle_char_store): Call get_string_cst.
7311 2018-01-03 Martin Liska <mliska@suse.cz>
7313 PR tree-optimization/83593
7314 * tree-ssa-strlen.c: Include tree-cfg.h.
7315 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
7316 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
7317 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
7319 (strlen_dom_walker::before_dom_children): Call
7320 gimple_purge_dead_eh_edges. Dump tranformation with details
7322 (strlen_dom_walker::before_dom_children): Update call by adding
7323 new argument cleanup_eh.
7324 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
7326 2018-01-03 Martin Liska <mliska@suse.cz>
7329 * cif-code.def (VARIADIC_THUNK): New enum value.
7330 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
7333 2018-01-03 Jan Beulich <jbeulich@suse.com>
7335 * sse.md (mov<mode>_internal): Tighten condition for when to use
7336 vmovdqu<ssescalarsize> for TI and OI modes.
7338 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7340 Update copyright years.
7342 2018-01-03 Martin Liska <mliska@suse.cz>
7345 * ipa-visibility.c (function_and_variable_visibility): Skip
7346 functions with noipa attribure.
7348 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7350 * gcc.c (process_command): Update copyright notice dates.
7351 * gcov-dump.c (print_version): Ditto.
7352 * gcov.c (print_version): Ditto.
7353 * gcov-tool.c (print_version): Ditto.
7354 * gengtype.c (create_file): Ditto.
7355 * doc/cpp.texi: Bump @copying's copyright year.
7356 * doc/cppinternals.texi: Ditto.
7357 * doc/gcc.texi: Ditto.
7358 * doc/gccint.texi: Ditto.
7359 * doc/gcov.texi: Ditto.
7360 * doc/install.texi: Ditto.
7361 * doc/invoke.texi: Ditto.
7363 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7365 * vector-builder.h (vector_builder::m_full_nelts): Change from
7366 unsigned int to poly_uint64.
7367 (vector_builder::full_nelts): Update prototype accordingly.
7368 (vector_builder::new_vector): Likewise.
7369 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
7370 (vector_builder::operator ==): Likewise.
7371 (vector_builder::finalize): Likewise.
7372 * int-vector-builder.h (int_vector_builder::int_vector_builder):
7373 Take the number of elements as a poly_uint64 rather than an
7375 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
7376 from unsigned int to poly_uint64.
7377 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
7378 (vec_perm_indices::new_vector): Likewise.
7379 (vec_perm_indices::length): Likewise.
7380 (vec_perm_indices::nelts_per_input): Likewise.
7381 (vec_perm_indices::input_nelts): Likewise.
7382 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
7383 number of elements per input as a poly_uint64 rather than an
7384 unsigned int. Use the original encoding for variable-length
7385 vectors, rather than clamping each individual element.
7386 For the second and subsequent elements in each pattern,
7387 clamp the step and base before clamping their sum.
7388 (vec_perm_indices::series_p): Handle polynomial element counts.
7389 (vec_perm_indices::all_in_range_p): Likewise.
7390 (vec_perm_indices_to_tree): Likewise.
7391 (vec_perm_indices_to_rtx): Likewise.
7392 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
7393 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
7394 (tree_vector_builder::new_binary_operation): Handle polynomial
7395 element counts. Return false if we need to know the number
7396 of elements at compile time.
7397 * fold-const.c (fold_vec_perm): Punt if the number of elements
7398 isn't known at compile time.
7400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7402 * vec-perm-indices.h (vec_perm_builder): Change element type
7403 from HOST_WIDE_INT to poly_int64.
7404 (vec_perm_indices::element_type): Update accordingly.
7405 (vec_perm_indices::clamp): Handle polynomial element_types.
7406 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7407 (vec_perm_indices::all_in_range_p): Likewise.
7408 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
7410 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
7411 polynomial vec_perm_indices element types.
7412 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
7413 * fold-const.c (fold_vec_perm): Likewise.
7414 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
7415 * tree-vect-generic.c (lower_vec_perm): Likewise.
7416 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7417 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
7418 element type to HOST_WIDE_INT.
7420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7421 Alan Hayward <alan.hayward@arm.com>
7422 David Sherwood <david.sherwood@arm.com>
7424 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
7425 rather than an int. Use plus_constant.
