1 2018-02-13 Jakub Jelinek <jakub@redhat.com>
4 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
5 operands[2] into a REG before using gen_lowpart on it.
7 2018-02-12 Jeff Law <law@redhat.com>
10 * config/sh/sh.c (find_barrier): Consider a sibling call
13 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
14 successfully back substituting a reg.
16 2018-02-12 Richard Biener <rguenther@suse.de>
18 PR tree-optimization/84037
19 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
20 parameter, move visited init to caller.
21 (vect_slp_analyze_operations): Separate cost from validity
22 check, initialize visited once for all instances.
23 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
25 * tree-vect-stmts.c (vect_model_simple_cost): Make early
27 (vect_model_promotion_demotion_cost): Likewise.
28 (vectorizable_bswap): Guard cost modeling with !slp_node
29 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
31 (vectorizable_call): Likewise.
32 (vectorizable_conversion): Likewise.
33 (vectorizable_assignment): Likewise.
34 (vectorizable_shift): Likewise.
35 (vectorizable_operation): Likewise.
36 (vectorizable_store): Likewise.
37 (vectorizable_load): Likewise.
38 (vectorizable_condition): Likewise.
39 (vectorizable_comparison): Likewise.
41 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
44 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
45 (ASAN_MARK): Fix fnspec to account for return value, change pointer
46 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
48 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
51 * params.def (inline-min-speedup): Increase from 8 to 15.
52 (max-inline-insns-auto): Decrease from 40 to 30.
53 * ipa-split.c (consider_split): Add some buffer for function to
54 be considered inlining candidate.
55 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
58 2018-02-12 Richard Biener <rguenther@suse.de>
60 PR tree-optimization/84037
61 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
62 matched stmts if we cannot swap the non-matched ones.
64 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
66 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
67 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
68 _mm_maskz_scalef_round_ss): New intrinsics.
69 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
70 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
71 __builtin_ia32_scalefss_round): Remove.
72 (__builtin_ia32_scalefsd_mask_round,
73 __builtin_ia32_scalefss_mask_round): New intrinsics.
74 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
75 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
76 ((match_operand:VF_128 2 "<round_nimm_predicate>"
77 "<round_constraint>")): Changed to ...
78 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
79 "<round_scalar_constraint>")): ... this.
80 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
81 %0, %1, %2<round_op3>}"): Changed to ...
82 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
83 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
84 %2<round_scalar_mask_op3>}"): ... this.
85 * config/i386/subst.md (round_scalar_nimm_predicate): New.
87 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
89 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
90 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
91 (_mm_maskz_sqrt_round_ss): New intrinsics.
92 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
93 (__builtin_ia32_sqrtsd_mask_round)
94 (__builtin_ia32_sqrtss_mask_round): New builtins.
95 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
96 (__builtin_ia32_sqrtss_round): Remove.
97 (__builtin_ia32_sqrtsd_mask_round)
98 (__builtin_ia32_sqrtss_mask_round): New builtins.
99 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
100 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
101 ((match_operand:VF_128 1 "vector_operand"
102 "xBm,<round_constraint>")): Changed to ...
103 ((match_operand:VF_128 1 "vector_operand"
104 "xBm,<round_scalar_constraint>")): ... this.
105 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
106 %0, %2, %<iptr>1<round_op3>}): Changed to ...
107 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
108 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
109 %<iptr>1<round_scalar_mask_op3>}): ... this.
110 ((set_attr "prefix" "<round_prefix>")): Changed to ...
111 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
113 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
116 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
117 Cast vec_cmpeq result to correct type.
118 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
119 Cast vec_cmpgt result to correct type.
121 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
123 * final.c (final_scan_insn_1): Renamed from...
124 (final_scan_insn): ... this. New wrapper, to recover
125 seen from the outermost call in recursive ones.
126 * config/sparc/sparc.c (output_return): Drop seen from call.
127 (output_sibcall): Likewise.
128 * config/visium/visium.c (output_branch): Likewise.
130 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
132 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
135 2018-02-10 Alan Modra <amodra@gmail.com>
138 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
139 Specify LR as an input.
141 2018-02-10 Jakub Jelinek <jakub@redhat.com>
144 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
145 remove_member_access_dummy_vars): New functions.
146 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
147 lower_omp_1, execute_lower_omp): Use them.
149 PR rtl-optimization/84308
150 * shrink-wrap.c (spread_components): Release todo vector.
152 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
154 PR rtl-optimization/57193
155 * ira-color.c (struct allocno_color_data): Add member
156 conflict_allocno_hard_prefs.
157 (update_conflict_allocno_hard_prefs): New.
158 (bucket_allocno_compare_func): Add a preference based on
159 conflict_allocno_hard_prefs.
160 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
161 (color_allocnos): Remove a dead code. Initiate
162 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
164 2018-02-09 Jakub Jelinek <jakub@redhat.com>
167 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
168 constraint from =wa to wa. Avoid a subreg on the output operand,
169 instead use a pseudo and subreg it in a move.
170 (p9_xxbrd_<mode>): Changed to ...
171 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
172 (p9_xxbrd_v2df): New expander.
173 (p9_xxbrw_<mode>): Changed to ...
174 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
175 (p9_xxbrw_v4sf): New expander.
177 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
179 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
181 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
184 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
185 multiply in 32-bit mode.
186 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
187 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
190 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
192 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
193 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
194 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
195 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
197 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
200 * dwarf2out.c (is_trivial_indirect_ref): New function.
201 (dwarf2out_late_global_decl): Do not generate a location
202 attribute for variables that have a non-trivial DECL_VALUE_EXPR
203 and that are not defined in the current unit.
205 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
207 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
208 instead of a libcall for UNORDERED.
210 2018-02-09 Tamar Christina <tamar.christina@arm.com>
213 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
214 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
216 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
219 * config/s390/s390.c (s390_set_current_function): Invoke
220 s390_indirect_branch_settings also if fndecl didn't change.
222 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
224 * config/rs6000/rs6000.md (blockage): Set length to zero.
226 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
228 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
230 2018-02-09 Jakub Jelinek <jakub@redhat.com>
233 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
234 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
238 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
239 PARALLEL incoming that failed vt_get_decl_and_offset check.
242 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
243 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
245 (get_variable_section): For decls in named .bss* sections pass true as
246 second argument to bss_initializer_p.
248 2018-02-09 Marek Polacek <polacek@redhat.com>
249 Jakub Jelinek <jakub@redhat.com>
252 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
253 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
254 Sync some changes from cxx_fold_indirect_ref.
256 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
258 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
260 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
261 (BLOCK_INLINE_ENTRY_LABEL): New.
262 (dwarf2out_var_location): Disregard inline entry markers.
263 (inline_entry_data): New struct.
264 (inline_entry_data_hasher): New hashtable type.
265 (inline_entry_data_hasher::hash): New.
266 (inline_entry_data_hasher::equal): New.
267 (inline_entry_data_table): New variable.
268 (add_high_low_attributes): Add DW_AT_entry_pc and
269 DW_AT_GNU_entry_view attributes if a pending entry is found
270 in inline_entry_data_table. Add old entry_pc attribute only
271 if debug nonbinding markers are disabled.
272 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
274 (block_within_block_p, dwarf2out_inline_entry): New.
275 (dwarf2out_finish): Check that no entries remained in
276 inline_entry_data_table.
277 * final.c (reemit_insn_block_notes): Handle inline entry notes.
278 (final_scan_insn, notice_source_line): Likewise.
279 (rest_of_clean_state): Skip inline entry markers.
280 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
282 * gimple.c (gimple_build_debug_inline_entry): New.
283 * gimple.h (enum gimple_debug_subcode): Add
284 GIMPLE_DEBUG_INLINE_ENTRY.
285 (gimple_build_debug_inline_entry): Declare.
286 (gimple_debug_inline_entry_p): New.
287 (gimple_debug_nonbind_marker_p): Adjust.
288 * insn-notes.def (INLINE_ENTRY): New.
289 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
290 inline entry marker notes.
291 (print_insn): Likewise.
292 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
293 (INSN_DEBUG_MARKER_KIND): Likewise.
294 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
295 * tree-inline.c (expand_call_inline): Build and insert
296 debug_inline_entry stmt.
297 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
298 inline entry blocks early, if nonbind markers are enabled.
299 (dump_scope_block): Dump fragment info.
300 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
301 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
302 (gimple_build_debug_inline_entry): New.
303 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
304 Enable/disable inline entry points too.
305 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
306 (DEBUG_INSN): Describe inline entry markers.
308 * common.opt (gvariable-location-views): New.
309 (gvariable-location-views=incompat5): New.
310 * config.in: Rebuilt.
311 * configure: Rebuilt.
312 * configure.ac: Test assembler for view support.
313 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
314 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
315 * dwarf2out.c (var_loc_view): New typedef.
316 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
317 (dwarf2out_locviews_in_attribute): New.
318 (dwarf2out_locviews_in_loclist): New.
319 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
320 (enum dw_line_info_opcode): Add LI_adv_address.
321 (struct dw_line_info_table): Add view.
322 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
323 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
324 (zero_view_p): New variable.
325 (ZERO_VIEW_P): New macro.
326 (output_asm_line_debug_info): New.
327 (struct var_loc_node): Add view.
328 (add_AT_view_list, AT_loc_list): New.
329 (add_var_loc_to_decl): Add view param. Test it against last.
330 (new_loc_list): Add view params. Record them.
331 (AT_loc_list_ptr): Handle loc and view lists.
332 (view_list_to_loc_list_val_node): New.
333 (print_dw_val): Handle dw_val_class_view_list.
334 (size_of_die): Likewise.
335 (value_format): Likewise.
336 (loc_list_has_views): New.
337 (gen_llsym): Set vl_symbol too.
338 (maybe_gen_llsym, skip_loc_list_entry): New.
339 (dwarf2out_maybe_output_loclist_view_pair): New.
340 (output_loc_list): Output view list or entries too.
341 (output_view_list_offset): New.
342 (output_die): Handle dw_val_class_view_list.
343 (output_dwarf_version): New.
344 (output_compilation_unit_header): Use it.
345 (output_skeleton_debug_sections): Likewise.
346 (output_rnglists, output_line_info): Likewise.
347 (output_pubnames, output_aranges): Update version comments.
348 (output_one_line_info_table): Output view numbers in asm comments.
349 (dw_loc_list): Determine current endview, pass it to new_loc_list.
350 Call maybe_gen_llsym.
351 (loc_list_from_tree_1): Adjust.
352 (add_AT_location_description): Create view list attribute if
353 needed, check it's absent otherwise.
354 (convert_cfa_to_fb_loc_list): Adjust.
355 (maybe_emit_file): Call output_asm_line_debug_info for test.
356 (dwarf2out_var_location): Reset views as needed. Precompute
357 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
359 (new_line_info_table): Reset next view.
360 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
361 (dwarf2out_source_line): Likewise. Output view resets and labels to
362 the assembler, or select appropriate line info opcodes.
363 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
364 (optimize_string_length): Catch it. Adjust.
365 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
366 dw_val_class_view_list, and remove it if no longer needed.
367 (hash_loc_list): Hash view numbers.
368 (loc_list_hasher::equal): Compare them.
369 (optimize_location_lists): Check whether a view list symbol is
370 needed, and whether the locview attribute is present, and
371 whether they match. Remove the locview attribute if no longer
373 (index_location_lists): Call skip_loc_list_entry for test.
374 (dwarf2out_finish): Call output_asm_line_debug_info for test.
375 Use output_dwarf_version.
376 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
377 (struct dw_val_node): Add val_view_list.
378 * final.c (SEEN_NEXT_VIEW): New.
379 (set_next_view_needed): New.
380 (clear_next_view_needed): New.
381 (maybe_output_next_view): New.
382 (final_start_function): Rename to...
383 (final_start_function_1): ... this. Take pointer to FIRST,
384 add SEEN parameter. Emit param bindings in the initial view.
385 (final_start_function): Reintroduce SEEN-less interface.
386 (final): Rename to...
387 (final_1): ... this. Take SEEN parameter. Output final pending
388 next view at the end.
389 (final): Reintroduce seen-less interface.
390 (final_scan_insn): Output pending next view before switching
391 sections or ending a block. Mark the next view as needed when
392 outputting variable locations. Notify debug backend of section
393 changes, and of location view changes.
394 (rest_of_handle_final): Adjust.
395 * toplev.c (process_options): Autodetect value for debug variable
396 location views option. Warn on incompat5 without -gdwarf-5.
397 * doc/invoke.texi (gvariable-location-views): New.
398 (gvariable-location-views=incompat5): New.
399 (gno-variable-location-views): New.
401 2018-02-08 David Malcolm <dmalcolm@redhat.com>
403 PR tree-optimization/84136
404 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
405 that the result of find_edge is non-NULL.
407 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
410 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
411 storing integer register in SImode. Fix cost of 256 and 512
412 byte aligned SSE register store.
414 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
416 * config/i386/i386.c (ix86_multiplication_cost): Fix
417 multiplication cost for TARGET_AVX512DQ.
419 2018-02-08 Marek Polacek <polacek@redhat.com>
421 PR tree-optimization/84238
422 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
425 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
427 PR tree-optimization/84265
428 * tree-vect-stmts.c (vectorizable_store): Don't treat
429 VMAT_CONTIGUOUS accesses as grouped.
430 (vectorizable_load): Likewise.
432 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
434 PR tree-optimization/81635
435 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
436 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
437 (test_round_for_mask): New functions.
438 (wide_int_cc_tests): Call test_round_for_mask.
439 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
440 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
441 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
442 range returned by get_range_info.
444 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
447 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
448 * symtab.c: Include builtins.h
449 (symtab_node::output_to_lto_symbol_table_p): Move here
450 from lto-streamer-out.c:output_symbol_p.
451 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
452 (output_symbol_p): Move all logic to symtab.c
453 (produce_symtab): Update.
455 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
457 * config/s390/s390-opts.h (enum indirect_branch): Define.
