1 2018-02-12 Jeff Law <law@redhat.com>
4 * config/sh/sh.c (find_barrier): Consider a sibling call
7 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
8 successfully back substituting a reg.
10 2018-02-12 Richard Biener <rguenther@suse.de>
12 PR tree-optimization/84037
13 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
14 parameter, move visited init to caller.
15 (vect_slp_analyze_operations): Separate cost from validity
16 check, initialize visited once for all instances.
17 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
19 * tree-vect-stmts.c (vect_model_simple_cost): Make early
21 (vect_model_promotion_demotion_cost): Likewise.
22 (vectorizable_bswap): Guard cost modeling with !slp_node
23 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
25 (vectorizable_call): Likewise.
26 (vectorizable_conversion): Likewise.
27 (vectorizable_assignment): Likewise.
28 (vectorizable_shift): Likewise.
29 (vectorizable_operation): Likewise.
30 (vectorizable_store): Likewise.
31 (vectorizable_load): Likewise.
32 (vectorizable_condition): Likewise.
33 (vectorizable_comparison): Likewise.
35 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
38 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
39 (ASAN_MARK): Fix fnspec to account for return value, change pointer
40 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
42 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
45 * params.def (inline-min-speedup): Increase from 8 to 15.
46 (max-inline-insns-auto): Decrease from 40 to 30.
47 * ipa-split.c (consider_split): Add some buffer for function to
48 be considered inlining candidate.
49 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
52 2018-02-12 Richard Biener <rguenther@suse.de>
54 PR tree-optimization/84037
55 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
56 matched stmts if we cannot swap the non-matched ones.
58 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
60 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
61 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
62 _mm_maskz_scalef_round_ss): New intrinsics.
63 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
64 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
65 __builtin_ia32_scalefss_round): Remove.
66 (__builtin_ia32_scalefsd_mask_round,
67 __builtin_ia32_scalefss_mask_round): New intrinsics.
68 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
69 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
70 ((match_operand:VF_128 2 "<round_nimm_predicate>"
71 "<round_constraint>")): Changed to ...
72 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
73 "<round_scalar_constraint>")): ... this.
74 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
75 %0, %1, %2<round_op3>}"): Changed to ...
76 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
77 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
78 %2<round_scalar_mask_op3>}"): ... this.
79 * config/i386/subst.md (round_scalar_nimm_predicate): New.
81 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
83 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
84 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
85 (_mm_maskz_sqrt_round_ss): New intrinsics.
86 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
87 (__builtin_ia32_sqrtsd_mask_round)
88 (__builtin_ia32_sqrtss_mask_round): New builtins.
89 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
90 (__builtin_ia32_sqrtss_round): Remove.
91 (__builtin_ia32_sqrtsd_mask_round)
92 (__builtin_ia32_sqrtss_mask_round): New builtins.
93 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
94 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
95 ((match_operand:VF_128 1 "vector_operand"
96 "xBm,<round_constraint>")): Changed to ...
97 ((match_operand:VF_128 1 "vector_operand"
98 "xBm,<round_scalar_constraint>")): ... this.
99 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
100 %0, %2, %<iptr>1<round_op3>}): Changed to ...
101 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
102 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
103 %<iptr>1<round_scalar_mask_op3>}): ... this.
104 ((set_attr "prefix" "<round_prefix>")): Changed to ...
105 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
107 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
110 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
111 Cast vec_cmpeq result to correct type.
112 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
113 Cast vec_cmpgt result to correct type.
115 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
117 * final.c (final_scan_insn_1): Renamed from...
118 (final_scan_insn): ... this. New wrapper, to recover
119 seen from the outermost call in recursive ones.
120 * config/sparc/sparc.c (output_return): Drop seen from call.
121 (output_sibcall): Likewise.
122 * config/visium/visium.c (output_branch): Likewise.
124 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
126 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
129 2018-02-10 Alan Modra <amodra@gmail.com>
132 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
133 Specify LR as an input.
135 2018-02-10 Jakub Jelinek <jakub@redhat.com>
138 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
139 remove_member_access_dummy_vars): New functions.
140 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
141 lower_omp_1, execute_lower_omp): Use them.
143 PR rtl-optimization/84308
144 * shrink-wrap.c (spread_components): Release todo vector.
146 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
148 PR rtl-optimization/57193
149 * ira-color.c (struct allocno_color_data): Add member
150 conflict_allocno_hard_prefs.
151 (update_conflict_allocno_hard_prefs): New.
152 (bucket_allocno_compare_func): Add a preference based on
153 conflict_allocno_hard_prefs.
154 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
155 (color_allocnos): Remove a dead code. Initiate
156 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
158 2018-02-09 Jakub Jelinek <jakub@redhat.com>
161 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
162 constraint from =wa to wa. Avoid a subreg on the output operand,
163 instead use a pseudo and subreg it in a move.
164 (p9_xxbrd_<mode>): Changed to ...
165 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
166 (p9_xxbrd_v2df): New expander.
167 (p9_xxbrw_<mode>): Changed to ...
168 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
169 (p9_xxbrw_v4sf): New expander.
171 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
173 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
175 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
178 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
179 multiply in 32-bit mode.
180 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
181 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
184 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
186 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
187 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
188 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
189 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
191 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
194 * dwarf2out.c (is_trivial_indirect_ref): New function.
195 (dwarf2out_late_global_decl): Do not generate a location
196 attribute for variables that have a non-trivial DECL_VALUE_EXPR
197 and that are not defined in the current unit.
199 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
201 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
202 instead of a libcall for UNORDERED.
204 2018-02-09 Tamar Christina <tamar.christina@arm.com>
207 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
208 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
210 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
213 * config/s390/s390.c (s390_set_current_function): Invoke
214 s390_indirect_branch_settings also if fndecl didn't change.
216 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
218 * config/rs6000/rs6000.md (blockage): Set length to zero.
220 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
222 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
224 2018-02-09 Jakub Jelinek <jakub@redhat.com>
227 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
228 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
232 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
233 PARALLEL incoming that failed vt_get_decl_and_offset check.
236 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
237 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
239 (get_variable_section): For decls in named .bss* sections pass true as
240 second argument to bss_initializer_p.
242 2018-02-09 Marek Polacek <polacek@redhat.com>
243 Jakub Jelinek <jakub@redhat.com>
246 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
247 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
248 Sync some changes from cxx_fold_indirect_ref.
250 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
252 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
254 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
255 (BLOCK_INLINE_ENTRY_LABEL): New.
256 (dwarf2out_var_location): Disregard inline entry markers.
257 (inline_entry_data): New struct.
258 (inline_entry_data_hasher): New hashtable type.
259 (inline_entry_data_hasher::hash): New.
260 (inline_entry_data_hasher::equal): New.
261 (inline_entry_data_table): New variable.
262 (add_high_low_attributes): Add DW_AT_entry_pc and
263 DW_AT_GNU_entry_view attributes if a pending entry is found
264 in inline_entry_data_table. Add old entry_pc attribute only
265 if debug nonbinding markers are disabled.
266 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
268 (block_within_block_p, dwarf2out_inline_entry): New.
269 (dwarf2out_finish): Check that no entries remained in
270 inline_entry_data_table.
271 * final.c (reemit_insn_block_notes): Handle inline entry notes.
272 (final_scan_insn, notice_source_line): Likewise.
273 (rest_of_clean_state): Skip inline entry markers.
274 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
276 * gimple.c (gimple_build_debug_inline_entry): New.
277 * gimple.h (enum gimple_debug_subcode): Add
278 GIMPLE_DEBUG_INLINE_ENTRY.
279 (gimple_build_debug_inline_entry): Declare.
280 (gimple_debug_inline_entry_p): New.
281 (gimple_debug_nonbind_marker_p): Adjust.
282 * insn-notes.def (INLINE_ENTRY): New.
283 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
284 inline entry marker notes.
285 (print_insn): Likewise.
286 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
287 (INSN_DEBUG_MARKER_KIND): Likewise.
288 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
289 * tree-inline.c (expand_call_inline): Build and insert
290 debug_inline_entry stmt.
291 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
292 inline entry blocks early, if nonbind markers are enabled.
293 (dump_scope_block): Dump fragment info.
294 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
295 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
296 (gimple_build_debug_inline_entry): New.
297 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
298 Enable/disable inline entry points too.
299 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
300 (DEBUG_INSN): Describe inline entry markers.
302 * common.opt (gvariable-location-views): New.
303 (gvariable-location-views=incompat5): New.
304 * config.in: Rebuilt.
305 * configure: Rebuilt.
306 * configure.ac: Test assembler for view support.
307 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
308 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
309 * dwarf2out.c (var_loc_view): New typedef.
310 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
311 (dwarf2out_locviews_in_attribute): New.
312 (dwarf2out_locviews_in_loclist): New.
313 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
314 (enum dw_line_info_opcode): Add LI_adv_address.
315 (struct dw_line_info_table): Add view.
316 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
317 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
318 (zero_view_p): New variable.
319 (ZERO_VIEW_P): New macro.
320 (output_asm_line_debug_info): New.
321 (struct var_loc_node): Add view.
322 (add_AT_view_list, AT_loc_list): New.
323 (add_var_loc_to_decl): Add view param. Test it against last.
324 (new_loc_list): Add view params. Record them.
325 (AT_loc_list_ptr): Handle loc and view lists.
326 (view_list_to_loc_list_val_node): New.
327 (print_dw_val): Handle dw_val_class_view_list.
328 (size_of_die): Likewise.
329 (value_format): Likewise.
330 (loc_list_has_views): New.
331 (gen_llsym): Set vl_symbol too.
332 (maybe_gen_llsym, skip_loc_list_entry): New.
333 (dwarf2out_maybe_output_loclist_view_pair): New.
334 (output_loc_list): Output view list or entries too.
335 (output_view_list_offset): New.
336 (output_die): Handle dw_val_class_view_list.
337 (output_dwarf_version): New.
338 (output_compilation_unit_header): Use it.
339 (output_skeleton_debug_sections): Likewise.
340 (output_rnglists, output_line_info): Likewise.
341 (output_pubnames, output_aranges): Update version comments.
342 (output_one_line_info_table): Output view numbers in asm comments.
343 (dw_loc_list): Determine current endview, pass it to new_loc_list.
344 Call maybe_gen_llsym.
345 (loc_list_from_tree_1): Adjust.
346 (add_AT_location_description): Create view list attribute if
347 needed, check it's absent otherwise.
348 (convert_cfa_to_fb_loc_list): Adjust.
349 (maybe_emit_file): Call output_asm_line_debug_info for test.
350 (dwarf2out_var_location): Reset views as needed. Precompute
351 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
353 (new_line_info_table): Reset next view.
354 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
355 (dwarf2out_source_line): Likewise. Output view resets and labels to
356 the assembler, or select appropriate line info opcodes.
357 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
358 (optimize_string_length): Catch it. Adjust.
359 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
360 dw_val_class_view_list, and remove it if no longer needed.
361 (hash_loc_list): Hash view numbers.
362 (loc_list_hasher::equal): Compare them.
363 (optimize_location_lists): Check whether a view list symbol is
364 needed, and whether the locview attribute is present, and
365 whether they match. Remove the locview attribute if no longer
367 (index_location_lists): Call skip_loc_list_entry for test.
368 (dwarf2out_finish): Call output_asm_line_debug_info for test.
369 Use output_dwarf_version.
370 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
371 (struct dw_val_node): Add val_view_list.
372 * final.c (SEEN_NEXT_VIEW): New.
373 (set_next_view_needed): New.
374 (clear_next_view_needed): New.
375 (maybe_output_next_view): New.
376 (final_start_function): Rename to...
377 (final_start_function_1): ... this. Take pointer to FIRST,
378 add SEEN parameter. Emit param bindings in the initial view.
379 (final_start_function): Reintroduce SEEN-less interface.
380 (final): Rename to...
381 (final_1): ... this. Take SEEN parameter. Output final pending
382 next view at the end.
383 (final): Reintroduce seen-less interface.
384 (final_scan_insn): Output pending next view before switching
385 sections or ending a block. Mark the next view as needed when
386 outputting variable locations. Notify debug backend of section
387 changes, and of location view changes.
388 (rest_of_handle_final): Adjust.
389 * toplev.c (process_options): Autodetect value for debug variable
390 location views option. Warn on incompat5 without -gdwarf-5.
391 * doc/invoke.texi (gvariable-location-views): New.
392 (gvariable-location-views=incompat5): New.
393 (gno-variable-location-views): New.
395 2018-02-08 David Malcolm <dmalcolm@redhat.com>
397 PR tree-optimization/84136
398 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
399 that the result of find_edge is non-NULL.
401 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
404 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
405 storing integer register in SImode. Fix cost of 256 and 512
406 byte aligned SSE register store.
408 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
410 * config/i386/i386.c (ix86_multiplication_cost): Fix
411 multiplication cost for TARGET_AVX512DQ.
413 2018-02-08 Marek Polacek <polacek@redhat.com>
415 PR tree-optimization/84238
416 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
419 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
421 PR tree-optimization/84265
422 * tree-vect-stmts.c (vectorizable_store): Don't treat
423 VMAT_CONTIGUOUS accesses as grouped.
424 (vectorizable_load): Likewise.
426 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
428 PR tree-optimization/81635
429 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
430 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
431 (test_round_for_mask): New functions.
432 (wide_int_cc_tests): Call test_round_for_mask.
433 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
434 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
435 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
436 range returned by get_range_info.
438 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
441 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
442 * symtab.c: Include builtins.h
443 (symtab_node::output_to_lto_symbol_table_p): Move here
444 from lto-streamer-out.c:output_symbol_p.
445 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
446 (output_symbol_p): Move all logic to symtab.c
447 (produce_symtab): Update.
449 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
451 * config/s390/s390-opts.h (enum indirect_branch): Define.
452 * config/s390/s390-protos.h (s390_return_addr_from_memory)
453 (s390_indirect_branch_via_thunk)
454 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
455 (enum s390_indirect_branch_type): Define.
456 * config/s390/s390.c (struct s390_frame_layout, struct
457 machine_function): Remove.
458 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
459 (indirect_branch_table_label_no, indirect_branch_table_name):
461 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
462 (enum s390_indirect_branch_option): Define.
463 (s390_return_addr_from_memory): New function.
464 (s390_handle_string_attribute): New function.
465 (s390_attribute_table): Add new attribute handler.
466 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
467 (s390_indirect_branch_via_thunk): New function.
468 (s390_indirect_branch_via_inline_thunk): New function.
469 (s390_function_ok_for_sibcall): When jumping via thunk disallow
470 sibling call optimization for non z10 compiles.
471 (s390_emit_call): Force indirect branch target to be a single
472 register. Add r1 clobber for non-z10 compiles.
473 (s390_emit_epilogue): Emit return jump via return_use expander.
474 (s390_reorg): Handle JUMP_INSNs as execute targets.
475 (s390_option_override_internal): Perform validity checks for the
476 new command line options.
477 (s390_indirect_branch_attrvalue): New function.
478 (s390_indirect_branch_settings): New function.
479 (s390_set_current_function): Invoke s390_indirect_branch_settings.
480 (s390_output_indirect_thunk_function): New function.
481 (s390_code_end): Implement target hook.
482 (s390_case_values_threshold): Implement target hook.
483 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
485 * config/s390/s390.h (struct s390_frame_layout)
486 (struct machine_function): Move here from s390.c.
487 (TARGET_INDIRECT_BRANCH_NOBP_RET)
488 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
489 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
490 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
491 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
492 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
493 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
494 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
495 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
496 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
497 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
498 (mnemonic attribute): Add values which aren't recognized
500 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
501 pattern for branch conversion. Fix mnemonic attribute.
502 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
503 indirect branch via thunk if requested.
504 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
505 ("*indirect_jump"): Disable for branch conversion using out of
507 ("indirect_jump_via_thunk<mode>_z10")
508 ("indirect_jump_via_thunk<mode>")
509 ("indirect_jump_via_inlinethunk<mode>_z10")
510 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
511 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
512 ("casesi_jump_via_inlinethunk<mode>_z10")
513 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
514 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
515 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
516 ("*indirect2_jump"): Disable for branch conversion.
517 ("casesi_jump"): Turn into expander and expand patterns for branch
519 ("return_use"): New expander.
520 ("*return"): Emit return via thunk and rename it to ...
521 ("*return<mode>"): ... this one.
522 * config/s390/s390.opt: Add new options and and enum for the
525 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
527 * lra-constraints.c (match_reload): Unconditionally use
528 gen_lowpart_SUBREG, rather than selecting between that
529 and equivalent gen_rtx_SUBREG code.
531 2018-02-08 Richard Biener <rguenther@suse.de>
533 PR tree-optimization/84233
534 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
535 changed flag instead of boguously re-using phi_inserted.
537 2018-02-08 Martin Jambor <mjambor@suse.cz>
539 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
540 static local variables.
542 2018-02-08 Richard Biener <rguenther@suse.de>
544 PR tree-optimization/84278
545 * tree-vect-stmts.c (vectorizable_store): When looking for
546 smaller vector types to perform grouped strided loads/stores
547 make sure the mode is supported by the target.
548 (vectorizable_load): Likewise.
550 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
552 * config/aarch64/aarch64.c (aarch64_components_for_bb):
553 Increase LDP/STP opportunities by adding adjacent callee-saves.
555 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
557 PR rtl-optimization/84068
558 PR rtl-optimization/83459
559 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
561 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
563 PR tree-optimization/84224
564 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
565 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
568 2018-02-07 Iain Sandoe <iain@codesourcery.com>
571 * config/rs6000/altivec.md (*restore_world): Remove LR use.
572 * config/rs6000/predicates.md (restore_world_operation): Adjust op
573 count, remove one USE.
575 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
577 * doc/install.texi (Configuration): Document the
578 --with-long-double-format={ibm,ieee} PowerPC configuration
582 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
583 Convert from define_expand to be define_insn_and_split. Rework
584 float/double/_Float128 conversions to QI/HI/SImode to work with
585 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
586 conversions to QI/HImode types did a store and then a load to
587 truncate the value. For conversions to VSX registers, don't split
588 the insn, instead emit the code directly. Use the code iterator
589 any_fix to combine signed and unsigned conversions.
590 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
591 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
592 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
593 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
594 (fix_<mode>di2_hw): Likewise.
595 (fixuns_<mode>di2_hw): Likewise.
596 (fix_<mode>si2_hw): Likewise.
597 (fixuns_<mode>si2_hw): Likewise.
598 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
599 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
600 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
601 fix<uns>_trunc<SFDF:mode>si2_p8.
602 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
604 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
605 (fix<uns>_<mode>_mem): Likewise.
606 (fctiw<u>z_<mode>_mem): Likewise.
607 (fix<uns>_<mode>_mem): Likewise.
608 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
609 the register allocator from doing a direct move to the GPRs to do
610 a store, and instead use the ISA 3.0 store byte/half-word from
611 vector register instruction. For IEEE 128-bit floating point,
612 also optimize stores of 32-bit ints.
613 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
615 2018-02-07 Alan Hayward <alan.hayward@arm.com>
617 * genextract.c (push_pathstr_operand): New function to support
619 (walk_rtx): Call push_pathstr_operand.
620 (print_path): Support [a-zA-Z].
622 2018-02-07 Richard Biener <rguenther@suse.de>
624 PR tree-optimization/84037
625 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
626 (cse_and_gimplify_to_preheader): Declare.
627 (vect_get_place_in_interleaving_chain): Likewise.
628 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
630 (_loop_vec_info::~_loop_vec_info): Delete it.
631 (cse_and_gimplify_to_preheader): New function.
632 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
633 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
634 (vectorizable_load): Likewise. For grouped stores always base
635 the IV on the first element.
636 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
637 condition before gimplifying.
639 2018-02-07 Jakub Jelinek <jakub@redhat.com>
641 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
642 *DIV_EXPR and *MOD_EXPR.
644 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
647 * config/i386/i386.c (ix86_option_override_internal): Mask out
648 the CF_SET bit when checking -fcf-protection.
650 2018-02-07 Tom de Vries <tom@codesourcery.com>
653 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
656 2018-02-07 Richard Biener <rguenther@suse.de>
658 PR tree-optimization/84204
659 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
662 PR tree-optimization/84205
663 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
664 special-case isl_ast_op_zdiv_r.
666 PR tree-optimization/84223
667 * graphite-scop-detection.c (gather_bbs::before_dom_children):
668 Only add conditions from within the region.
669 (gather_bbs::after_dom_children): Adjust.
671 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
674 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
675 * config/avr/avr.md: Only post-reload split REG-REG moves if
676 either register is REGERAL_REG_P.
678 2018-02-07 Jakub Jelinek <jakub@redhat.com>
680 PR tree-optimization/84235
681 * tree-ssa-scopedtables.c
682 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
683 if the subtraction is performed in floating point type where NaNs are
684 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
685 build 1. Formatting fix.
687 2018-02-06 Jakub Jelinek <jakub@redhat.com>
690 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
691 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
692 and skip it regardless of bb boundaries. Use CALL_P macro,
693 don't test INSN_P (insn) together with CALL_P or JUMP_P check
694 unnecessarily, formatting fix.
696 2018-02-06 Michael Collison <michael.collison@arm.com>
698 * config/arm/thumb2.md:
699 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
700 (*thumb_mov_notscc): Ditto.
702 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
705 * config/rs6000/rs6000.md (su code attribute): Use "u" for
706 unsigned_fix, not "s".
708 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
710 * configure.ac (gcc_fn_eh_frame_ro): New function.
711 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
712 correct .eh_frame permissions.
713 * configure: Regenerate.
715 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
717 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
720 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
722 * config/rs6000/rs6000.c (rs6000_option_override_internal):
723 Display warning message for -mno-speculate-indirect-jumps.
725 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
727 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
729 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
731 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
733 PR tree-optimization/84225
734 * tree-eh.c (find_trapping_overflow): Only call
735 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
737 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
740 * config/i386/i386.c: Reimplement the check of possible options
741 -mibt/-mshstk conbination. Change error messages.
742 * doc/invoke.texi: Fix a typo: remove extra '='.
744 2018-02-06 Marek Polacek <polacek@redhat.com>
746 PR tree-optimization/84228
747 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
749 2018-02-06 Tamar Christina <tamar.christina@arm.com>
752 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
753 emitted arch directives.
754 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
755 __ARM_FEATURE_COPROC before changing architectures.
757 2018-02-06 Richard Biener <rguenther@suse.de>
759 * config/i386/i386.c (print_reg): Fix typo.
760 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
762 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
764 * configure: Regenerate.
766 2018-02-05 Martin Sebor <msebor@redhat.com>
768 PR tree-optimization/83369
769 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
772 2018-02-05 Martin Liska <mliska@suse.cz>
774 * doc/invoke.texi: Cherry-pick upstream r323995.
776 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
778 * ira.c (ira_init_register_move_cost): Adjust comment.
780 2018-02-05 Martin Liska <mliska@suse.cz>
782 PR gcov-profile/84137
783 * doc/gcov.texi: Fix typo in documentation.
785 2018-02-05 Martin Liska <mliska@suse.cz>
787 PR gcov-profile/83879
788 * doc/gcov.texi: Document necessity of --dynamic-list-data when
789 using dlopen functionality.
791 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
793 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
794 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
795 _mm_maskz_range_ss, _mm_mask_range_round_ss,
796 _mm_maskz_range_round_ss): New intrinsics.
797 (__builtin_ia32_rangesd128_round)
798 (__builtin_ia32_rangess128_round): Remove.
799 (__builtin_ia32_rangesd128_mask_round,
800 __builtin_ia32_rangess128_mask_round): New builtins.
801 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
802 __builtin_ia32_rangess128_round): Remove.
803 (__builtin_ia32_rangesd128_mask_round,
804 __builtin_ia32_rangess128_mask_round): New builtins.
805 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
806 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
807 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
808 "<round_saeonly_constraint>")): Changed to ...
809 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
810 "<round_saeonly_scalar_constraint>")): ... this.
811 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
812 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
813 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
814 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
815 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
817 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
819 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
821 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
822 Remove all values except native, 8540 and 8548.
824 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
826 * config/i386/i386.c (ix86_output_function_return): Pass
827 INVALID_REGNUM, instead of -1, as invalid register number to
828 indirect_thunk_name and output_indirect_thunk.
830 2018-02-02 Julia Koval <julia.koval@intel.com>
832 * config.gcc: Add -march=icelake.