7426 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
7427 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
7429 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7430 Alan Hayward <alan.hayward@arm.com>
7431 David Sherwood <david.sherwood@arm.com>
7433 * calls.c (emit_call_1, expand_call): Change struct_value_size from
7434 a HOST_WIDE_INT to a poly_int64.
7436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7437 Alan Hayward <alan.hayward@arm.com>
7438 David Sherwood <david.sherwood@arm.com>
7440 * calls.c (load_register_parameters): Cope with polynomial
7441 mode sizes. Require a constant size for BLKmode parameters
7442 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
7443 forces a parameter to be padded at the lsb end in order to
7444 fill a complete number of words, require the parameter size
7445 to be ordered wrt UNITS_PER_WORD.
7447 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7448 Alan Hayward <alan.hayward@arm.com>
7449 David Sherwood <david.sherwood@arm.com>
7451 * reload1.c (spill_stack_slot_width): Change element type
7452 from unsigned int to poly_uint64_pod.
7453 (alter_reg): Treat mode sizes as polynomial.
7455 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7456 Alan Hayward <alan.hayward@arm.com>
7457 David Sherwood <david.sherwood@arm.com>
7459 * reload.c (complex_word_subreg_p): New function.
7460 (reload_inner_reg_of_subreg, push_reload): Use it.
7462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7463 Alan Hayward <alan.hayward@arm.com>
7464 David Sherwood <david.sherwood@arm.com>
7466 * lra-constraints.c (process_alt_operands): Reject matched
7467 operands whose sizes aren't ordered.
7468 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
7475 that the mode size is in the set {1, 2, 4, 8, 16}.
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 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
7482 Use plus_constant instead of gen_rtx_PLUS.
7484 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7485 Alan Hayward <alan.hayward@arm.com>
7486 David Sherwood <david.sherwood@arm.com>
7488 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
7489 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
7490 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
7491 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
7492 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
7493 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
7494 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
7495 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
7496 * config/i386/i386.c (ix86_push_rounding): ...this new function.
7497 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
7499 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
7500 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
7501 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
7502 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
7503 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
7504 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
7505 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
7506 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
7507 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
7508 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
7510 * expr.c (emit_move_resolve_push): Treat the input and result
7511 of PUSH_ROUNDING as a poly_int64.
7512 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
7513 (emit_push_insn): Likewise.
7514 * lra-eliminations.c (mark_not_eliminable): Likewise.
7515 * recog.c (push_operand): Likewise.
7516 * reload1.c (elimination_effects): Likewise.
7517 * rtlanal.c (nonzero_bits1): Likewise.
7518 * calls.c (store_one_arg): Likewise. Require the padding to be
7519 known at compile time.
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 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7526 Use plus_constant instead of gen_rtx_PLUS.
7528 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7529 Alan Hayward <alan.hayward@arm.com>
7530 David Sherwood <david.sherwood@arm.com>
7532 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7535 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7536 Alan Hayward <alan.hayward@arm.com>
7537 David Sherwood <david.sherwood@arm.com>
7539 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7540 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7541 via stack temporaries. Treat the mode size as polynomial too.
7543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7544 Alan Hayward <alan.hayward@arm.com>
7545 David Sherwood <david.sherwood@arm.com>
7547 * expr.c (expand_expr_real_2): When handling conversions involving
7548 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7549 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7550 as a poly_uint64 too.
7552 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7553 Alan Hayward <alan.hayward@arm.com>
7554 David Sherwood <david.sherwood@arm.com>
7556 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
7558 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7559 Alan Hayward <alan.hayward@arm.com>
7560 David Sherwood <david.sherwood@arm.com>
7562 * combine.c (can_change_dest_mode): Handle polynomial
7563 REGMODE_NATURAL_SIZE.
7564 * expmed.c (store_bit_field_1): Likewise.
7565 * expr.c (store_constructor): Likewise.
7566 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7567 and polynomial REGMODE_NATURAL_SIZE.
7568 (gen_lowpart_common): Likewise.
7569 * reginfo.c (record_subregs_of_mode): Likewise.
7570 * rtlanal.c (read_modify_subreg_p): Likewise.
7572 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7573 Alan Hayward <alan.hayward@arm.com>
7574 David Sherwood <david.sherwood@arm.com>
7576 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7577 numbers of elements.