458 * config/s390/s390-protos.h (s390_return_addr_from_memory)
459 (s390_indirect_branch_via_thunk)
460 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
461 (enum s390_indirect_branch_type): Define.
462 * config/s390/s390.c (struct s390_frame_layout, struct
463 machine_function): Remove.
464 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
465 (indirect_branch_table_label_no, indirect_branch_table_name):
467 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
468 (enum s390_indirect_branch_option): Define.
469 (s390_return_addr_from_memory): New function.
470 (s390_handle_string_attribute): New function.
471 (s390_attribute_table): Add new attribute handler.
472 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
473 (s390_indirect_branch_via_thunk): New function.
474 (s390_indirect_branch_via_inline_thunk): New function.
475 (s390_function_ok_for_sibcall): When jumping via thunk disallow
476 sibling call optimization for non z10 compiles.
477 (s390_emit_call): Force indirect branch target to be a single
478 register. Add r1 clobber for non-z10 compiles.
479 (s390_emit_epilogue): Emit return jump via return_use expander.
480 (s390_reorg): Handle JUMP_INSNs as execute targets.
481 (s390_option_override_internal): Perform validity checks for the
482 new command line options.
483 (s390_indirect_branch_attrvalue): New function.
484 (s390_indirect_branch_settings): New function.
485 (s390_set_current_function): Invoke s390_indirect_branch_settings.
486 (s390_output_indirect_thunk_function): New function.
487 (s390_code_end): Implement target hook.
488 (s390_case_values_threshold): Implement target hook.
489 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
491 * config/s390/s390.h (struct s390_frame_layout)
492 (struct machine_function): Move here from s390.c.
493 (TARGET_INDIRECT_BRANCH_NOBP_RET)
494 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
495 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
496 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
497 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
498 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
499 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
500 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
501 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
502 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
503 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
504 (mnemonic attribute): Add values which aren't recognized
506 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
507 pattern for branch conversion. Fix mnemonic attribute.
508 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
509 indirect branch via thunk if requested.
510 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
511 ("*indirect_jump"): Disable for branch conversion using out of
513 ("indirect_jump_via_thunk<mode>_z10")
514 ("indirect_jump_via_thunk<mode>")
515 ("indirect_jump_via_inlinethunk<mode>_z10")
516 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
517 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
518 ("casesi_jump_via_inlinethunk<mode>_z10")
519 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
520 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
521 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
522 ("*indirect2_jump"): Disable for branch conversion.
523 ("casesi_jump"): Turn into expander and expand patterns for branch
525 ("return_use"): New expander.
526 ("*return"): Emit return via thunk and rename it to ...
527 ("*return<mode>"): ... this one.
528 * config/s390/s390.opt: Add new options and and enum for the
531 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
533 * lra-constraints.c (match_reload): Unconditionally use
534 gen_lowpart_SUBREG, rather than selecting between that
535 and equivalent gen_rtx_SUBREG code.
537 2018-02-08 Richard Biener <rguenther@suse.de>
539 PR tree-optimization/84233
540 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
541 changed flag instead of boguously re-using phi_inserted.
543 2018-02-08 Martin Jambor <mjambor@suse.cz>
545 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
546 static local variables.
548 2018-02-08 Richard Biener <rguenther@suse.de>
550 PR tree-optimization/84278
551 * tree-vect-stmts.c (vectorizable_store): When looking for
552 smaller vector types to perform grouped strided loads/stores
553 make sure the mode is supported by the target.
554 (vectorizable_load): Likewise.
556 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
558 * config/aarch64/aarch64.c (aarch64_components_for_bb):
559 Increase LDP/STP opportunities by adding adjacent callee-saves.
561 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
563 PR rtl-optimization/84068
564 PR rtl-optimization/83459
565 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
567 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
569 PR tree-optimization/84224
570 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
571 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
574 2018-02-07 Iain Sandoe <iain@codesourcery.com>
577 * config/rs6000/altivec.md (*restore_world): Remove LR use.
578 * config/rs6000/predicates.md (restore_world_operation): Adjust op
579 count, remove one USE.
581 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
583 * doc/install.texi (Configuration): Document the
584 --with-long-double-format={ibm,ieee} PowerPC configuration
588 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
589 Convert from define_expand to be define_insn_and_split. Rework
590 float/double/_Float128 conversions to QI/HI/SImode to work with
591 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
592 conversions to QI/HImode types did a store and then a load to
593 truncate the value. For conversions to VSX registers, don't split
594 the insn, instead emit the code directly. Use the code iterator
595 any_fix to combine signed and unsigned conversions.
596 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
597 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
598 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
599 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
600 (fix_<mode>di2_hw): Likewise.
601 (fixuns_<mode>di2_hw): Likewise.
602 (fix_<mode>si2_hw): Likewise.
603 (fixuns_<mode>si2_hw): Likewise.
604 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
605 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
606 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
607 fix<uns>_trunc<SFDF:mode>si2_p8.
608 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
610 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
611 (fix<uns>_<mode>_mem): Likewise.
612 (fctiw<u>z_<mode>_mem): Likewise.
613 (fix<uns>_<mode>_mem): Likewise.
614 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
615 the register allocator from doing a direct move to the GPRs to do
616 a store, and instead use the ISA 3.0 store byte/half-word from
617 vector register instruction. For IEEE 128-bit floating point,
618 also optimize stores of 32-bit ints.
619 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
621 2018-02-07 Alan Hayward <alan.hayward@arm.com>
623 * genextract.c (push_pathstr_operand): New function to support
625 (walk_rtx): Call push_pathstr_operand.
626 (print_path): Support [a-zA-Z].
628 2018-02-07 Richard Biener <rguenther@suse.de>
630 PR tree-optimization/84037
631 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
632 (cse_and_gimplify_to_preheader): Declare.
633 (vect_get_place_in_interleaving_chain): Likewise.
634 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
636 (_loop_vec_info::~_loop_vec_info): Delete it.
637 (cse_and_gimplify_to_preheader): New function.
638 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
639 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
640 (vectorizable_load): Likewise. For grouped stores always base
641 the IV on the first element.
642 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
643 condition before gimplifying.
645 2018-02-07 Jakub Jelinek <jakub@redhat.com>
647 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
648 *DIV_EXPR and *MOD_EXPR.
650 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
653 * config/i386/i386.c (ix86_option_override_internal): Mask out
654 the CF_SET bit when checking -fcf-protection.
656 2018-02-07 Tom de Vries <tom@codesourcery.com>
659 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
662 2018-02-07 Richard Biener <rguenther@suse.de>
664 PR tree-optimization/84204
665 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
668 PR tree-optimization/84205
669 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
670 special-case isl_ast_op_zdiv_r.
672 PR tree-optimization/84223
673 * graphite-scop-detection.c (gather_bbs::before_dom_children):
674 Only add conditions from within the region.
675 (gather_bbs::after_dom_children): Adjust.
677 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
680 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
681 * config/avr/avr.md: Only post-reload split REG-REG moves if
682 either register is REGERAL_REG_P.
684 2018-02-07 Jakub Jelinek <jakub@redhat.com>
686 PR tree-optimization/84235
687 * tree-ssa-scopedtables.c
688 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
689 if the subtraction is performed in floating point type where NaNs are
690 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
691 build 1. Formatting fix.
693 2018-02-06 Jakub Jelinek <jakub@redhat.com>
696 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
697 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
698 and skip it regardless of bb boundaries. Use CALL_P macro,
699 don't test INSN_P (insn) together with CALL_P or JUMP_P check
700 unnecessarily, formatting fix.
702 2018-02-06 Michael Collison <michael.collison@arm.com>
704 * config/arm/thumb2.md:
705 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
706 (*thumb_mov_notscc): Ditto.
708 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
711 * config/rs6000/rs6000.md (su code attribute): Use "u" for
712 unsigned_fix, not "s".
714 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
716 * configure.ac (gcc_fn_eh_frame_ro): New function.
717 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
718 correct .eh_frame permissions.
719 * configure: Regenerate.
721 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
723 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
726 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
728 * config/rs6000/rs6000.c (rs6000_option_override_internal):
729 Display warning message for -mno-speculate-indirect-jumps.
731 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
733 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
735 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
737 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
739 PR tree-optimization/84225
740 * tree-eh.c (find_trapping_overflow): Only call
741 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
743 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
746 * config/i386/i386.c: Reimplement the check of possible options
747 -mibt/-mshstk conbination. Change error messages.
748 * doc/invoke.texi: Fix a typo: remove extra '='.
750 2018-02-06 Marek Polacek <polacek@redhat.com>
752 PR tree-optimization/84228
753 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
755 2018-02-06 Tamar Christina <tamar.christina@arm.com>
758 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
759 emitted arch directives.
760 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
761 __ARM_FEATURE_COPROC before changing architectures.
763 2018-02-06 Richard Biener <rguenther@suse.de>
765 * config/i386/i386.c (print_reg): Fix typo.
766 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
768 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
770 * configure: Regenerate.
772 2018-02-05 Martin Sebor <msebor@redhat.com>
774 PR tree-optimization/83369
775 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
778 2018-02-05 Martin Liska <mliska@suse.cz>
780 * doc/invoke.texi: Cherry-pick upstream r323995.
782 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
784 * ira.c (ira_init_register_move_cost): Adjust comment.
786 2018-02-05 Martin Liska <mliska@suse.cz>
788 PR gcov-profile/84137
789 * doc/gcov.texi: Fix typo in documentation.
791 2018-02-05 Martin Liska <mliska@suse.cz>
793 PR gcov-profile/83879
794 * doc/gcov.texi: Document necessity of --dynamic-list-data when
795 using dlopen functionality.
797 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
799 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
800 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
801 _mm_maskz_range_ss, _mm_mask_range_round_ss,
802 _mm_maskz_range_round_ss): New intrinsics.
803 (__builtin_ia32_rangesd128_round)
804 (__builtin_ia32_rangess128_round): Remove.
805 (__builtin_ia32_rangesd128_mask_round,
806 __builtin_ia32_rangess128_mask_round): New builtins.
807 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
808 __builtin_ia32_rangess128_round): Remove.
809 (__builtin_ia32_rangesd128_mask_round,
810 __builtin_ia32_rangess128_mask_round): New builtins.
811 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
812 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
813 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
814 "<round_saeonly_constraint>")): Changed to ...
815 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
816 "<round_saeonly_scalar_constraint>")): ... this.
817 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
818 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
819 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
820 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
821 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
823 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
825 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
827 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
828 Remove all values except native, 8540 and 8548.
830 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
832 * config/i386/i386.c (ix86_output_function_return): Pass
833 INVALID_REGNUM, instead of -1, as invalid register number to
834 indirect_thunk_name and output_indirect_thunk.
836 2018-02-02 Julia Koval <julia.koval@intel.com>
838 * config.gcc: Add -march=icelake.
839 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
840 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
841 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
842 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
843 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
844 (processor_target_table): Add icelake.
845 (ix86_option_override_internal): Handle new PTAs.
846 (get_builtin_code_for_version): Handle icelake.
847 (M_INTEL_COREI7_ICELAKE): New.
848 (fold_builtin_cpu): Handle icelake.
849 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
850 * doc/invoke.texi: Add -march=icelake.
852 2018-02-02 Julia Koval <julia.koval@intel.com>
854 * config/i386/i386.c (ix86_option_override_internal): Change flags type
856 * wide-int-bitmask.h: New.
858 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
861 * config/i386/i386.md: Replace Pmode with word_mode in
862 builtin_setjmp_setup and builtin_longjmp to support x32.
864 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
868 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
870 (rs6000_supported_cpu_names): New static variable.
871 (linux_cpu_translation_table): Likewise.
872 (elf_platform) <cpu>: Define new static variable and use it.
873 Translate kernel AT_PLATFORM name to canonical name if needed.
874 Error if platform name is unknown.
876 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
879 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
881 2018-02-01 Jeff Law <law@redhat.com>
884 * config/i386/i386.c (release_scratch_register_on_entry): Add new
885 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
886 the scratch if RELEASE_VIA_POP is false.
887 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
888 If we have to save a temporary register, decrement SIZE appropriately.
889 Pass new arguments to release_scratch_register_on_entry.
890 (ix86_adjust_stack_and_probe): Likewise.
891 (ix86_emit_probe_stack_range): Pass new arguments to
892 release_scratch_register_on_entry.
894 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
896 PR rtl-optimization/84157
897 * combine.c (change_zero_ext): Use REG_P predicate in
898 front of HARD_REGISTER_P predicate.
900 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
902 * config/avr/avr.c (avr_option_override): Move disabling of
903 -fdelete-null-pointer-checks to...
904 * common/config/avr/avr-common.c (avr_option_optimization_table):
907 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
909 PR tree-optimization/81635
910 * tree-data-ref.c (split_constant_offset_1): For types that
911 wrap on overflow, try to use range info to prove that wrapping
914 2018-02-01 Renlin Li <renlin.li@arm.com>
917 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
919 (aarch64_register_move_cost): Likewise.
920 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
922 (REG_CLASS_NAMES): Likewise.
923 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
924 TAILCALL_ADDR_REGS. Remove IP registers.
925 * config/aarch64/aarch64.md (Ucs): Update register constraint.
927 2018-02-01 Richard Biener <rguenther@suse.de>
929 * domwalk.h (dom_walker::dom_walker): Add additional constructor
930 for specifying RPO order and allow NULL for that.
931 * domwalk.c (dom_walker::dom_walker): Likewise.
932 (dom_walker::walk): Handle NULL RPO order.
933 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
935 (rewrite_update_dom_walker): Likewise.
936 (mark_def_dom_walker): Likewise.
938 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
940 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
941 (aarch64_maybe_expand_sve_subreg_move): Declare.
942 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
943 * config/aarch64/predicates.md (aarch64_any_register_operand): New
945 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
946 that are semantically a reverse operation.
947 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
948 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
949 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
951 (aarch64_can_change_mode_class): For big-endian, forbid changes
952 between two SVE modes if they have different element sizes.