833 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
834 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
835 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
836 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
837 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
838 (processor_target_table): Add icelake.
839 (ix86_option_override_internal): Handle new PTAs.
840 (get_builtin_code_for_version): Handle icelake.
841 (M_INTEL_COREI7_ICELAKE): New.
842 (fold_builtin_cpu): Handle icelake.
843 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
844 * doc/invoke.texi: Add -march=icelake.
846 2018-02-02 Julia Koval <julia.koval@intel.com>
848 * config/i386/i386.c (ix86_option_override_internal): Change flags type
850 * wide-int-bitmask.h: New.
852 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
855 * config/i386/i386.md: Replace Pmode with word_mode in
856 builtin_setjmp_setup and builtin_longjmp to support x32.
858 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
862 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
864 (rs6000_supported_cpu_names): New static variable.
865 (linux_cpu_translation_table): Likewise.
866 (elf_platform) <cpu>: Define new static variable and use it.
867 Translate kernel AT_PLATFORM name to canonical name if needed.
868 Error if platform name is unknown.
870 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
873 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
875 2018-02-01 Jeff Law <law@redhat.com>
878 * config/i386/i386.c (release_scratch_register_on_entry): Add new
879 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
880 the scratch if RELEASE_VIA_POP is false.
881 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
882 If we have to save a temporary register, decrement SIZE appropriately.
883 Pass new arguments to release_scratch_register_on_entry.
884 (ix86_adjust_stack_and_probe): Likewise.
885 (ix86_emit_probe_stack_range): Pass new arguments to
886 release_scratch_register_on_entry.
888 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
890 PR rtl-optimization/84157
891 * combine.c (change_zero_ext): Use REG_P predicate in
892 front of HARD_REGISTER_P predicate.
894 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
896 * config/avr/avr.c (avr_option_override): Move disabling of
897 -fdelete-null-pointer-checks to...
898 * common/config/avr/avr-common.c (avr_option_optimization_table):
901 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
903 PR tree-optimization/81635
904 * tree-data-ref.c (split_constant_offset_1): For types that
905 wrap on overflow, try to use range info to prove that wrapping
908 2018-02-01 Renlin Li <renlin.li@arm.com>
911 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
913 (aarch64_register_move_cost): Likewise.
914 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
916 (REG_CLASS_NAMES): Likewise.
917 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
918 TAILCALL_ADDR_REGS. Remove IP registers.
919 * config/aarch64/aarch64.md (Ucs): Update register constraint.
921 2018-02-01 Richard Biener <rguenther@suse.de>
923 * domwalk.h (dom_walker::dom_walker): Add additional constructor
924 for specifying RPO order and allow NULL for that.
925 * domwalk.c (dom_walker::dom_walker): Likewise.
926 (dom_walker::walk): Handle NULL RPO order.
927 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
929 (rewrite_update_dom_walker): Likewise.
930 (mark_def_dom_walker): Likewise.
932 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
934 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
935 (aarch64_maybe_expand_sve_subreg_move): Declare.
936 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
937 * config/aarch64/predicates.md (aarch64_any_register_operand): New
939 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
940 that are semantically a reverse operation.
941 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
942 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
943 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
945 (aarch64_can_change_mode_class): For big-endian, forbid changes
946 between two SVE modes if they have different element sizes.
948 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
950 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
951 the TImode handling for big-endian targets.
953 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
955 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
956 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
958 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
959 Remove BSWAP handing for big-endian targets and use the form of
960 LD1RQ appropariate for the mode.
962 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
964 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
965 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
968 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
971 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
972 check for operands that need to go through aarch64_sve_reload_be.
974 2018-02-01 Jakub Jelinek <jakub@redhat.com>
976 PR tree-optimization/81661
977 PR tree-optimization/84117
978 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
979 * tree-eh.c: Include gimplify.h.
980 (find_trapping_overflow, replace_trapping_overflow,
981 rewrite_to_non_trapping_overflow): New functions.
982 * tree-vect-loop.c: Include tree-eh.h.
983 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
984 * tree-data-ref.c: Include tree-eh.h.
985 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
987 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
989 PR rtl-optimization/84123
990 * combine.c (change_zero_ext): Check if hard register satisfies
991 can_change_dest_mode before calling gen_lowpart_SUBREG.
993 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
996 * ira.c (ira_init_register_move_cost): Remove assert.
998 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1000 PR rtl-optimization/84071
1001 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1002 * doc/tm.texi: Regenerate.
1004 2018-01-31 Richard Biener <rguenther@suse.de>
1006 PR tree-optimization/84132
1007 * tree-data-ref.c (analyze_miv_subscript): Properly
1008 check whether evolution_function_is_affine_multivariate_p
1009 before calling gcd_of_steps_may_divide_p.
1011 2018-01-31 Julia Koval <julia.koval@intel.com>
1014 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1015 * config/i386/i386.md (rdpid_rex64) New.
1016 (rdpid): Make 32bit only.
1018 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1021 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1022 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1024 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1027 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1029 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1031 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1033 PR rtl-optimization/84071
1034 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1035 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1037 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1039 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1040 (arc_attribute_table): Add 'aux' attribute.
1041 (arc_in_small_data_p): Consider aux like variables.
1042 (arc_is_aux_reg_p): New function.
1043 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1044 (arc_get_aux_arg): New function.
1045 (prepare_move_operands): Handle aux-register access.
1046 (arc_handle_aux_attribute): New function.
1047 * doc/extend.texi (ARC Variable attributes): Add subsection.
1049 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1051 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1052 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1053 (arc_attribute_table): Add 'uncached' attribute.
1054 (arc_print_operand): Print '.di' flag for uncached memory
1056 (arc_in_small_data_p): Do not consider for small data the uncached
1058 (arc_is_uncached_mem_p): New function.
1059 * config/arc/predicates.md (compact_store_memory_operand): Check
1060 for uncached memory accesses.
1061 (nonvol_nonimm_operand): Likewise.
1062 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1064 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1067 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1068 falign-loops=): Add Optimization flag.
1070 2018-01-30 Jeff Law <law@redhat.com>
1073 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1074 INT_REGISTERS_SAVED. Check it prior to calling
1075 get_scratch_register_on_entry.
1076 (ix86_adjust_stack_and_probe): Similarly.
1077 (ix86_emit_probe_stack_range): Similarly.
1078 (ix86_expand_prologue): Corresponding changes.
1080 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1083 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1084 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1086 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1089 * lra-constraints.c (curr_insn_transform): Process AND in the
1092 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1094 PR rtl-optimization/83986
1095 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1096 dependence against last_pending_memory_flush in addition to
1099 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1101 PR tree-optimization/81611
1102 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1105 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1107 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1110 2018-01-30 Richard Biener <rguenther@suse.de>
1111 Jakub Jelinek <jakub@redhat.com>
1113 PR tree-optimization/84111
1114 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1115 inner loops added during recursion, as they don't have up-to-date
1118 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1121 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1122 (can_inline_edge_by_limits_p): ... here.
1123 (can_early_inline_edge_p, check_callers,
1124 update_caller_keys, update_callee_keys, recursive_inlining,
1125 add_new_edges_to_heap, speculation_useful_p,
1126 inline_small_functions,
1127 inline_small_functions, flatten_function,
1128 inline_to_all_callers_1): Update.
1130 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1132 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1135 2018-01-30 Richard Biener <rguenther@suse.de>
1137 PR tree-optimization/83008
1138 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1139 invariant and constant vector uses in stmts when they need
1142 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1145 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1146 * configure: Regenerate.
1148 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1150 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1152 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1153 Use gen_rtx_REG rather than gen_lowpart.
1155 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1157 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1158 rather than 0 when creating partial subregs.
1160 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1162 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1165 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1168 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1169 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1170 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1171 flags. This restores the settings used before the 2017-07-24.
1172 Turning off pre increment/decrement/modify allows IVOPTS to
1173 optimize DF/SF loops where the index is an int.
1175 2018-01-29 Richard Biener <rguenther@suse.de>
1176 Kelvin Nilsen <kelvin@gcc.gnu.org>
1179 * tree-vect-stmts.c (vectorizable_call): Don't call
1180 targetm.vectorize_builtin_md_vectorized_function if callee is
1183 2018-01-22 Carl Love <cel@us.ibm.com>
1185 * doc/extend.tex: Fix typo in second arg in
1186 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1188 2018-01-29 Richard Biener <rguenther@suse.de>
1190 PR tree-optimization/84086
1191 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1192 (flush_ssaname_freelist): When SSA names were released reset
1193 the SCEV hash table.
1195 2018-01-29 Richard Biener <rguenther@suse.de>
1197 PR tree-optimization/84057
1198 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1199 removed paths when removing edges.
1201 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1203 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1204 -mfunction-return=@var{choice}.
1206 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1209 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1211 (layout::print_source_line): Likewise.
1212 (test_get_line_width_without_trailing_whitespace): Add test cases.
1214 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1217 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1220 2018-01-26 Jim Wilson <jimw@sifive.com>
1222 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1224 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1227 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1229 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1230 and CMP + SUB-immediate -> SUBS.
1232 2018-01-26 Martin Sebor <msebor@redhat.com>
1234 PR tree-optimization/83896
1235 * tree-ssa-strlen.c (get_string_len): Rename...
1236 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1237 Avoid assuming length is constant.
1238 (handle_char_store): Use HOST_WIDE_INT for string length.
1240 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1243 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1244 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1246 2018-01-26 Richard Biener <rguenther@suse.de>
1248 PR rtl-optimization/84003
1249 * dse.c (record_store): Only record redundant stores when
1250 the earlier store aliases at least all accesses the later one does.
1252 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1254 PR rtl-optimization/83985
1255 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1256 REG_CFA_RESTORE insns.
1257 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1258 REG_CFA_RESTORE insns here.
1261 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1262 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1264 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1266 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1267 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1268 (arc_init): Likewise.
1269 (arc_override_options): Likewise.
1270 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1272 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1274 * config/arc/arc.h (TARGET_DBNZ): Define.
1275 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1276 properly set the tune attribute.
1277 (dbnz): Use TARGET_DBNZ guard.
1278 * config/arc/arc.opt (mtune): Add core3 option.
1280 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1282 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1283 recognize new pic like addresses.
1284 (arc_delegitimize_address): Clean up.
1286 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1288 * config/arc/arc-arches.def: Option mrf16 valid for all
1290 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1291 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1292 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1293 * config/arc/arc-tables.opt: Regenerate.
1294 * config/arc/arc.c (arc_conditional_register_usage): Handle
1295 reduced register file case.
1296 (arc_file_start): Set must have build attributes.
1297 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1299 * config/arc/arc.opt (mrf16): Add new option.
1300 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1301 * config/arc/genmultilib.awk: Handle new mrf16 option.
1302 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1303 * config/arc/t-multilib: Regenerate.
1304 * doc/invoke.texi (ARC Options): Document mrf16 option.
1306 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1308 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1309 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1310 (arc_attribute_table): Add 'secure_call' attribute.
1311 (arc_print_operand): Print secure call operand.
1312 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1314 (arc_is_secure_call_p): New function. * config/arc/arc.md
1315 (call_i): Add support for sjli instruction.
1316 (call_value_i): Likewise.
1317 * config/arc/constraints.md (Csc): New constraint.
1319 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1320 John Eric Martin <John.Martin@emmicro-us.com>
1322 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1323 * config/arc/arc.c (_arc_jli_section): New struct.
1324 (arc_jli_section): New type.
1325 (rc_jli_sections): New static variable.
1326 (arc_handle_jli_attribute): New function.
1327 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1328 (arc_file_end): New function.
1329 (TARGET_ASM_FILE_END): Define.
1330 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1331 (arc_add_jli_section): New function.
1332 (jli_call_scan): Likewise.
1333 (arc_reorg): Call jli_call_scan.
1334 (arc_output_addsi): Remove 'S' from printing asm operand.
1335 (arc_is_jli_call_p): New function.
1336 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1338 (movhi_insn): Likewise.
1339 (movsi_insn): Likewise.
1340 (movsi_set_cc_insn): Likewise.
1341 (loadqi_update): Likewise.
1342 (load_zeroextendqisi_update): Likewise.
1343 (load_signextendqisi_update): Likewise.
1344 (loadhi_update): Likewise.
1345 (load_zeroextendhisi_update): Likewise.
1346 (load_signextendhisi_update): Likewise.
1347 (loadsi_update): Likewise.
1348 (loadsf_update): Likewise.
1349 (movsicc_insn): Likewise.
1350 (bset_insn): Likewise.
1351 (bxor_insn): Likewise.
1352 (bclr_insn): Likewise.
1353 (bmsk_insn): Likewise.
1354 (bicsi3_insn): Likewise.
1355 (cmpsi_cc_c_insn): Likewise.
1356 (movsi_ne): Likewise.
1357 (movsi_cond_exec): Likewise.
1358 (clrsbsi2): Likewise.
1369 (call_i): Remove 'S' asm letter, add jli instruction.
1370 (call_value_i): Likewise.
1371 * config/arc/arc.op (mjli-always): New option.
1372 * config/arc/constraints.md (Cji): New constraint.
1373 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1375 (subsf3_fpx): Likewise.
1376 (mulsf3_fpx): Likewise.
1377 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1379 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1381 * doc/invoke.texi (ARC): Document mjli-always option.
1383 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1385 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1386 avoid addition with 0 and use incw and decw where possible.
1388 2018-01-26 Richard Biener <rguenther@suse.de>
1390 PR tree-optimization/81082
1391 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1392 association if it requires casting to unsigned.
1393 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1394 from fold_plusminus_mult_expr to catch important cases late when
1395 range info is available.
1397 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1399 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1400 * configure.ac (hidden_linkonce): New test.
1401 * configure: Regenerate.
1402 * config.in: Regenerate.
1404 2018-01-26 Julia Koval <julia.koval@intel.com>
1406 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1407 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1408 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1409 _mm_mask_bitshuffle_epi64_mask): Fix type.
1410 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1411 USI_FTYPE_V4DI_V4DI_USI): Remove.
1412 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1413 __builtin_ia32_vpshufbitqmb256_mask,
1414 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1415 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1416 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1418 2018-01-26 Alan Modra <amodra@gmail.com>
1421 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1422 UNSPEC_VBPERMQ. Sort other unspecs.
1424 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1426 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1428 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1431 * predict.c (drop_profile): Do not push/pop cfun; update also
1433 (handle_missing_profiles): Fix logic looking for zero profiles.
1435 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1438 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1439 on functions with #pragma omp declare simd or functions with simd
1441 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1442 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1443 Remove trailing \n from warning_at calls.
1445 2018-01-25 Tom de Vries <tom@codesourcery.com>
1448 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1449 for neutered workers.
1451 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1454 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1455 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1457 2018-01-24 Jeff Law <law@redhat.com>
1460 * i386.c (get_probe_interval): Move to earlier point.
1461 (ix86_compute_frame_layout): If -fstack-clash-protection and
1462 the frame is larger than the probe interval, then use pushes
1463 to save registers rather than reg->mem moves.
1464 (ix86_expand_prologue): Remove conditional for int_registers_saved
1467 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1470 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1471 min/max for never referenced object.
1473 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1476 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1478 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1479 attributes from DECL_ATTRIBUTES (decl) without affecting
1480 DECL_ATTRIBUTES (current_function_decl).
1481 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1482 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1484 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1486 PR tree-optimization/83979
1487 * fold-const.c (fold_comparison): Use constant_boolean_node
1488 instead of boolean_{true,false}_node.
1490 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1492 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1495 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1497 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1498 Simplify the clause that sets the length attribute.
1499 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1500 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1501 clause that sets the length attribute.
1502 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1504 2018-01-24 Tom de Vries <tom@codesourcery.com>
1507 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1508 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1509 Add strict parameter.
1510 (prevent_branch_around_nothing): Insert dummy insn between branch to
1511 label and label with no ptx insn inbetween.
1512 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1514 2018-01-24 Tom de Vries <tom@codesourcery.com>
1517 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1518 for neutered threads in warp.
1519 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1521 2018-01-24 Richard Biener <rguenther@suse.de>
1523 PR tree-optimization/83176
1524 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1527 2018-01-24 Richard Biener <rguenther@suse.de>
1529 PR tree-optimization/82819
1530 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1531 code generating pluses that are no-ops in the target precision.
1533 2018-01-24 Richard Biener <rguenther@suse.de>
1536 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1538 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1540 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1541 to merge probabilities.
1542 * predict.c (probably_never_executed): Also mark as cold functions
1543 with global 0 profile and guessed local profile.
1544 * profile-count.c (profile_probability::combine_with_count): New
1546 * profile-count.h (profile_probability::operator*,
1547 profile_probability::operator*=, profile_probability::operator/,
1548 profile_probability::operator/=): Reduce precision to adjusted
1549 and set value to guessed on contradictory divisions.
1550 (profile_probability::combine_with_freq): Remove.
1551 (profile_probability::combine_wiht_count): Declare.
1552 (profile_count::force_nonzero):: Set to adjusted.
1553 (profile_count::probability_in):: Set quality to adjusted.
1554 * tree-ssa-tail-merge.c (replace_block_by): Use
1557 2018-01-23 Andrew Waterman <andrew@sifive.com>
1558 Jim Wilson <jimw@sifive.com>
1560 * config/riscv/riscv.c (riscv_stack_boundary): New.
1561 (riscv_option_override): Set riscv_stack_boundary. Handle
1562 riscv_preferred_stack_boundary_arg.
1563 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1564 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1565 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1566 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1567 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1568 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1570 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1573 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1574 of struct ix86_frame.
1575 (ix86_expand_epilogue): Likewise. Add a local variable for
1576 the reg_save_offset field in struct ix86_frame.
1578 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1580 PR tree-optimization/82604
1581 * tree-loop-distribution.c (enum partition_kind): New enum item
1582 PKIND_PARTIAL_MEMSET.
1583 (partition_builtin_p): Support above new enum item.
1584 (generate_code_for_partition): Ditto.
1585 (compute_access_range): Differentiate cases that equality can be
1586 proven at all loops, the innermost loops or no loops.
1587 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1588 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1589 (finalize_partitions, distribute_loop): Don't fuse partition of
1590 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1591 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1594 2018-01-23 Martin Liska <mliska@suse.cz>
1596 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1597 order to ignore the predictor.
1598 (PRED_POLYMORPHIC_CALL): Likewise.
1599 (PRED_RECURSIVE_CALL): Likewise.
1601 2018-01-23 Martin Liska <mliska@suse.cz>
1603 * tree-profile.c (tree_profiling): Print function header to
1604 aware reader which function we are working on.
1605 * value-prof.c (gimple_find_values_to_profile): Do not print
1606 not interesting value histograms.
1608 2018-01-23 Martin Liska <mliska@suse.cz>
1610 * profile-count.h (enum profile_quality): Add
1611 profile_uninitialized as the first value. Do not number values
1612 as they are zero based.
1613 (profile_count::verify): Update sanity check.
1614 (profile_probability::verify): Likewise.
1616 2018-01-23 Nathan Sidwell <nathan@acm.org>
1618 * doc/invoke.texi (ffor-scope): Deprecate.
1620 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1622 PR tree-optimization/83510
1623 * domwalk.c (set_all_edges_as_executable): New function.
1624 (dom_walker::dom_walker): Convert bool param
1625 "skip_unreachable_blocks" to enum reachability. Move setup of
1626 edge flags to set_all_edges_as_executable and only do it when
1627 reachability is REACHABLE_BLOCKS.
1628 * domwalk.h (enum dom_walker::reachability): New enum.
1629 (dom_walker::dom_walker): Convert bool param
1630 "skip_unreachable_blocks" to enum reachability.
1631 (set_all_edges_as_executable): New decl.
1632 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1633 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1635 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1636 but converting true to REACHABLE_BLOCKS.
1637 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1639 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1640 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1641 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1643 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1644 if check_all_array_refs will be called.
1646 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1648 * tree.c (selftest::test_location_wrappers): Add more test
1651 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1653 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1654 (selftest::test_bit_in_range): Likewise.
1656 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1659 * doc/sourcebuild.texi (vect_float): Say that the selector
1660 only describes the situation when -funsafe-math-optimizations is on.
1661 (vect_float_strict): Document.
1663 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1665 PR tree-optimization/83965
1666 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1667 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1668 instead of checking only for a reduction.
1669 (vect_recog_widen_sum_pattern): Likewise.
1671 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1673 * predict.c (probably_never_executed): Only use precise profile info.
1674 (compute_function_frequency): Skip after inlining hack since we now
1675 have quality checking.
1677 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1679 * profile-count.h (profile_probability::very_unlikely,
1680 profile_probability::unlikely, profile_probability::even): Set
1681 precision to guessed.
1683 2018-01-23 Richard Biener <rguenther@suse.de>
1685 PR tree-optimization/83963
1686 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1687 Properly terminate dominator walk when crossing the exit edge not
1688 when visiting its source block.
1690 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1693 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1694 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1696 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1698 PR tree-optimization/83957
1699 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1700 semicolon after for body surrounded by braces.
1702 PR tree-optimization/83081
1703 * profile-count.h (profile_probability::split): New method.
1704 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1705 Use profile_probability::split.
1706 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1707 when splitting a single conditional jump into 2.
1709 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1711 PR tree-optimization/69452
1712 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1715 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1717 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1718 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1719 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1721 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1723 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1724 * config/rl78/rl78.md (movdi): New define_expand.
1725 * config/rl78/rl78.c (rl78_split_movdi): New function.
1727 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1730 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1732 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1733 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1734 128-bit to produce an UNSPEC move to get the double word with the
1735 signbit and then a shift directly to do signbit.
1736 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1737 implementation with a new version that just does either a direct
1738 move or a regular move. Move memory interface to separate insns.
1739 Move insns so they are next to the expander.
1740 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1741 with signbit move. Split big and little endian case.
1742 (signbit<mode>2_dm_mem_le): Likewise.
1743 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1744 (signbit<mode>2_dm2): Likewise.
1746 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1748 * config/rl78/rl78.md (anddi3): New define_expand.
1750 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1752 * config/rl78/rl78.md (umindi3): New define_expand.
1754 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1756 * config/rl78/rl78.md (smindi3): New define_expand.
1758 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1760 * config/rl78/rl78.md (smaxdi3): New define_expand.
1762 2018-01-22 Carl Love <cel@us.ibm.com>
1764 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1765 LVX_V1TI): Add macro expansion.
1766 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1767 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1768 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1769 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1770 Change check to determine if the instruction is a byte reversing
1771 entry. Fix typo in comment.
1772 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1773 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1774 Add def_builtin calls for new builtins.
1775 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1776 Add define_insn expansion.
1778 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1780 * config/rl78/rl78.md (umaxdi3): New define_expand.
1782 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1784 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
1785 for non-QImode registers.
1787 2018-01-22 Richard Biener <rguenther@suse.de>
1789 PR tree-optimization/83963
1790 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1791 including the loop exit block.
1792 (scop_detection::merge_sese): Likewise.
1793 (scop_detection::add_scop): Do it here instead.
1795 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1797 * doc/sourcebuild.texi (arm_softfloat): Document.
1799 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1802 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1803 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1804 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1806 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1807 David Edelsohn <dje.gcc@gmail.com>
1810 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1811 Change "crset eq" to "crset 2".
1812 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1813 (*call_indirect_aix<mode>_nospec): Likewise.
1814 (*call_value_indirect_aix<mode>_nospec): Likewise.
1815 (*call_indirect_elfv2<mode>_nospec): Likewise.
1816 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1817 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1818 change assembly output from . to $.
1819 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1820 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1821 (*tablejump<mode>_internal1_nospec): Likewise.
1823 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1826 * config/sh/sh_optimize_sett_clrt.cc:
1827 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1829 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1831 PR tree-optimization/83940
1832 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1833 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1834 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1835 use it to pass back the definition type.
1836 (vect_check_store_rhs): Likewise rhs_dt_out.
1837 (vect_build_gather_load_calls): Add a mask_dt argument and use
1838 it instead of a call to vect_is_simple_use.
1839 (vectorizable_store): Update calls to vect_check_load_store_mask
1840 and vect_check_store_rhs. Use the dt returned by the latter instead
1841 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1842 instead of calls to vect_is_simple_use. Pass the scalar rather
1843 than the vector operand to vect_is_simple_use when handling
1844 second and subsequent copies of an rhs value.