7579 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7580 Alan Hayward <alan.hayward@arm.com>
7581 David Sherwood <david.sherwood@arm.com>
7583 * match.pd: Cope with polynomial numbers of vector elements.
7585 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7586 Alan Hayward <alan.hayward@arm.com>
7587 David Sherwood <david.sherwood@arm.com>
7589 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7590 in a POINTER_PLUS_EXPR.
7592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7593 Alan Hayward <alan.hayward@arm.com>
7594 David Sherwood <david.sherwood@arm.com>
7596 * omp-simd-clone.c (simd_clone_subparts): New function.
7597 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7598 (ipa_simd_modify_function_body): Likewise.
7600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7601 Alan Hayward <alan.hayward@arm.com>
7602 David Sherwood <david.sherwood@arm.com>
7604 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7605 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7606 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7607 (expand_vector_condition, vector_element): Likewise.
7608 (subparts_gt): New function.
7609 (get_compute_type): Use subparts_gt.
7610 (count_type_subparts): Delete.
7611 (expand_vector_operations_1): Use subparts_gt instead of
7612 count_type_subparts.
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 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7619 (vect_compile_time_alias): ...this new function. Do the calculation
7620 on poly_ints rather than trees.
7621 (vect_prune_runtime_alias_test_list): Update call accordingly.
7623 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7624 Alan Hayward <alan.hayward@arm.com>
7625 David Sherwood <david.sherwood@arm.com>
7627 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7629 (vect_schedule_slp_instance): Likewise.
7631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7632 Alan Hayward <alan.hayward@arm.com>
7633 David Sherwood <david.sherwood@arm.com>
7635 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7636 constant and extern definitions for variable-length vectors.
7637 (vect_get_constant_vectors): Note that the number of units
7638 is known to be constant.
7640 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7641 Alan Hayward <alan.hayward@arm.com>
7642 David Sherwood <david.sherwood@arm.com>
7644 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7645 of units as polynomial. Choose between WIDE and NARROW based
7648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7649 Alan Hayward <alan.hayward@arm.com>
7650 David Sherwood <david.sherwood@arm.com>
7652 * tree-vect-stmts.c (simd_clone_subparts): New function.
7653 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7655 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7656 Alan Hayward <alan.hayward@arm.com>
7657 David Sherwood <david.sherwood@arm.com>
7659 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7660 vectors as polynomial. Use build_index_vector for
7663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7664 Alan Hayward <alan.hayward@arm.com>
7665 David Sherwood <david.sherwood@arm.com>
7667 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7668 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7669 for variable-length vectors.
7670 (vectorizable_mask_load_store): Treat the number of units as
7671 polynomial, asserting that it is constant if the condition has
7672 already been enforced.
7673 (vectorizable_store, vectorizable_load): Likewise.
7675 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7676 Alan Hayward <alan.hayward@arm.com>
7677 David Sherwood <david.sherwood@arm.com>
7679 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7680 of units as polynomial. Punt if we can't tell at compile time
7681 which vector contains the final result.
7683 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7684 Alan Hayward <alan.hayward@arm.com>
7685 David Sherwood <david.sherwood@arm.com>
7687 * tree-vect-loop.c (vectorizable_induction): Treat the number
7688 of units as polynomial. Punt on SLP inductions. Use an integer
7689 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7690 cast of such a series for variable-length floating-point
7693 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7694 Alan Hayward <alan.hayward@arm.com>
7695 David Sherwood <david.sherwood@arm.com>
7697 * tree.h (build_index_vector): Declare.
7698 * tree.c (build_index_vector): New function.
7699 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7700 of units as polynomial, forcibly converting it to a constant if
7701 vectorizable_reduction has already enforced the condition.
7702 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7703 to create a {1,2,3,...} vector.
7704 (vectorizable_reduction): Treat the number of units as polynomial.
7705 Choose vectype_in based on the largest scalar element size rather
7706 than the smallest number of units. Enforce the restrictions
7709 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7710 Alan Hayward <alan.hayward@arm.com>
7711 David Sherwood <david.sherwood@arm.com>
7713 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7714 number of units as polynomial.
7716 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7717 Alan Hayward <alan.hayward@arm.com>
7718 David Sherwood <david.sherwood@arm.com>
7720 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7721 * target.def (autovectorize_vector_sizes): Return the vector sizes
7722 by pointer, using vector_sizes rather than a bitmask.