954 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
956 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
957 the TImode handling for big-endian targets.
959 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
961 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
962 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
964 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
965 Remove BSWAP handing for big-endian targets and use the form of
966 LD1RQ appropariate for the mode.
968 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
970 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
971 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
974 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
977 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
978 check for operands that need to go through aarch64_sve_reload_be.
980 2018-02-01 Jakub Jelinek <jakub@redhat.com>
982 PR tree-optimization/81661
983 PR tree-optimization/84117
984 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
985 * tree-eh.c: Include gimplify.h.
986 (find_trapping_overflow, replace_trapping_overflow,
987 rewrite_to_non_trapping_overflow): New functions.
988 * tree-vect-loop.c: Include tree-eh.h.
989 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
990 * tree-data-ref.c: Include tree-eh.h.
991 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
993 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
995 PR rtl-optimization/84123
996 * combine.c (change_zero_ext): Check if hard register satisfies
997 can_change_dest_mode before calling gen_lowpart_SUBREG.
999 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1002 * ira.c (ira_init_register_move_cost): Remove assert.
1004 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1006 PR rtl-optimization/84071
1007 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1008 * doc/tm.texi: Regenerate.
1010 2018-01-31 Richard Biener <rguenther@suse.de>
1012 PR tree-optimization/84132
1013 * tree-data-ref.c (analyze_miv_subscript): Properly
1014 check whether evolution_function_is_affine_multivariate_p
1015 before calling gcd_of_steps_may_divide_p.
1017 2018-01-31 Julia Koval <julia.koval@intel.com>
1020 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1021 * config/i386/i386.md (rdpid_rex64) New.
1022 (rdpid): Make 32bit only.
1024 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1027 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1028 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1030 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1033 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1035 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1037 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1039 PR rtl-optimization/84071
1040 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1041 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1043 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1045 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1046 (arc_attribute_table): Add 'aux' attribute.
1047 (arc_in_small_data_p): Consider aux like variables.
1048 (arc_is_aux_reg_p): New function.
1049 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1050 (arc_get_aux_arg): New function.
1051 (prepare_move_operands): Handle aux-register access.
1052 (arc_handle_aux_attribute): New function.
1053 * doc/extend.texi (ARC Variable attributes): Add subsection.
1055 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1057 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1058 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1059 (arc_attribute_table): Add 'uncached' attribute.
1060 (arc_print_operand): Print '.di' flag for uncached memory
1062 (arc_in_small_data_p): Do not consider for small data the uncached
1064 (arc_is_uncached_mem_p): New function.
1065 * config/arc/predicates.md (compact_store_memory_operand): Check
1066 for uncached memory accesses.
1067 (nonvol_nonimm_operand): Likewise.
1068 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1070 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1073 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1074 falign-loops=): Add Optimization flag.
1076 2018-01-30 Jeff Law <law@redhat.com>
1079 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1080 INT_REGISTERS_SAVED. Check it prior to calling
1081 get_scratch_register_on_entry.
1082 (ix86_adjust_stack_and_probe): Similarly.
1083 (ix86_emit_probe_stack_range): Similarly.
1084 (ix86_expand_prologue): Corresponding changes.
1086 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1089 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1090 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1092 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1095 * lra-constraints.c (curr_insn_transform): Process AND in the
1098 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1100 PR rtl-optimization/83986
1101 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1102 dependence against last_pending_memory_flush in addition to
1105 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1107 PR tree-optimization/81611
1108 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1111 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1113 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1116 2018-01-30 Richard Biener <rguenther@suse.de>
1117 Jakub Jelinek <jakub@redhat.com>
1119 PR tree-optimization/84111
1120 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1121 inner loops added during recursion, as they don't have up-to-date
1124 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1127 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1128 (can_inline_edge_by_limits_p): ... here.
1129 (can_early_inline_edge_p, check_callers,
1130 update_caller_keys, update_callee_keys, recursive_inlining,
1131 add_new_edges_to_heap, speculation_useful_p,
1132 inline_small_functions,
1133 inline_small_functions, flatten_function,
1134 inline_to_all_callers_1): Update.
1136 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1138 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1141 2018-01-30 Richard Biener <rguenther@suse.de>
1143 PR tree-optimization/83008
1144 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1145 invariant and constant vector uses in stmts when they need
1148 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1151 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1152 * configure: Regenerate.
1154 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1156 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1158 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1159 Use gen_rtx_REG rather than gen_lowpart.
1161 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1163 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1164 rather than 0 when creating partial subregs.
1166 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1168 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1171 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1174 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1175 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1176 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1177 flags. This restores the settings used before the 2017-07-24.
1178 Turning off pre increment/decrement/modify allows IVOPTS to
1179 optimize DF/SF loops where the index is an int.
1181 2018-01-29 Richard Biener <rguenther@suse.de>
1182 Kelvin Nilsen <kelvin@gcc.gnu.org>
1185 * tree-vect-stmts.c (vectorizable_call): Don't call
1186 targetm.vectorize_builtin_md_vectorized_function if callee is
1189 2018-01-22 Carl Love <cel@us.ibm.com>
1191 * doc/extend.tex: Fix typo in second arg in
1192 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1194 2018-01-29 Richard Biener <rguenther@suse.de>
1196 PR tree-optimization/84086
1197 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1198 (flush_ssaname_freelist): When SSA names were released reset
1199 the SCEV hash table.
1201 2018-01-29 Richard Biener <rguenther@suse.de>
1203 PR tree-optimization/84057
1204 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1205 removed paths when removing edges.
1207 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1209 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1210 -mfunction-return=@var{choice}.
1212 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1215 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1217 (layout::print_source_line): Likewise.
1218 (test_get_line_width_without_trailing_whitespace): Add test cases.
1220 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1223 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1226 2018-01-26 Jim Wilson <jimw@sifive.com>
1228 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1230 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1233 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1235 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1236 and CMP + SUB-immediate -> SUBS.
1238 2018-01-26 Martin Sebor <msebor@redhat.com>
1240 PR tree-optimization/83896
1241 * tree-ssa-strlen.c (get_string_len): Rename...
1242 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1243 Avoid assuming length is constant.
1244 (handle_char_store): Use HOST_WIDE_INT for string length.
1246 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1249 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1250 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1252 2018-01-26 Richard Biener <rguenther@suse.de>
1254 PR rtl-optimization/84003
1255 * dse.c (record_store): Only record redundant stores when
1256 the earlier store aliases at least all accesses the later one does.
1258 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1260 PR rtl-optimization/83985
1261 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1262 REG_CFA_RESTORE insns.
1263 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1264 REG_CFA_RESTORE insns here.
1267 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1268 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1270 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1272 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1273 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1274 (arc_init): Likewise.
1275 (arc_override_options): Likewise.
1276 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1278 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1280 * config/arc/arc.h (TARGET_DBNZ): Define.
1281 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1282 properly set the tune attribute.
1283 (dbnz): Use TARGET_DBNZ guard.
1284 * config/arc/arc.opt (mtune): Add core3 option.
1286 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1288 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1289 recognize new pic like addresses.
1290 (arc_delegitimize_address): Clean up.
1292 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1294 * config/arc/arc-arches.def: Option mrf16 valid for all
1296 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1297 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1298 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1299 * config/arc/arc-tables.opt: Regenerate.
1300 * config/arc/arc.c (arc_conditional_register_usage): Handle
1301 reduced register file case.
1302 (arc_file_start): Set must have build attributes.
1303 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1305 * config/arc/arc.opt (mrf16): Add new option.
1306 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1307 * config/arc/genmultilib.awk: Handle new mrf16 option.
1308 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1309 * config/arc/t-multilib: Regenerate.
1310 * doc/invoke.texi (ARC Options): Document mrf16 option.
1312 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1314 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1315 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1316 (arc_attribute_table): Add 'secure_call' attribute.
1317 (arc_print_operand): Print secure call operand.
1318 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1320 (arc_is_secure_call_p): New function. * config/arc/arc.md
1321 (call_i): Add support for sjli instruction.
1322 (call_value_i): Likewise.
1323 * config/arc/constraints.md (Csc): New constraint.
1325 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1326 John Eric Martin <John.Martin@emmicro-us.com>
1328 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1329 * config/arc/arc.c (_arc_jli_section): New struct.
1330 (arc_jli_section): New type.
1331 (rc_jli_sections): New static variable.
1332 (arc_handle_jli_attribute): New function.
1333 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1334 (arc_file_end): New function.
1335 (TARGET_ASM_FILE_END): Define.
1336 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1337 (arc_add_jli_section): New function.
1338 (jli_call_scan): Likewise.
1339 (arc_reorg): Call jli_call_scan.
1340 (arc_output_addsi): Remove 'S' from printing asm operand.
1341 (arc_is_jli_call_p): New function.
1342 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1344 (movhi_insn): Likewise.
1345 (movsi_insn): Likewise.
1346 (movsi_set_cc_insn): Likewise.
1347 (loadqi_update): Likewise.
1348 (load_zeroextendqisi_update): Likewise.
1349 (load_signextendqisi_update): Likewise.
1350 (loadhi_update): Likewise.
1351 (load_zeroextendhisi_update): Likewise.
1352 (load_signextendhisi_update): Likewise.
1353 (loadsi_update): Likewise.
1354 (loadsf_update): Likewise.
1355 (movsicc_insn): Likewise.
1356 (bset_insn): Likewise.
1357 (bxor_insn): Likewise.
1358 (bclr_insn): Likewise.
1359 (bmsk_insn): Likewise.
1360 (bicsi3_insn): Likewise.
1361 (cmpsi_cc_c_insn): Likewise.
1362 (movsi_ne): Likewise.
1363 (movsi_cond_exec): Likewise.
1364 (clrsbsi2): Likewise.
1375 (call_i): Remove 'S' asm letter, add jli instruction.
1376 (call_value_i): Likewise.
1377 * config/arc/arc.op (mjli-always): New option.
1378 * config/arc/constraints.md (Cji): New constraint.
1379 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1381 (subsf3_fpx): Likewise.
1382 (mulsf3_fpx): Likewise.
1383 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1385 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1387 * doc/invoke.texi (ARC): Document mjli-always option.
1389 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1391 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1392 avoid addition with 0 and use incw and decw where possible.
1394 2018-01-26 Richard Biener <rguenther@suse.de>
1396 PR tree-optimization/81082
1397 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1398 association if it requires casting to unsigned.
1399 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1400 from fold_plusminus_mult_expr to catch important cases late when
1401 range info is available.
1403 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1405 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1406 * configure.ac (hidden_linkonce): New test.
1407 * configure: Regenerate.
1408 * config.in: Regenerate.
1410 2018-01-26 Julia Koval <julia.koval@intel.com>
1412 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1413 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1414 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1415 _mm_mask_bitshuffle_epi64_mask): Fix type.
1416 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1417 USI_FTYPE_V4DI_V4DI_USI): Remove.
1418 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1419 __builtin_ia32_vpshufbitqmb256_mask,
1420 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1421 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1422 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1424 2018-01-26 Alan Modra <amodra@gmail.com>
1427 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1428 UNSPEC_VBPERMQ. Sort other unspecs.
1430 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1432 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1434 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1437 * predict.c (drop_profile): Do not push/pop cfun; update also
1439 (handle_missing_profiles): Fix logic looking for zero profiles.
1441 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1444 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1445 on functions with #pragma omp declare simd or functions with simd
1447 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1448 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1449 Remove trailing \n from warning_at calls.
1451 2018-01-25 Tom de Vries <tom@codesourcery.com>
1454 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1455 for neutered workers.
1457 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1460 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1461 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1463 2018-01-24 Jeff Law <law@redhat.com>
1466 * i386.c (get_probe_interval): Move to earlier point.
1467 (ix86_compute_frame_layout): If -fstack-clash-protection and
1468 the frame is larger than the probe interval, then use pushes
1469 to save registers rather than reg->mem moves.
1470 (ix86_expand_prologue): Remove conditional for int_registers_saved
1473 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1476 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1477 min/max for never referenced object.
1479 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1482 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1484 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1485 attributes from DECL_ATTRIBUTES (decl) without affecting
1486 DECL_ATTRIBUTES (current_function_decl).
1487 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1488 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1490 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1492 PR tree-optimization/83979
1493 * fold-const.c (fold_comparison): Use constant_boolean_node
1494 instead of boolean_{true,false}_node.
1496 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1498 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1501 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1503 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1504 Simplify the clause that sets the length attribute.
1505 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1506 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1507 clause that sets the length attribute.
1508 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1510 2018-01-24 Tom de Vries <tom@codesourcery.com>
1513 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1514 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1515 Add strict parameter.
1516 (prevent_branch_around_nothing): Insert dummy insn between branch to
1517 label and label with no ptx insn inbetween.
1518 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1520 2018-01-24 Tom de Vries <tom@codesourcery.com>
1523 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1524 for neutered threads in warp.
1525 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1527 2018-01-24 Richard Biener <rguenther@suse.de>
1529 PR tree-optimization/83176
1530 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1533 2018-01-24 Richard Biener <rguenther@suse.de>
1535 PR tree-optimization/82819
1536 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1537 code generating pluses that are no-ops in the target precision.
1539 2018-01-24 Richard Biener <rguenther@suse.de>
1542 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1544 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1546 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1547 to merge probabilities.
1548 * predict.c (probably_never_executed): Also mark as cold functions
1549 with global 0 profile and guessed local profile.
1550 * profile-count.c (profile_probability::combine_with_count): New
1552 * profile-count.h (profile_probability::operator*,
1553 profile_probability::operator*=, profile_probability::operator/,
1554 profile_probability::operator/=): Reduce precision to adjusted
1555 and set value to guessed on contradictory divisions.
1556 (profile_probability::combine_with_freq): Remove.
1557 (profile_probability::combine_wiht_count): Declare.
1558 (profile_count::force_nonzero):: Set to adjusted.