1845 (vectorizable_load): Update calls to vect_check_load_store_mask
1846 and vect_build_gather_load_calls. Use the cached mask_dt and
1847 gs_info.offset_dt instead of calls to vect_is_simple_use.
1849 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1852 * tree-emutls.c: Include gimplify.h.
1853 (lower_emutls_2): New function.
1854 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1855 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1856 it before further processing.
1859 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1860 UINTVAL (trueop1) instead of INTVAL (op1).
1862 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1866 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1867 INCOMING_FRAME_SP_OFFSET if not defined.
1868 (scan_trace): Add ENTRY argument. If true and
1869 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1870 emit a note to adjust the CFA offset.
1871 (create_cfi_notes): Adjust scan_trace callers.
1872 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1873 INCOMING_FRAME_SP_OFFSET in the CIE.
1874 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1875 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1877 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1878 * doc/tm.texi: Regenerated.
1880 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1882 PR rtl-optimization/83147
1883 * lra-constraints.c (remove_inheritance_pseudos): Use
1884 lra_substitute_pseudo_within_insn.
1886 2018-01-19 Tom de Vries <tom@codesourcery.com>
1887 Cesar Philippidis <cesar@codesourcery.com>
1890 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1892 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1895 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1896 spaces for function labels.
1898 2018-01-19 Martin Liska <mliska@suse.cz>
1900 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1901 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1902 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1903 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1904 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1905 (PRED_CONST_RETURN): Change from 69 to 65.
1906 (PRED_NULL_RETURN): Change from 91 to 71.
1907 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1908 (PRED_LOOP_GUARD): Change from 66 to 73.
1910 2018-01-19 Martin Liska <mliska@suse.cz>
1912 * predict.c (predict_insn_def): Add new assert.
1913 (struct branch_predictor): Change type to signed integer.
1914 (test_prediction_value_range): Amend test to cover
1916 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1917 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1918 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1919 (PRED_LOOP_IV_COMPARE): Likewise.
1920 * predict.h (PROB_UNINITIALIZED): Define new constant.
1922 2018-01-19 Martin Liska <mliska@suse.cz>
1924 * predict.c (dump_prediction): Add new format for
1925 analyze_brprob.py script which is enabled with -details
1927 * profile-count.h (precise_p): New function.
1929 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1931 PR tree-optimization/83922
1932 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1933 there are no statements that need masking.
1934 (vect_active_double_reduction_p): New function.
1935 (vect_analyze_loop_operations): Use it when handling phis that
1936 are not in the loop header.
1938 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1940 PR tree-optimization/83914
1941 * tree-vect-loop.c (vectorizable_induction): Don't convert
1942 init_expr or apply the peeling adjustment for inductions
1943 that are nested within the vectorized loop.
1945 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1947 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1950 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1954 * function.h (gimplify_parameters): Add gimple_seq * argument.
1955 * function.c: Include gimple.h and options.h.
1956 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1957 for the added local temporaries if needed.
1958 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1959 if there are any parameter cleanups, wrap whole body into a
1960 try/finally with the cleanups.
1962 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1965 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1966 Use GET_MODE_CLASS for scalar floating point.
1968 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1972 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1973 Fix call of call_cgraph_insertion_hooks.
1975 2018-01-18 Martin Sebor <msebor@redhat.com>
1977 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1979 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1982 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1985 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1988 * common.opt: (-ffile-prefix-map): New option.
1989 * opts.c (common_handle_option): Defer it.
1990 * opts-global.c (handle_common_deferred_options): Handle it.
1991 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1992 * file-prefix-map.h: New file.
1993 (remap_debug_filename, add_debug_prefix_map): ...here.
1994 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1995 * final.c (debug_prefix_map, add_debug_prefix_map
1996 remap_debug_filename): Move to...
1997 * file-prefix-map.c: New file.
1998 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1999 generalize, get rid of alloca(), use strrchr() instead of strchr().
2000 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2001 Implement in terms of add_prefix_map().
2002 (remap_macro_filename, remap_debug_filename): Implement in term of
2004 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2005 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2006 * dbxout.c: Include file-prefix-map.h.
2007 * varasm.c: Likewise.
2008 * vmsdbgout.c: Likewise.
2009 * xcoffout.c: Likewise.
2010 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2011 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2012 * doc/invoke.texi (-ffile-prefix-map): Document.
2013 (-fdebug-prefix-map): Update description.
2015 2018-01-18 Martin Liska <mliska@suse.cz>
2017 * config/i386/i386.c (indirect_thunk_name): Document that also
2019 (output_indirect_thunk): Document why both instructions
2020 (pause and lfence) are generated.
2022 2018-01-18 Richard Biener <rguenther@suse.de>
2024 PR tree-optimization/83887
2025 * graphite-scop-detection.c
2026 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2027 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2028 (scop_detection::merge_sese): Re-implement with a flood-fill
2029 algorithm that properly finds a SESE region if it exists.
2031 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2034 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2035 pointer_diff optimizations use view_convert instead of convert.
2037 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2039 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2040 Generate different code for -mno-speculate-indirect-jumps.
2041 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2042 (*call_indirect_aix<mode>): Disable for
2043 -mno-speculate-indirect-jumps.
2044 (*call_indirect_aix<mode>_nospec): New define_insn.
2045 (*call_value_indirect_aix<mode>): Disable for
2046 -mno-speculate-indirect-jumps.
2047 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2048 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2049 -mno-speculate-indirect-jumps.
2050 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2052 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2054 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2055 long double type, set the flags for noting the default long double
2056 type, even if we don't pass or return a long double type.
2058 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2061 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2064 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2066 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2067 support for merge[hl].
2068 (fold_mergehl_helper): New helper function.
2069 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2070 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2071 (altivec_vmrglw_direct): Add xxmrglw insn.
2073 2018-01-17 Andrew Waterman <andrew@sifive.com>
2075 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2076 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2078 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2081 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2082 call the lto_location_cache before reading the
2083 DECL_SOURCE_LOCATION of the types.
2085 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2086 Richard Sandiford <richard.sandiford@linaro.org>
2088 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2089 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2090 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2091 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2092 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2094 * config/aarch64/constraints.md (aarch64_movti_operand):
2096 * config/aarch64/predicates.md (Uti): Add new constraint.
2098 2018-01-17 Carl Love <cel@us.ibm.com>
2099 * config/rs6000/vsx.md (define_expand xl_len_r,
2100 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2101 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2103 (define_expand, define_insn): Move the shift left from the
2104 define_insn to the define_expand for lxvl and stxvl instructions.
2105 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2106 and XL_LEN_R definitions to PURE.
2108 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2110 * config/i386/i386.c (indirect_thunk_name): Declare regno
2111 as unsigned int. Compare regno with INVALID_REGNUM.
2112 (output_indirect_thunk): Ditto.
2113 (output_indirect_thunk_function): Ditto.
2114 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2115 in the call to output_indirect_thunk_function.
2117 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2120 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2121 rather than the size of inner_type to determine the stack slot size
2122 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2124 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2127 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2130 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2132 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2133 endian Linux systems to optionally enable multilibs for selecting
2134 the long double type if the user configured an explicit type.
2135 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2136 have no long double multilibs if not defined.
2137 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2138 warn if the user used -mabi={ieee,ibm}longdouble and we built
2139 multilibs for long double.
2140 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2141 appropriate multilib option.
2142 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2144 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2145 for building long double multilibs.
2146 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2148 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2150 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2153 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2155 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2158 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2161 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2164 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2166 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2167 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2169 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2171 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2172 different rtl trees depending on TARGET_64BIT.
2173 (rs6000_gen_lvx): Likewise.
2175 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2177 * config/visium/visium.md (nop): Tweak comment.
2178 (hazard_nop): Likewise.
2180 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2182 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2183 -mspeculate-indirect-jumps.
2184 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2185 for -mno-speculate-indirect-jumps.
2186 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2187 (*call_value_indirect_elfv2<mode>): Disable for
2188 -mno-speculate-indirect-jumps.
2189 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2190 (indirect_jump): Emit different RTL for
2191 -mno-speculate-indirect-jumps.
2192 (*indirect_jump<mode>): Disable for
2193 -mno-speculate-indirect-jumps.
2194 (*indirect_jump<mode>_nospec): New define_insn.
2195 (tablejump): Emit different RTL for
2196 -mno-speculate-indirect-jumps.
2197 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2198 (tablejumpsi_nospec): New define_expand.
2199 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2200 (tablejumpdi_nospec): New define_expand.
2201 (*tablejump<mode>_internal1): Disable for
2202 -mno-speculate-indirect-jumps.
2203 (*tablejump<mode>_internal1_nospec): New define_insn.
2204 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2207 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2209 * caller-save.c (insert_save): Drop unnecessary parameter. All
2212 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2213 Richard Biener <rguenth@suse.de>
2216 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2217 return early, inline manually is_gimple_sizepos. Make sure if we
2218 call gimplify_expr we don't end up with a gimple constant.
2219 * tree.c (variably_modified_type_p): Don't return true for
2220 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2221 * gimplify.h (is_gimple_sizepos): Remove.
2223 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2225 PR tree-optimization/83857
2226 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2227 vectorizable_live_operation for pure SLP statements.
2228 (vectorizable_live_operation): Handle PHIs.
2230 2018-01-16 Richard Biener <rguenther@suse.de>
2232 PR tree-optimization/83867
2233 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2234 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2236 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2239 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2240 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2241 If off is not INTEGER_CST, issue a may not be aligned warning
2242 rather than isn't aligned. Use isn%'t rather than isn't.
2243 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2245 <case MULT_EXPR>: Improve the case when bottom and one of the
2246 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2247 operand, in that case check if the other operand is multiple of
2248 bottom divided by the INTEGER_CST operand.
2250 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2253 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2254 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2255 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2256 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2257 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2258 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2259 (pa_function_arg_size): New function.
2261 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2263 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2264 in a separate statement.
2266 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2268 PR tree-optimization/83847
2269 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2270 group gathers and scatters.
2272 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2274 PR rtl-optimization/86620
2275 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2277 PR rtl-optimization/83213
2278 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2279 to last if both are JUMP_INSNs.
2281 PR tree-optimization/83843
2282 * gimple-ssa-store-merging.c
2283 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2284 store_immediate_info for bswap/nop orig_stores.
2286 2018-01-15 Andrew Waterman <andrew@sifive.com>
2288 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2290 <UDIV>: Increase cost if !TARGET_DIV.
2292 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2294 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2295 (define_attr "cr_logical_3op"): New.
2296 (cceq_ior_compare): Adjust.
2297 (cceq_ior_compare_complement): Adjust.
2298 (*cceq_rev_compare): Adjust.
2299 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2300 (is_cracked_insn): Adjust.
2301 (insn_must_be_first_in_group): Adjust.
2302 * config/rs6000/40x.md: Adjust.
2303 * config/rs6000/440.md: Adjust.
2304 * config/rs6000/476.md: Adjust.
2305 * config/rs6000/601.md: Adjust.
2306 * config/rs6000/603.md: Adjust.
2307 * config/rs6000/6xx.md: Adjust.
2308 * config/rs6000/7450.md: Adjust.
2309 * config/rs6000/7xx.md: Adjust.
2310 * config/rs6000/8540.md: Adjust.
2311 * config/rs6000/cell.md: Adjust.
2312 * config/rs6000/e300c2c3.md: Adjust.
2313 * config/rs6000/e500mc.md: Adjust.
2314 * config/rs6000/e500mc64.md: Adjust.
2315 * config/rs6000/e5500.md: Adjust.
2316 * config/rs6000/e6500.md: Adjust.
2317 * config/rs6000/mpc.md: Adjust.
2318 * config/rs6000/power4.md: Adjust.
2319 * config/rs6000/power5.md: Adjust.
2320 * config/rs6000/power6.md: Adjust.
2321 * config/rs6000/power7.md: Adjust.
2322 * config/rs6000/power8.md: Adjust.
2323 * config/rs6000/power9.md: Adjust.
2324 * config/rs6000/rs64.md: Adjust.
2325 * config/rs6000/titan.md: Adjust.
2327 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2329 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2330 ix86_indirect_branch_register logic.
2332 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2334 * config/i386/constraints.md (Bs): Update
2335 ix86_indirect_branch_register check. Don't check
2336 ix86_indirect_branch_register with GOT_memory_operand.
2338 * config/i386/predicates.md (GOT_memory_operand): Don't check
2339 ix86_indirect_branch_register here.
2340 (GOT32_symbol_operand): Likewise.
2342 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2344 * config/i386/predicates.md (constant_call_address_operand):
2345 Rewrite ix86_indirect_branch_register logic.
2346 (sibcall_insn_operand): Likewise.
2348 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2350 * config/i386/constraints.md (Bs): Replace
2351 ix86_indirect_branch_thunk_register with
2352 ix86_indirect_branch_register.
2354 * config/i386/i386.md (indirect_jump): Likewise.
2355 (tablejump): Likewise.
2356 (*sibcall_memory): Likewise.
2357 (*sibcall_value_memory): Likewise.
2358 Peepholes of indirect call and jump via memory: Likewise.
2359 * config/i386/i386.opt: Likewise.
2360 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2361 (GOT_memory_operand): Likewise.
2362 (call_insn_operand): Likewise.
2363 (sibcall_insn_operand): Likewise.
2364 (GOT32_symbol_operand): Likewise.
2366 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2369 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2370 type rather than type addr's type points to.
2371 (expand_omp_atomic_mutex): Likewise.
2372 (expand_omp_atomic): Likewise.
2374 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2377 * config/i386/i386.c (output_indirect_thunk_function): Use
2378 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2379 for __x86_return_thunk.
2381 2018-01-15 Richard Biener <rguenther@suse.de>
2384 * expmed.c (extract_bit_field_1): Fix typo.
2386 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2389 * config/arm/iterators.md (VF): New mode iterator.
2390 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2391 Remove integer-related logic from pattern.
2392 (neon_vabd<mode>_3): Likewise.
2394 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2397 * common.opt (fstrict-overflow): No longer an alias.
2398 (fwrapv-pointer): New option.
2399 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2400 also for pointer types based on flag_wrapv_pointer.
2401 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2402 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2403 opts->x_flag_wrapv got set.
2404 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2405 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2406 POINTER_TYPE_OVERFLOW_UNDEFINED.
2407 * match.pd: Likewise in address comparison pattern.
2408 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2410 2018-01-15 Richard Biener <rguenther@suse.de>
2413 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2414 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2415 Reset type names to their identifier if their TYPE_DECL doesn't
2416 have linkage (and thus is used for ODR and devirt).
2417 (save_debug_info_for_decl): Remove.
2418 (save_debug_info_for_type): Likewise.
2419 (add_tree_to_fld_list): Adjust.
2420 * tree-pretty-print.c (dump_generic_node): Make dumping of
2421 type names more robust.
2423 2018-01-15 Richard Biener <rguenther@suse.de>
2425 * BASE-VER: Bump to 8.0.1.
2427 2018-01-14 Martin Sebor <msebor@redhat.com>
2430 * builtins.c (check_access): Avoid warning when the no-warning bit
2433 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2435 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2436 * ira-color (allocno_hard_regs_compare): Likewise.
2438 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2441 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2442 Use .pushsection/.popsection.
2444 2018-01-14 Martin Sebor <msebor@redhat.com>
2447 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2449 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2451 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2452 entry from extra_headers.
2453 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2454 extra_headers, make the list bitwise identical to the i?86-*-* one.
2456 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2458 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2459 -mcmodel=large with -mindirect-branch=thunk,
2460 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2461 -mfunction-return=thunk-extern.
2462 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2463 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2464 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2466 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2468 * config/i386/i386.c (print_reg): Print the name of the full
2469 integer register without '%'.
2470 (ix86_print_operand): Handle 'V'.
2471 * doc/extend.texi: Document 'V' modifier.
2473 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2475 * config/i386/constraints.md (Bs): Disallow memory operand for
2476 -mindirect-branch-register.
2478 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2479 (GOT_memory_operand): Likewise.
2480 (call_insn_operand): Likewise.
2481 (sibcall_insn_operand): Likewise.
2482 (GOT32_symbol_operand): Likewise.
2483 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2484 for -mindirect-branch-register.
2485 (tablejump): Likewise.
2486 (*sibcall_memory): Likewise.
2487 (*sibcall_value_memory): Likewise.
2488 Disallow peepholes of indirect call and jump via memory for
2489 -mindirect-branch-register.
2490 (*call_pop): Replace m with Bw.
2491 (*call_value_pop): Likewise.
2492 (*sibcall_pop_memory): Replace m with Bs.
2493 * config/i386/i386.opt (mindirect-branch-register): New option.
2494 * doc/invoke.texi: Document -mindirect-branch-register option.
2496 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2498 * config/i386/i386-protos.h (ix86_output_function_return): New.
2499 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2500 set function_return_type.
2501 (indirect_thunk_name): Add ret_p to indicate thunk for function
2503 (output_indirect_thunk_function): Pass false to
2504 indirect_thunk_name.
2505 (ix86_output_indirect_branch_via_reg): Likewise.
2506 (ix86_output_indirect_branch_via_push): Likewise.
2507 (output_indirect_thunk_function): Create alias for function
2508 return thunk if regno < 0.
2509 (ix86_output_function_return): New function.
2510 (ix86_handle_fndecl_attribute): Handle function_return.
2511 (ix86_attribute_table): Add function_return.
2512 * config/i386/i386.h (machine_function): Add
2513 function_return_type.
2514 * config/i386/i386.md (simple_return_internal): Use
2515 ix86_output_function_return.
2516 (simple_return_internal_long): Likewise.
2517 * config/i386/i386.opt (mfunction-return=): New option.
2518 (indirect_branch): Mention -mfunction-return=.
2519 * doc/extend.texi: Document function_return function attribute.
2520 * doc/invoke.texi: Document -mfunction-return= option.
2522 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2524 * config/i386/i386-opts.h (indirect_branch): New.
2525 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2526 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2527 with local indirect jump when converting indirect call and jump.
2528 (ix86_set_indirect_branch_type): New.
2529 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2530 (indirectlabelno): New.
2531 (indirect_thunk_needed): Likewise.
2532 (indirect_thunk_bnd_needed): Likewise.
2533 (indirect_thunks_used): Likewise.
2534 (indirect_thunks_bnd_used): Likewise.
2535 (INDIRECT_LABEL): Likewise.
2536 (indirect_thunk_name): Likewise.
2537 (output_indirect_thunk): Likewise.
2538 (output_indirect_thunk_function): Likewise.
2539 (ix86_output_indirect_branch_via_reg): Likewise.
2540 (ix86_output_indirect_branch_via_push): Likewise.
2541 (ix86_output_indirect_branch): Likewise.
2542 (ix86_output_indirect_jmp): Likewise.
2543 (ix86_code_end): Call output_indirect_thunk_function if needed.
2544 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2546 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2547 (ix86_attribute_table): Add indirect_branch.
2548 * config/i386/i386.h (machine_function): Add indirect_branch_type
2549 and has_local_indirect_jump.
2550 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2552 (tablejump): Likewise.
2553 (*indirect_jump): Use ix86_output_indirect_jmp.
2554 (*tablejump_1): Likewise.
2555 (simple_return_indirect_internal): Likewise.
2556 * config/i386/i386.opt (mindirect-branch=): New option.
2557 (indirect_branch): New.
2560 (thunk-inline): Likewise.
2561 (thunk-extern): Likewise.
2562 * doc/extend.texi: Document indirect_branch function attribute.
2563 * doc/invoke.texi: Document -mindirect-branch= option.
2565 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2568 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2570 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2572 * ipa-inline.c (want_inline_small_function_p): Return false if
2573 inlining has already failed with CIF_FINAL_ERROR.
2574 (update_caller_keys): Call want_inline_small_function_p before
2576 (update_callee_keys): Likewise.
2578 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2580 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2582 (rs6000_quadword_masked_address_p): Likewise.
2583 (quad_aligned_load_p): Likewise.
2584 (quad_aligned_store_p): Likewise.
2585 (const_load_sequence_p): Add comment to describe the outer-most loop.
2586 (mimic_memory_attributes_and_flags): New function.
2587 (rs6000_gen_stvx): Likewise.
2588 (replace_swapped_aligned_store): Likewise.
2589 (rs6000_gen_lvx): Likewise.
2590 (replace_swapped_aligned_load): Likewise.
2591 (replace_swapped_load_constant): Capitalize argument name in
2592 comment describing this function.
2593 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2594 and stores that access quad-word aligned addresses and replace
2595 with stvx or lvx instructions when appropriate.
2596 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2597 New function prototype.
2598 (rs6000_quadword_masked_address_p): Likewise.
2599 (rs6000_gen_lvx): Likewise.
2600 (rs6000_gen_stvx): Likewise.
2601 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2602 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2603 when memory address is aligned.
2604 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2605 this split to select lvx instruction when memory address is aligned.
2606 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2607 instruction when memory address is aligned.
2608 (*vsx_le_perm_load_v16qi): Likewise.
2609 (four unnamed splitters): Modify to select the stvx instruction
2610 when memory is aligned.
2612 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2614 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2615 which appears in the queue multiple times.
2617 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2618 Alan Hayward <alan.hayward@arm.com>
2619 David Sherwood <david.sherwood@arm.com>
2621 * tree-vectorizer.h (vec_lower_bound): New structure.
2622 (_loop_vec_info): Add check_nonzero and lower_bounds.
2623 (LOOP_VINFO_CHECK_NONZERO): New macro.
2624 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2625 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2626 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2627 fields. Make seg_len the distance travelled, not including the
2629 (dr_direction_indicator): Declare.
2630 (dr_zero_step_indicator): Likewise.
2631 (dr_known_forward_stride_p): Likewise.
2632 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2634 (runtime_alias_check_p): Allow runtime alias checks with
2636 (operator ==): Compare access_size and align.
2637 (prune_runtime_alias_test_list): Rework for new distinction between
2638 the access_size and seg_len.
2639 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2641 (get_segment_min_max): New function.
2642 (create_intersect_range_checks): Use it.
2643 (dr_step_indicator): New function.
2644 (dr_direction_indicator): Likewise.
2645 (dr_zero_step_indicator): Likewise.
2646 (dr_known_forward_stride_p): Likewise.
2647 * tree-loop-distribution.c (data_ref_segment_size): Return
2648 DR_STEP * (niters - 1).
2649 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2651 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2652 (vect_preserves_scalar_order_p): New function, split out from...
2653 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2654 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2655 (vect_vfa_access_size): New function.
2656 (vect_vfa_align): Likewise.
2657 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2658 (dump_lower_bound): New function.
2659 (vect_check_lower_bound): Likewise.
2660 (vect_small_gap_p): Likewise.
2661 (vectorizable_with_step_bound_p): Likewise.
2662 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2663 depencies if the vectorization factor is 1. Convert the checks
2664 for nonzero steps into checks on the bounds of DR_STEP. Try using
2665 a bunds check for variable steps if the minimum required step is
2666 relatively small. Update calls to the dr_with_seg_len
2667 constructor and to vect_compile_time_alias.
2668 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2670 (vect_loop_versioning): Call it.
2671 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2673 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2675 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2676 Alan Hayward <alan.hayward@arm.com>
2677 David Sherwood <david.sherwood@arm.com>
2679 * doc/sourcebuild.texi (vect_scatter_store): Document.
2680 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2682 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2684 * genopinit.c (main): Add supports_vec_scatter_store and
2685 supports_vec_scatter_store_cached to target_optabs.
2686 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2687 IFN_MASK_SCATTER_STORE.
2688 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2690 * internal-fn.h (internal_store_fn_p): Declare.
2691 (internal_fn_stored_value_index): Likewise.
2692 * internal-fn.c (scatter_store_direct): New macro.
2693 (expand_scatter_store_optab_fn): New function.
2694 (direct_scatter_store_optab_supported_p): New macro.
2695 (internal_store_fn_p): New function.
2696 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2697 IFN_MASK_SCATTER_STORE.
2698 (internal_fn_mask_index): Likewise.
2699 (internal_fn_stored_value_index): New function.
2700 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2702 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2703 * optabs-query.c (supports_vec_scatter_store_p): New function.
2704 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2705 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2706 true for scatter stores.
2707 (vect_gather_scatter_fn_p): Handle scatter stores too.
2708 (vect_check_gather_scatter): Consider using scatter stores if
2709 supports_vec_scatter_store_p.