7723 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7724 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7725 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7727 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7728 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7729 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7730 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7731 * omp-general.c (omp_max_vf): Likewise.
7732 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7733 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7734 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7735 * tree-vect-slp.c (vect_slp_bb): Likewise.
7736 * doc/tm.texi: Regenerate.
7737 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7739 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7740 the vector size as a poly_uint64 rather than an unsigned int.
7741 (current_vector_size): Change from an unsigned int to a poly_uint64.
7742 (get_vectype_for_scalar_type): Update accordingly.
7743 * tree.h (build_truth_vector_type): Take the size and number of
7744 units as a poly_uint64 rather than an unsigned int.
7745 (build_vector_type): Add a temporary overload that takes
7746 the number of units as a poly_uint64 rather than an unsigned int.
7747 * tree.c (make_vector_type): Likewise.
7748 (build_truth_vector_type): Take the number of units as a poly_uint64
7749 rather than an unsigned int.
7751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7752 Alan Hayward <alan.hayward@arm.com>
7753 David Sherwood <david.sherwood@arm.com>
7755 * target.def (get_mask_mode): Take the number of units and length
7756 as poly_uint64s rather than unsigned ints.
7757 * targhooks.h (default_get_mask_mode): Update accordingly.
7758 * targhooks.c (default_get_mask_mode): Likewise.
7759 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7760 * doc/tm.texi: Regenerate.
7762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7763 Alan Hayward <alan.hayward@arm.com>
7764 David Sherwood <david.sherwood@arm.com>
7766 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7767 * omp-general.c (omp_max_vf): Likewise.
7768 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7769 (expand_omp_simd): Handle polynomial safelen.
7770 * omp-low.c (omplow_simd_context): Add a default constructor.
7771 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7772 (lower_rec_simd_input_clauses): Update accordingly.
7773 (lower_rec_input_clauses): Likewise.
7775 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7776 Alan Hayward <alan.hayward@arm.com>
7777 David Sherwood <david.sherwood@arm.com>
7779 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7780 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7781 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7782 (vect_analyze_slp_cost): Likewise.
7783 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7784 (vect_model_load_cost): Likewise.
7786 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7787 Alan Hayward <alan.hayward@arm.com>
7788 David Sherwood <david.sherwood@arm.com>
7790 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7791 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7792 from an unsigned int * to a poly_uint64_pod *.
7793 (calculate_unrolling_factor): New function.
7794 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7797 Alan Hayward <alan.hayward@arm.com>
7798 David Sherwood <david.sherwood@arm.com>
7800 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7801 from an unsigned int to a poly_uint64.
7802 (_loop_vec_info::slp_unrolling_factor): Likewise.
7803 (_loop_vec_info::vectorization_factor): Change from an int
7805 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7806 (vect_get_num_vectors): New function.
7807 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7808 (vect_get_num_copies): Use vect_get_num_vectors.
7809 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7810 to an unsigned int *.
7811 (vect_analyze_data_refs): Change min_vf from an int * to a
7813 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7814 than an unsigned HOST_WIDE_INT.
7815 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7816 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7817 to an unsigned int *.
7818 (vect_analyze_data_ref_dependences): Likewise.
7819 (vect_compute_data_ref_alignment): Handle polynomial vf.
7820 (vect_enhance_data_refs_alignment): Likewise.
7821 (vect_prune_runtime_alias_test_list): Likewise.
7822 (vect_shift_permute_load_chain): Likewise.
7823 (vect_supportable_dr_alignment): Likewise.
7824 (dependence_distance_ge_vf): Take the vectorization factor as a
7825 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7826 (vect_analyze_data_refs): Change min_vf from an int * to a
7828 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7829 vfm1 as a poly_uint64 rather than an int. Make the same change
7830 for the returned bound_scalar.
7831 (vect_gen_vector_loop_niters): Handle polynomial vf.
7832 (vect_do_peeling): Likewise. Update call to
7833 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7834 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7836 * tree-vect-loop.c (vect_determine_vectorization_factor)
7837 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7838 (vect_get_known_peeling_cost): Likewise.