1559 (profile_count::probability_in):: Set quality to adjusted.
1560 * tree-ssa-tail-merge.c (replace_block_by): Use
1563 2018-01-23 Andrew Waterman <andrew@sifive.com>
1564 Jim Wilson <jimw@sifive.com>
1566 * config/riscv/riscv.c (riscv_stack_boundary): New.
1567 (riscv_option_override): Set riscv_stack_boundary. Handle
1568 riscv_preferred_stack_boundary_arg.
1569 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1570 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1571 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1572 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1573 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1574 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1576 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1579 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1580 of struct ix86_frame.
1581 (ix86_expand_epilogue): Likewise. Add a local variable for
1582 the reg_save_offset field in struct ix86_frame.
1584 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1586 PR tree-optimization/82604
1587 * tree-loop-distribution.c (enum partition_kind): New enum item
1588 PKIND_PARTIAL_MEMSET.
1589 (partition_builtin_p): Support above new enum item.
1590 (generate_code_for_partition): Ditto.
1591 (compute_access_range): Differentiate cases that equality can be
1592 proven at all loops, the innermost loops or no loops.
1593 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1594 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1595 (finalize_partitions, distribute_loop): Don't fuse partition of
1596 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1597 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1600 2018-01-23 Martin Liska <mliska@suse.cz>
1602 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1603 order to ignore the predictor.
1604 (PRED_POLYMORPHIC_CALL): Likewise.
1605 (PRED_RECURSIVE_CALL): Likewise.
1607 2018-01-23 Martin Liska <mliska@suse.cz>
1609 * tree-profile.c (tree_profiling): Print function header to
1610 aware reader which function we are working on.
1611 * value-prof.c (gimple_find_values_to_profile): Do not print
1612 not interesting value histograms.
1614 2018-01-23 Martin Liska <mliska@suse.cz>
1616 * profile-count.h (enum profile_quality): Add
1617 profile_uninitialized as the first value. Do not number values
1618 as they are zero based.
1619 (profile_count::verify): Update sanity check.
1620 (profile_probability::verify): Likewise.
1622 2018-01-23 Nathan Sidwell <nathan@acm.org>
1624 * doc/invoke.texi (ffor-scope): Deprecate.
1626 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1628 PR tree-optimization/83510
1629 * domwalk.c (set_all_edges_as_executable): New function.
1630 (dom_walker::dom_walker): Convert bool param
1631 "skip_unreachable_blocks" to enum reachability. Move setup of
1632 edge flags to set_all_edges_as_executable and only do it when
1633 reachability is REACHABLE_BLOCKS.
1634 * domwalk.h (enum dom_walker::reachability): New enum.
1635 (dom_walker::dom_walker): Convert bool param
1636 "skip_unreachable_blocks" to enum reachability.
1637 (set_all_edges_as_executable): New decl.
1638 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1639 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1641 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1642 but converting true to REACHABLE_BLOCKS.
1643 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1645 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1646 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1647 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1649 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1650 if check_all_array_refs will be called.
1652 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1654 * tree.c (selftest::test_location_wrappers): Add more test
1657 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1659 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1660 (selftest::test_bit_in_range): Likewise.
1662 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1665 * doc/sourcebuild.texi (vect_float): Say that the selector
1666 only describes the situation when -funsafe-math-optimizations is on.
1667 (vect_float_strict): Document.
1669 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1671 PR tree-optimization/83965
1672 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1673 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1674 instead of checking only for a reduction.
1675 (vect_recog_widen_sum_pattern): Likewise.
1677 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1679 * predict.c (probably_never_executed): Only use precise profile info.
1680 (compute_function_frequency): Skip after inlining hack since we now
1681 have quality checking.
1683 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1685 * profile-count.h (profile_probability::very_unlikely,
1686 profile_probability::unlikely, profile_probability::even): Set
1687 precision to guessed.
1689 2018-01-23 Richard Biener <rguenther@suse.de>
1691 PR tree-optimization/83963
1692 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1693 Properly terminate dominator walk when crossing the exit edge not
1694 when visiting its source block.
1696 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1699 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1700 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1702 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1704 PR tree-optimization/83957
1705 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1706 semicolon after for body surrounded by braces.
1708 PR tree-optimization/83081
1709 * profile-count.h (profile_probability::split): New method.
1710 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1711 Use profile_probability::split.
1712 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1713 when splitting a single conditional jump into 2.
1715 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1717 PR tree-optimization/69452
1718 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1721 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1723 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1724 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1725 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1727 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1729 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1730 * config/rl78/rl78.md (movdi): New define_expand.
1731 * config/rl78/rl78.c (rl78_split_movdi): New function.
1733 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1736 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1738 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1739 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1740 128-bit to produce an UNSPEC move to get the double word with the
1741 signbit and then a shift directly to do signbit.
1742 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1743 implementation with a new version that just does either a direct
1744 move or a regular move. Move memory interface to separate insns.
1745 Move insns so they are next to the expander.
1746 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1747 with signbit move. Split big and little endian case.
1748 (signbit<mode>2_dm_mem_le): Likewise.
1749 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1750 (signbit<mode>2_dm2): Likewise.
1752 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1754 * config/rl78/rl78.md (anddi3): New define_expand.
1756 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1758 * config/rl78/rl78.md (umindi3): New define_expand.
1760 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1762 * config/rl78/rl78.md (smindi3): New define_expand.
1764 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1766 * config/rl78/rl78.md (smaxdi3): New define_expand.
1768 2018-01-22 Carl Love <cel@us.ibm.com>
1770 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1771 LVX_V1TI): Add macro expansion.
1772 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1773 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1774 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1775 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1776 Change check to determine if the instruction is a byte reversing
1777 entry. Fix typo in comment.
1778 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1779 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1780 Add def_builtin calls for new builtins.
1781 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1782 Add define_insn expansion.
1784 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1786 * config/rl78/rl78.md (umaxdi3): New define_expand.
1788 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1790 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
1791 for non-QImode registers.
1793 2018-01-22 Richard Biener <rguenther@suse.de>
1795 PR tree-optimization/83963
1796 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1797 including the loop exit block.
1798 (scop_detection::merge_sese): Likewise.
1799 (scop_detection::add_scop): Do it here instead.
1801 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1803 * doc/sourcebuild.texi (arm_softfloat): Document.
1805 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1808 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1809 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1810 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1812 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1813 David Edelsohn <dje.gcc@gmail.com>
1816 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1817 Change "crset eq" to "crset 2".
1818 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1819 (*call_indirect_aix<mode>_nospec): Likewise.
1820 (*call_value_indirect_aix<mode>_nospec): Likewise.
1821 (*call_indirect_elfv2<mode>_nospec): Likewise.
1822 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1823 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1824 change assembly output from . to $.
1825 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1826 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1827 (*tablejump<mode>_internal1_nospec): Likewise.
1829 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1832 * config/sh/sh_optimize_sett_clrt.cc:
1833 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1835 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1837 PR tree-optimization/83940
1838 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1839 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1840 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1841 use it to pass back the definition type.
1842 (vect_check_store_rhs): Likewise rhs_dt_out.
1843 (vect_build_gather_load_calls): Add a mask_dt argument and use
1844 it instead of a call to vect_is_simple_use.
1845 (vectorizable_store): Update calls to vect_check_load_store_mask
1846 and vect_check_store_rhs. Use the dt returned by the latter instead
1847 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1848 instead of calls to vect_is_simple_use. Pass the scalar rather
1849 than the vector operand to vect_is_simple_use when handling
1850 second and subsequent copies of an rhs value.
1851 (vectorizable_load): Update calls to vect_check_load_store_mask
1852 and vect_build_gather_load_calls. Use the cached mask_dt and
1853 gs_info.offset_dt instead of calls to vect_is_simple_use.
1855 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1858 * tree-emutls.c: Include gimplify.h.
1859 (lower_emutls_2): New function.
1860 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1861 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1862 it before further processing.
1865 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1866 UINTVAL (trueop1) instead of INTVAL (op1).
1868 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1872 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1873 INCOMING_FRAME_SP_OFFSET if not defined.
1874 (scan_trace): Add ENTRY argument. If true and
1875 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1876 emit a note to adjust the CFA offset.
1877 (create_cfi_notes): Adjust scan_trace callers.
1878 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1879 INCOMING_FRAME_SP_OFFSET in the CIE.
1880 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1881 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1883 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1884 * doc/tm.texi: Regenerated.
1886 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1888 PR rtl-optimization/83147
1889 * lra-constraints.c (remove_inheritance_pseudos): Use
1890 lra_substitute_pseudo_within_insn.
1892 2018-01-19 Tom de Vries <tom@codesourcery.com>
1893 Cesar Philippidis <cesar@codesourcery.com>
1896 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1898 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1901 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1902 spaces for function labels.
1904 2018-01-19 Martin Liska <mliska@suse.cz>
1906 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1907 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1908 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1909 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1910 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1911 (PRED_CONST_RETURN): Change from 69 to 65.
1912 (PRED_NULL_RETURN): Change from 91 to 71.
1913 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1914 (PRED_LOOP_GUARD): Change from 66 to 73.
1916 2018-01-19 Martin Liska <mliska@suse.cz>
1918 * predict.c (predict_insn_def): Add new assert.
1919 (struct branch_predictor): Change type to signed integer.
1920 (test_prediction_value_range): Amend test to cover
1922 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1923 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1924 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1925 (PRED_LOOP_IV_COMPARE): Likewise.
1926 * predict.h (PROB_UNINITIALIZED): Define new constant.
1928 2018-01-19 Martin Liska <mliska@suse.cz>
1930 * predict.c (dump_prediction): Add new format for
1931 analyze_brprob.py script which is enabled with -details
1933 * profile-count.h (precise_p): New function.
1935 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1937 PR tree-optimization/83922
1938 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1939 there are no statements that need masking.
1940 (vect_active_double_reduction_p): New function.
1941 (vect_analyze_loop_operations): Use it when handling phis that
1942 are not in the loop header.
1944 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1946 PR tree-optimization/83914
1947 * tree-vect-loop.c (vectorizable_induction): Don't convert
1948 init_expr or apply the peeling adjustment for inductions
1949 that are nested within the vectorized loop.
1951 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1953 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1956 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1960 * function.h (gimplify_parameters): Add gimple_seq * argument.
1961 * function.c: Include gimple.h and options.h.
1962 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1963 for the added local temporaries if needed.
1964 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1965 if there are any parameter cleanups, wrap whole body into a
1966 try/finally with the cleanups.
1968 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1971 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1972 Use GET_MODE_CLASS for scalar floating point.
1974 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1978 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1979 Fix call of call_cgraph_insertion_hooks.
1981 2018-01-18 Martin Sebor <msebor@redhat.com>
1983 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1985 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1988 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1991 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1994 * common.opt: (-ffile-prefix-map): New option.
1995 * opts.c (common_handle_option): Defer it.
1996 * opts-global.c (handle_common_deferred_options): Handle it.
1997 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1998 * file-prefix-map.h: New file.
1999 (remap_debug_filename, add_debug_prefix_map): ...here.
2000 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2001 * final.c (debug_prefix_map, add_debug_prefix_map
2002 remap_debug_filename): Move to...
2003 * file-prefix-map.c: New file.
2004 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2005 generalize, get rid of alloca(), use strrchr() instead of strchr().
2006 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2007 Implement in terms of add_prefix_map().
2008 (remap_macro_filename, remap_debug_filename): Implement in term of
2010 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2011 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2012 * dbxout.c: Include file-prefix-map.h.
2013 * varasm.c: Likewise.
2014 * vmsdbgout.c: Likewise.
2015 * xcoffout.c: Likewise.
2016 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2017 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2018 * doc/invoke.texi (-ffile-prefix-map): Document.
2019 (-fdebug-prefix-map): Update description.
2021 2018-01-18 Martin Liska <mliska@suse.cz>
2023 * config/i386/i386.c (indirect_thunk_name): Document that also
2025 (output_indirect_thunk): Document why both instructions
2026 (pause and lfence) are generated.
2028 2018-01-18 Richard Biener <rguenther@suse.de>
2030 PR tree-optimization/83887
2031 * graphite-scop-detection.c
2032 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2033 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2034 (scop_detection::merge_sese): Re-implement with a flood-fill
2035 algorithm that properly finds a SESE region if it exists.
2037 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2040 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2041 pointer_diff optimizations use view_convert instead of convert.
2043 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2045 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2046 Generate different code for -mno-speculate-indirect-jumps.
2047 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2048 (*call_indirect_aix<mode>): Disable for
2049 -mno-speculate-indirect-jumps.
2050 (*call_indirect_aix<mode>_nospec): New define_insn.
2051 (*call_value_indirect_aix<mode>): Disable for
2052 -mno-speculate-indirect-jumps.
2053 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2054 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2055 -mno-speculate-indirect-jumps.
2056 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2058 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2060 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2061 long double type, set the flags for noting the default long double
2062 type, even if we don't pass or return a long double type.
2064 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2067 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2070 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2072 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2073 support for merge[hl].
2074 (fold_mergehl_helper): New helper function.
2075 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2076 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2077 (altivec_vmrglw_direct): Add xxmrglw insn.
2079 2018-01-17 Andrew Waterman <andrew@sifive.com>
2081 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2082 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2084 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2087 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2088 call the lto_location_cache before reading the
2089 DECL_SOURCE_LOCATION of the types.
2091 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2092 Richard Sandiford <richard.sandiford@linaro.org>
2094 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2095 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2096 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2097 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2098 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2100 * config/aarch64/constraints.md (aarch64_movti_operand):
2102 * config/aarch64/predicates.md (Uti): Add new constraint.
2104 2018-01-17 Carl Love <cel@us.ibm.com>
2105 * config/rs6000/vsx.md (define_expand xl_len_r,
2106 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2107 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2109 (define_expand, define_insn): Move the shift left from the
2110 define_insn to the define_expand for lxvl and stxvl instructions.
2111 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2112 and XL_LEN_R definitions to PURE.