2710 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2712 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2713 internal_fn_stored_value_index.
2714 (check_load_store_masking): Handle scatter stores too.
2715 (vect_get_store_rhs): Make public.
2716 (vectorizable_call): Use internal_store_fn_p.
2717 (vectorizable_store): Handle scatter store internal functions.
2718 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2719 when deciding whether the end of the group has been reached.
2720 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2721 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2722 (mask_scatter_store<mode>): New insns.
2724 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2725 Alan Hayward <alan.hayward@arm.com>
2726 David Sherwood <david.sherwood@arm.com>
2728 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2729 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2730 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2732 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2733 Use vect_truncate_gather_scatter_offset if we can't treat the
2734 operation as a normal gather load or scatter store.
2735 (get_group_load_store_type): Take the gather_scatter_info
2736 as argument. Try using a gather load or scatter store for
2737 single-element groups.
2738 (get_load_store_type): Update calls to get_group_load_store_type
2739 and vect_use_strided_gather_scatters_p.
2741 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2742 Alan Hayward <alan.hayward@arm.com>
2743 David Sherwood <david.sherwood@arm.com>
2745 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2746 optional tree argument.
2747 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2749 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2750 but continue to use the current value as a fallback.
2751 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2752 to compare the updates.
2753 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2754 (get_load_store_type): Use it when handling a strided access.
2755 (vect_get_strided_load_store_ops): New function.
2756 (vect_get_data_ptr_increment): Likewise.
2757 (vectorizable_load): Handle strided gather loads. Always pass
2758 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2760 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2761 Alan Hayward <alan.hayward@arm.com>
2762 David Sherwood <david.sherwood@arm.com>
2764 * doc/md.texi (gather_load@var{m}): Document.
2765 (mask_gather_load@var{m}): Likewise.
2766 * genopinit.c (main): Add supports_vec_gather_load and
2767 supports_vec_gather_load_cached to target_optabs.
2768 * optabs-tree.c (init_tree_optimization_optabs): Use
2769 ggc_cleared_alloc to allocate target_optabs.
2770 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2771 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2773 * internal-fn.h (internal_load_fn_p): Declare.
2774 (internal_gather_scatter_fn_p): Likewise.
2775 (internal_fn_mask_index): Likewise.
2776 (internal_gather_scatter_fn_supported_p): Likewise.
2777 * internal-fn.c (gather_load_direct): New macro.
2778 (expand_gather_load_optab_fn): New function.
2779 (direct_gather_load_optab_supported_p): New macro.
2780 (direct_internal_fn_optab): New function.
2781 (internal_load_fn_p): Likewise.
2782 (internal_gather_scatter_fn_p): Likewise.
2783 (internal_fn_mask_index): Likewise.
2784 (internal_gather_scatter_fn_supported_p): Likewise.
2785 * optabs-query.c (supports_at_least_one_mode_p): New function.
2786 (supports_vec_gather_load_p): Likewise.
2787 * optabs-query.h (supports_vec_gather_load_p): Declare.
2788 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2789 and memory_type field.
2790 (NUM_PATTERNS): Bump to 15.
2791 * tree-vect-data-refs.c: Include internal-fn.h.
2792 (vect_gather_scatter_fn_p): New function.
2793 (vect_describe_gather_scatter_call): Likewise.
2794 (vect_check_gather_scatter): Try using internal functions for
2795 gather loads. Recognize existing calls to a gather load function.
2796 (vect_analyze_data_refs): Consider using gather loads if
2797 supports_vec_gather_load_p.
2798 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2799 (vect_get_gather_scatter_offset_type): Likewise.
2800 (vect_convert_mask_for_vectype): Likewise.
2801 (vect_add_conversion_to_patterm): Likewise.
2802 (vect_try_gather_scatter_pattern): Likewise.
2803 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2804 (vect_vect_recog_func_ptrs): Add it.
2805 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2806 internal_fn_mask_index and internal_gather_scatter_fn_p.
2807 (check_load_store_masking): Take the gather_scatter_info as an
2808 argument and handle gather loads.
2809 (vect_get_gather_scatter_ops): New function.
2810 (vectorizable_call): Check internal_load_fn_p.
2811 (vectorizable_load): Likewise. Handle gather load internal
2813 (vectorizable_store): Update call to check_load_store_masking.
2814 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2815 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2816 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2817 (aarch64_gather_scale_operand_d): New predicates.
2818 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2819 (mask_gather_load<mode>): New insns.
2821 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2822 Alan Hayward <alan.hayward@arm.com>
2823 David Sherwood <david.sherwood@arm.com>
2825 * optabs.def (fold_left_plus_optab): New optab.
2826 * doc/md.texi (fold_left_plus_@var{m}): Document.
2827 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2828 * internal-fn.c (fold_left_direct): Define.
2829 (expand_fold_left_optab_fn): Likewise.
2830 (direct_fold_left_optab_supported_p): Likewise.
2831 * fold-const-call.c (fold_const_fold_left): New function.
2832 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2833 * tree-parloops.c (valid_reduction_p): New function.
2834 (gather_scalar_reductions): Use it.
2835 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2836 (vect_finish_replace_stmt): Declare.
2837 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2838 (needs_fold_left_reduction_p): New function, split out from...
2839 (vect_is_simple_reduction): ...here. Accept reductions that
2840 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2841 (vect_force_simple_reduction): Also store the reduction type in
2842 the assignment's STMT_VINFO_REDUC_TYPE.
2843 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2844 (merge_with_identity): New function.
2845 (vect_expand_fold_left): Likewise.
2846 (vectorize_fold_left_reduction): Likewise.
2847 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2848 scalar phi in place for it. Check for target support and reject
2849 cases that would reassociate the operation. Defer the transform
2850 phase to vectorize_fold_left_reduction.
2851 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2852 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2853 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2855 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2857 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2858 call to ifc_temp_var.
2860 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2861 Alan Hayward <alan.hayward@arm.com>
2862 David Sherwood <david.sherwood@arm.com>
2864 * target.def (legitimize_address_displacement): Take the original
2865 offset as a poly_int.
2866 * targhooks.h (default_legitimize_address_displacement): Update
2868 * targhooks.c (default_legitimize_address_displacement): Likewise.
2869 * doc/tm.texi: Regenerate.
2870 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2871 as an argument, moving assert of ad->disp == ad->disp_term to...
2872 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2873 Try calling targetm.legitimize_address_displacement before expanding
2874 the address rather than afterwards, and adjust for the new interface.
2875 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2876 Match the new hook interface. Handle SVE addresses.
2877 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2880 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2882 * Makefile.in (OBJS): Add early-remat.o.
2883 * target.def (select_early_remat_modes): New hook.
2884 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2885 * doc/tm.texi: Regenerate.
2886 * targhooks.h (default_select_early_remat_modes): Declare.
2887 * targhooks.c (default_select_early_remat_modes): New function.
2888 * timevar.def (TV_EARLY_REMAT): New timevar.
2889 * passes.def (pass_early_remat): New pass.
2890 * tree-pass.h (make_pass_early_remat): Declare.
2891 * early-remat.c: New file.
2892 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2894 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2896 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2897 Alan Hayward <alan.hayward@arm.com>
2898 David Sherwood <david.sherwood@arm.com>
2900 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2901 vfm1 with a bound_epilog parameter.
2902 (vect_do_peeling): Update calls accordingly, and move the prologue
2903 call earlier in the function. Treat the base bound_epilog as 0 for
2904 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2905 this base when peeling for gaps.
2906 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2907 with fully-masked loops.
2908 (vect_estimate_min_profitable_iters): Handle the single peeled
2909 iteration in that case.
2911 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2912 Alan Hayward <alan.hayward@arm.com>
2913 David Sherwood <david.sherwood@arm.com>
2915 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2916 single-element interleaving even if the size is not a power of 2.
2917 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2918 accesses for single-element interleaving if the group size is
2921 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2922 Alan Hayward <alan.hayward@arm.com>
2923 David Sherwood <david.sherwood@arm.com>
2925 * doc/md.texi (fold_extract_last_@var{m}): Document.
2926 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2927 * optabs.def (fold_extract_last_optab): New optab.
2928 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2929 * internal-fn.c (fold_extract_direct): New macro.
2930 (expand_fold_extract_optab_fn): Likewise.
2931 (direct_fold_extract_optab_supported_p): Likewise.
2932 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2933 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2934 EXTRACT_LAST_REDUCTION.
2935 (get_initial_def_for_reduction): Do not create an initial vector
2936 for EXTRACT_LAST_REDUCTION reductions.
2937 (vectorizable_reduction): Leave the scalar phi in place for
2938 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2939 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2940 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2941 transform phase to vectorizable_condition.
2942 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2944 (vect_finish_stmt_generation): ...here.
2945 (vect_finish_replace_stmt): New function.
2946 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2947 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2949 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2951 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2952 Alan Hayward <alan.hayward@arm.com>
2953 David Sherwood <david.sherwood@arm.com>
2955 * doc/md.texi (extract_last_@var{m}): Document.
2956 * optabs.def (extract_last_optab): New optab.
2957 * internal-fn.def (EXTRACT_LAST): New internal function.
2958 * internal-fn.c (cond_unary_direct): New macro.
2959 (expand_cond_unary_optab_fn): Likewise.
2960 (direct_cond_unary_optab_supported_p): Likewise.
2961 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2962 loops using EXTRACT_LAST.
2963 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2964 (extract_last_<mode>): ...this optab.
2965 (vec_extract<mode><Vel>): Update accordingly.
2967 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2968 Alan Hayward <alan.hayward@arm.com>
2969 David Sherwood <david.sherwood@arm.com>
2971 * target.def (empty_mask_is_expensive): New hook.
2972 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2973 * doc/tm.texi: Regenerate.
2974 * targhooks.h (default_empty_mask_is_expensive): Declare.
2975 * targhooks.c (default_empty_mask_is_expensive): New function.
2976 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2977 if the target says that empty masks are expensive.
2978 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2980 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2982 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2983 Alan Hayward <alan.hayward@arm.com>
2984 David Sherwood <david.sherwood@arm.com>
2986 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2987 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2988 (vect_use_loop_mask_for_alignment_p): New function.
2989 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2990 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2991 niters_skip argument. Make sure that the first niters_skip elements
2992 of the first iteration are inactive.
2993 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2994 Update call to vect_set_loop_masks_directly.
2995 (get_misalign_in_elems): New function, split out from...
2996 (vect_gen_prolog_loop_niters): ...here.
2997 (vect_update_init_of_dr): Take a code argument that specifies whether
2998 the adjustment should be added or subtracted.
2999 (vect_update_init_of_drs): Likewise.
3000 (vect_prepare_for_masked_peels): New function.
3001 (vect_do_peeling): Skip prologue peeling if we're using a mask
3002 instead. Update call to vect_update_inits_of_drs.
3003 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3005 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3006 alignment. Do not include the number of peeled iterations in
3007 the minimum threshold in that case.
3008 (vectorizable_induction): Adjust the start value down by
3009 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3010 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3011 Take the number of skipped iterations into account when calculating
3013 * tree-vect-stmts.c (vect_gen_while_not): New function.
3015 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3016 Alan Hayward <alan.hayward@arm.com>
3017 David Sherwood <david.sherwood@arm.com>
3019 * doc/sourcebuild.texi (vect_fully_masked): Document.
3020 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3022 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3024 (vect_analyze_loop_2): ...here. Don't check the vectorization
3025 factor against the number of loop iterations if the loop is
3028 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3029 Alan Hayward <alan.hayward@arm.com>
3030 David Sherwood <david.sherwood@arm.com>
3032 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3033 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3034 (dump_groups): Update accordingly.
3035 (iv_use::mem_type): New member variable.
3036 (address_p): New function.
3037 (record_use): Add a mem_type argument and initialize the new
3039 (record_group_use): Add a mem_type argument. Use address_p.
3040 Remove obsolete null checks of base_object. Update call to record_use.
3041 (find_interesting_uses_op): Update call to record_group_use.
3042 (find_interesting_uses_cond): Likewise.
3043 (find_interesting_uses_address): Likewise.
3044 (get_mem_type_for_internal_fn): New function.
3045 (find_address_like_use): Likewise.
3046 (find_interesting_uses_stmt): Try find_address_like_use before
3047 calling find_interesting_uses_op.
3048 (addr_offset_valid_p): Use the iv mem_type field as the type
3049 of the addressed memory.
3050 (add_autoinc_candidates): Likewise.
3051 (get_address_cost): Likewise.
3052 (split_small_address_groups_p): Use address_p.
3053 (split_address_groups): Likewise.
3054 (add_iv_candidate_for_use): Likewise.
3055 (autoinc_possible_for_pair): Likewise.
3056 (rewrite_groups): Likewise.
3057 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3058 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3059 (get_alias_ptr_type_for_ptr_address): New function.
3060 (rewrite_use_address): Rewrite address uses in calls that were
3061 identified by find_address_like_use.
3063 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3064 Alan Hayward <alan.hayward@arm.com>
3065 David Sherwood <david.sherwood@arm.com>
3067 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3069 * gimple-expr.h (is_gimple_addressable: Likewise.
3070 * gimple-expr.c (is_gimple_address): Likewise.
3071 * internal-fn.c (expand_call_mem_ref): New function.
3072 (expand_mask_load_optab_fn): Use it.
3073 (expand_mask_store_optab_fn): Likewise.
3075 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3076 Alan Hayward <alan.hayward@arm.com>
3077 David Sherwood <david.sherwood@arm.com>
3079 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3080 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3081 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3082 (cond_umax@var{mode}): Document.
3083 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3084 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3085 (cond_umin_optab, cond_umax_optab): New optabs.
3086 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3087 (COND_IOR, COND_XOR): New internal functions.
3088 * internal-fn.h (get_conditional_internal_fn): Declare.
3089 * internal-fn.c (cond_binary_direct): New macro.
3090 (expand_cond_binary_optab_fn): Likewise.
3091 (direct_cond_binary_optab_supported_p): Likewise.
3092 (get_conditional_internal_fn): New function.
3093 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3094 Cope with reduction statements that are vectorized as calls rather
3096 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3097 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3098 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3099 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3100 (UNSPEC_COND_EOR): New unspecs.
3101 (optab): Add mappings for them.
3102 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3103 (sve_int_op, sve_fp_op): New int attributes.
3105 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3106 Alan Hayward <alan.hayward@arm.com>
3107 David Sherwood <david.sherwood@arm.com>
3109 * optabs.def (while_ult_optab): New optab.
3110 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3111 * internal-fn.def (WHILE_ULT): New internal function.
3112 * internal-fn.h (direct_internal_fn_supported_p): New override
3113 that takes two types as argument.
3114 * internal-fn.c (while_direct): New macro.
3115 (expand_while_optab_fn): New function.
3116 (convert_optab_supported_p): Likewise.
3117 (direct_while_optab_supported_p): New macro.
3118 * wide-int.h (wi::udiv_ceil): New function.
3119 * tree-vectorizer.h (rgroup_masks): New structure.
3120 (vec_loop_masks): New typedef.
3121 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3123 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3124 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3125 (vect_max_vf): New function.
3126 (slpeel_make_loop_iterate_ntimes): Delete.
3127 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3128 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3129 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3130 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3131 internal-fn.h, stor-layout.h and optabs-query.h.
3132 (vect_set_loop_mask): New function.
3133 (add_preheader_seq): Likewise.
3134 (add_header_seq): Likewise.
3135 (interleave_supported_p): Likewise.
3136 (vect_maybe_permute_loop_masks): Likewise.
3137 (vect_set_loop_masks_directly): Likewise.
3138 (vect_set_loop_condition_masked): Likewise.
3139 (vect_set_loop_condition_unmasked): New function, split out from
3140 slpeel_make_loop_iterate_ntimes.
3141 (slpeel_make_loop_iterate_ntimes): Rename to..
3142 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3143 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3144 (vect_do_peeling): Update call accordingly.
3145 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3147 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3148 mask_compare_type, can_fully_mask_p and fully_masked_p.
3149 (release_vec_loop_masks): New function.
3150 (_loop_vec_info): Use it to free the loop masks.
3151 (can_produce_all_loop_masks_p): New function.
3152 (vect_get_max_nscalars_per_iter): Likewise.
3153 (vect_verify_full_masking): Likewise.
3154 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3155 retries, and free the mask rgroups before retrying. Check loop-wide
3156 reasons for disallowing fully-masked loops. Make the final decision
3157 about whether use a fully-masked loop or not.
3158 (vect_estimate_min_profitable_iters): Do not assume that peeling
3159 for the number of iterations will be needed for fully-masked loops.
3160 (vectorizable_reduction): Disable fully-masked loops.
3161 (vectorizable_live_operation): Likewise.
3162 (vect_halve_mask_nunits): New function.
3163 (vect_double_mask_nunits): Likewise.
3164 (vect_record_loop_mask): Likewise.
3165 (vect_get_loop_mask): Likewise.
3166 (vect_transform_loop): Handle the case in which the final loop
3167 iteration might handle a partial vector. Call vect_set_loop_condition
3168 instead of slpeel_make_loop_iterate_ntimes.
3169 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3170 (check_load_store_masking): New function.
3171 (prepare_load_store_mask): Likewise.
3172 (vectorizable_store): Handle fully-masked loops.
3173 (vectorizable_load): Likewise.
3174 (supportable_widening_operation): Use vect_halve_mask_nunits for
3176 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3177 (vect_gen_while): New function.
3178 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3179 (aarch64_uqdec<mode>): New insn.
3181 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3182 Alan Hayward <alan.hayward@arm.com>
3183 David Sherwood <david.sherwood@arm.com>
3185 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3186 (reduc_xor_scal_optab): New optabs.
3187 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3188 (reduc_xor_scal_@var{m}): Document.
3189 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3190 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3192 * fold-const-call.c (fold_const_call): Handle them.
3193 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3194 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3195 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3196 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3197 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3198 (UNSPEC_XORV): New unspecs.
3199 (optab): Add entries for them.
3200 (BITWISEV): New int iterator.
3201 (bit_reduc_op): New int attributes.
3203 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3204 Alan Hayward <alan.hayward@arm.com>
3205 David Sherwood <david.sherwood@arm.com>
3207 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3208 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3209 * optabs.def (vec_shl_insert_optab): New optab.
3210 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3211 (duplicate_and_interleave): Likewise.
3212 * tree-vect-loop.c: Include internal-fn.h.
3213 (neutral_op_for_slp_reduction): New function, split out from
3214 get_initial_defs_for_reduction.
3215 (get_initial_def_for_reduction): Handle option 2 for variable-length
3216 vectors by loading the neutral value into a vector and then shifting
3217 the initial value into element 0.
3218 (get_initial_defs_for_reduction): Replace the code argument with
3219 the neutral value calculated by neutral_op_for_slp_reduction.
3220 Use gimple_build_vector for constant-length vectors.
3221 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3222 but the first group_size elements have a neutral value.
3223 Use duplicate_and_interleave otherwise.
3224 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3225 Update call to get_initial_defs_for_reduction. Handle SLP
3226 reductions for variable-length vectors by creating one vector
3227 result for each scalar result, with the elements associated
3228 with other scalar results stubbed out with the neutral value.
3229 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3230 Require IFN_VEC_SHL_INSERT for double reductions on
3231 variable-length vectors, or SLP reductions that have
3232 a neutral value. Require can_duplicate_and_interleave_p
3233 support for variable-length unchained SLP reductions if there
3234 is no neutral value, such as for MIN/MAX reductions. Also require
3235 the number of vector elements to be a multiple of the number of
3236 SLP statements when doing variable-length unchained SLP reductions.
3237 Update call to vect_create_epilog_for_reduction.
3238 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3239 and remove initial values.
3240 (duplicate_and_interleave): Make public.
3241 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3242 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3244 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3245 Alan Hayward <alan.hayward@arm.com>
3246 David Sherwood <david.sherwood@arm.com>
3248 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3249 (can_duplicate_and_interleave_p): New function.
3250 (vect_get_and_check_slp_defs): Take the vector of statements
3251 rather than just the current one. Remove excess parentheses.
3252 Restriction rejectinon of vect_constant_def and vect_external_def
3253 for variable-length vectors to boolean types, or types for which
3254 can_duplicate_and_interleave_p is false.
3255 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3256 (duplicate_and_interleave): New function.
3257 (vect_get_constant_vectors): Use gimple_build_vector for
3258 constant-length vectors and suitable variable-length constant
3259 vectors. Use duplicate_and_interleave for other variable-length
3260 vectors. Don't defer the update when inserting new statements.
3262 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3263 Alan Hayward <alan.hayward@arm.com>
3264 David Sherwood <david.sherwood@arm.com>
3266 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3267 min_profitable_iters doesn't go negative.
3269 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3270 Alan Hayward <alan.hayward@arm.com>
3271 David Sherwood <david.sherwood@arm.com>
3273 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3274 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3275 * optabs.def (vec_mask_load_lanes_optab): New optab.
3276 (vec_mask_store_lanes_optab): Likewise.
3277 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3278 (MASK_STORE_LANES): Likewise.
3279 * internal-fn.c (mask_load_lanes_direct): New macro.
3280 (mask_store_lanes_direct): Likewise.
3281 (expand_mask_load_optab_fn): Handle masked operations.
3282 (expand_mask_load_lanes_optab_fn): New macro.
3283 (expand_mask_store_optab_fn): Handle masked operations.
3284 (expand_mask_store_lanes_optab_fn): New macro.
3285 (direct_mask_load_lanes_optab_supported_p): Likewise.
3286 (direct_mask_store_lanes_optab_supported_p): Likewise.
3287 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3289 (vect_load_lanes_supported): Likewise.
3290 * tree-vect-data-refs.c (strip_conversion): New function.
3291 (can_group_stmts_p): Likewise.
3292 (vect_analyze_data_ref_accesses): Use it instead of checking
3293 for a pair of assignments.
3294 (vect_store_lanes_supported): Take a masked_p parameter.
3295 (vect_load_lanes_supported): Likewise.
3296 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3297 vect_store_lanes_supported and vect_load_lanes_supported.
3298 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3299 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3300 parameter. Don't allow gaps for masked accesses.
3301 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3302 and vect_load_lanes_supported.
3303 (get_load_store_type): Take a masked_p parameter and update
3304 call to get_group_load_store_type.
3305 (vectorizable_store): Update call to get_load_store_type.
3306 Handle IFN_MASK_STORE_LANES.
3307 (vectorizable_load): Update call to get_load_store_type.
3308 Handle IFN_MASK_LOAD_LANES.
3310 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3311 Alan Hayward <alan.hayward@arm.com>
3312 David Sherwood <david.sherwood@arm.com>
3314 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3316 * config/aarch64/aarch64-protos.h
3317 (aarch64_sve_struct_memory_operand_p): Declare.
3318 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3319 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3320 (VPRED, vpred): Handle SVE structure modes.
3321 * config/aarch64/constraints.md (Utx): New constraint.
3322 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3323 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3324 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3325 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3326 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3327 structure modes. Split into pieces after RA.
3328 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3329 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3331 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3332 SVE structure modes.
3333 (aarch64_classify_address): Likewise.
3334 (sizetochar): Move earlier in file.
3335 (aarch64_print_operand): Handle SVE register lists.
3336 (aarch64_array_mode): New function.
3337 (aarch64_sve_struct_memory_operand_p): Likewise.
3338 (TARGET_ARRAY_MODE): Redefine.
3340 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3341 Alan Hayward <alan.hayward@arm.com>
3342 David Sherwood <david.sherwood@arm.com>
3344 * target.def (array_mode): New target hook.
3345 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3346 * doc/tm.texi: Regenerate.
3347 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3348 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3349 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3351 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3354 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3355 Alan Hayward <alan.hayward@arm.com>
3356 David Sherwood <david.sherwood@arm.com>
3358 * fold-const.c (fold_binary_loc): Check the argument types
3359 rather than the result type when testing for a vector operation.
3361 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3363 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3364 * doc/tm.texi: Regenerate.
3366 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3367 Alan Hayward <alan.hayward@arm.com>
3368 David Sherwood <david.sherwood@arm.com>
3370 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3371 (sve): Document new AArch64 extension.
3372 * doc/md.texi (w): Extend the description of the AArch64
3373 constraint to include SVE vectors.
3374 (Upl, Upa): Document new AArch64 predicate constraints.
3375 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3377 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3378 (msve-vector-bits=): New option.