7839 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7840 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7841 (vect_transform_loop): Likewise. Use the lowest possible VF when
7842 updating the upper bounds of the loop.
7843 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7845 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7846 polynomial unroll factors.
7847 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7848 (vect_make_slp_decision): Likewise.
7849 (vect_supported_load_permutation_p): Likewise, and polynomial
7851 (vect_analyze_slp_cost): Handle polynomial vf.
7852 (vect_slp_analyze_node_operations): Likewise.
7853 (vect_slp_analyze_bb_1): Likewise.
7854 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7855 than an unsigned HOST_WIDE_INT.
7856 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7857 (vectorizable_load): Handle polynomial vf.
7858 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7860 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7862 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7863 Alan Hayward <alan.hayward@arm.com>
7864 David Sherwood <david.sherwood@arm.com>
7866 * match.pd: Handle bit operations involving three constants
7867 and try to fold one pair.
7869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7871 * tree-vect-loop-manip.c: Include gimple-fold.h.
7872 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7873 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7874 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7875 Add a path that uses a step of VF instead of 1, but disable it
7877 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7878 and niters_no_overflow parameters. Update calls to
7879 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7880 Create a new SSA name if the latter choses to use a ste other
7881 than zero, and return it via niters_vector_mult_vf_var.
7882 * tree-vect-loop.c (vect_transform_loop): Update calls to
7883 vect_do_peeling, vect_gen_vector_loop_niters and
7884 slpeel_make_loop_iterate_ntimes.
7885 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7886 (vect_gen_vector_loop_niters): Update declarations after above changes.
7888 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7890 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7891 128-bit round to integer instructions.
7892 (ceil<mode>2): Likewise.
7893 (btrunc<mode>2): Likewise.
7894 (round<mode>2): Likewise.
7896 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7898 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7899 unaligned VSX load/store on P8/P9.
7900 (expand_block_clear): Allow the use of unaligned VSX
7901 load/store on P8/P9.
7903 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7905 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7907 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7908 swap associated with both a load and a store.
7910 2018-01-02 Andrew Waterman <andrew@sifive.com>
7912 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7913 * config/riscv/riscv.md (clear_cache): Use it.
7915 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7917 * web.c: Remove out-of-date comment.
7919 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7921 * expr.c (fixup_args_size_notes): Check that any existing
7922 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7923 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7924 (emit_single_push_insn): ...here.
7926 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7928 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7929 (const_vector_encoded_nelts): New function.
7930 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7931 (const_vector_int_elt, const_vector_elt): Declare.
7932 * emit-rtl.c (const_vector_int_elt_1): New function.
7933 (const_vector_elt): Likewise.
7934 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7935 of CONST_VECTOR_ELT.
7937 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7939 * expr.c: Include rtx-vector-builder.h.
7940 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7941 directly on the tree encoding.
7942 (const_vector_from_tree): Likewise.
7943 * optabs.c: Include rtx-vector-builder.h.
7944 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7945 sequence of "u" values.
7946 * vec-perm-indices.c: Include rtx-vector-builder.h.
7947 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7948 directly on the vec_perm_indices encoding.
7950 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7952 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7953 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7954 * rtx-vector-builder.h: New file.
7955 * rtx-vector-builder.c: Likewise.
7956 * rtl.h (rtx_def::u2): Add a const_vector field.
7957 (CONST_VECTOR_NPATTERNS): New macro.
7958 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7959 (CONST_VECTOR_DUPLICATE_P): Likewise.
7960 (CONST_VECTOR_STEPPED_P): Likewise.
7961 (CONST_VECTOR_ENCODED_ELT): Likewise.
7962 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7963 (unwrap_const_vec_duplicate): Likewise.
7964 (const_vec_series_p): Check for a non-duplicated vector encoding.
7965 Say that the function only returns true for integer vectors.
7966 * emit-rtl.c: Include rtx-vector-builder.h.
7967 (gen_const_vec_duplicate_1): Delete.
7968 (gen_const_vector): Call gen_const_vec_duplicate instead of
7969 gen_const_vec_duplicate_1.
7970 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7971 (gen_const_vec_duplicate): Use rtx_vector_builder.
7972 (gen_const_vec_series): Likewise.
7973 (gen_rtx_CONST_VECTOR): Likewise.
7974 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7975 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7976 Build a new vector rather than modifying a CONST_VECTOR in-place.