2114 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2116 * config/i386/i386.c (indirect_thunk_name): Declare regno
2117 as unsigned int. Compare regno with INVALID_REGNUM.
2118 (output_indirect_thunk): Ditto.
2119 (output_indirect_thunk_function): Ditto.
2120 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2121 in the call to output_indirect_thunk_function.
2123 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2126 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2127 rather than the size of inner_type to determine the stack slot size
2128 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2130 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2133 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2136 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2138 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2139 endian Linux systems to optionally enable multilibs for selecting
2140 the long double type if the user configured an explicit type.
2141 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2142 have no long double multilibs if not defined.
2143 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2144 warn if the user used -mabi={ieee,ibm}longdouble and we built
2145 multilibs for long double.
2146 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2147 appropriate multilib option.
2148 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2150 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2151 for building long double multilibs.
2152 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2154 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2156 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2159 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2161 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2164 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2167 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2170 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2172 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2173 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2175 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2177 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2178 different rtl trees depending on TARGET_64BIT.
2179 (rs6000_gen_lvx): Likewise.
2181 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2183 * config/visium/visium.md (nop): Tweak comment.
2184 (hazard_nop): Likewise.
2186 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2188 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2189 -mspeculate-indirect-jumps.
2190 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2191 for -mno-speculate-indirect-jumps.
2192 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2193 (*call_value_indirect_elfv2<mode>): Disable for
2194 -mno-speculate-indirect-jumps.
2195 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2196 (indirect_jump): Emit different RTL for
2197 -mno-speculate-indirect-jumps.
2198 (*indirect_jump<mode>): Disable for
2199 -mno-speculate-indirect-jumps.
2200 (*indirect_jump<mode>_nospec): New define_insn.
2201 (tablejump): Emit different RTL for
2202 -mno-speculate-indirect-jumps.
2203 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2204 (tablejumpsi_nospec): New define_expand.
2205 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2206 (tablejumpdi_nospec): New define_expand.
2207 (*tablejump<mode>_internal1): Disable for
2208 -mno-speculate-indirect-jumps.
2209 (*tablejump<mode>_internal1_nospec): New define_insn.
2210 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2213 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2215 * caller-save.c (insert_save): Drop unnecessary parameter. All
2218 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2219 Richard Biener <rguenth@suse.de>
2222 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2223 return early, inline manually is_gimple_sizepos. Make sure if we
2224 call gimplify_expr we don't end up with a gimple constant.
2225 * tree.c (variably_modified_type_p): Don't return true for
2226 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2227 * gimplify.h (is_gimple_sizepos): Remove.
2229 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2231 PR tree-optimization/83857
2232 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2233 vectorizable_live_operation for pure SLP statements.
2234 (vectorizable_live_operation): Handle PHIs.
2236 2018-01-16 Richard Biener <rguenther@suse.de>
2238 PR tree-optimization/83867
2239 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2240 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2242 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2245 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2246 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2247 If off is not INTEGER_CST, issue a may not be aligned warning
2248 rather than isn't aligned. Use isn%'t rather than isn't.
2249 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2251 <case MULT_EXPR>: Improve the case when bottom and one of the
2252 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2253 operand, in that case check if the other operand is multiple of
2254 bottom divided by the INTEGER_CST operand.
2256 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2259 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2260 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2261 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2262 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2263 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2264 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2265 (pa_function_arg_size): New function.
2267 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2269 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2270 in a separate statement.
2272 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2274 PR tree-optimization/83847
2275 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2276 group gathers and scatters.
2278 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2280 PR rtl-optimization/86620
2281 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2283 PR rtl-optimization/83213
2284 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2285 to last if both are JUMP_INSNs.
2287 PR tree-optimization/83843
2288 * gimple-ssa-store-merging.c
2289 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2290 store_immediate_info for bswap/nop orig_stores.
2292 2018-01-15 Andrew Waterman <andrew@sifive.com>
2294 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2296 <UDIV>: Increase cost if !TARGET_DIV.
2298 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2300 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2301 (define_attr "cr_logical_3op"): New.
2302 (cceq_ior_compare): Adjust.
2303 (cceq_ior_compare_complement): Adjust.
2304 (*cceq_rev_compare): Adjust.
2305 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2306 (is_cracked_insn): Adjust.
2307 (insn_must_be_first_in_group): Adjust.
2308 * config/rs6000/40x.md: Adjust.
2309 * config/rs6000/440.md: Adjust.
2310 * config/rs6000/476.md: Adjust.
2311 * config/rs6000/601.md: Adjust.
2312 * config/rs6000/603.md: Adjust.
2313 * config/rs6000/6xx.md: Adjust.
2314 * config/rs6000/7450.md: Adjust.
2315 * config/rs6000/7xx.md: Adjust.
2316 * config/rs6000/8540.md: Adjust.
2317 * config/rs6000/cell.md: Adjust.
2318 * config/rs6000/e300c2c3.md: Adjust.
2319 * config/rs6000/e500mc.md: Adjust.
2320 * config/rs6000/e500mc64.md: Adjust.
2321 * config/rs6000/e5500.md: Adjust.
2322 * config/rs6000/e6500.md: Adjust.
2323 * config/rs6000/mpc.md: Adjust.
2324 * config/rs6000/power4.md: Adjust.
2325 * config/rs6000/power5.md: Adjust.
2326 * config/rs6000/power6.md: Adjust.
2327 * config/rs6000/power7.md: Adjust.
2328 * config/rs6000/power8.md: Adjust.
2329 * config/rs6000/power9.md: Adjust.
2330 * config/rs6000/rs64.md: Adjust.
2331 * config/rs6000/titan.md: Adjust.
2333 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2335 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2336 ix86_indirect_branch_register logic.
2338 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2340 * config/i386/constraints.md (Bs): Update
2341 ix86_indirect_branch_register check. Don't check
2342 ix86_indirect_branch_register with GOT_memory_operand.
2344 * config/i386/predicates.md (GOT_memory_operand): Don't check
2345 ix86_indirect_branch_register here.
2346 (GOT32_symbol_operand): Likewise.
2348 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2350 * config/i386/predicates.md (constant_call_address_operand):
2351 Rewrite ix86_indirect_branch_register logic.
2352 (sibcall_insn_operand): Likewise.
2354 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2356 * config/i386/constraints.md (Bs): Replace
2357 ix86_indirect_branch_thunk_register with
2358 ix86_indirect_branch_register.
2360 * config/i386/i386.md (indirect_jump): Likewise.
2361 (tablejump): Likewise.
2362 (*sibcall_memory): Likewise.
2363 (*sibcall_value_memory): Likewise.
2364 Peepholes of indirect call and jump via memory: Likewise.
2365 * config/i386/i386.opt: Likewise.
2366 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2367 (GOT_memory_operand): Likewise.
2368 (call_insn_operand): Likewise.
2369 (sibcall_insn_operand): Likewise.
2370 (GOT32_symbol_operand): Likewise.
2372 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2375 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2376 type rather than type addr's type points to.
2377 (expand_omp_atomic_mutex): Likewise.
2378 (expand_omp_atomic): Likewise.
2380 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2383 * config/i386/i386.c (output_indirect_thunk_function): Use
2384 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2385 for __x86_return_thunk.
2387 2018-01-15 Richard Biener <rguenther@suse.de>
2390 * expmed.c (extract_bit_field_1): Fix typo.
2392 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2395 * config/arm/iterators.md (VF): New mode iterator.
2396 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2397 Remove integer-related logic from pattern.
2398 (neon_vabd<mode>_3): Likewise.
2400 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2403 * common.opt (fstrict-overflow): No longer an alias.
2404 (fwrapv-pointer): New option.
2405 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2406 also for pointer types based on flag_wrapv_pointer.
2407 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2408 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2409 opts->x_flag_wrapv got set.
2410 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2411 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2412 POINTER_TYPE_OVERFLOW_UNDEFINED.
2413 * match.pd: Likewise in address comparison pattern.
2414 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2416 2018-01-15 Richard Biener <rguenther@suse.de>
2419 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2420 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2421 Reset type names to their identifier if their TYPE_DECL doesn't
2422 have linkage (and thus is used for ODR and devirt).
2423 (save_debug_info_for_decl): Remove.
2424 (save_debug_info_for_type): Likewise.
2425 (add_tree_to_fld_list): Adjust.
2426 * tree-pretty-print.c (dump_generic_node): Make dumping of
2427 type names more robust.
2429 2018-01-15 Richard Biener <rguenther@suse.de>
2431 * BASE-VER: Bump to 8.0.1.
2433 2018-01-14 Martin Sebor <msebor@redhat.com>
2436 * builtins.c (check_access): Avoid warning when the no-warning bit
2439 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2441 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2442 * ira-color (allocno_hard_regs_compare): Likewise.
2444 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2447 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2448 Use .pushsection/.popsection.
2450 2018-01-14 Martin Sebor <msebor@redhat.com>
2453 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2455 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2457 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2458 entry from extra_headers.
2459 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2460 extra_headers, make the list bitwise identical to the i?86-*-* one.
2462 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2464 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2465 -mcmodel=large with -mindirect-branch=thunk,
2466 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2467 -mfunction-return=thunk-extern.
2468 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2469 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2470 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2472 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2474 * config/i386/i386.c (print_reg): Print the name of the full
2475 integer register without '%'.
2476 (ix86_print_operand): Handle 'V'.
2477 * doc/extend.texi: Document 'V' modifier.
2479 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2481 * config/i386/constraints.md (Bs): Disallow memory operand for
2482 -mindirect-branch-register.
2484 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2485 (GOT_memory_operand): Likewise.
2486 (call_insn_operand): Likewise.
2487 (sibcall_insn_operand): Likewise.
2488 (GOT32_symbol_operand): Likewise.
2489 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2490 for -mindirect-branch-register.
2491 (tablejump): Likewise.
2492 (*sibcall_memory): Likewise.
2493 (*sibcall_value_memory): Likewise.
2494 Disallow peepholes of indirect call and jump via memory for
2495 -mindirect-branch-register.
2496 (*call_pop): Replace m with Bw.
2497 (*call_value_pop): Likewise.
2498 (*sibcall_pop_memory): Replace m with Bs.
2499 * config/i386/i386.opt (mindirect-branch-register): New option.
2500 * doc/invoke.texi: Document -mindirect-branch-register option.
2502 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2504 * config/i386/i386-protos.h (ix86_output_function_return): New.
2505 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2506 set function_return_type.
2507 (indirect_thunk_name): Add ret_p to indicate thunk for function
2509 (output_indirect_thunk_function): Pass false to
2510 indirect_thunk_name.
2511 (ix86_output_indirect_branch_via_reg): Likewise.
2512 (ix86_output_indirect_branch_via_push): Likewise.
2513 (output_indirect_thunk_function): Create alias for function
2514 return thunk if regno < 0.
2515 (ix86_output_function_return): New function.
2516 (ix86_handle_fndecl_attribute): Handle function_return.
2517 (ix86_attribute_table): Add function_return.
2518 * config/i386/i386.h (machine_function): Add
2519 function_return_type.
2520 * config/i386/i386.md (simple_return_internal): Use
2521 ix86_output_function_return.
2522 (simple_return_internal_long): Likewise.
2523 * config/i386/i386.opt (mfunction-return=): New option.
2524 (indirect_branch): Mention -mfunction-return=.
2525 * doc/extend.texi: Document function_return function attribute.
2526 * doc/invoke.texi: Document -mfunction-return= option.
2528 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2530 * config/i386/i386-opts.h (indirect_branch): New.
2531 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2532 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2533 with local indirect jump when converting indirect call and jump.
2534 (ix86_set_indirect_branch_type): New.
2535 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2536 (indirectlabelno): New.
2537 (indirect_thunk_needed): Likewise.
2538 (indirect_thunk_bnd_needed): Likewise.
2539 (indirect_thunks_used): Likewise.
2540 (indirect_thunks_bnd_used): Likewise.
2541 (INDIRECT_LABEL): Likewise.
2542 (indirect_thunk_name): Likewise.
2543 (output_indirect_thunk): Likewise.
2544 (output_indirect_thunk_function): Likewise.
2545 (ix86_output_indirect_branch_via_reg): Likewise.
2546 (ix86_output_indirect_branch_via_push): Likewise.
2547 (ix86_output_indirect_branch): Likewise.
2548 (ix86_output_indirect_jmp): Likewise.
2549 (ix86_code_end): Call output_indirect_thunk_function if needed.
2550 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2552 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2553 (ix86_attribute_table): Add indirect_branch.
2554 * config/i386/i386.h (machine_function): Add indirect_branch_type
2555 and has_local_indirect_jump.
2556 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2558 (tablejump): Likewise.
2559 (*indirect_jump): Use ix86_output_indirect_jmp.
2560 (*tablejump_1): Likewise.
2561 (simple_return_indirect_internal): Likewise.
2562 * config/i386/i386.opt (mindirect-branch=): New option.
2563 (indirect_branch): New.
2566 (thunk-inline): Likewise.
2567 (thunk-extern): Likewise.
2568 * doc/extend.texi: Document indirect_branch function attribute.
2569 * doc/invoke.texi: Document -mindirect-branch= option.
2571 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2574 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2576 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2578 * ipa-inline.c (want_inline_small_function_p): Return false if
2579 inlining has already failed with CIF_FINAL_ERROR.
2580 (update_caller_keys): Call want_inline_small_function_p before
2582 (update_callee_keys): Likewise.
2584 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2586 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2588 (rs6000_quadword_masked_address_p): Likewise.
2589 (quad_aligned_load_p): Likewise.
2590 (quad_aligned_store_p): Likewise.
2591 (const_load_sequence_p): Add comment to describe the outer-most loop.
2592 (mimic_memory_attributes_and_flags): New function.
2593 (rs6000_gen_stvx): Likewise.
2594 (replace_swapped_aligned_store): Likewise.
2595 (rs6000_gen_lvx): Likewise.