3379 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3380 SVE when these are disabled.
3381 (sve): New extension.
3382 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3383 modes. Adjust their number of units based on aarch64_sve_vg.
3384 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3385 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3386 aarch64_addr_query_type.
3387 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3388 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3389 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3390 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3391 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3392 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3393 (aarch64_simd_imm_zero_p): Delete.
3394 (aarch64_check_zero_based_sve_index_immediate): Declare.
3395 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3396 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3397 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3398 (aarch64_sve_float_mul_immediate_p): Likewise.
3399 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3401 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3402 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3403 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3404 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3405 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3406 (aarch64_regmode_natural_size): Likewise.
3407 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3408 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3410 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3411 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3412 for VG and the SVE predicate registers.
3413 (V_ALIASES): Add a "z"-prefixed alias.
3414 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3415 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3416 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3417 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3418 (REG_CLASS_NAMES): Add entries for them.
3419 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3420 and the predicate registers.
3421 (aarch64_sve_vg): Declare.
3422 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3423 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3424 (REGMODE_NATURAL_SIZE): Define.
3425 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3427 * config/aarch64/aarch64.c: Include cfgrtl.h.
3428 (simd_immediate_info): Add a constructor for series vectors,
3429 and an associated step field.
3430 (aarch64_sve_vg): New variable.
3431 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3432 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3433 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3434 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3435 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3436 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3437 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3438 (aarch64_get_mask_mode): New functions.
3439 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3440 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3441 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3442 predicate modes and predicate registers. Explicitly restrict
3443 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3444 to store a vector mode if it is recognized by
3445 aarch64_classify_vector_mode.
3446 (aarch64_regmode_natural_size): New function.
3447 (aarch64_hard_regno_caller_save_mode): Return the original mode
3449 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3450 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3451 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3452 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3454 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3455 does not overlap dest if the function is frame-related. Handle
3457 (aarch64_split_add_offset): New function.
3458 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3459 them aarch64_add_offset.
3460 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3461 and update call to aarch64_sub_sp.
3462 (aarch64_add_cfa_expression): New function.
3463 (aarch64_expand_prologue): Pass extra temporary registers to the
3464 functions above. Handle the case in which we need to emit new
3465 DW_CFA_expressions for registers that were originally saved
3466 relative to the stack pointer, but now have to be expressed
3467 relative to the frame pointer.
3468 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3470 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3471 IP0 and IP1 values for SVE frames.
3472 (aarch64_expand_vec_series): New function.
3473 (aarch64_expand_sve_widened_duplicate): Likewise.
3474 (aarch64_expand_sve_const_vector): Likewise.
3475 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3476 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3477 into the register, rather than emitting a SET directly.
3478 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3479 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3480 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3481 (offset_9bit_signed_scaled_p): New functions.
3482 (aarch64_replicate_bitmask_imm): New function.
3483 (aarch64_bitmask_imm): Use it.
3484 (aarch64_cannot_force_const_mem): Reject expressions involving
3485 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3486 (aarch64_classify_index): Handle SVE indices, by requiring
3487 a plain register index with a scale that matches the element size.
3488 (aarch64_classify_address): Handle SVE addresses. Assert that
3489 the mode of the address is VOIDmode or an integer mode.
3490 Update call to aarch64_classify_symbol.
3491 (aarch64_classify_symbolic_expression): Update call to
3492 aarch64_classify_symbol.
3493 (aarch64_const_vec_all_in_range_p): New function.
3494 (aarch64_print_vector_float_operand): Likewise.
3495 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3496 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3497 and the FP immediates 1.0 and 0.5.
3498 (aarch64_print_address_internal): Handle SVE addresses.
3499 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3500 (aarch64_regno_regclass): Handle predicate registers.
3501 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3503 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3504 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3505 (aarch64_convert_sve_vector_bits): New function.
3506 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3507 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3509 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3510 Handle SVE vector and predicate modes. Accept VL-based constants
3511 that need only one temporary register, and VL offsets that require
3512 no temporary registers.
3513 (aarch64_conditional_register_usage): Mark the predicate registers
3514 as fixed if SVE isn't available.
3515 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3516 Return true for SVE vector and predicate modes.
3517 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3518 rather than an unsigned int. Handle SVE modes.
3519 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3521 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3523 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3524 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3525 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3526 (aarch64_sve_float_mul_immediate_p): New functions.
3527 (aarch64_sve_valid_immediate): New function.
3528 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3529 Explicitly reject structure modes. Check for INDEX constants.
3530 Handle PTRUE and PFALSE constants.
3531 (aarch64_check_zero_based_sve_index_immediate): New function.
3532 (aarch64_simd_imm_zero_p): Delete.
3533 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3534 vector modes. Accept constants in the range of CNT[BHWD].
3535 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3536 ask for an Advanced SIMD mode.
3537 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3538 (aarch64_simd_vector_alignment): Handle SVE predicates.
3539 (aarch64_vectorize_preferred_vector_alignment): New function.
3540 (aarch64_simd_vector_alignment_reachable): Use it instead of
3542 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3543 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3545 (MAX_VECT_LEN): Delete.
3546 (expand_vec_perm_d): Add a vec_flags field.
3547 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3548 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3549 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3551 (aarch64_evpc_rev): Rename to...
3552 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3553 (aarch64_evpc_rev_global): New function.
3554 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3555 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3557 (aarch64_evpc_sve_tbl): New function.
3558 (aarch64_expand_vec_perm_const_1): Update after rename of
3559 aarch64_evpc_rev. Handle SVE permutes too, trying
3560 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3561 than aarch64_evpc_tbl.
3562 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3563 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3564 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3565 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3566 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3567 (aarch64_expand_sve_vcond): New functions.
3568 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3569 of aarch64_vector_mode_p.
3570 (aarch64_dwarf_poly_indeterminate_value): New function.
3571 (aarch64_compute_pressure_classes): Likewise.
3572 (aarch64_can_change_mode_class): Likewise.
3573 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3574 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3575 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3576 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3577 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3578 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3579 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3580 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3582 (Dn, Dl, Dr): Accept const as well as const_vector.
3583 (Dz): Likewise. Compare against CONST0_RTX.
3584 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3585 of "vector" where appropriate.
3586 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3587 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3588 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3589 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3590 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3591 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3592 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3593 (v_int_equiv): Extend to SVE modes.
3594 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3596 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3597 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3598 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3599 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3600 (SVE_COND_FP_CMP): New int iterators.
3601 (perm_hilo): Handle the new unpack unspecs.
3602 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3604 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3605 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3606 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3607 (aarch64_equality_operator, aarch64_constant_vector_operand)
3608 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3609 (aarch64_sve_nonimmediate_operand): Likewise.
3610 (aarch64_sve_general_operand): Likewise.
3611 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3612 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3613 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3614 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3615 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3616 (aarch64_sve_float_arith_immediate): Likewise.
3617 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3618 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3619 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3620 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3621 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3622 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3623 (aarch64_sve_float_arith_operand): Likewise.
3624 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3625 (aarch64_sve_float_mul_operand): Likewise.
3626 (aarch64_sve_vec_perm_operand): Likewise.
3627 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3628 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3629 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3630 as well as const_vector.
3631 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3632 in file. Use CONST0_RTX and CONSTM1_RTX.
3633 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3634 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3635 Use aarch64_simd_imm_zero.
3636 * config/aarch64/aarch64-sve.md: New file.
3637 * config/aarch64/aarch64.md: Include it.
3638 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3639 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3640 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3641 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3642 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3643 (sve): New attribute.
3644 (enabled): Disable instructions with the sve attribute unless
3646 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3647 aarch64_expand_mov_immediate.
3648 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3649 CNT[BHSD] immediates.
3650 (movti): Split CONST_POLY_INT moves into two halves.
3651 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3652 Split additions that need a temporary here if the destination
3653 is the stack pointer.
3654 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3655 (*add<mode>3_poly_1): New instruction.
3656 (set_clobber_cc): New expander.
3658 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3660 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3661 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3662 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3663 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3664 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3665 Change innermode from fixed_mode_size to machine_mode.
3666 (simplify_subreg): Update call accordingly. Handle a constant-sized
3667 subreg of a variable-length CONST_VECTOR.
3669 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3670 Alan Hayward <alan.hayward@arm.com>
3671 David Sherwood <david.sherwood@arm.com>
3673 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3674 (add_offset_to_base): New function, split out from...
3675 (create_mem_ref): ...here. When handling a scale other than 1,
3676 check first whether the address is valid without the offset.
3677 Add it into the base if so, leaving the index and scale as-is.
3679 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3682 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3683 fold_for_warn before checking if arg2 is INTEGER_CST.
3685 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3687 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3688 (store_multiple_operation): Delete.
3689 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3690 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3691 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3692 guarded by TARGET_STRING.
3693 (rs6000_output_load_multiple): Delete.
3694 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3695 OPTION_MASK_STRING / TARGET_STRING handling.
3696 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3697 (const rs6000_opt_masks) <"string">: Change mask to 0.
3698 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3699 (MASK_STRING): Delete.
3700 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3702 (load_multiple): Delete.
3709 (store_multiple): Delete.
3716 (movmemsi_8reg): Delete.
3717 (corresponding unnamed define_insn): Delete.
3718 (movmemsi_6reg): Delete.
3719 (corresponding unnamed define_insn): Delete.
3720 (movmemsi_4reg): Delete.
3721 (corresponding unnamed define_insn): Delete.
3722 (movmemsi_2reg): Delete.
3723 (corresponding unnamed define_insn): Delete.
3724 (movmemsi_1reg): Delete.
3725 (corresponding unnamed define_insn): Delete.
3726 * config/rs6000/rs6000.opt (mno-string): New.
3727 (mstring): Replace by deprecation warning stub.
3728 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3730 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3732 * regrename.c (regrename_do_replace): If replacing the same
3733 reg multiple times, try to reuse last created gen_raw_REG.
3736 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3737 main to workaround a bug in GDB.
3739 2018-01-12 Tom de Vries <tom@codesourcery.com>
3742 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3744 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3746 PR rtl-optimization/80481
3747 * ira-color.c (get_cap_member): New function.
3748 (allocnos_conflict_by_live_ranges_p): Use it.
3749 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3750 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3752 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3755 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3756 (*saddl_se_1): Ditto.
3758 (*ssubl_se_1): Ditto.
3760 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3762 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3763 rather than wi::to_widest for DR_INITs.
3764 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3765 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3766 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3768 (vect_analyze_group_access_1): Note that here.
3770 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3772 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3773 polynomial type sizes.
3775 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3777 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3778 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3779 (gimple_add_tmp_var): Likewise.
3781 2018-01-12 Martin Liska <mliska@suse.cz>
3783 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3784 (gimple_alloc_sizes): Likewise.
3785 (dump_gimple_statistics): Use PRIu64 in printf format.
3786 * gimple.h: Change uint64_t to int.
3788 2018-01-12 Martin Liska <mliska@suse.cz>
3790 * tree-core.h: Use uint64_t instead of int.
3791 * tree.c (tree_node_counts): Likewise.
3792 (tree_node_sizes): Likewise.
3793 (dump_tree_statistics): Use PRIu64 in printf format.
3795 2018-01-12 Martin Liska <mliska@suse.cz>
3797 * Makefile.in: As qsort_chk is implemented in vec.c, add
3798 vec.o to linkage of gencfn-macros.
3799 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3800 passing the info to record_node_allocation_statistics.
3801 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3803 * ggc-common.c (struct ggc_usage): Add operator== and use
3804 it in operator< and compare function.
3805 * mem-stats.h (struct mem_usage): Likewise.
3806 * vec.c (struct vec_usage): Remove operator< and compare
3807 function. Can be simply inherited.
3809 2018-01-12 Martin Jambor <mjambor@suse.cz>
3812 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3813 * tree-ssa-math-opts.c: Include domwalk.h.
3814 (convert_mult_to_fma_1): New function.
3815 (fma_transformation_info): New type.
3816 (fma_deferring_state): Likewise.
3817 (cancel_fma_deferring): New function.
3818 (result_of_phi): Likewise.
3819 (last_fma_candidate_feeds_initial_phi): Likewise.
3820 (convert_mult_to_fma): Added deferring logic, split actual
3821 transformation to convert_mult_to_fma_1.
3822 (math_opts_dom_walker): New type.
3823 (math_opts_dom_walker::after_dom_children): New method, body moved
3824 here from pass_optimize_widening_mul::execute, added deferring logic
3826 (pass_optimize_widening_mul::execute): Moved most of code to
3827 math_opts_dom_walker::after_dom_children.
3828 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3829 * config/i386/i386.c (ix86_option_override_internal): Added
3830 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3832 2018-01-12 Richard Biener <rguenther@suse.de>
3835 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3836 inline instance vars.
3838 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3841 * config/rx/rx.c (rx_is_restricted_memory_address):
3844 2018-01-12 Richard Biener <rguenther@suse.de>
3846 PR tree-optimization/80846
3847 * target.def (split_reduction): New target hook.
3848 * targhooks.c (default_split_reduction): New function.
3849 * targhooks.h (default_split_reduction): Declare.
3850 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3851 target requests first reduce vectors by combining low and high
3853 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3854 (get_vectype_for_scalar_type_and_size): Export.
3855 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3856 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3857 * doc/tm.texi: Regenerate.
3858 * config/i386/i386.c (ix86_split_reduction): Implement
3859 TARGET_VECTORIZE_SPLIT_REDUCTION.
3861 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3864 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3865 in PIC mode except for TARGET_VXWORKS_RTP.
3866 * config/sparc/sparc.c: Include cfgrtl.h.
3867 (TARGET_INIT_PIC_REG): Define.
3868 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3869 (sparc_pic_register_p): New predicate.
3870 (sparc_legitimate_address_p): Use it.
3871 (sparc_legitimize_pic_address): Likewise.
3872 (sparc_delegitimize_address): Likewise.
3873 (sparc_mode_dependent_address_p): Likewise.
3874 (gen_load_pcrel_sym): Remove 4th parameter.
3875 (load_got_register): Adjust call to above. Remove obsolete stuff.
3876 (sparc_expand_prologue): Do not call load_got_register here.
3877 (sparc_flat_expand_prologue): Likewise.
3878 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3879 (sparc_use_pseudo_pic_reg): New function.
3880 (sparc_init_pic_reg): Likewise.
3881 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3882 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3884 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3886 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3887 Add item for branch_cost.
3889 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3891 PR rtl-optimization/83565
3892 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3893 not extend the result to a larger mode for rotate operations.
3894 (num_sign_bit_copies1): Likewise.
3896 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3899 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3901 Use values-Xc.o for -pedantic.
3902 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3904 2018-01-12 Martin Liska <mliska@suse.cz>
3907 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3909 (possible_polymorphic_call_targets): Use it.
3910 (ipa_devirt): Likewise.
3912 2018-01-12 Martin Liska <mliska@suse.cz>
3914 * profile-count.h (enum profile_quality): Use 0 as invalid
3915 enum value of profile_quality.
3917 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3919 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3920 -mext-string options.
3922 2018-01-12 Richard Biener <rguenther@suse.de>
3924 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3925 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3926 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3928 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3930 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3932 * configure.ac (--with-long-double-format): Add support for the
3933 configuration option to change the default long double format on
3935 * config.gcc (powerpc*-linux*-*): Likewise.
3936 * configure: Regenerate.
3937 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3938 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3939 used without modification.
3941 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3943 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3944 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3945 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3946 MISC_BUILTIN_SPEC_BARRIER.
3947 (rs6000_init_builtins): Likewise.
3948 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3950 (speculation_barrier): New define_insn.
3951 * doc/extend.texi: Document __builtin_speculation_barrier.
3953 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3956 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3957 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3958 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3960 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3961 integral modes instead of "ss" and "sd".
3962 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3963 vectors with 32-bit and 64-bit elements.
3964 (vecdupssescalarmodesuffix): New mode attribute.
3965 (vec_dup<mode>): Use it.
3967 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3970 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3971 frame if argument is passed on stack.
3973 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3976 * ree.c (combine_reaching_defs): Optimize also
3977 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3978 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3980 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3983 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3985 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3988 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3989 after they are computed.
3991 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3993 PR tree-optimization/83695
3994 * gimple-loop-linterchange.cc
3995 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3996 reset cached scev information after interchange.
3997 (pass_linterchange::execute): Remove call to scev_reset_htab.
3999 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4001 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4002 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4003 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4004 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4005 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4006 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4007 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4008 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4009 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4010 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4011 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4012 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4013 (V_lane_reg): Likewise.
4014 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4016 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4017 (vfmal_lane_low<mode>_intrinsic,
4018 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4019 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4020 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4021 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4022 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4023 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4025 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4027 * config/arm/arm-cpus.in (fp16fml): New feature.
4028 (ALL_SIMD): Add fp16fml.
4029 (armv8.2-a): Add fp16fml as an option.
4030 (armv8.3-a): Likewise.
4031 (armv8.4-a): Add fp16fml as part of fp16.
4032 * config/arm/arm.h (TARGET_FP16FML): Define.
4033 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4035 * config/arm/arm-modes.def (V2HF): Define.
4036 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4037 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4038 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4039 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4040 vfmsl_low, vfmsl_high): New set of builtins.
4041 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4042 (vfml_op): New code attribute.
4043 (VFMLHALVES): New int iterator.
4044 (VFML, VFMLSEL): New mode attributes.
4045 (V_reg): Define mapping for V2HF.
4046 (V_hi, V_lo): New mode attributes.
4047 (VF_constraint): Likewise.
4048 (vfml_half, vfml_half_selector): New int attributes.
4049 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4051 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4052 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4054 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4055 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4056 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4057 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4059 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4060 Document new effective target and option set.
4062 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4064 * config/arm/arm-cpus.in (armv8_4): New feature.
4065 (ARMv8_4a): New fgroup.
4066 (armv8.4-a): New arch.
4067 * config/arm/arm-tables.opt: Regenerate.
4068 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4069 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4070 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4071 Add matching rules for -march=armv8.4-a and extensions.
4072 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4074 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4077 * config/rx/rx.md (BW): New mode attribute.
4078 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4080 2018-01-11 Richard Biener <rguenther@suse.de>
4082 PR tree-optimization/83435
4083 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4084 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4085 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4087 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4088 Alan Hayward <alan.hayward@arm.com>
4089 David Sherwood <david.sherwood@arm.com>
4091 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4093 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4094 (aarch64_print_address_internal): Use it to check for a zero offset.
4096 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4097 Alan Hayward <alan.hayward@arm.com>
4098 David Sherwood <david.sherwood@arm.com>
4100 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4101 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4102 Return a poly_int64 rather than a HOST_WIDE_INT.
4103 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4104 rather than a HOST_WIDE_INT.
4105 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4106 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4107 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4108 final_offset from HOST_WIDE_INT to poly_int64.
4109 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4110 to_constant when getting the number of units in an Advanced SIMD
4112 (aarch64_builtin_vectorized_function): Check for a constant number
4114 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4116 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4117 attribute instead of GET_MODE_NUNITS.
4118 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4119 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4120 GET_MODE_SIZE for fixed-size registers.
4121 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4122 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4123 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4124 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4125 (aarch64_print_operand, aarch64_print_address_internal)
4126 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4127 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4128 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4129 Handle polynomial GET_MODE_SIZE.
4130 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4131 wider than SImode without modification.
4132 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4133 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4134 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4135 passing and returning SVE modes.
4136 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4137 rather than GEN_INT.
4138 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4139 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4140 (aarch64_allocate_and_probe_stack_space): Likewise.
4141 (aarch64_layout_frame): Cope with polynomial offsets.
4142 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4143 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4145 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4146 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4147 poly_int64 rather than a HOST_WIDE_INT.
4148 (aarch64_get_separate_components, aarch64_process_components)
4149 (aarch64_expand_prologue, aarch64_expand_epilogue)
4150 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4151 (aarch64_anchor_offset): New function, split out from...
4152 (aarch64_legitimize_address): ...here.
4153 (aarch64_builtin_vectorization_cost): Handle polynomial
4154 TYPE_VECTOR_SUBPARTS.
4155 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4157 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4158 number of elements from the PARALLEL rather than the mode.
4159 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4160 rather than GET_MODE_BITSIZE.
4161 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4162 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4163 (aarch64_expand_vec_perm_const_1): Handle polynomial
4164 d->perm.length () and d->perm elements.
4165 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4166 Apply to_constant to d->perm elements.
4167 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4168 polynomial CONST_VECTOR_NUNITS.
4169 (aarch64_move_pointer): Take amount as a poly_int64 rather
4171 (aarch64_progress_pointer): Avoid temporary variable.
4172 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4173 the mode attribute instead of GET_MODE.
4175 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4176 Alan Hayward <alan.hayward@arm.com>
4177 David Sherwood <david.sherwood@arm.com>
4179 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4180 x exists before using it.
4181 (aarch64_add_constant_internal): Rename to...
4182 (aarch64_add_offset_1): ...this. Replace regnum with separate
4183 src and dest rtxes. Handle the case in which they're different,
4184 including when the offset is zero. Replace scratchreg with an rtx.
4185 Use 2 additions if there is no spare register into which we can
4186 move a 16-bit constant.
4187 (aarch64_add_constant): Delete.
4188 (aarch64_add_offset): Replace reg with separate src and dest
4189 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4190 Use aarch64_add_offset_1.
4191 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4192 an rtx rather than an int. Take the delta as a poly_int64
4193 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4194 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4195 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4196 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4197 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4199 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4200 aarch64_add_constant.
4202 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4204 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4205 Use scalar_float_mode.
4207 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4209 * config/aarch64/aarch64-simd.md
4210 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4211 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4212 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4213 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4214 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4215 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4216 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4217 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4218 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4219 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4221 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4224 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4225 targ_options->x_arm_arch_string is non NULL.
4227 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4229 * config/aarch64/aarch64.h
4230 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4232 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4235 * expmed.c (emit_store_flag_force): Swap if const op0
4236 and change VOIDmode to mode of op0.
4238 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4240 PR rtl-optimization/83761
4241 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4242 than bytes to mode_for_size.
4244 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4247 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4248 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4251 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4254 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4255 when in layout mode.
4256 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4257 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4260 2018-01-10 Michael Collison <michael.collison@arm.com>
4262 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4263 * config/aarch64/aarch64-option-extension.def: Add
4264 AARCH64_OPT_EXTENSION of 'fp16fml'.
4265 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4266 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4267 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4268 * config/aarch64/constraints.md (Ui7): New constraint.
4269 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4270 (VFMLA_SEL_W): Ditto.
4273 (VFMLA16_LOW): New int iterator.
4274 (VFMLA16_HIGH): Ditto.
4275 (UNSPEC_FMLAL): New unspec.
4276 (UNSPEC_FMLSL): Ditto.
4277 (UNSPEC_FMLAL2): Ditto.
4278 (UNSPEC_FMLSL2): Ditto.
4279 (f16mac): New code attribute.
4280 * config/aarch64/aarch64-simd-builtins.def
4281 (aarch64_fmlal_lowv2sf): Ditto.
4282 (aarch64_fmlsl_lowv2sf): Ditto.
4283 (aarch64_fmlalq_lowv4sf): Ditto.
4284 (aarch64_fmlslq_lowv4sf): Ditto.
4285 (aarch64_fmlal_highv2sf): Ditto.
4286 (aarch64_fmlsl_highv2sf): Ditto.
4287 (aarch64_fmlalq_highv4sf): Ditto.
4288 (aarch64_fmlslq_highv4sf): Ditto.
4289 (aarch64_fmlal_lane_lowv2sf): Ditto.
4290 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4291 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4292 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4293 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4294 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4295 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4296 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4297 (aarch64_fmlal_lane_highv2sf): Ditto.
4298 (aarch64_fmlsl_lane_highv2sf): Ditto.
4299 (aarch64_fmlal_laneq_highv2sf): Ditto.
4300 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4301 (aarch64_fmlalq_lane_highv4sf): Ditto.
4302 (aarch64_fmlsl_lane_highv4sf): Ditto.
4303 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4304 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4305 * config/aarch64/aarch64-simd.md:
4306 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4307 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4308 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4309 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4310 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4311 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4312 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4313 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4314 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4315 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4316 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4317 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4318 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4319 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4320 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4321 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4322 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4323 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4324 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4325 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4326 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4327 (vfmlsl_low_u32): Ditto.
4328 (vfmlalq_low_u32): Ditto.