7977 (handle_special_swappables): Update call accordingly.
7978 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7979 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7980 Build a new vector rather than modifying a CONST_VECTOR in-place.
7981 (handle_special_swappables): Update call accordingly.
7983 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7985 * simplify-rtx.c (simplify_const_binary_operation): Use
7986 CONST_VECTOR_ELT instead of XVECEXP.
7988 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7990 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7991 the selector elements to be different from the data elements
7992 if the selector is a VECTOR_CST.
7993 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7994 ssizetype for the selector.
7996 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7998 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7999 before testing each element individually.
8000 * tree-vect-generic.c (lower_vec_perm): Likewise.
8002 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8004 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
8005 * selftest-run-tests.c (selftest::run_tests): Call it.
8006 * vector-builder.h (vector_builder::operator ==): New function.
8007 (vector_builder::operator !=): Likewise.
8008 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
8009 (vec_perm_indices::all_from_input_p): New function.
8010 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
8011 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
8012 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
8013 instead of reading the VECTOR_CST directly. Detect whether both
8014 vector inputs are the same before constructing the vec_perm_indices,
8015 and update the number of inputs argument accordingly. Use the
8016 utility functions added above. Only construct sel2 if we need to.
8018 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8020 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
8021 the broadcast of the low byte.
8022 (expand_mult_highpart): Use an explicit encoding for the permutes.
8023 * optabs-query.c (can_mult_highpart_p): Likewise.
8024 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
8025 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
8026 (vectorizable_bswap): Likewise.
8027 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
8028 explicit encoding for the power-of-2 permutes.
8029 (vect_permute_store_chain): Likewise.
8030 (vect_grouped_load_supported): Likewise.
8031 (vect_permute_load_chain): Likewise.
8033 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8035 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
8036 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
8037 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
8038 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
8039 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
8040 (vect_gen_perm_mask_any): Likewise.
8042 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8044 * int-vector-builder.h: New file.
8045 * vec-perm-indices.h: Include int-vector-builder.h.
8046 (vec_perm_indices): Redefine as an int_vector_builder.
8047 (auto_vec_perm_indices): Delete.
8048 (vec_perm_builder): Redefine as a stand-alone class.
8049 (vec_perm_indices::vec_perm_indices): New function.
8050 (vec_perm_indices::clamp): Likewise.
8051 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
8052 (vec_perm_indices::new_vector): New function.
8053 (vec_perm_indices::new_expanded_vector): Update for new
8054 vec_perm_indices class.
8055 (vec_perm_indices::rotate_inputs): New function.
8056 (vec_perm_indices::all_in_range_p): Operate directly on the
8057 encoded form, without computing elided elements.
8058 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
8059 encoding. Update for new vec_perm_indices class.
8060 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
8061 the given vec_perm_builder.
8062 (expand_vec_perm_var): Update vec_perm_builder constructor.
8063 (expand_mult_highpart): Use vec_perm_builder instead of
8064 auto_vec_perm_indices.
8065 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
8066 vec_perm_indices instead of auto_vec_perm_indices. Use a single
8067 or double series encoding as appropriate.
8068 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
8069 vec_perm_indices instead of auto_vec_perm_indices.
8070 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
8071 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
8072 (vect_permute_store_chain): Likewise.
8073 (vect_grouped_load_supported): Likewise.
8074 (vect_permute_load_chain): Likewise.
8075 (vect_shift_permute_load_chain): Likewise.
8076 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
8077 (vect_transform_slp_perm_load): Likewise.
8078 (vect_schedule_slp_instance): Likewise.
8079 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
8080 (vectorizable_mask_load_store): Likewise.
8081 (vectorizable_bswap): Likewise.
8082 (vectorizable_store): Likewise.
8083 (vectorizable_load): Likewise.
8084 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
8085 vec_perm_indices instead of auto_vec_perm_indices. Use
8086 tree_to_vec_perm_builder to read the vector from a tree.
8087 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
8088 vec_perm_builder instead of a vec_perm_indices.
8089 (have_whole_vector_shift): Use vec_perm_builder and
8090 vec_perm_indices instead of auto_vec_perm_indices. Leave the
8091 truncation to calc_vec_perm_mask_for_shift.
8092 (vect_create_epilog_for_reduction): Likewise.