2596 (replace_swapped_aligned_load): Likewise.
2597 (replace_swapped_load_constant): Capitalize argument name in
2598 comment describing this function.
2599 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2600 and stores that access quad-word aligned addresses and replace
2601 with stvx or lvx instructions when appropriate.
2602 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2603 New function prototype.
2604 (rs6000_quadword_masked_address_p): Likewise.
2605 (rs6000_gen_lvx): Likewise.
2606 (rs6000_gen_stvx): Likewise.
2607 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2608 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2609 when memory address is aligned.
2610 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2611 this split to select lvx instruction when memory address is aligned.
2612 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2613 instruction when memory address is aligned.
2614 (*vsx_le_perm_load_v16qi): Likewise.
2615 (four unnamed splitters): Modify to select the stvx instruction
2616 when memory is aligned.
2618 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2620 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2621 which appears in the queue multiple times.
2623 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2624 Alan Hayward <alan.hayward@arm.com>
2625 David Sherwood <david.sherwood@arm.com>
2627 * tree-vectorizer.h (vec_lower_bound): New structure.
2628 (_loop_vec_info): Add check_nonzero and lower_bounds.
2629 (LOOP_VINFO_CHECK_NONZERO): New macro.
2630 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2631 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2632 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2633 fields. Make seg_len the distance travelled, not including the
2635 (dr_direction_indicator): Declare.
2636 (dr_zero_step_indicator): Likewise.
2637 (dr_known_forward_stride_p): Likewise.
2638 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2640 (runtime_alias_check_p): Allow runtime alias checks with
2642 (operator ==): Compare access_size and align.
2643 (prune_runtime_alias_test_list): Rework for new distinction between
2644 the access_size and seg_len.
2645 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2647 (get_segment_min_max): New function.
2648 (create_intersect_range_checks): Use it.
2649 (dr_step_indicator): New function.
2650 (dr_direction_indicator): Likewise.
2651 (dr_zero_step_indicator): Likewise.
2652 (dr_known_forward_stride_p): Likewise.
2653 * tree-loop-distribution.c (data_ref_segment_size): Return
2654 DR_STEP * (niters - 1).
2655 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2657 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2658 (vect_preserves_scalar_order_p): New function, split out from...
2659 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2660 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2661 (vect_vfa_access_size): New function.
2662 (vect_vfa_align): Likewise.
2663 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2664 (dump_lower_bound): New function.
2665 (vect_check_lower_bound): Likewise.
2666 (vect_small_gap_p): Likewise.
2667 (vectorizable_with_step_bound_p): Likewise.
2668 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2669 depencies if the vectorization factor is 1. Convert the checks
2670 for nonzero steps into checks on the bounds of DR_STEP. Try using
2671 a bunds check for variable steps if the minimum required step is
2672 relatively small. Update calls to the dr_with_seg_len
2673 constructor and to vect_compile_time_alias.
2674 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2676 (vect_loop_versioning): Call it.
2677 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2679 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2681 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2682 Alan Hayward <alan.hayward@arm.com>
2683 David Sherwood <david.sherwood@arm.com>
2685 * doc/sourcebuild.texi (vect_scatter_store): Document.
2686 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2688 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2690 * genopinit.c (main): Add supports_vec_scatter_store and
2691 supports_vec_scatter_store_cached to target_optabs.
2692 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2693 IFN_MASK_SCATTER_STORE.
2694 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2696 * internal-fn.h (internal_store_fn_p): Declare.
2697 (internal_fn_stored_value_index): Likewise.
2698 * internal-fn.c (scatter_store_direct): New macro.
2699 (expand_scatter_store_optab_fn): New function.
2700 (direct_scatter_store_optab_supported_p): New macro.
2701 (internal_store_fn_p): New function.
2702 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2703 IFN_MASK_SCATTER_STORE.
2704 (internal_fn_mask_index): Likewise.
2705 (internal_fn_stored_value_index): New function.
2706 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2708 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2709 * optabs-query.c (supports_vec_scatter_store_p): New function.
2710 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2711 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2712 true for scatter stores.
2713 (vect_gather_scatter_fn_p): Handle scatter stores too.
2714 (vect_check_gather_scatter): Consider using scatter stores if
2715 supports_vec_scatter_store_p.
2716 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2718 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2719 internal_fn_stored_value_index.
2720 (check_load_store_masking): Handle scatter stores too.
2721 (vect_get_store_rhs): Make public.
2722 (vectorizable_call): Use internal_store_fn_p.
2723 (vectorizable_store): Handle scatter store internal functions.
2724 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2725 when deciding whether the end of the group has been reached.
2726 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2727 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2728 (mask_scatter_store<mode>): New insns.
2730 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2731 Alan Hayward <alan.hayward@arm.com>
2732 David Sherwood <david.sherwood@arm.com>
2734 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2735 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2736 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2738 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2739 Use vect_truncate_gather_scatter_offset if we can't treat the
2740 operation as a normal gather load or scatter store.
2741 (get_group_load_store_type): Take the gather_scatter_info
2742 as argument. Try using a gather load or scatter store for
2743 single-element groups.
2744 (get_load_store_type): Update calls to get_group_load_store_type
2745 and vect_use_strided_gather_scatters_p.
2747 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2748 Alan Hayward <alan.hayward@arm.com>
2749 David Sherwood <david.sherwood@arm.com>
2751 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2752 optional tree argument.
2753 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2755 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2756 but continue to use the current value as a fallback.
2757 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2758 to compare the updates.
2759 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2760 (get_load_store_type): Use it when handling a strided access.
2761 (vect_get_strided_load_store_ops): New function.
2762 (vect_get_data_ptr_increment): Likewise.
2763 (vectorizable_load): Handle strided gather loads. Always pass
2764 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2766 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2767 Alan Hayward <alan.hayward@arm.com>
2768 David Sherwood <david.sherwood@arm.com>
2770 * doc/md.texi (gather_load@var{m}): Document.
2771 (mask_gather_load@var{m}): Likewise.
2772 * genopinit.c (main): Add supports_vec_gather_load and
2773 supports_vec_gather_load_cached to target_optabs.
2774 * optabs-tree.c (init_tree_optimization_optabs): Use
2775 ggc_cleared_alloc to allocate target_optabs.
2776 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2777 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2779 * internal-fn.h (internal_load_fn_p): Declare.
2780 (internal_gather_scatter_fn_p): Likewise.
2781 (internal_fn_mask_index): Likewise.
2782 (internal_gather_scatter_fn_supported_p): Likewise.
2783 * internal-fn.c (gather_load_direct): New macro.
2784 (expand_gather_load_optab_fn): New function.
2785 (direct_gather_load_optab_supported_p): New macro.
2786 (direct_internal_fn_optab): New function.
2787 (internal_load_fn_p): Likewise.
2788 (internal_gather_scatter_fn_p): Likewise.
2789 (internal_fn_mask_index): Likewise.
2790 (internal_gather_scatter_fn_supported_p): Likewise.
2791 * optabs-query.c (supports_at_least_one_mode_p): New function.
2792 (supports_vec_gather_load_p): Likewise.
2793 * optabs-query.h (supports_vec_gather_load_p): Declare.
2794 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2795 and memory_type field.
2796 (NUM_PATTERNS): Bump to 15.
2797 * tree-vect-data-refs.c: Include internal-fn.h.
2798 (vect_gather_scatter_fn_p): New function.
2799 (vect_describe_gather_scatter_call): Likewise.
2800 (vect_check_gather_scatter): Try using internal functions for
2801 gather loads. Recognize existing calls to a gather load function.
2802 (vect_analyze_data_refs): Consider using gather loads if
2803 supports_vec_gather_load_p.
2804 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2805 (vect_get_gather_scatter_offset_type): Likewise.
2806 (vect_convert_mask_for_vectype): Likewise.
2807 (vect_add_conversion_to_patterm): Likewise.
2808 (vect_try_gather_scatter_pattern): Likewise.
2809 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2810 (vect_vect_recog_func_ptrs): Add it.
2811 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2812 internal_fn_mask_index and internal_gather_scatter_fn_p.
2813 (check_load_store_masking): Take the gather_scatter_info as an
2814 argument and handle gather loads.
2815 (vect_get_gather_scatter_ops): New function.
2816 (vectorizable_call): Check internal_load_fn_p.
2817 (vectorizable_load): Likewise. Handle gather load internal
2819 (vectorizable_store): Update call to check_load_store_masking.
2820 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2821 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2822 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2823 (aarch64_gather_scale_operand_d): New predicates.
2824 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2825 (mask_gather_load<mode>): New insns.
2827 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2828 Alan Hayward <alan.hayward@arm.com>
2829 David Sherwood <david.sherwood@arm.com>
2831 * optabs.def (fold_left_plus_optab): New optab.
2832 * doc/md.texi (fold_left_plus_@var{m}): Document.
2833 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2834 * internal-fn.c (fold_left_direct): Define.
2835 (expand_fold_left_optab_fn): Likewise.
2836 (direct_fold_left_optab_supported_p): Likewise.
2837 * fold-const-call.c (fold_const_fold_left): New function.
2838 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2839 * tree-parloops.c (valid_reduction_p): New function.
2840 (gather_scalar_reductions): Use it.
2841 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2842 (vect_finish_replace_stmt): Declare.
2843 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2844 (needs_fold_left_reduction_p): New function, split out from...
2845 (vect_is_simple_reduction): ...here. Accept reductions that
2846 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2847 (vect_force_simple_reduction): Also store the reduction type in
2848 the assignment's STMT_VINFO_REDUC_TYPE.
2849 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2850 (merge_with_identity): New function.
2851 (vect_expand_fold_left): Likewise.
2852 (vectorize_fold_left_reduction): Likewise.
2853 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2854 scalar phi in place for it. Check for target support and reject
2855 cases that would reassociate the operation. Defer the transform
2856 phase to vectorize_fold_left_reduction.
2857 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2858 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2859 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2861 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2863 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2864 call to ifc_temp_var.
2866 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2867 Alan Hayward <alan.hayward@arm.com>
2868 David Sherwood <david.sherwood@arm.com>
2870 * target.def (legitimize_address_displacement): Take the original
2871 offset as a poly_int.
2872 * targhooks.h (default_legitimize_address_displacement): Update
2874 * targhooks.c (default_legitimize_address_displacement): Likewise.
2875 * doc/tm.texi: Regenerate.
2876 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2877 as an argument, moving assert of ad->disp == ad->disp_term to...
2878 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2879 Try calling targetm.legitimize_address_displacement before expanding
2880 the address rather than afterwards, and adjust for the new interface.
2881 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2882 Match the new hook interface. Handle SVE addresses.
2883 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2886 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2888 * Makefile.in (OBJS): Add early-remat.o.
2889 * target.def (select_early_remat_modes): New hook.
2890 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2891 * doc/tm.texi: Regenerate.
2892 * targhooks.h (default_select_early_remat_modes): Declare.
2893 * targhooks.c (default_select_early_remat_modes): New function.
2894 * timevar.def (TV_EARLY_REMAT): New timevar.
2895 * passes.def (pass_early_remat): New pass.
2896 * tree-pass.h (make_pass_early_remat): Declare.
2897 * early-remat.c: New file.
2898 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2900 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2902 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2903 Alan Hayward <alan.hayward@arm.com>
2904 David Sherwood <david.sherwood@arm.com>
2906 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2907 vfm1 with a bound_epilog parameter.
2908 (vect_do_peeling): Update calls accordingly, and move the prologue
2909 call earlier in the function. Treat the base bound_epilog as 0 for
2910 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2911 this base when peeling for gaps.
2912 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2913 with fully-masked loops.
2914 (vect_estimate_min_profitable_iters): Handle the single peeled
2915 iteration in that case.
2917 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2918 Alan Hayward <alan.hayward@arm.com>
2919 David Sherwood <david.sherwood@arm.com>
2921 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2922 single-element interleaving even if the size is not a power of 2.
2923 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2924 accesses for single-element interleaving if the group size is
2927 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2928 Alan Hayward <alan.hayward@arm.com>
2929 David Sherwood <david.sherwood@arm.com>
2931 * doc/md.texi (fold_extract_last_@var{m}): Document.
2932 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2933 * optabs.def (fold_extract_last_optab): New optab.
2934 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2935 * internal-fn.c (fold_extract_direct): New macro.
2936 (expand_fold_extract_optab_fn): Likewise.
2937 (direct_fold_extract_optab_supported_p): Likewise.
2938 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2939 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2940 EXTRACT_LAST_REDUCTION.
2941 (get_initial_def_for_reduction): Do not create an initial vector
2942 for EXTRACT_LAST_REDUCTION reductions.
2943 (vectorizable_reduction): Leave the scalar phi in place for
2944 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2945 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2946 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2947 transform phase to vectorizable_condition.
2948 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2950 (vect_finish_stmt_generation): ...here.
2951 (vect_finish_replace_stmt): New function.
2952 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2953 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2955 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2957 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2958 Alan Hayward <alan.hayward@arm.com>
2959 David Sherwood <david.sherwood@arm.com>
2961 * doc/md.texi (extract_last_@var{m}): Document.
2962 * optabs.def (extract_last_optab): New optab.
2963 * internal-fn.def (EXTRACT_LAST): New internal function.
2964 * internal-fn.c (cond_unary_direct): New macro.
2965 (expand_cond_unary_optab_fn): Likewise.
2966 (direct_cond_unary_optab_supported_p): Likewise.
2967 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2968 loops using EXTRACT_LAST.
2969 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2970 (extract_last_<mode>): ...this optab.
2971 (vec_extract<mode><Vel>): Update accordingly.
2973 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2974 Alan Hayward <alan.hayward@arm.com>
2975 David Sherwood <david.sherwood@arm.com>
2977 * target.def (empty_mask_is_expensive): New hook.
2978 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2979 * doc/tm.texi: Regenerate.
2980 * targhooks.h (default_empty_mask_is_expensive): Declare.
2981 * targhooks.c (default_empty_mask_is_expensive): New function.