4329 (vfmlslq_low_u32): Ditto.
4330 (vfmlal_high_u32): Ditto.
4331 (vfmlsl_high_u32): Ditto.
4332 (vfmlalq_high_u32): Ditto.
4333 (vfmlslq_high_u32): Ditto.
4334 (vfmlal_lane_low_u32): Ditto.
4335 (vfmlsl_lane_low_u32): Ditto.
4336 (vfmlal_laneq_low_u32): Ditto.
4337 (vfmlsl_laneq_low_u32): Ditto.
4338 (vfmlalq_lane_low_u32): Ditto.
4339 (vfmlslq_lane_low_u32): Ditto.
4340 (vfmlalq_laneq_low_u32): Ditto.
4341 (vfmlslq_laneq_low_u32): Ditto.
4342 (vfmlal_lane_high_u32): Ditto.
4343 (vfmlsl_lane_high_u32): Ditto.
4344 (vfmlal_laneq_high_u32): Ditto.
4345 (vfmlsl_laneq_high_u32): Ditto.
4346 (vfmlalq_lane_high_u32): Ditto.
4347 (vfmlslq_lane_high_u32): Ditto.
4348 (vfmlalq_laneq_high_u32): Ditto.
4349 (vfmlslq_laneq_high_u32): Ditto.
4350 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4351 (AARCH64_FL_FOR_ARCH8_4): New.
4352 (AARCH64_ISA_F16FML): New ISA flag.
4353 (TARGET_F16FML): New feature flag for fp16fml.
4354 (doc/invoke.texi): Document new fp16fml option.
4356 2018-01-10 Michael Collison <michael.collison@arm.com>
4358 * config/aarch64/aarch64-builtins.c:
4359 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4360 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4361 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4362 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4363 (AARCH64_ISA_SHA3): New ISA flag.
4364 (TARGET_SHA3): New feature flag for sha3.
4365 * config/aarch64/iterators.md (sha512_op): New int attribute.
4366 (CRYPTO_SHA512): New int iterator.
4367 (UNSPEC_SHA512H): New unspec.
4368 (UNSPEC_SHA512H2): Ditto.
4369 (UNSPEC_SHA512SU0): Ditto.
4370 (UNSPEC_SHA512SU1): Ditto.
4371 * config/aarch64/aarch64-simd-builtins.def
4372 (aarch64_crypto_sha512hqv2di): New builtin.
4373 (aarch64_crypto_sha512h2qv2di): Ditto.
4374 (aarch64_crypto_sha512su0qv2di): Ditto.
4375 (aarch64_crypto_sha512su1qv2di): Ditto.
4376 (aarch64_eor3qv8hi): Ditto.
4377 (aarch64_rax1qv2di): Ditto.
4378 (aarch64_xarqv2di): Ditto.
4379 (aarch64_bcaxqv8hi): Ditto.
4380 * config/aarch64/aarch64-simd.md:
4381 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4382 (aarch64_crypto_sha512su0qv2di): Ditto.
4383 (aarch64_crypto_sha512su1qv2di): Ditto.
4384 (aarch64_eor3qv8hi): Ditto.
4385 (aarch64_rax1qv2di): Ditto.
4386 (aarch64_xarqv2di): Ditto.
4387 (aarch64_bcaxqv8hi): Ditto.
4388 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4389 (vsha512h2q_u64): Ditto.
4390 (vsha512su0q_u64): Ditto.
4391 (vsha512su1q_u64): Ditto.
4392 (veor3q_u16): Ditto.
4393 (vrax1q_u64): Ditto.
4395 (vbcaxq_u16): Ditto.
4396 * config/arm/types.md (crypto_sha512): New type attribute.
4397 (crypto_sha3): Ditto.
4398 (doc/invoke.texi): Document new sha3 option.
4400 2018-01-10 Michael Collison <michael.collison@arm.com>
4402 * config/aarch64/aarch64-builtins.c:
4403 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4404 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4405 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4406 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4407 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4408 (AARCH64_ISA_SM4): New ISA flag.
4409 (TARGET_SM4): New feature flag for sm4.
4410 * config/aarch64/aarch64-simd-builtins.def
4411 (aarch64_sm3ss1qv4si): Ditto.
4412 (aarch64_sm3tt1aq4si): Ditto.
4413 (aarch64_sm3tt1bq4si): Ditto.
4414 (aarch64_sm3tt2aq4si): Ditto.
4415 (aarch64_sm3tt2bq4si): Ditto.
4416 (aarch64_sm3partw1qv4si): Ditto.
4417 (aarch64_sm3partw2qv4si): Ditto.
4418 (aarch64_sm4eqv4si): Ditto.
4419 (aarch64_sm4ekeyqv4si): Ditto.
4420 * config/aarch64/aarch64-simd.md:
4421 (aarch64_sm3ss1qv4si): Ditto.
4422 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4423 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4424 (aarch64_sm4eqv4si): Ditto.
4425 (aarch64_sm4ekeyqv4si): Ditto.
4426 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4427 (sm3part_op): Ditto.
4428 (CRYPTO_SM3TT): Ditto.
4429 (CRYPTO_SM3PART): Ditto.
4430 (UNSPEC_SM3SS1): New unspec.
4431 (UNSPEC_SM3TT1A): Ditto.
4432 (UNSPEC_SM3TT1B): Ditto.
4433 (UNSPEC_SM3TT2A): Ditto.
4434 (UNSPEC_SM3TT2B): Ditto.
4435 (UNSPEC_SM3PARTW1): Ditto.
4436 (UNSPEC_SM3PARTW2): Ditto.
4437 (UNSPEC_SM4E): Ditto.
4438 (UNSPEC_SM4EKEY): Ditto.
4439 * config/aarch64/constraints.md (Ui2): New constraint.
4440 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4441 * config/arm/types.md (crypto_sm3): New type attribute.
4442 (crypto_sm4): Ditto.
4443 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4444 (vsm3tt1aq_u32): Ditto.
4445 (vsm3tt1bq_u32): Ditto.
4446 (vsm3tt2aq_u32): Ditto.
4447 (vsm3tt2bq_u32): Ditto.
4448 (vsm3partw1q_u32): Ditto.
4449 (vsm3partw2q_u32): Ditto.
4450 (vsm4eq_u32): Ditto.
4451 (vsm4ekeyq_u32): Ditto.
4452 (doc/invoke.texi): Document new sm4 option.
4454 2018-01-10 Michael Collison <michael.collison@arm.com>
4456 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4457 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4458 (AARCH64_FL_FOR_ARCH8_4): New.
4459 (AARCH64_FL_V8_4): New flag.
4460 (doc/invoke.texi): Document new armv8.4-a option.
4462 2018-01-10 Michael Collison <michael.collison@arm.com>
4464 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4465 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4466 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4467 * config/aarch64/aarch64-option-extension.def: Add
4468 AARCH64_OPT_EXTENSION of 'sha2'.
4469 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4470 (crypto): Disable sha2 and aes if crypto disabled.
4471 (crypto): Enable aes and sha2 if enabled.
4472 (simd): Disable sha2 and aes if simd disabled.
4473 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4475 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4476 (TARGET_SHA2): New feature flag for sha2.
4477 (TARGET_AES): New feature flag for aes.
4478 * config/aarch64/aarch64-simd.md:
4479 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4480 conditional on TARGET_AES.
4481 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4482 (aarch64_crypto_sha1hsi): Make pattern conditional
4484 (aarch64_crypto_sha1hv4si): Ditto.
4485 (aarch64_be_crypto_sha1hv4si): Ditto.
4486 (aarch64_crypto_sha1su1v4si): Ditto.
4487 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4488 (aarch64_crypto_sha1su0v4si): Ditto.
4489 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4490 (aarch64_crypto_sha256su0v4si): Ditto.
4491 (aarch64_crypto_sha256su1v4si): Ditto.
4492 (doc/invoke.texi): Document new aes and sha2 options.
4494 2018-01-10 Martin Sebor <msebor@redhat.com>
4496 PR tree-optimization/83781
4497 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4500 2018-01-11 Martin Sebor <msebor@gmail.com>
4501 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4503 PR tree-optimization/83501
4504 PR tree-optimization/81703
4506 * tree-ssa-strlen.c (get_string_cst): Rename...
4507 (get_string_len): ...to this. Handle global constants.
4508 (handle_char_store): Adjust.
4510 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4511 Jim Wilson <jimw@sifive.com>
4513 * config/riscv/riscv-protos.h (riscv_output_return): New.
4514 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4515 (riscv_attribute_table, riscv_output_return),
4516 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4517 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4518 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4520 (riscv_expand_prologue): Add early return for naked function.
4521 (riscv_expand_epilogue): Likewise.
4522 (riscv_function_ok_for_sibcall): Return false for naked function.
4523 (riscv_set_current_function): New.
4524 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4525 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4526 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4527 * doc/extend.texi (RISC-V Function Attributes): New.
4529 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4531 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4532 check for 128-bit long double before checking TCmode.
4533 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4534 128-bit long doubles before checking TFmode or TCmode.
4535 (FLOAT128_IBM_P): Likewise.
4537 2018-01-10 Martin Sebor <msebor@redhat.com>
4539 PR tree-optimization/83671
4540 * builtins.c (c_strlen): Unconditionally return zero for the empty
4542 Use -Warray-bounds for warnings.
4543 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4544 for non-constant array indices with COMPONENT_REF, arrays of
4545 arrays, and pointers to arrays.
4546 (gimple_fold_builtin_strlen): Determine and set length range for
4547 non-constant character arrays.
4549 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4552 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4555 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4557 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4559 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4562 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4563 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4564 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4565 indexed_or_indirect_operand predicate.
4566 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4567 (*vsx_le_perm_load_v8hi): Likewise.
4568 (*vsx_le_perm_load_v16qi): Likewise.
4569 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4570 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4571 (*vsx_le_perm_store_v8hi): Likewise.
4572 (*vsx_le_perm_store_v16qi): Likewise.
4573 (eight unnamed splitters): Likewise.
4575 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4577 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4578 * config/rs6000/emmintrin.h: Likewise.
4579 * config/rs6000/mmintrin.h: Likewise.
4580 * config/rs6000/xmmintrin.h: Likewise.
4582 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4585 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4587 * tree.c (tree_nop_conversion): Return true for location wrapper
4589 (maybe_wrap_with_location): New function.
4590 (selftest::check_strip_nops): New function.
4591 (selftest::test_location_wrappers): New function.
4592 (selftest::tree_c_tests): Call it.
4593 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4594 (maybe_wrap_with_location): New decl.
4595 (EXPR_LOCATION_WRAPPER_P): New macro.
4596 (location_wrapper_p): New inline function.
4597 (tree_strip_any_location_wrapper): New inline function.
4599 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4602 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4603 stack_realign_offset for the largest alignment of stack slot
4605 (ix86_find_max_used_stack_alignment): New function.
4606 (ix86_finalize_stack_frame_flags): Use it. Set
4607 max_used_stack_alignment if we don't realign stack.
4608 * config/i386/i386.h (machine_function): Add
4609 max_used_stack_alignment.
4611 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4613 * config/arm/arm.opt (-mbranch-cost): New option.
4614 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4617 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4620 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4621 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4623 2018-01-10 Richard Biener <rguenther@suse.de>
4626 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4627 early out so it also covers the case where we have a non-NULL
4630 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4632 PR tree-optimization/83753
4633 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4634 for non-strided grouped accesses if the number of elements is 1.
4636 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4639 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4640 * i386.h (TARGET_USE_GATHER): Define.
4641 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4643 2018-01-10 Martin Liska <mliska@suse.cz>
4646 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4647 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4649 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4650 CLEANUP_NO_PARTITIONING is not set.
4652 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4654 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4655 for vectors, as a partial revert of r254296.
4656 * rtl.h (const_vec_p): Delete.
4657 (const_vec_duplicate_p): Don't test for vector CONSTs.
4658 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4659 * expmed.c (make_tree): Likewise.
4662 * common.md (E, F): Use CONSTANT_P instead of checking for
4664 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4665 checking for CONST_VECTOR.
4667 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4670 * predict.c (force_edge_cold): Handle in more sane way edges
4673 2018-01-09 Carl Love <cel@us.ibm.com>
4675 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4677 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4678 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4679 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4680 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4681 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4682 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4683 * config/rs6000/rs6000-protos.h: Add extern defition for
4684 rs6000_generate_float2_double_code.
4685 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4687 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4688 (float2_v2df): Add define_expand.
4690 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4693 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4694 op_mode in the force_to_mode call.
4696 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4698 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4699 instead of checking each element individually.
4700 (aarch64_evpc_uzp): Likewise.
4701 (aarch64_evpc_zip): Likewise.
4702 (aarch64_evpc_ext): Likewise.
4703 (aarch64_evpc_rev): Likewise.
4704 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4705 instead of checking each element individually. Return true without
4707 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4708 whether all selected elements come from the same input, instead of
4709 checking each element individually. Remove calls to gen_rtx_REG,
4710 start_sequence and end_sequence and instead assert that no rtl is
4713 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4715 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4716 order of HIGH and CONST checks.
4718 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4720 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4721 if the destination isn't an SSA_NAME.
4723 2018-01-09 Richard Biener <rguenther@suse.de>
4725 PR tree-optimization/83668
4726 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4728 (canonicalize_loop_form): ... here, renamed from ...
4729 (canonicalize_loop_closed_ssa_form): ... this and amended to
4730 swap successor edges for loop exit blocks to make us use
4731 the RPO order we need for initial schedule generation.
4733 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4735 PR tree-optimization/64811
4736 * match.pd: When optimizing comparisons with Inf, avoid
4737 introducing or losing exceptions from comparisons with NaN.
4739 2018-01-09 Martin Liska <mliska@suse.cz>
4742 * asan.c (shadow_mem_size): Add gcc_assert.
4744 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4746 Don't save registers in main().
4749 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4750 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4751 * config/avr/avr.c (avr_set_current_function): Don't error if
4752 naked, OS_task or OS_main are specified at the same time.
4753 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4755 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4757 * common/config/avr/avr-common.c (avr_option_optimization_table):
4758 Switch on -mmain-is-OS_task for optimizing compilations.
4760 2018-01-09 Richard Biener <rguenther@suse.de>
4762 PR tree-optimization/83572
4763 * graphite.c: Include cfganal.h.
4764 (graphite_transform_loops): Connect infinite loops to exit
4765 and remove fake edges at the end.
4767 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4769 * ipa-inline.c (edge_badness): Revert accidental checkin.
4771 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4774 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4775 symbols; not inline clones.
4777 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4780 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4781 hard registers. Formatting fixes.
4783 PR preprocessor/83722
4784 * gcc.c (try_generate_repro): Pass
4785 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4786 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4789 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4790 Kito Cheng <kito.cheng@gmail.com>
4792 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4793 (riscv_leaf_function_p): Delete.
4794 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4796 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4798 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4800 (do_ifelse): New function.
4801 (do_isel): New function.
4802 (do_sub3): New function.
4803 (do_add3): New function.
4804 (do_load_mask_compare): New function.
4805 (do_overlap_load_compare): New function.
4806 (expand_compare_loop): New function.
4807 (expand_block_compare): Call expand_compare_loop() when appropriate.
4808 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4810 (-mblock-compare-inline-loop-limit): New option.
4812 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4815 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4816 Reverse order of second and third operands in first alternative.
4817 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4818 of first and second elements in UNSPEC_VPERMR vector.
4819 (altivec_expand_vec_perm_le): Likewise.
4821 2018-01-08 Jeff Law <law@redhat.com>
4823 PR rtl-optimizatin/81308
4824 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4825 (process_switch): If group_case_labels makes a change, then set
4827 (pass_convert_switch::execute): If a switch is converted, then
4828 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4830 PR rtl-optimization/81308
4831 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4834 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4836 PR target/83663 - Revert r255946
4837 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4838 generation for cases where splatting a value is not useful.
4839 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4840 across a vec_duplicate and a paradoxical subreg forming a vector
4841 mode to a vec_concat.
4843 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4845 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4846 -march=armv8.3-a variants.
4847 * config/arm/t-multilib: Likewise.
4848 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4850 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4852 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4854 (cceq_ior_compare_complement): Give it a name so I can use it, and
4855 change boolean_or_operator predicate to boolean_operator so it can
4856 be used to generate a crand.
4857 (eqne): New code iterator.
4858 (bd/bd_neg): New code_attrs.
4859 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4860 a single define_insn.
4861 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4862 decrement (bdnzt/bdnzf/bdzt/bdzf).
4863 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4864 with the new names of the branch decrement patterns, and added the
4865 names of the branch decrement conditional patterns.
4867 2018-01-08 Richard Biener <rguenther@suse.de>
4869 PR tree-optimization/83563
4870 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4873 2018-01-08 Richard Biener <rguenther@suse.de>
4876 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4878 2018-01-08 Richard Biener <rguenther@suse.de>
4880 PR tree-optimization/83685
4881 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4882 references to abnormals.
4884 2018-01-08 Richard Biener <rguenther@suse.de>
4887 * dwarf2out.c (output_indirect_strings): Handle empty
4888 skeleton_debug_str_hash.
4889 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4891 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4893 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4894 (emit_store_direct): Likewise.
4895 (arc_trampoline_adjust_address): Likewise.
4896 (arc_asm_trampoline_template): New function.
4897 (arc_initialize_trampoline): Use asm_trampoline_template.
4898 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4899 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4900 * config/arc/arc.md (flush_icache): Delete pattern.
4902 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4904 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4905 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4908 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4911 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4912 by not USED_FOR_TARGET.
4913 (make_pass_resolve_sw_modes): Likewise.
4915 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4917 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4920 2018-01-08 Richard Biener <rguenther@suse.de>
4923 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4925 2018-01-08 Richard Biener <rguenther@suse.de>
4928 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4930 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4933 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4934 basic blocks with a small number of successors.
4935 (convert_control_dep_chain_into_preds): Improve handling of
4937 (dump_predicates): Split apart into...
4938 (dump_pred_chain): ...here...
4939 (dump_pred_info): ...and here.
4940 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4941 (can_chain_union_be_invalidated_p): Improve check for invalidation
4943 (uninit_uses_cannot_happen): Avoid unnecessary if
4944 convert_control_dep_chain_into_preds yielded nothing.
4946 2018-01-06 Martin Sebor <msebor@redhat.com>
4948 PR tree-optimization/83640
4949 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4950 subtracting negative offset from size.
4951 (builtin_access::overlap): Adjust offset bounds of the access to fall
4952 within the size of the object if possible.
4954 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4956 PR rtl-optimization/83699
4957 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4958 extract_bit_field_as_subreg to cases in which the extracted
4959 value is also a vector.
4961 * lra-constraints.c (process_alt_operands): Test for the equivalence
4962 substitutions when detecting a possible reload cycle.
4964 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4967 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4968 by default if flag_selective_schedling{,2}. Formatting fixes.
4970 PR rtl-optimization/83682
4971 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4972 if it has non-VECTOR_MODE element mode.
4973 (vec_duplicate_p): Likewise.
4976 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4977 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4979 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4982 * config/i386/i386-builtin.def
4983 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4984 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4985 Require also OPTION_MASK_ISA_AVX512F in addition to
4986 OPTION_MASK_ISA_GFNI.
4987 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4988 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4989 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4990 to OPTION_MASK_ISA_GFNI.
4991 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4992 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4993 OPTION_MASK_ISA_AVX512BW.
4994 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4995 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4996 addition to OPTION_MASK_ISA_GFNI.
4997 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4998 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4999 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5000 to OPTION_MASK_ISA_GFNI.
5001 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5002 a requirement for all ISAs rather than any of them with a few
5004 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5006 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5007 bitmasks to be enabled with 3 exceptions, instead of requiring any
5008 enabled ISA with lots of exceptions.
5009 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5010 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5011 Change avx512bw in isa attribute to avx512f.
5012 * config/i386/sgxintrin.h: Add license boilerplate.
5013 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5014 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5015 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5016 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5018 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5019 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5020 temporarily sse2 rather than sse if not enabled already.
5023 * config/i386/sse.md (VI248_VLBW): Rename to ...
5024 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5025 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5026 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5027 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5028 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5029 mode iterator instead of VI248_VLBW.
5031 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5033 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5034 (record_modified): Skip clobbers; add debug output.
5035 (param_change_prob): Use sreal frequencies.
5037 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5039 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5040 punt for user-aligned variables.
5042 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5044 * tree-chrec.c (chrec_contains_symbols): Return true for
5047 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5050 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5051 of (x|y) == x for BICS pattern.
5053 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5055 PR tree-optimization/83605
5056 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5057 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5060 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5062 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5063 * config/epiphany/rtems.h: New file.
5065 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5066 Uros Bizjak <ubizjak@gmail.com>
5069 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5070 QIreg_operand instead of register_operand predicate.
5071 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5072 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5073 comments instead of -fmitigate[-_]rop.
5075 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5078 * cgraphunit.c (symbol_table::compile): Switch to text_section
5079 before calling assembly_start debug hook.
5080 * run-rtl-passes.c (run_rtl_passes): Likewise.
5083 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5085 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5086 range_int_cst_p rather than !symbolic_range_p before calling
5087 extract_range_from_multiplicative_op_1.
5089 2018-01-04 Jeff Law <law@redhat.com>
5091 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5092 redundant test in assertion.
5094 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5096 * doc/rtl.texi: Document machine_mode wrapper classes.
5098 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5100 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5103 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5105 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5106 the VEC_PERM_EXPR fold to fail.
5108 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5111 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5112 to switched_sections.
5114 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5117 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5120 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5123 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5124 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5126 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5129 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5130 is BLKmode and bitpos not zero or mode change is needed.
5132 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5135 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5138 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5141 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5142 instead of MULT rtx. Update all corresponding splitters.
5144 (*ssub<modesuffix>): Ditto.
5146 (*cmp_sadd_di): Update split patterns.
5147 (*cmp_sadd_si): Ditto.
5148 (*cmp_sadd_sidi): Ditto.
5149 (*cmp_ssub_di): Ditto.
5150 (*cmp_ssub_si): Ditto.
5151 (*cmp_ssub_sidi): Ditto.
5152 * config/alpha/predicates.md (const23_operand): New predicate.
5153 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5154 Look for ASHIFT, not MULT inner operand.
5155 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5157 2018-01-04 Martin Liska <mliska@suse.cz>
5159 PR gcov-profile/83669
5160 * gcov.c (output_intermediate_file): Add version to intermediate
5162 * doc/gcov.texi: Document new field 'version' in intermediate
5163 file format. Fix location of '-k' option of gcov command.
5165 2018-01-04 Martin Liska <mliska@suse.cz>
5168 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5170 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5172 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5174 2018-01-03 Martin Sebor <msebor@redhat.com>
5176 PR tree-optimization/83655
5177 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5178 checking calls with invalid arguments.
5180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5182 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5183 (vectorizable_mask_load_store): Delete.
5184 (vectorizable_call): Return false for masked loads and stores.
5185 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5186 instead of gimple_assign_rhs1.
5187 (vectorizable_load): Handle IFN_MASK_LOAD.
5188 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5190 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5192 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5194 (vectorizable_mask_load_store): ...here.
5195 (vectorizable_load): ...and here.
5197 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5199 * tree-vect-stmts.c (vect_build_all_ones_mask)
5200 (vect_build_zero_merge_argument): New functions, split out from...
5201 (vectorizable_load): ...here.
5203 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5205 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5207 (vectorizable_mask_load_store): ...here.
5208 (vectorizable_store): ...and here.
5210 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5212 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5214 (vectorizable_mask_load_store): ...here.
5216 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5218 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5219 (vect_model_store_cost): Take a vec_load_store_type instead of a
5221 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5222 (vect_model_store_cost): Take a vec_load_store_type instead of a
5224 (vectorizable_mask_load_store): Update accordingly.
5225 (vectorizable_store): Likewise.
5226 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5228 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5230 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5231 IFN_MASK_LOAD calls here rather than...
5232 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5235 Alan Hayward <alan.hayward@arm.com>
5236 David Sherwood <david.sherwood@arm.com>
5238 * expmed.c (extract_bit_field_1): For vector extracts,
5239 fall back to extract_bit_field_as_subreg if vec_extract
5242 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5243 Alan Hayward <alan.hayward@arm.com>
5244 David Sherwood <david.sherwood@arm.com>
5246 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5247 they are variable or constant sized.