8093 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
8094 from auto_vec_perm_indices to vec_perm_indices.
8095 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
8096 instead of changing individual elements.
8097 (aarch64_vectorize_vec_perm_const): Use new_vector to install
8098 the vector in d.perm.
8099 * config/arm/arm.c (expand_vec_perm_d::perm): Change
8100 from auto_vec_perm_indices to vec_perm_indices.
8101 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
8102 instead of changing individual elements.
8103 (arm_vectorize_vec_perm_const): Use new_vector to install
8104 the vector in d.perm.
8105 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
8106 Update vec_perm_builder constructor.
8107 (rs6000_expand_interleave): Likewise.
8108 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
8109 (rs6000_expand_interleave): Likewise.
8111 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8113 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
8114 to qimode could truncate the indices.
8115 * optabs.c (expand_vec_perm_var): Likewise.
8117 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8119 * Makefile.in (OBJS): Add vec-perm-indices.o.
8120 * vec-perm-indices.h: New file.
8121 * vec-perm-indices.c: Likewise.
8122 * target.h (vec_perm_indices): Replace with a forward class
8124 (auto_vec_perm_indices): Move to vec-perm-indices.h.
8125 * optabs.h: Include vec-perm-indices.h.
8126 (expand_vec_perm): Delete.
8127 (selector_fits_mode_p, expand_vec_perm_var): Declare.
8128 (expand_vec_perm_const): Declare.
8129 * target.def (vec_perm_const_ok): Replace with...
8130 (vec_perm_const): ...this new hook.
8131 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
8132 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
8133 * doc/tm.texi: Regenerate.
8134 * optabs.def (vec_perm_const): Delete.
8135 * doc/md.texi (vec_perm_const): Likewise.
8136 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
8137 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
8138 expand_vec_perm for constant permutation vectors. Assert that
8139 the mode of variable permutation vectors is the integer equivalent
8140 of the mode that is being permuted.
8141 * optabs-query.h (selector_fits_mode_p): Declare.
8142 * optabs-query.c: Include vec-perm-indices.h.
8143 (selector_fits_mode_p): New function.
8144 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
8145 is defined, instead of checking whether the vec_perm_const_optab
8146 exists. Use targetm.vectorize.vec_perm_const instead of
8147 targetm.vectorize.vec_perm_const_ok. Check whether the indices
8148 fit in the vector mode before using a variable permute.
8149 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
8150 vec_perm_indices instead of an rtx.
8151 (expand_vec_perm): Replace with...
8152 (expand_vec_perm_const): ...this new function. Take the selector
8153 as a vec_perm_indices rather than an rtx. Also take the mode of
8154 the selector. Update call to shift_amt_for_vec_perm_mask.
8155 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
8156 Use vec_perm_indices::new_expanded_vector to expand the original
8157 selector into bytes. Check whether the indices fit in the vector
8158 mode before using a variable permute.
8159 (expand_vec_perm_var): Make global.
8160 (expand_mult_highpart): Use expand_vec_perm_const.
8161 * fold-const.c: Includes vec-perm-indices.h.
8162 * tree-ssa-forwprop.c: Likewise.
8163 * tree-vect-data-refs.c: Likewise.
8164 * tree-vect-generic.c: Likewise.
8165 * tree-vect-loop.c: Likewise.
8166 * tree-vect-slp.c: Likewise.
8167 * tree-vect-stmts.c: Likewise.
8168 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
8170 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
8171 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
8172 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
8173 (aarch64_vectorize_vec_perm_const): ...this new function.
8174 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8175 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8176 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
8177 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
8178 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8179 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8180 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
8182 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
8183 check for NEON modes.
8184 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
8185 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
8186 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
8187 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
8189 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
8190 the old VEC_PERM_CONST conditions.
8191 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
8192 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
8193 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
8194 (ia64_vectorize_vec_perm_const_ok): Merge into...
8195 (ia64_vectorize_vec_perm_const): ...this new function.
8196 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
8197 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
8198 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
8199 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
8200 * config/mips/mips.c (mips_expand_vec_perm_const)
8201 (mips_vectorize_vec_perm_const_ok): Merge into...
8202 (mips_vectorize_vec_perm_const): ...this new function.