2982 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2983 if the target says that empty masks are expensive.
2984 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2986 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2988 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2989 Alan Hayward <alan.hayward@arm.com>
2990 David Sherwood <david.sherwood@arm.com>
2992 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2993 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2994 (vect_use_loop_mask_for_alignment_p): New function.
2995 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2996 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2997 niters_skip argument. Make sure that the first niters_skip elements
2998 of the first iteration are inactive.
2999 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3000 Update call to vect_set_loop_masks_directly.
3001 (get_misalign_in_elems): New function, split out from...
3002 (vect_gen_prolog_loop_niters): ...here.
3003 (vect_update_init_of_dr): Take a code argument that specifies whether
3004 the adjustment should be added or subtracted.
3005 (vect_update_init_of_drs): Likewise.
3006 (vect_prepare_for_masked_peels): New function.
3007 (vect_do_peeling): Skip prologue peeling if we're using a mask
3008 instead. Update call to vect_update_inits_of_drs.
3009 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3011 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3012 alignment. Do not include the number of peeled iterations in
3013 the minimum threshold in that case.
3014 (vectorizable_induction): Adjust the start value down by
3015 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3016 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3017 Take the number of skipped iterations into account when calculating
3019 * tree-vect-stmts.c (vect_gen_while_not): New function.
3021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3022 Alan Hayward <alan.hayward@arm.com>
3023 David Sherwood <david.sherwood@arm.com>
3025 * doc/sourcebuild.texi (vect_fully_masked): Document.
3026 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3028 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3030 (vect_analyze_loop_2): ...here. Don't check the vectorization
3031 factor against the number of loop iterations if the loop is
3034 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3035 Alan Hayward <alan.hayward@arm.com>
3036 David Sherwood <david.sherwood@arm.com>
3038 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3039 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3040 (dump_groups): Update accordingly.
3041 (iv_use::mem_type): New member variable.
3042 (address_p): New function.
3043 (record_use): Add a mem_type argument and initialize the new
3045 (record_group_use): Add a mem_type argument. Use address_p.
3046 Remove obsolete null checks of base_object. Update call to record_use.
3047 (find_interesting_uses_op): Update call to record_group_use.
3048 (find_interesting_uses_cond): Likewise.
3049 (find_interesting_uses_address): Likewise.
3050 (get_mem_type_for_internal_fn): New function.
3051 (find_address_like_use): Likewise.
3052 (find_interesting_uses_stmt): Try find_address_like_use before
3053 calling find_interesting_uses_op.
3054 (addr_offset_valid_p): Use the iv mem_type field as the type
3055 of the addressed memory.
3056 (add_autoinc_candidates): Likewise.
3057 (get_address_cost): Likewise.
3058 (split_small_address_groups_p): Use address_p.
3059 (split_address_groups): Likewise.
3060 (add_iv_candidate_for_use): Likewise.
3061 (autoinc_possible_for_pair): Likewise.
3062 (rewrite_groups): Likewise.
3063 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3064 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3065 (get_alias_ptr_type_for_ptr_address): New function.
3066 (rewrite_use_address): Rewrite address uses in calls that were
3067 identified by find_address_like_use.
3069 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3070 Alan Hayward <alan.hayward@arm.com>
3071 David Sherwood <david.sherwood@arm.com>
3073 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3075 * gimple-expr.h (is_gimple_addressable: Likewise.
3076 * gimple-expr.c (is_gimple_address): Likewise.
3077 * internal-fn.c (expand_call_mem_ref): New function.
3078 (expand_mask_load_optab_fn): Use it.
3079 (expand_mask_store_optab_fn): Likewise.
3081 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3082 Alan Hayward <alan.hayward@arm.com>
3083 David Sherwood <david.sherwood@arm.com>
3085 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3086 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3087 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3088 (cond_umax@var{mode}): Document.
3089 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3090 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3091 (cond_umin_optab, cond_umax_optab): New optabs.
3092 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3093 (COND_IOR, COND_XOR): New internal functions.
3094 * internal-fn.h (get_conditional_internal_fn): Declare.
3095 * internal-fn.c (cond_binary_direct): New macro.
3096 (expand_cond_binary_optab_fn): Likewise.
3097 (direct_cond_binary_optab_supported_p): Likewise.
3098 (get_conditional_internal_fn): New function.
3099 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3100 Cope with reduction statements that are vectorized as calls rather
3102 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3103 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3104 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3105 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3106 (UNSPEC_COND_EOR): New unspecs.
3107 (optab): Add mappings for them.
3108 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3109 (sve_int_op, sve_fp_op): New int attributes.
3111 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3112 Alan Hayward <alan.hayward@arm.com>
3113 David Sherwood <david.sherwood@arm.com>
3115 * optabs.def (while_ult_optab): New optab.
3116 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3117 * internal-fn.def (WHILE_ULT): New internal function.
3118 * internal-fn.h (direct_internal_fn_supported_p): New override
3119 that takes two types as argument.
3120 * internal-fn.c (while_direct): New macro.
3121 (expand_while_optab_fn): New function.
3122 (convert_optab_supported_p): Likewise.
3123 (direct_while_optab_supported_p): New macro.
3124 * wide-int.h (wi::udiv_ceil): New function.
3125 * tree-vectorizer.h (rgroup_masks): New structure.
3126 (vec_loop_masks): New typedef.
3127 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3129 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3130 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3131 (vect_max_vf): New function.
3132 (slpeel_make_loop_iterate_ntimes): Delete.
3133 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3134 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3135 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3136 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3137 internal-fn.h, stor-layout.h and optabs-query.h.
3138 (vect_set_loop_mask): New function.
3139 (add_preheader_seq): Likewise.
3140 (add_header_seq): Likewise.
3141 (interleave_supported_p): Likewise.
3142 (vect_maybe_permute_loop_masks): Likewise.
3143 (vect_set_loop_masks_directly): Likewise.
3144 (vect_set_loop_condition_masked): Likewise.
3145 (vect_set_loop_condition_unmasked): New function, split out from
3146 slpeel_make_loop_iterate_ntimes.
3147 (slpeel_make_loop_iterate_ntimes): Rename to..
3148 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3149 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3150 (vect_do_peeling): Update call accordingly.
3151 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3153 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3154 mask_compare_type, can_fully_mask_p and fully_masked_p.
3155 (release_vec_loop_masks): New function.
3156 (_loop_vec_info): Use it to free the loop masks.
3157 (can_produce_all_loop_masks_p): New function.
3158 (vect_get_max_nscalars_per_iter): Likewise.
3159 (vect_verify_full_masking): Likewise.
3160 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3161 retries, and free the mask rgroups before retrying. Check loop-wide
3162 reasons for disallowing fully-masked loops. Make the final decision
3163 about whether use a fully-masked loop or not.
3164 (vect_estimate_min_profitable_iters): Do not assume that peeling
3165 for the number of iterations will be needed for fully-masked loops.
3166 (vectorizable_reduction): Disable fully-masked loops.
3167 (vectorizable_live_operation): Likewise.
3168 (vect_halve_mask_nunits): New function.
3169 (vect_double_mask_nunits): Likewise.
3170 (vect_record_loop_mask): Likewise.
3171 (vect_get_loop_mask): Likewise.
3172 (vect_transform_loop): Handle the case in which the final loop
3173 iteration might handle a partial vector. Call vect_set_loop_condition
3174 instead of slpeel_make_loop_iterate_ntimes.
3175 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3176 (check_load_store_masking): New function.
3177 (prepare_load_store_mask): Likewise.
3178 (vectorizable_store): Handle fully-masked loops.
3179 (vectorizable_load): Likewise.
3180 (supportable_widening_operation): Use vect_halve_mask_nunits for
3182 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3183 (vect_gen_while): New function.
3184 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3185 (aarch64_uqdec<mode>): New insn.
3187 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3188 Alan Hayward <alan.hayward@arm.com>
3189 David Sherwood <david.sherwood@arm.com>
3191 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3192 (reduc_xor_scal_optab): New optabs.
3193 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3194 (reduc_xor_scal_@var{m}): Document.
3195 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3196 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3198 * fold-const-call.c (fold_const_call): Handle them.
3199 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3200 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3201 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3202 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3203 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3204 (UNSPEC_XORV): New unspecs.
3205 (optab): Add entries for them.
3206 (BITWISEV): New int iterator.
3207 (bit_reduc_op): New int attributes.
3209 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3210 Alan Hayward <alan.hayward@arm.com>
3211 David Sherwood <david.sherwood@arm.com>
3213 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3214 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3215 * optabs.def (vec_shl_insert_optab): New optab.
3216 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3217 (duplicate_and_interleave): Likewise.
3218 * tree-vect-loop.c: Include internal-fn.h.
3219 (neutral_op_for_slp_reduction): New function, split out from
3220 get_initial_defs_for_reduction.
3221 (get_initial_def_for_reduction): Handle option 2 for variable-length
3222 vectors by loading the neutral value into a vector and then shifting
3223 the initial value into element 0.
3224 (get_initial_defs_for_reduction): Replace the code argument with
3225 the neutral value calculated by neutral_op_for_slp_reduction.
3226 Use gimple_build_vector for constant-length vectors.
3227 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3228 but the first group_size elements have a neutral value.
3229 Use duplicate_and_interleave otherwise.
3230 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3231 Update call to get_initial_defs_for_reduction. Handle SLP
3232 reductions for variable-length vectors by creating one vector
3233 result for each scalar result, with the elements associated
3234 with other scalar results stubbed out with the neutral value.
3235 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3236 Require IFN_VEC_SHL_INSERT for double reductions on
3237 variable-length vectors, or SLP reductions that have
3238 a neutral value. Require can_duplicate_and_interleave_p
3239 support for variable-length unchained SLP reductions if there
3240 is no neutral value, such as for MIN/MAX reductions. Also require
3241 the number of vector elements to be a multiple of the number of
3242 SLP statements when doing variable-length unchained SLP reductions.
3243 Update call to vect_create_epilog_for_reduction.
3244 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3245 and remove initial values.
3246 (duplicate_and_interleave): Make public.
3247 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3248 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3250 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3251 Alan Hayward <alan.hayward@arm.com>
3252 David Sherwood <david.sherwood@arm.com>
3254 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3255 (can_duplicate_and_interleave_p): New function.
3256 (vect_get_and_check_slp_defs): Take the vector of statements
3257 rather than just the current one. Remove excess parentheses.
3258 Restriction rejectinon of vect_constant_def and vect_external_def
3259 for variable-length vectors to boolean types, or types for which
3260 can_duplicate_and_interleave_p is false.
3261 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3262 (duplicate_and_interleave): New function.
3263 (vect_get_constant_vectors): Use gimple_build_vector for
3264 constant-length vectors and suitable variable-length constant
3265 vectors. Use duplicate_and_interleave for other variable-length
3266 vectors. Don't defer the update when inserting new statements.
3268 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3269 Alan Hayward <alan.hayward@arm.com>
3270 David Sherwood <david.sherwood@arm.com>
3272 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3273 min_profitable_iters doesn't go negative.
3275 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3276 Alan Hayward <alan.hayward@arm.com>
3277 David Sherwood <david.sherwood@arm.com>
3279 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3280 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3281 * optabs.def (vec_mask_load_lanes_optab): New optab.
3282 (vec_mask_store_lanes_optab): Likewise.
3283 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3284 (MASK_STORE_LANES): Likewise.
3285 * internal-fn.c (mask_load_lanes_direct): New macro.
3286 (mask_store_lanes_direct): Likewise.
3287 (expand_mask_load_optab_fn): Handle masked operations.
3288 (expand_mask_load_lanes_optab_fn): New macro.
3289 (expand_mask_store_optab_fn): Handle masked operations.
3290 (expand_mask_store_lanes_optab_fn): New macro.
3291 (direct_mask_load_lanes_optab_supported_p): Likewise.
3292 (direct_mask_store_lanes_optab_supported_p): Likewise.
3293 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3295 (vect_load_lanes_supported): Likewise.
3296 * tree-vect-data-refs.c (strip_conversion): New function.
3297 (can_group_stmts_p): Likewise.
3298 (vect_analyze_data_ref_accesses): Use it instead of checking
3299 for a pair of assignments.
3300 (vect_store_lanes_supported): Take a masked_p parameter.
3301 (vect_load_lanes_supported): Likewise.
3302 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3303 vect_store_lanes_supported and vect_load_lanes_supported.
3304 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3305 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3306 parameter. Don't allow gaps for masked accesses.
3307 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3308 and vect_load_lanes_supported.
3309 (get_load_store_type): Take a masked_p parameter and update
3310 call to get_group_load_store_type.
3311 (vectorizable_store): Update call to get_load_store_type.
3312 Handle IFN_MASK_STORE_LANES.
3313 (vectorizable_load): Update call to get_load_store_type.
3314 Handle IFN_MASK_LOAD_LANES.
3316 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3317 Alan Hayward <alan.hayward@arm.com>
3318 David Sherwood <david.sherwood@arm.com>
3320 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3322 * config/aarch64/aarch64-protos.h
3323 (aarch64_sve_struct_memory_operand_p): Declare.
3324 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3325 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3326 (VPRED, vpred): Handle SVE structure modes.
3327 * config/aarch64/constraints.md (Utx): New constraint.
3328 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3329 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3330 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3331 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3332 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3333 structure modes. Split into pieces after RA.
3334 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3335 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3337 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3338 SVE structure modes.
3339 (aarch64_classify_address): Likewise.
3340 (sizetochar): Move earlier in file.
3341 (aarch64_print_operand): Handle SVE register lists.
3342 (aarch64_array_mode): New function.
3343 (aarch64_sve_struct_memory_operand_p): Likewise.
3344 (TARGET_ARRAY_MODE): Redefine.
3346 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3347 Alan Hayward <alan.hayward@arm.com>
3348 David Sherwood <david.sherwood@arm.com>
3350 * target.def (array_mode): New target hook.