5248 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5249 slots for constant-sized data.
5251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5252 Alan Hayward <alan.hayward@arm.com>
5253 David Sherwood <david.sherwood@arm.com>
5255 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5256 handling COND_EXPRs with boolean comparisons, try to find a better
5257 basis for the mask type than the boolean itself.
5259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5261 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5262 is calculated and how it can be overridden.
5263 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5264 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5266 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5269 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5270 Alan Hayward <alan.hayward@arm.com>
5271 David Sherwood <david.sherwood@arm.com>
5273 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5274 Remove the mode argument.
5275 (aarch64_simd_valid_immediate): Remove the mode and inverse
5277 * config/aarch64/iterators.md (bitsize): New iterator.
5278 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5279 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5280 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5281 aarch64_simd_valid_immediate.
5282 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5283 (aarch64_reg_or_bic_imm): Likewise.
5284 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5285 with an insn_type enum and msl with a modifier_type enum.
5286 Replace element_width with a scalar_mode. Change the shift
5287 to unsigned int. Add constructors for scalar_float_mode and
5288 scalar_int_mode elements.
5289 (aarch64_vect_float_const_representable_p): Delete.
5290 (aarch64_can_const_movi_rtx_p)
5291 (aarch64_simd_scalar_immediate_valid_for_move)
5292 (aarch64_simd_make_constant): Update call to
5293 aarch64_simd_valid_immediate.
5294 (aarch64_advsimd_valid_immediate_hs): New function.
5295 (aarch64_advsimd_valid_immediate): Likewise.
5296 (aarch64_simd_valid_immediate): Remove mode and inverse
5297 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5298 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5299 and aarch64_float_const_representable_p on the result.
5300 (aarch64_output_simd_mov_immediate): Remove mode argument.
5301 Update call to aarch64_simd_valid_immediate and use of
5302 simd_immediate_info.
5303 (aarch64_output_scalar_simd_mov_immediate): Update call
5306 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5307 Alan Hayward <alan.hayward@arm.com>
5308 David Sherwood <david.sherwood@arm.com>
5310 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5311 (mode_nunits): Likewise CONST_MODE_NUNITS.
5312 * machmode.def (ADJUST_NUNITS): Document.
5313 * genmodes.c (mode_data::need_nunits_adj): New field.
5314 (blank_mode): Update accordingly.
5315 (adj_nunits): New variable.
5316 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5318 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5319 listed in adj_nunits.
5320 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5321 listed in adj_nunits. Don't emit case statements for such modes.
5322 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5323 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5324 nothing if adj_nunits is nonnull.
5325 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5326 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5327 (emit_mode_fbit): Update use of print_maybe_const_decl.
5328 (emit_move_size): Likewise. Treat the array as non-const
5330 (emit_mode_adjustments): Handle adj_nunits.
5332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5334 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5335 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5336 (VECTOR_MODES): Use it.
5337 (make_vector_modes): Take the prefix as an argument.
5339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5340 Alan Hayward <alan.hayward@arm.com>
5341 David Sherwood <david.sherwood@arm.com>
5343 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5344 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5345 for MODE_VECTOR_BOOL.
5346 * machmode.def (VECTOR_BOOL_MODE): Document.
5347 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5348 (make_vector_bool_mode): New function.
5349 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5351 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5352 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5354 * stor-layout.c (int_mode_for_mode): Likewise.
5355 * tree.c (build_vector_type_for_mode): Likewise.
5356 * varasm.c (output_constant_pool_2): Likewise.
5357 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5358 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5359 for MODE_VECTOR_BOOL.
5360 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5361 of mode class checks.
5362 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5363 instead of a list of mode class checks.
5364 (expand_vector_scalar_condition): Likewise.
5365 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5367 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5368 Alan Hayward <alan.hayward@arm.com>
5369 David Sherwood <david.sherwood@arm.com>
5371 * machmode.h (mode_size): Change from unsigned short to
5373 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5374 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5375 or if measurement_type is not polynomial.
5376 (fixed_size_mode::includes_p): Check for constant-sized modes.
5377 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5378 return a poly_uint16 rather than an unsigned short.
5379 (emit_mode_size): Change the type of mode_size from unsigned short
5380 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5381 (emit_mode_adjustments): Cope with polynomial vector sizes.
5382 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5384 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5386 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5387 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5388 * caller-save.c (setup_save_areas): Likewise.
5389 (replace_reg_with_saved_mem): Likewise.
5390 * calls.c (emit_library_call_value_1): Likewise.
5391 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5392 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5393 (gen_lowpart_for_combine): Likewise.
5394 * convert.c (convert_to_integer_1): Likewise.
5395 * cse.c (equiv_constant, cse_insn): Likewise.
5396 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5397 (cselib_subst_to_values): Likewise.
5398 * dce.c (word_dce_process_block): Likewise.
5399 * df-problems.c (df_word_lr_mark_ref): Likewise.
5400 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5401 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5402 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5403 (rtl_for_decl_location): Likewise.
5404 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5405 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5406 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5407 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5408 (expand_expr_real_1): Likewise.
5409 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5410 (pad_below): Likewise.
5411 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5412 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5413 * ira.c (get_subreg_tracking_sizes): Likewise.
5414 * ira-build.c (ira_create_allocno_objects): Likewise.
5415 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5416 (ira_sort_regnos_for_alter_reg): Likewise.
5417 * ira-costs.c (record_operand_costs): Likewise.
5418 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5419 (resolve_simple_move): Likewise.
5420 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5421 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5422 (lra_constraints): Likewise.
5423 (CONST_POOL_OK_P): Reject variable-sized modes.
5424 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5425 (add_pseudo_to_slot, lra_spill): Likewise.
5426 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5427 * optabs-query.c (get_best_extraction_insn): Likewise.
5428 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5429 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5430 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5431 * recog.c (offsettable_address_addr_space_p): Likewise.
5432 * regcprop.c (maybe_mode_change): Likewise.
5433 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5434 * regrename.c (build_def_use): Likewise.
5435 * regstat.c (dump_reg_info): Likewise.
5436 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5437 (find_reloads, find_reloads_subreg_address): Likewise.
5438 * reload1.c (eliminate_regs_1): Likewise.
5439 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5440 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5441 (simplify_binary_operation_1, simplify_subreg): Likewise.
5442 * targhooks.c (default_function_arg_padding): Likewise.
5443 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5444 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5445 (verify_gimple_assign_ternary): Likewise.
5446 * tree-inline.c (estimate_move_cost): Likewise.
5447 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5448 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5449 (get_address_cost_ainc): Likewise.
5450 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5451 (vect_supportable_dr_alignment): Likewise.
5452 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5453 (vectorizable_reduction): Likewise.
5454 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5455 (vectorizable_operation, vectorizable_load): Likewise.
5456 * tree.c (build_same_sized_truth_vector_type): Likewise.
5457 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5458 * var-tracking.c (emit_note_insn_var_location): Likewise.
5459 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5460 (ADDR_VEC_ALIGN): Likewise.
5462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5463 Alan Hayward <alan.hayward@arm.com>
5464 David Sherwood <david.sherwood@arm.com>
5466 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5468 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5469 or if measurement_type is polynomial.
5470 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5471 * combine.c (make_extraction): Likewise.
5472 * dse.c (find_shift_sequence): Likewise.
5473 * dwarf2out.c (mem_loc_descriptor): Likewise.
5474 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5475 (extract_bit_field, extract_low_bits): Likewise.
5476 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5477 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5478 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5479 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5480 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5481 * reload.c (find_reloads): Likewise.
5482 * reload1.c (alter_reg): Likewise.
5483 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5484 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5485 * tree-if-conv.c (predicate_mem_writes): Likewise.
5486 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5487 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5488 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5489 * valtrack.c (dead_debug_insert_temp): Likewise.
5490 * varasm.c (mergeable_constant_section): Likewise.
5491 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5493 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5494 Alan Hayward <alan.hayward@arm.com>
5495 David Sherwood <david.sherwood@arm.com>
5497 * expr.c (expand_assignment): Cope with polynomial mode sizes
5498 when assigning to a CONCAT.
5500 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5501 Alan Hayward <alan.hayward@arm.com>
5502 David Sherwood <david.sherwood@arm.com>
5504 * machmode.h (mode_precision): Change from unsigned short to
5506 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5508 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5509 or if measurement_type is not polynomial.
5510 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5511 in which the mode is already known to be a scalar_int_mode.
5512 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5513 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5515 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5516 for GET_MODE_PRECISION.
5517 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5518 for GET_MODE_PRECISION.
5519 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5521 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5522 (expand_field_assignment, make_extraction): Likewise.
5523 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5524 (get_last_value): Likewise.
5525 * convert.c (convert_to_integer_1): Likewise.
5526 * cse.c (cse_insn): Likewise.
5527 * expr.c (expand_expr_real_1): Likewise.
5528 * lra-constraints.c (simplify_operand_subreg): Likewise.
5529 * optabs-query.c (can_atomic_load_p): Likewise.
5530 * optabs.c (expand_atomic_load): Likewise.
5531 (expand_atomic_store): Likewise.
5532 * ree.c (combine_reaching_defs): Likewise.
5533 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5534 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5535 * tree.h (type_has_mode_precision_p): Likewise.
5536 * ubsan.c (instrument_si_overflow): Likewise.
5538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5539 Alan Hayward <alan.hayward@arm.com>
5540 David Sherwood <david.sherwood@arm.com>
5542 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5543 polynomial numbers of units.
5544 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5545 (valid_vector_subparts_p): New function.
5546 (build_vector_type): Remove temporary shim and take the number
5547 of units as a poly_uint64 rather than an int.
5548 (build_opaque_vector_type): Take the number of units as a
5549 poly_uint64 rather than an int.
5550 * tree.c (build_vector_from_ctor): Handle polynomial
5551 TYPE_VECTOR_SUBPARTS.
5552 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5553 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5554 (build_vector_from_val): If the number of units is variable,
5555 use build_vec_duplicate_cst for constant operands and
5556 VEC_DUPLICATE_EXPR otherwise.
5557 (make_vector_type): Remove temporary is_constant ().
5558 (build_vector_type, build_opaque_vector_type): Take the number of
5559 units as a poly_uint64 rather than an int.
5560 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5562 * cfgexpand.c (expand_debug_expr): Likewise.
5563 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5564 (store_constructor, expand_expr_real_1): Likewise.
5565 (const_scalar_mask_from_tree): Likewise.
5566 * fold-const-call.c (fold_const_reduction): Likewise.
5567 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5568 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5569 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5570 (fold_relational_const): Likewise.
5571 (native_interpret_vector): Likewise. Change the size from an
5572 int to an unsigned int.
5573 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5574 TYPE_VECTOR_SUBPARTS.
5575 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5576 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5577 duplicating a non-constant operand into a variable-length vector.
5578 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5579 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5580 * ipa-icf.c (sem_variable::equals): Likewise.
5581 * match.pd: Likewise.
5582 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5583 * print-tree.c (print_node): Likewise.
5584 * stor-layout.c (layout_type): Likewise.
5585 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5586 * tree-cfg.c (verify_gimple_comparison): Likewise.
5587 (verify_gimple_assign_binary): Likewise.
5588 (verify_gimple_assign_ternary): Likewise.
5589 (verify_gimple_assign_single): Likewise.
5590 * tree-pretty-print.c (dump_generic_node): Likewise.
5591 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5592 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5593 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5594 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5595 (vect_shift_permute_load_chain): Likewise.
5596 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5597 (expand_vector_condition, optimize_vector_constructor): Likewise.
5598 (lower_vec_perm, get_compute_type): Likewise.
5599 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5600 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5601 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5602 (vect_recog_mask_conversion_pattern): Likewise.
5603 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5604 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5605 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5606 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5607 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5608 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5609 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5610 (supportable_widening_operation): Likewise.
5611 (supportable_narrowing_operation): Likewise.
5612 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5614 * varasm.c (output_constant): Likewise.
5616 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5617 Alan Hayward <alan.hayward@arm.com>
5618 David Sherwood <david.sherwood@arm.com>
5620 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5621 so that both the length == 3 and length != 3 cases set up their
5622 own permute vectors. Add comments explaining why we know the
5623 number of elements is constant.
5624 (vect_permute_load_chain): Likewise.
5626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5627 Alan Hayward <alan.hayward@arm.com>
5628 David Sherwood <david.sherwood@arm.com>
5630 * machmode.h (mode_nunits): Change from unsigned char to
5632 (ONLY_FIXED_SIZE_MODES): New macro.
5633 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5634 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5635 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5637 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5638 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5639 or if measurement_type is not polynomial.
5640 * genmodes.c (ZERO_COEFFS): New macro.
5641 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5643 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5644 Use ZERO_COEFFS when emitting initializers.
5645 * data-streamer.h (bp_pack_poly_value): New function.
5646 (bp_unpack_poly_value): Likewise.
5647 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5648 for GET_MODE_NUNITS.
5649 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5650 for GET_MODE_NUNITS.
5651 * tree.c (make_vector_type): Remove temporary shim and make
5652 the real function take the number of units as a poly_uint64
5654 (build_vector_type_for_mode): Handle polynomial nunits.
5655 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5656 * emit-rtl.c (const_vec_series_p_1): Likewise.
5657 (gen_rtx_CONST_VECTOR): Likewise.
5658 * fold-const.c (test_vec_duplicate_folding): Likewise.
5659 * genrecog.c (validate_pattern): Likewise.
5660 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5661 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5662 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5663 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5664 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5665 * rtlanal.c (subreg_get_info): Likewise.
5666 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5667 (vect_grouped_load_supported): Likewise.
5668 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5669 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5670 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5671 (simplify_const_unary_operation, simplify_binary_operation_1)
5672 (simplify_const_binary_operation, simplify_ternary_operation)
5673 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5674 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5675 instead of CONST_VECTOR_NUNITS.
5676 * varasm.c (output_constant_pool_2): Likewise.
5677 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5678 explicit-encoded elements in the XVEC for variable-length vectors.
5680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5682 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5685 Alan Hayward <alan.hayward@arm.com>
5686 David Sherwood <david.sherwood@arm.com>
5688 * coretypes.h (fixed_size_mode): Declare.
5689 (fixed_size_mode_pod): New typedef.
5690 * builtins.h (target_builtins::x_apply_args_mode)
5691 (target_builtins::x_apply_result_mode): Change type to
5692 fixed_size_mode_pod.
5693 * builtins.c (apply_args_size, apply_result_size, result_vector)
5694 (expand_builtin_apply_args_1, expand_builtin_apply)
5695 (expand_builtin_return): Update accordingly.
5697 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5699 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5700 * cselib.c (cselib_hash_rtx): Likewise.
5701 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5702 CONST_VECTOR encoding.
5704 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5705 Jeff Law <law@redhat.com>
5708 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5709 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5710 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5711 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5714 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5715 explicitly probe *sp in a noreturn function if there were any callee
5716 register saves or frame pointer is needed.
5718 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5721 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5722 BLKmode for ternary, binary or unary expressions.
5725 * var-tracking.c (delete_vta_debug_insn): New inline function.
5726 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5727 insns from get_insns () to NULL instead of each bb separately.
5728 Use delete_vta_debug_insn. No longer static.
5729 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5730 delete_vta_debug_insns callers.
5731 * rtl.h (delete_vta_debug_insns): Declare.
5732 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5733 instead of variable_tracking_main.
5735 2018-01-03 Martin Sebor <msebor@redhat.com>
5737 PR tree-optimization/83603
5738 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5739 arguments past the endof the argument list in functions declared
5740 without a prototype.
5741 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5742 Avoid checking when arguments are null.
5744 2018-01-03 Martin Sebor <msebor@redhat.com>
5747 * doc/extend.texi (attribute const): Fix a typo.
5748 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5749 issuing -Wsuggest-attribute for void functions.
5751 2018-01-03 Martin Sebor <msebor@redhat.com>
5753 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5754 offset_int::from instead of wide_int::to_shwi.
5755 (maybe_diag_overlap): Remove assertion.
5756 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5757 * gimple-ssa-sprintf.c (format_directive): Same.
5758 (parse_directive): Same.
5759 (sprintf_dom_walker::compute_format_length): Same.
5760 (try_substitute_return_value): Same.
5762 2018-01-03 Jeff Law <law@redhat.com>
5765 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5766 non-constant residual for zero at runtime and avoid probing in
5767 that case. Reorganize code for trailing problem to mirror handling
5770 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5772 PR tree-optimization/83501
5773 * tree-ssa-strlen.c (get_string_cst): New.
5774 (handle_char_store): Call get_string_cst.
5776 2018-01-03 Martin Liska <mliska@suse.cz>
5778 PR tree-optimization/83593
5779 * tree-ssa-strlen.c: Include tree-cfg.h.
5780 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5781 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5782 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5784 (strlen_dom_walker::before_dom_children): Call
5785 gimple_purge_dead_eh_edges. Dump tranformation with details
5787 (strlen_dom_walker::before_dom_children): Update call by adding
5788 new argument cleanup_eh.
5789 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5791 2018-01-03 Martin Liska <mliska@suse.cz>
5794 * cif-code.def (VARIADIC_THUNK): New enum value.
5795 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5798 2018-01-03 Jan Beulich <jbeulich@suse.com>
5800 * sse.md (mov<mode>_internal): Tighten condition for when to use
5801 vmovdqu<ssescalarsize> for TI and OI modes.
5803 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5805 Update copyright years.
5807 2018-01-03 Martin Liska <mliska@suse.cz>
5810 * ipa-visibility.c (function_and_variable_visibility): Skip
5811 functions with noipa attribure.
5813 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5815 * gcc.c (process_command): Update copyright notice dates.
5816 * gcov-dump.c (print_version): Ditto.
5817 * gcov.c (print_version): Ditto.
5818 * gcov-tool.c (print_version): Ditto.
5819 * gengtype.c (create_file): Ditto.
5820 * doc/cpp.texi: Bump @copying's copyright year.
5821 * doc/cppinternals.texi: Ditto.
5822 * doc/gcc.texi: Ditto.
5823 * doc/gccint.texi: Ditto.
5824 * doc/gcov.texi: Ditto.
5825 * doc/install.texi: Ditto.
5826 * doc/invoke.texi: Ditto.
5828 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5830 * vector-builder.h (vector_builder::m_full_nelts): Change from
5831 unsigned int to poly_uint64.
5832 (vector_builder::full_nelts): Update prototype accordingly.
5833 (vector_builder::new_vector): Likewise.
5834 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5835 (vector_builder::operator ==): Likewise.
5836 (vector_builder::finalize): Likewise.
5837 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5838 Take the number of elements as a poly_uint64 rather than an
5840 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5841 from unsigned int to poly_uint64.
5842 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5843 (vec_perm_indices::new_vector): Likewise.
5844 (vec_perm_indices::length): Likewise.
5845 (vec_perm_indices::nelts_per_input): Likewise.
5846 (vec_perm_indices::input_nelts): Likewise.
5847 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5848 number of elements per input as a poly_uint64 rather than an
5849 unsigned int. Use the original encoding for variable-length
5850 vectors, rather than clamping each individual element.
5851 For the second and subsequent elements in each pattern,
5852 clamp the step and base before clamping their sum.
5853 (vec_perm_indices::series_p): Handle polynomial element counts.
5854 (vec_perm_indices::all_in_range_p): Likewise.
5855 (vec_perm_indices_to_tree): Likewise.
5856 (vec_perm_indices_to_rtx): Likewise.
5857 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5858 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5859 (tree_vector_builder::new_binary_operation): Handle polynomial
5860 element counts. Return false if we need to know the number
5861 of elements at compile time.
5862 * fold-const.c (fold_vec_perm): Punt if the number of elements
5863 isn't known at compile time.
5865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5867 * vec-perm-indices.h (vec_perm_builder): Change element type
5868 from HOST_WIDE_INT to poly_int64.
5869 (vec_perm_indices::element_type): Update accordingly.
5870 (vec_perm_indices::clamp): Handle polynomial element_types.
5871 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5872 (vec_perm_indices::all_in_range_p): Likewise.
5873 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5875 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5876 polynomial vec_perm_indices element types.
5877 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5878 * fold-const.c (fold_vec_perm): Likewise.
5879 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5880 * tree-vect-generic.c (lower_vec_perm): Likewise.
5881 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5882 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5883 element type to HOST_WIDE_INT.
5885 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5886 Alan Hayward <alan.hayward@arm.com>
5887 David Sherwood <david.sherwood@arm.com>
5889 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5890 rather than an int. Use plus_constant.
5891 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5892 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5894 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5895 Alan Hayward <alan.hayward@arm.com>
5896 David Sherwood <david.sherwood@arm.com>
5898 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5899 a HOST_WIDE_INT to a poly_int64.
5901 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5902 Alan Hayward <alan.hayward@arm.com>
5903 David Sherwood <david.sherwood@arm.com>
5905 * calls.c (load_register_parameters): Cope with polynomial
5906 mode sizes. Require a constant size for BLKmode parameters
5907 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5908 forces a parameter to be padded at the lsb end in order to
5909 fill a complete number of words, require the parameter size
5910 to be ordered wrt UNITS_PER_WORD.
5912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5913 Alan Hayward <alan.hayward@arm.com>
5914 David Sherwood <david.sherwood@arm.com>
5916 * reload1.c (spill_stack_slot_width): Change element type
5917 from unsigned int to poly_uint64_pod.
5918 (alter_reg): Treat mode sizes as polynomial.
5920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5921 Alan Hayward <alan.hayward@arm.com>
5922 David Sherwood <david.sherwood@arm.com>
5924 * reload.c (complex_word_subreg_p): New function.
5925 (reload_inner_reg_of_subreg, push_reload): Use it.
5927 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5928 Alan Hayward <alan.hayward@arm.com>
5929 David Sherwood <david.sherwood@arm.com>
5931 * lra-constraints.c (process_alt_operands): Reject matched
5932 operands whose sizes aren't ordered.
5933 (match_reload): Refer to this check here.
5935 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5936 Alan Hayward <alan.hayward@arm.com>
5937 David Sherwood <david.sherwood@arm.com>
5939 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5940 that the mode size is in the set {1, 2, 4, 8, 16}.
5942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5943 Alan Hayward <alan.hayward@arm.com>
5944 David Sherwood <david.sherwood@arm.com>
5946 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5947 Use plus_constant instead of gen_rtx_PLUS.
5949 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5950 Alan Hayward <alan.hayward@arm.com>
5951 David Sherwood <david.sherwood@arm.com>
5953 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5954 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5955 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5956 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5957 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5958 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5959 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5960 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5961 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5962 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5964 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5965 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5966 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5967 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5968 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5969 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5970 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5971 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5972 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5973 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5975 * expr.c (emit_move_resolve_push): Treat the input and result
5976 of PUSH_ROUNDING as a poly_int64.
5977 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5978 (emit_push_insn): Likewise.
5979 * lra-eliminations.c (mark_not_eliminable): Likewise.
5980 * recog.c (push_operand): Likewise.
5981 * reload1.c (elimination_effects): Likewise.
5982 * rtlanal.c (nonzero_bits1): Likewise.
5983 * calls.c (store_one_arg): Likewise. Require the padding to be
5984 known at compile time.
5986 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5987 Alan Hayward <alan.hayward@arm.com>
5988 David Sherwood <david.sherwood@arm.com>
5990 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5991 Use plus_constant instead of gen_rtx_PLUS.
5993 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5994 Alan Hayward <alan.hayward@arm.com>
5995 David Sherwood <david.sherwood@arm.com>
5997 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6001 Alan Hayward <alan.hayward@arm.com>
6002 David Sherwood <david.sherwood@arm.com>
6004 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6005 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6006 via stack temporaries. Treat the mode size as polynomial too.
6008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6009 Alan Hayward <alan.hayward@arm.com>
6010 David Sherwood <david.sherwood@arm.com>
6012 * expr.c (expand_expr_real_2): When handling conversions involving
6013 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6014 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6015 as a poly_uint64 too.
6017 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6018 Alan Hayward <alan.hayward@arm.com>
6019 David Sherwood <david.sherwood@arm.com>
6021 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6024 Alan Hayward <alan.hayward@arm.com>
6025 David Sherwood <david.sherwood@arm.com>
6027 * combine.c (can_change_dest_mode): Handle polynomial
6028 REGMODE_NATURAL_SIZE.
6029 * expmed.c (store_bit_field_1): Likewise.