8203 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
8204 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
8205 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
8206 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
8207 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
8208 (rs6000_expand_vec_perm_const): Delete.
8209 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
8211 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8212 (altivec_expand_vec_perm_const_le): Take each operand individually.
8213 Operate on constant selectors rather than rtxes.
8214 (altivec_expand_vec_perm_const): Likewise. Update call to
8215 altivec_expand_vec_perm_const_le.
8216 (rs6000_expand_vec_perm_const): Delete.
8217 (rs6000_vectorize_vec_perm_const_ok): Delete.
8218 (rs6000_vectorize_vec_perm_const): New function.
8219 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
8220 an element count and rtx array.
8221 (rs6000_expand_extract_even): Update call accordingly.
8222 (rs6000_expand_interleave): Likewise.
8223 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
8224 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
8225 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
8226 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
8227 (rs6000_expand_vec_perm_const): Delete.
8228 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8229 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8230 (altivec_expand_vec_perm_const_le): Take each operand individually.
8231 Operate on constant selectors rather than rtxes.
8232 (altivec_expand_vec_perm_const): Likewise. Update call to
8233 altivec_expand_vec_perm_const_le.
8234 (rs6000_expand_vec_perm_const): Delete.
8235 (rs6000_vectorize_vec_perm_const_ok): Delete.
8236 (rs6000_vectorize_vec_perm_const): New function. Remove stray
8237 reference to the SPE evmerge intructions.
8238 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
8239 an element count and rtx array.
8240 (rs6000_expand_extract_even): Update call accordingly.
8241 (rs6000_expand_interleave): Likewise.
8242 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
8243 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
8245 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8247 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8249 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
8250 vector mode and that that mode matches the mode of the data
8252 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
8253 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
8254 directly using expand_vec_perm_1 when forcing selectors into
8256 (expand_vec_perm_var): New function, split out from expand_vec_perm.
8258 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8260 * optabs-query.h (can_vec_perm_p): Delete.
8261 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
8262 * optabs-query.c (can_vec_perm_p): Split into...
8263 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
8264 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
8265 particular selector is valid.
8266 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
8267 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
8268 (vect_grouped_load_supported): Likewise.
8269 (vect_shift_permute_load_chain): Likewise.
8270 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
8271 (vect_transform_slp_perm_load): Likewise.
8272 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
8273 (vectorizable_bswap): Likewise.
8274 (vect_gen_perm_mask_checked): Likewise.
8275 * fold-const.c (fold_ternary_loc): Likewise. Don't take
8276 implementations of variable permutation vectors into account
8277 when deciding which selector to use.
8278 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
8279 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
8280 with a false third argument.
8281 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
8282 to test whether the constant selector is valid and can_vec_perm_var_p
8283 to test whether a variable selector is valid.
8285 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8287 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
8288 * optabs-query.c (can_vec_perm_p): Likewise.
8289 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
8290 instead of vec_perm_indices.
8291 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
8292 (vect_gen_perm_mask_checked): Likewise,
8293 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
8294 (vect_gen_perm_mask_checked): Likewise,
8296 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8298 * optabs-query.h (qimode_for_vec_perm): Declare.
8299 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
8300 (qimode_for_vec_perm): ...this new function.
8301 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
8303 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
8305 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
8306 does not have a conditional at the top.
8308 2018-01-02 Richard Biener <rguenther@suse.de>
8310 * ipa-inline.c (big_speedup_p): Fix expression.
8312 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
8315 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
8318 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
8322 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
8323 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
8324 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
8325 cond_taken_branch_cost 3->4.
8327 2018-01-01 Jakub Jelinek <jakub@redhat.com>
8329 PR tree-optimization/83581
8330 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
8331 TODO_cleanup_cfg if any changes have been made.
8334 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
8335 convert_modes if target mode has the right side, but different mode
8339 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
8340 last argument when extracting from CONCAT. If either from_real or
8341 from_imag is NULL, use expansion through memory. If result is not
8342 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
8343 the parts directly to inner mode, if even that fails, use expansion
8347 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
8348 check for bswap in mode rather than HImode and use that in expand_unop
8351 Copyright (C) 2018 Free Software Foundation, Inc.
8353 Copying and distribution of this file, with or without modification,
8354 are permitted in any medium without royalty provided the copyright
8355 notice and this notice are preserved.