6030 * expr.c (store_constructor): Likewise.
6031 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6032 and polynomial REGMODE_NATURAL_SIZE.
6033 (gen_lowpart_common): Likewise.
6034 * reginfo.c (record_subregs_of_mode): Likewise.
6035 * rtlanal.c (read_modify_subreg_p): Likewise.
6037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6038 Alan Hayward <alan.hayward@arm.com>
6039 David Sherwood <david.sherwood@arm.com>
6041 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6042 numbers of elements.
6044 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6045 Alan Hayward <alan.hayward@arm.com>
6046 David Sherwood <david.sherwood@arm.com>
6048 * match.pd: Cope with polynomial numbers of vector elements.
6050 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6051 Alan Hayward <alan.hayward@arm.com>
6052 David Sherwood <david.sherwood@arm.com>
6054 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6055 in a POINTER_PLUS_EXPR.
6057 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6058 Alan Hayward <alan.hayward@arm.com>
6059 David Sherwood <david.sherwood@arm.com>
6061 * omp-simd-clone.c (simd_clone_subparts): New function.
6062 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6063 (ipa_simd_modify_function_body): Likewise.
6065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6066 Alan Hayward <alan.hayward@arm.com>
6067 David Sherwood <david.sherwood@arm.com>
6069 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6070 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6071 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6072 (expand_vector_condition, vector_element): Likewise.
6073 (subparts_gt): New function.
6074 (get_compute_type): Use subparts_gt.
6075 (count_type_subparts): Delete.
6076 (expand_vector_operations_1): Use subparts_gt instead of
6077 count_type_subparts.
6079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6080 Alan Hayward <alan.hayward@arm.com>
6081 David Sherwood <david.sherwood@arm.com>
6083 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6084 (vect_compile_time_alias): ...this new function. Do the calculation
6085 on poly_ints rather than trees.
6086 (vect_prune_runtime_alias_test_list): Update call accordingly.
6088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6089 Alan Hayward <alan.hayward@arm.com>
6090 David Sherwood <david.sherwood@arm.com>
6092 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6094 (vect_schedule_slp_instance): Likewise.
6096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6097 Alan Hayward <alan.hayward@arm.com>
6098 David Sherwood <david.sherwood@arm.com>
6100 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6101 constant and extern definitions for variable-length vectors.
6102 (vect_get_constant_vectors): Note that the number of units
6103 is known to be constant.
6105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6106 Alan Hayward <alan.hayward@arm.com>
6107 David Sherwood <david.sherwood@arm.com>
6109 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6110 of units as polynomial. Choose between WIDE and NARROW based
6113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6114 Alan Hayward <alan.hayward@arm.com>
6115 David Sherwood <david.sherwood@arm.com>
6117 * tree-vect-stmts.c (simd_clone_subparts): New function.
6118 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6121 Alan Hayward <alan.hayward@arm.com>
6122 David Sherwood <david.sherwood@arm.com>
6124 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6125 vectors as polynomial. Use build_index_vector for
6128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6129 Alan Hayward <alan.hayward@arm.com>
6130 David Sherwood <david.sherwood@arm.com>
6132 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6133 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6134 for variable-length vectors.
6135 (vectorizable_mask_load_store): Treat the number of units as
6136 polynomial, asserting that it is constant if the condition has
6137 already been enforced.
6138 (vectorizable_store, vectorizable_load): Likewise.
6140 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6141 Alan Hayward <alan.hayward@arm.com>
6142 David Sherwood <david.sherwood@arm.com>
6144 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6145 of units as polynomial. Punt if we can't tell at compile time
6146 which vector contains the final result.
6148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6149 Alan Hayward <alan.hayward@arm.com>
6150 David Sherwood <david.sherwood@arm.com>
6152 * tree-vect-loop.c (vectorizable_induction): Treat the number
6153 of units as polynomial. Punt on SLP inductions. Use an integer
6154 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6155 cast of such a series for variable-length floating-point
6158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6159 Alan Hayward <alan.hayward@arm.com>
6160 David Sherwood <david.sherwood@arm.com>
6162 * tree.h (build_index_vector): Declare.
6163 * tree.c (build_index_vector): New function.
6164 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6165 of units as polynomial, forcibly converting it to a constant if
6166 vectorizable_reduction has already enforced the condition.
6167 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6168 to create a {1,2,3,...} vector.
6169 (vectorizable_reduction): Treat the number of units as polynomial.
6170 Choose vectype_in based on the largest scalar element size rather
6171 than the smallest number of units. Enforce the restrictions
6174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6175 Alan Hayward <alan.hayward@arm.com>
6176 David Sherwood <david.sherwood@arm.com>
6178 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6179 number of units as polynomial.
6181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6182 Alan Hayward <alan.hayward@arm.com>
6183 David Sherwood <david.sherwood@arm.com>
6185 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6186 * target.def (autovectorize_vector_sizes): Return the vector sizes
6187 by pointer, using vector_sizes rather than a bitmask.
6188 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6189 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6190 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6192 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6193 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6194 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6195 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6196 * omp-general.c (omp_max_vf): Likewise.
6197 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6198 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6199 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6200 * tree-vect-slp.c (vect_slp_bb): Likewise.
6201 * doc/tm.texi: Regenerate.
6202 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6204 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6205 the vector size as a poly_uint64 rather than an unsigned int.
6206 (current_vector_size): Change from an unsigned int to a poly_uint64.
6207 (get_vectype_for_scalar_type): Update accordingly.
6208 * tree.h (build_truth_vector_type): Take the size and number of
6209 units as a poly_uint64 rather than an unsigned int.
6210 (build_vector_type): Add a temporary overload that takes
6211 the number of units as a poly_uint64 rather than an unsigned int.
6212 * tree.c (make_vector_type): Likewise.
6213 (build_truth_vector_type): Take the number of units as a poly_uint64
6214 rather than an unsigned int.
6216 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6217 Alan Hayward <alan.hayward@arm.com>
6218 David Sherwood <david.sherwood@arm.com>
6220 * target.def (get_mask_mode): Take the number of units and length
6221 as poly_uint64s rather than unsigned ints.
6222 * targhooks.h (default_get_mask_mode): Update accordingly.
6223 * targhooks.c (default_get_mask_mode): Likewise.
6224 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6225 * doc/tm.texi: Regenerate.
6227 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6228 Alan Hayward <alan.hayward@arm.com>
6229 David Sherwood <david.sherwood@arm.com>
6231 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6232 * omp-general.c (omp_max_vf): Likewise.
6233 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6234 (expand_omp_simd): Handle polynomial safelen.
6235 * omp-low.c (omplow_simd_context): Add a default constructor.
6236 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6237 (lower_rec_simd_input_clauses): Update accordingly.
6238 (lower_rec_input_clauses): Likewise.
6240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6241 Alan Hayward <alan.hayward@arm.com>
6242 David Sherwood <david.sherwood@arm.com>
6244 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6245 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6246 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6247 (vect_analyze_slp_cost): Likewise.
6248 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6249 (vect_model_load_cost): Likewise.
6251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6252 Alan Hayward <alan.hayward@arm.com>
6253 David Sherwood <david.sherwood@arm.com>
6255 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6256 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6257 from an unsigned int * to a poly_uint64_pod *.
6258 (calculate_unrolling_factor): New function.
6259 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6262 Alan Hayward <alan.hayward@arm.com>
6263 David Sherwood <david.sherwood@arm.com>
6265 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6266 from an unsigned int to a poly_uint64.
6267 (_loop_vec_info::slp_unrolling_factor): Likewise.
6268 (_loop_vec_info::vectorization_factor): Change from an int
6270 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6271 (vect_get_num_vectors): New function.
6272 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6273 (vect_get_num_copies): Use vect_get_num_vectors.
6274 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6275 to an unsigned int *.
6276 (vect_analyze_data_refs): Change min_vf from an int * to a
6278 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6279 than an unsigned HOST_WIDE_INT.
6280 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6281 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6282 to an unsigned int *.
6283 (vect_analyze_data_ref_dependences): Likewise.
6284 (vect_compute_data_ref_alignment): Handle polynomial vf.
6285 (vect_enhance_data_refs_alignment): Likewise.
6286 (vect_prune_runtime_alias_test_list): Likewise.
6287 (vect_shift_permute_load_chain): Likewise.
6288 (vect_supportable_dr_alignment): Likewise.
6289 (dependence_distance_ge_vf): Take the vectorization factor as a
6290 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6291 (vect_analyze_data_refs): Change min_vf from an int * to a
6293 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6294 vfm1 as a poly_uint64 rather than an int. Make the same change
6295 for the returned bound_scalar.
6296 (vect_gen_vector_loop_niters): Handle polynomial vf.
6297 (vect_do_peeling): Likewise. Update call to
6298 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6299 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6301 * tree-vect-loop.c (vect_determine_vectorization_factor)
6302 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6303 (vect_get_known_peeling_cost): Likewise.
6304 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6305 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6306 (vect_transform_loop): Likewise. Use the lowest possible VF when
6307 updating the upper bounds of the loop.
6308 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6310 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6311 polynomial unroll factors.
6312 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6313 (vect_make_slp_decision): Likewise.
6314 (vect_supported_load_permutation_p): Likewise, and polynomial
6316 (vect_analyze_slp_cost): Handle polynomial vf.
6317 (vect_slp_analyze_node_operations): Likewise.
6318 (vect_slp_analyze_bb_1): Likewise.
6319 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6320 than an unsigned HOST_WIDE_INT.
6321 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6322 (vectorizable_load): Handle polynomial vf.
6323 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6325 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6328 Alan Hayward <alan.hayward@arm.com>
6329 David Sherwood <david.sherwood@arm.com>
6331 * match.pd: Handle bit operations involving three constants
6332 and try to fold one pair.
6334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6336 * tree-vect-loop-manip.c: Include gimple-fold.h.
6337 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6338 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6339 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6340 Add a path that uses a step of VF instead of 1, but disable it
6342 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6343 and niters_no_overflow parameters. Update calls to
6344 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6345 Create a new SSA name if the latter choses to use a ste other
6346 than zero, and return it via niters_vector_mult_vf_var.
6347 * tree-vect-loop.c (vect_transform_loop): Update calls to
6348 vect_do_peeling, vect_gen_vector_loop_niters and
6349 slpeel_make_loop_iterate_ntimes.
6350 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6351 (vect_gen_vector_loop_niters): Update declarations after above changes.
6353 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6355 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6356 128-bit round to integer instructions.
6357 (ceil<mode>2): Likewise.
6358 (btrunc<mode>2): Likewise.
6359 (round<mode>2): Likewise.
6361 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6363 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6364 unaligned VSX load/store on P8/P9.
6365 (expand_block_clear): Allow the use of unaligned VSX
6366 load/store on P8/P9.
6368 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6370 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6372 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6373 swap associated with both a load and a store.
6375 2018-01-02 Andrew Waterman <andrew@sifive.com>
6377 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6378 * config/riscv/riscv.md (clear_cache): Use it.
6380 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6382 * web.c: Remove out-of-date comment.
6384 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6386 * expr.c (fixup_args_size_notes): Check that any existing
6387 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6388 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6389 (emit_single_push_insn): ...here.
6391 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6393 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6394 (const_vector_encoded_nelts): New function.
6395 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6396 (const_vector_int_elt, const_vector_elt): Declare.
6397 * emit-rtl.c (const_vector_int_elt_1): New function.
6398 (const_vector_elt): Likewise.
6399 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6400 of CONST_VECTOR_ELT.
6402 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6404 * expr.c: Include rtx-vector-builder.h.
6405 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6406 directly on the tree encoding.
6407 (const_vector_from_tree): Likewise.
6408 * optabs.c: Include rtx-vector-builder.h.
6409 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6410 sequence of "u" values.
6411 * vec-perm-indices.c: Include rtx-vector-builder.h.
6412 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6413 directly on the vec_perm_indices encoding.
6415 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6417 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6418 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6419 * rtx-vector-builder.h: New file.
6420 * rtx-vector-builder.c: Likewise.
6421 * rtl.h (rtx_def::u2): Add a const_vector field.
6422 (CONST_VECTOR_NPATTERNS): New macro.
6423 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6424 (CONST_VECTOR_DUPLICATE_P): Likewise.
6425 (CONST_VECTOR_STEPPED_P): Likewise.
6426 (CONST_VECTOR_ENCODED_ELT): Likewise.
6427 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6428 (unwrap_const_vec_duplicate): Likewise.
6429 (const_vec_series_p): Check for a non-duplicated vector encoding.
6430 Say that the function only returns true for integer vectors.
6431 * emit-rtl.c: Include rtx-vector-builder.h.
6432 (gen_const_vec_duplicate_1): Delete.
6433 (gen_const_vector): Call gen_const_vec_duplicate instead of
6434 gen_const_vec_duplicate_1.
6435 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6436 (gen_const_vec_duplicate): Use rtx_vector_builder.
6437 (gen_const_vec_series): Likewise.
6438 (gen_rtx_CONST_VECTOR): Likewise.
6439 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6440 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6441 Build a new vector rather than modifying a CONST_VECTOR in-place.
6442 (handle_special_swappables): Update call accordingly.
6443 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6444 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6445 Build a new vector rather than modifying a CONST_VECTOR in-place.
6446 (handle_special_swappables): Update call accordingly.
6448 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6450 * simplify-rtx.c (simplify_const_binary_operation): Use
6451 CONST_VECTOR_ELT instead of XVECEXP.
6453 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6455 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6456 the selector elements to be different from the data elements
6457 if the selector is a VECTOR_CST.
6458 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6459 ssizetype for the selector.
6461 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6463 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6464 before testing each element individually.
6465 * tree-vect-generic.c (lower_vec_perm): Likewise.
6467 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6469 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6470 * selftest-run-tests.c (selftest::run_tests): Call it.
6471 * vector-builder.h (vector_builder::operator ==): New function.
6472 (vector_builder::operator !=): Likewise.
6473 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6474 (vec_perm_indices::all_from_input_p): New function.
6475 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6476 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6477 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6478 instead of reading the VECTOR_CST directly. Detect whether both
6479 vector inputs are the same before constructing the vec_perm_indices,
6480 and update the number of inputs argument accordingly. Use the
6481 utility functions added above. Only construct sel2 if we need to.
6483 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6485 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6486 the broadcast of the low byte.
6487 (expand_mult_highpart): Use an explicit encoding for the permutes.
6488 * optabs-query.c (can_mult_highpart_p): Likewise.
6489 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6490 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6491 (vectorizable_bswap): Likewise.
6492 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6493 explicit encoding for the power-of-2 permutes.
6494 (vect_permute_store_chain): Likewise.
6495 (vect_grouped_load_supported): Likewise.
6496 (vect_permute_load_chain): Likewise.
6498 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6500 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6501 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6502 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6503 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6504 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6505 (vect_gen_perm_mask_any): Likewise.
6507 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6509 * int-vector-builder.h: New file.
6510 * vec-perm-indices.h: Include int-vector-builder.h.
6511 (vec_perm_indices): Redefine as an int_vector_builder.
6512 (auto_vec_perm_indices): Delete.
6513 (vec_perm_builder): Redefine as a stand-alone class.
6514 (vec_perm_indices::vec_perm_indices): New function.
6515 (vec_perm_indices::clamp): Likewise.
6516 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6517 (vec_perm_indices::new_vector): New function.
6518 (vec_perm_indices::new_expanded_vector): Update for new
6519 vec_perm_indices class.
6520 (vec_perm_indices::rotate_inputs): New function.
6521 (vec_perm_indices::all_in_range_p): Operate directly on the
6522 encoded form, without computing elided elements.
6523 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6524 encoding. Update for new vec_perm_indices class.
6525 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6526 the given vec_perm_builder.
6527 (expand_vec_perm_var): Update vec_perm_builder constructor.
6528 (expand_mult_highpart): Use vec_perm_builder instead of
6529 auto_vec_perm_indices.
6530 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6531 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6532 or double series encoding as appropriate.
6533 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6534 vec_perm_indices instead of auto_vec_perm_indices.
6535 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6536 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6537 (vect_permute_store_chain): Likewise.
6538 (vect_grouped_load_supported): Likewise.
6539 (vect_permute_load_chain): Likewise.
6540 (vect_shift_permute_load_chain): Likewise.
6541 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6542 (vect_transform_slp_perm_load): Likewise.
6543 (vect_schedule_slp_instance): Likewise.
6544 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6545 (vectorizable_mask_load_store): Likewise.
6546 (vectorizable_bswap): Likewise.
6547 (vectorizable_store): Likewise.
6548 (vectorizable_load): Likewise.
6549 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6550 vec_perm_indices instead of auto_vec_perm_indices. Use
6551 tree_to_vec_perm_builder to read the vector from a tree.
6552 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6553 vec_perm_builder instead of a vec_perm_indices.
6554 (have_whole_vector_shift): Use vec_perm_builder and
6555 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6556 truncation to calc_vec_perm_mask_for_shift.
6557 (vect_create_epilog_for_reduction): Likewise.
6558 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6559 from auto_vec_perm_indices to vec_perm_indices.
6560 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6561 instead of changing individual elements.
6562 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6563 the vector in d.perm.
6564 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6565 from auto_vec_perm_indices to vec_perm_indices.
6566 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6567 instead of changing individual elements.
6568 (arm_vectorize_vec_perm_const): Use new_vector to install
6569 the vector in d.perm.
6570 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6571 Update vec_perm_builder constructor.
6572 (rs6000_expand_interleave): Likewise.
6573 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6574 (rs6000_expand_interleave): Likewise.
6576 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6578 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6579 to qimode could truncate the indices.
6580 * optabs.c (expand_vec_perm_var): Likewise.
6582 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6584 * Makefile.in (OBJS): Add vec-perm-indices.o.
6585 * vec-perm-indices.h: New file.
6586 * vec-perm-indices.c: Likewise.
6587 * target.h (vec_perm_indices): Replace with a forward class
6589 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6590 * optabs.h: Include vec-perm-indices.h.
6591 (expand_vec_perm): Delete.
6592 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6593 (expand_vec_perm_const): Declare.
6594 * target.def (vec_perm_const_ok): Replace with...
6595 (vec_perm_const): ...this new hook.
6596 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6597 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6598 * doc/tm.texi: Regenerate.
6599 * optabs.def (vec_perm_const): Delete.
6600 * doc/md.texi (vec_perm_const): Likewise.
6601 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6602 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6603 expand_vec_perm for constant permutation vectors. Assert that
6604 the mode of variable permutation vectors is the integer equivalent
6605 of the mode that is being permuted.
6606 * optabs-query.h (selector_fits_mode_p): Declare.
6607 * optabs-query.c: Include vec-perm-indices.h.
6608 (selector_fits_mode_p): New function.
6609 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6610 is defined, instead of checking whether the vec_perm_const_optab
6611 exists. Use targetm.vectorize.vec_perm_const instead of
6612 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6613 fit in the vector mode before using a variable permute.
6614 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6615 vec_perm_indices instead of an rtx.
6616 (expand_vec_perm): Replace with...
6617 (expand_vec_perm_const): ...this new function. Take the selector
6618 as a vec_perm_indices rather than an rtx. Also take the mode of
6619 the selector. Update call to shift_amt_for_vec_perm_mask.
6620 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6621 Use vec_perm_indices::new_expanded_vector to expand the original
6622 selector into bytes. Check whether the indices fit in the vector
6623 mode before using a variable permute.
6624 (expand_vec_perm_var): Make global.
6625 (expand_mult_highpart): Use expand_vec_perm_const.
6626 * fold-const.c: Includes vec-perm-indices.h.
6627 * tree-ssa-forwprop.c: Likewise.
6628 * tree-vect-data-refs.c: Likewise.
6629 * tree-vect-generic.c: Likewise.
6630 * tree-vect-loop.c: Likewise.
6631 * tree-vect-slp.c: Likewise.
6632 * tree-vect-stmts.c: Likewise.
6633 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6635 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6636 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6637 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6638 (aarch64_vectorize_vec_perm_const): ...this new function.
6639 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6640 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6641 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6642 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6643 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6644 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6645 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6647 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6648 check for NEON modes.
6649 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6650 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6651 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6652 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6654 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6655 the old VEC_PERM_CONST conditions.
6656 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6657 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6658 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6659 (ia64_vectorize_vec_perm_const_ok): Merge into...
6660 (ia64_vectorize_vec_perm_const): ...this new function.
6661 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6662 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6663 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6664 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6665 * config/mips/mips.c (mips_expand_vec_perm_const)
6666 (mips_vectorize_vec_perm_const_ok): Merge into...
6667 (mips_vectorize_vec_perm_const): ...this new function.
6668 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6669 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6670 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6671 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6672 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6673 (rs6000_expand_vec_perm_const): Delete.
6674 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6676 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6677 (altivec_expand_vec_perm_const_le): Take each operand individually.
6678 Operate on constant selectors rather than rtxes.
6679 (altivec_expand_vec_perm_const): Likewise. Update call to
6680 altivec_expand_vec_perm_const_le.
6681 (rs6000_expand_vec_perm_const): Delete.
6682 (rs6000_vectorize_vec_perm_const_ok): Delete.
6683 (rs6000_vectorize_vec_perm_const): New function.
6684 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6685 an element count and rtx array.
6686 (rs6000_expand_extract_even): Update call accordingly.
6687 (rs6000_expand_interleave): Likewise.
6688 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6689 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6690 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6691 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6692 (rs6000_expand_vec_perm_const): Delete.
6693 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6694 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6695 (altivec_expand_vec_perm_const_le): Take each operand individually.
6696 Operate on constant selectors rather than rtxes.
6697 (altivec_expand_vec_perm_const): Likewise. Update call to
6698 altivec_expand_vec_perm_const_le.
6699 (rs6000_expand_vec_perm_const): Delete.
6700 (rs6000_vectorize_vec_perm_const_ok): Delete.
6701 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6702 reference to the SPE evmerge intructions.
6703 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6704 an element count and rtx array.
6705 (rs6000_expand_extract_even): Update call accordingly.
6706 (rs6000_expand_interleave): Likewise.
6707 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6708 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6710 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6712 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6714 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6715 vector mode and that that mode matches the mode of the data
6717 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6718 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6719 directly using expand_vec_perm_1 when forcing selectors into
6721 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6723 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6725 * optabs-query.h (can_vec_perm_p): Delete.
6726 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6727 * optabs-query.c (can_vec_perm_p): Split into...
6728 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6729 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6730 particular selector is valid.
6731 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6732 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6733 (vect_grouped_load_supported): Likewise.
6734 (vect_shift_permute_load_chain): Likewise.
6735 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6736 (vect_transform_slp_perm_load): Likewise.
6737 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6738 (vectorizable_bswap): Likewise.
6739 (vect_gen_perm_mask_checked): Likewise.
6740 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6741 implementations of variable permutation vectors into account
6742 when deciding which selector to use.
6743 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6744 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6745 with a false third argument.
6746 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6747 to test whether the constant selector is valid and can_vec_perm_var_p
6748 to test whether a variable selector is valid.
6750 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6752 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6753 * optabs-query.c (can_vec_perm_p): Likewise.
6754 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6755 instead of vec_perm_indices.
6756 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6757 (vect_gen_perm_mask_checked): Likewise,
6758 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6759 (vect_gen_perm_mask_checked): Likewise,
6761 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6763 * optabs-query.h (qimode_for_vec_perm): Declare.
6764 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6765 (qimode_for_vec_perm): ...this new function.
6766 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6768 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6770 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6771 does not have a conditional at the top.
6773 2018-01-02 Richard Biener <rguenther@suse.de>
6775 * ipa-inline.c (big_speedup_p): Fix expression.
6777 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6780 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6783 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6787 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6788 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6789 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6790 cond_taken_branch_cost 3->4.
6792 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6794 PR tree-optimization/83581
6795 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6796 TODO_cleanup_cfg if any changes have been made.
6799 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6800 convert_modes if target mode has the right side, but different mode
6804 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6805 last argument when extracting from CONCAT. If either from_real or
6806 from_imag is NULL, use expansion through memory. If result is not
6807 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6808 the parts directly to inner mode, if even that fails, use expansion
6812 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6813 check for bswap in mode rather than HImode and use that in expand_unop
6816 Copyright (C) 2018 Free Software Foundation, Inc.
6818 Copying and distribution of this file, with or without modification,
6819 are permitted in any medium without royalty provided the copyright
6820 notice and this notice are preserved.