3351 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3352 * doc/tm.texi: Regenerate.
3353 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3354 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3355 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3357 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3360 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3361 Alan Hayward <alan.hayward@arm.com>
3362 David Sherwood <david.sherwood@arm.com>
3364 * fold-const.c (fold_binary_loc): Check the argument types
3365 rather than the result type when testing for a vector operation.
3367 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3369 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3370 * doc/tm.texi: Regenerate.
3372 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3373 Alan Hayward <alan.hayward@arm.com>
3374 David Sherwood <david.sherwood@arm.com>
3376 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3377 (sve): Document new AArch64 extension.
3378 * doc/md.texi (w): Extend the description of the AArch64
3379 constraint to include SVE vectors.
3380 (Upl, Upa): Document new AArch64 predicate constraints.
3381 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3383 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3384 (msve-vector-bits=): New option.
3385 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3386 SVE when these are disabled.
3387 (sve): New extension.
3388 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3389 modes. Adjust their number of units based on aarch64_sve_vg.
3390 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3391 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3392 aarch64_addr_query_type.
3393 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3394 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3395 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3396 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3397 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3398 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3399 (aarch64_simd_imm_zero_p): Delete.
3400 (aarch64_check_zero_based_sve_index_immediate): Declare.
3401 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3402 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3403 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3404 (aarch64_sve_float_mul_immediate_p): Likewise.
3405 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3407 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3408 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3409 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3410 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3411 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3412 (aarch64_regmode_natural_size): Likewise.
3413 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3414 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3416 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3417 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3418 for VG and the SVE predicate registers.
3419 (V_ALIASES): Add a "z"-prefixed alias.
3420 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3421 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3422 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3423 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3424 (REG_CLASS_NAMES): Add entries for them.
3425 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3426 and the predicate registers.
3427 (aarch64_sve_vg): Declare.
3428 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3429 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3430 (REGMODE_NATURAL_SIZE): Define.
3431 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3433 * config/aarch64/aarch64.c: Include cfgrtl.h.
3434 (simd_immediate_info): Add a constructor for series vectors,
3435 and an associated step field.
3436 (aarch64_sve_vg): New variable.
3437 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3438 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3439 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3440 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3441 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3442 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3443 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3444 (aarch64_get_mask_mode): New functions.
3445 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3446 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3447 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3448 predicate modes and predicate registers. Explicitly restrict
3449 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3450 to store a vector mode if it is recognized by
3451 aarch64_classify_vector_mode.
3452 (aarch64_regmode_natural_size): New function.
3453 (aarch64_hard_regno_caller_save_mode): Return the original mode
3455 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3456 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3457 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3458 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3460 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3461 does not overlap dest if the function is frame-related. Handle
3463 (aarch64_split_add_offset): New function.
3464 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3465 them aarch64_add_offset.
3466 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3467 and update call to aarch64_sub_sp.
3468 (aarch64_add_cfa_expression): New function.
3469 (aarch64_expand_prologue): Pass extra temporary registers to the
3470 functions above. Handle the case in which we need to emit new
3471 DW_CFA_expressions for registers that were originally saved
3472 relative to the stack pointer, but now have to be expressed
3473 relative to the frame pointer.
3474 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3476 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3477 IP0 and IP1 values for SVE frames.
3478 (aarch64_expand_vec_series): New function.
3479 (aarch64_expand_sve_widened_duplicate): Likewise.
3480 (aarch64_expand_sve_const_vector): Likewise.
3481 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3482 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3483 into the register, rather than emitting a SET directly.
3484 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3485 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3486 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3487 (offset_9bit_signed_scaled_p): New functions.
3488 (aarch64_replicate_bitmask_imm): New function.
3489 (aarch64_bitmask_imm): Use it.
3490 (aarch64_cannot_force_const_mem): Reject expressions involving
3491 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3492 (aarch64_classify_index): Handle SVE indices, by requiring
3493 a plain register index with a scale that matches the element size.
3494 (aarch64_classify_address): Handle SVE addresses. Assert that
3495 the mode of the address is VOIDmode or an integer mode.
3496 Update call to aarch64_classify_symbol.
3497 (aarch64_classify_symbolic_expression): Update call to
3498 aarch64_classify_symbol.
3499 (aarch64_const_vec_all_in_range_p): New function.
3500 (aarch64_print_vector_float_operand): Likewise.
3501 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3502 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3503 and the FP immediates 1.0 and 0.5.
3504 (aarch64_print_address_internal): Handle SVE addresses.
3505 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3506 (aarch64_regno_regclass): Handle predicate registers.
3507 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3509 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3510 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3511 (aarch64_convert_sve_vector_bits): New function.
3512 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3513 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3515 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3516 Handle SVE vector and predicate modes. Accept VL-based constants
3517 that need only one temporary register, and VL offsets that require
3518 no temporary registers.
3519 (aarch64_conditional_register_usage): Mark the predicate registers
3520 as fixed if SVE isn't available.
3521 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3522 Return true for SVE vector and predicate modes.
3523 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3524 rather than an unsigned int. Handle SVE modes.
3525 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3527 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3529 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3530 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3531 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3532 (aarch64_sve_float_mul_immediate_p): New functions.
3533 (aarch64_sve_valid_immediate): New function.
3534 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3535 Explicitly reject structure modes. Check for INDEX constants.
3536 Handle PTRUE and PFALSE constants.
3537 (aarch64_check_zero_based_sve_index_immediate): New function.
3538 (aarch64_simd_imm_zero_p): Delete.
3539 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3540 vector modes. Accept constants in the range of CNT[BHWD].
3541 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3542 ask for an Advanced SIMD mode.
3543 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3544 (aarch64_simd_vector_alignment): Handle SVE predicates.
3545 (aarch64_vectorize_preferred_vector_alignment): New function.
3546 (aarch64_simd_vector_alignment_reachable): Use it instead of
3548 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3549 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3551 (MAX_VECT_LEN): Delete.
3552 (expand_vec_perm_d): Add a vec_flags field.
3553 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3554 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3555 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3557 (aarch64_evpc_rev): Rename to...
3558 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3559 (aarch64_evpc_rev_global): New function.
3560 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3561 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3563 (aarch64_evpc_sve_tbl): New function.
3564 (aarch64_expand_vec_perm_const_1): Update after rename of
3565 aarch64_evpc_rev. Handle SVE permutes too, trying
3566 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3567 than aarch64_evpc_tbl.
3568 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3569 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3570 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3571 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3572 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3573 (aarch64_expand_sve_vcond): New functions.
3574 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3575 of aarch64_vector_mode_p.
3576 (aarch64_dwarf_poly_indeterminate_value): New function.
3577 (aarch64_compute_pressure_classes): Likewise.
3578 (aarch64_can_change_mode_class): Likewise.
3579 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3580 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3581 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3582 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3583 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3584 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3585 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3586 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3588 (Dn, Dl, Dr): Accept const as well as const_vector.
3589 (Dz): Likewise. Compare against CONST0_RTX.
3590 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3591 of "vector" where appropriate.
3592 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3593 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3594 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3595 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3596 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3597 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3598 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3599 (v_int_equiv): Extend to SVE modes.
3600 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3602 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3603 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3604 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3605 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3606 (SVE_COND_FP_CMP): New int iterators.
3607 (perm_hilo): Handle the new unpack unspecs.
3608 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3610 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3611 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3612 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3613 (aarch64_equality_operator, aarch64_constant_vector_operand)
3614 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3615 (aarch64_sve_nonimmediate_operand): Likewise.
3616 (aarch64_sve_general_operand): Likewise.
3617 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3618 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3619 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3620 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3621 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3622 (aarch64_sve_float_arith_immediate): Likewise.
3623 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3624 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3625 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3626 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3627 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3628 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3629 (aarch64_sve_float_arith_operand): Likewise.
3630 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3631 (aarch64_sve_float_mul_operand): Likewise.
3632 (aarch64_sve_vec_perm_operand): Likewise.
3633 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3634 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3635 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3636 as well as const_vector.
3637 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3638 in file. Use CONST0_RTX and CONSTM1_RTX.
3639 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3640 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3641 Use aarch64_simd_imm_zero.
3642 * config/aarch64/aarch64-sve.md: New file.
3643 * config/aarch64/aarch64.md: Include it.
3644 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3645 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3646 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3647 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3648 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3649 (sve): New attribute.
3650 (enabled): Disable instructions with the sve attribute unless
3652 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3653 aarch64_expand_mov_immediate.
3654 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3655 CNT[BHSD] immediates.
3656 (movti): Split CONST_POLY_INT moves into two halves.
3657 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3658 Split additions that need a temporary here if the destination
3659 is the stack pointer.
3660 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3661 (*add<mode>3_poly_1): New instruction.
3662 (set_clobber_cc): New expander.
3664 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3666 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3667 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3668 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3669 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3670 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3671 Change innermode from fixed_mode_size to machine_mode.
3672 (simplify_subreg): Update call accordingly. Handle a constant-sized
3673 subreg of a variable-length CONST_VECTOR.
3675 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3676 Alan Hayward <alan.hayward@arm.com>
3677 David Sherwood <david.sherwood@arm.com>
3679 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3680 (add_offset_to_base): New function, split out from...
3681 (create_mem_ref): ...here. When handling a scale other than 1,
3682 check first whether the address is valid without the offset.
3683 Add it into the base if so, leaving the index and scale as-is.
3685 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3688 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3689 fold_for_warn before checking if arg2 is INTEGER_CST.
3691 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3693 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3694 (store_multiple_operation): Delete.
3695 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3696 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3697 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3698 guarded by TARGET_STRING.
3699 (rs6000_output_load_multiple): Delete.
3700 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3701 OPTION_MASK_STRING / TARGET_STRING handling.
3702 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3703 (const rs6000_opt_masks) <"string">: Change mask to 0.
3704 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3705 (MASK_STRING): Delete.
3706 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3708 (load_multiple): Delete.
3715 (store_multiple): Delete.
3722 (movmemsi_8reg): Delete.
3723 (corresponding unnamed define_insn): Delete.
3724 (movmemsi_6reg): Delete.
3725 (corresponding unnamed define_insn): Delete.
3726 (movmemsi_4reg): Delete.
3727 (corresponding unnamed define_insn): Delete.
3728 (movmemsi_2reg): Delete.
3729 (corresponding unnamed define_insn): Delete.
3730 (movmemsi_1reg): Delete.
3731 (corresponding unnamed define_insn): Delete.
3732 * config/rs6000/rs6000.opt (mno-string): New.
3733 (mstring): Replace by deprecation warning stub.
3734 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3736 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3738 * regrename.c (regrename_do_replace): If replacing the same
3739 reg multiple times, try to reuse last created gen_raw_REG.
3742 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3743 main to workaround a bug in GDB.
3745 2018-01-12 Tom de Vries <tom@codesourcery.com>
3748 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3750 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3752 PR rtl-optimization/80481
3753 * ira-color.c (get_cap_member): New function.
3754 (allocnos_conflict_by_live_ranges_p): Use it.
3755 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3756 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3758 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3761 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3762 (*saddl_se_1): Ditto.
3764 (*ssubl_se_1): Ditto.
3766 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3768 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3769 rather than wi::to_widest for DR_INITs.
3770 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3771 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3772 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3774 (vect_analyze_group_access_1): Note that here.
3776 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3778 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3779 polynomial type sizes.
3781 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3783 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3784 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3785 (gimple_add_tmp_var): Likewise.
3787 2018-01-12 Martin Liska <mliska@suse.cz>
3789 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3790 (gimple_alloc_sizes): Likewise.
3791 (dump_gimple_statistics): Use PRIu64 in printf format.
3792 * gimple.h: Change uint64_t to int.
3794 2018-01-12 Martin Liska <mliska@suse.cz>
3796 * tree-core.h: Use uint64_t instead of int.
3797 * tree.c (tree_node_counts): Likewise.
3798 (tree_node_sizes): Likewise.
3799 (dump_tree_statistics): Use PRIu64 in printf format.
3801 2018-01-12 Martin Liska <mliska@suse.cz>
3803 * Makefile.in: As qsort_chk is implemented in vec.c, add
3804 vec.o to linkage of gencfn-macros.
3805 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3806 passing the info to record_node_allocation_statistics.
3807 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3809 * ggc-common.c (struct ggc_usage): Add operator== and use
3810 it in operator< and compare function.
3811 * mem-stats.h (struct mem_usage): Likewise.
3812 * vec.c (struct vec_usage): Remove operator< and compare
3813 function. Can be simply inherited.
3815 2018-01-12 Martin Jambor <mjambor@suse.cz>
3818 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3819 * tree-ssa-math-opts.c: Include domwalk.h.
3820 (convert_mult_to_fma_1): New function.
3821 (fma_transformation_info): New type.
3822 (fma_deferring_state): Likewise.
3823 (cancel_fma_deferring): New function.
3824 (result_of_phi): Likewise.
3825 (last_fma_candidate_feeds_initial_phi): Likewise.
3826 (convert_mult_to_fma): Added deferring logic, split actual
3827 transformation to convert_mult_to_fma_1.
3828 (math_opts_dom_walker): New type.
3829 (math_opts_dom_walker::after_dom_children): New method, body moved
3830 here from pass_optimize_widening_mul::execute, added deferring logic
3832 (pass_optimize_widening_mul::execute): Moved most of code to
3833 math_opts_dom_walker::after_dom_children.
3834 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3835 * config/i386/i386.c (ix86_option_override_internal): Added
3836 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3838 2018-01-12 Richard Biener <rguenther@suse.de>
3841 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3842 inline instance vars.
3844 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3847 * config/rx/rx.c (rx_is_restricted_memory_address):
3850 2018-01-12 Richard Biener <rguenther@suse.de>
3852 PR tree-optimization/80846
3853 * target.def (split_reduction): New target hook.
3854 * targhooks.c (default_split_reduction): New function.
3855 * targhooks.h (default_split_reduction): Declare.
3856 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3857 target requests first reduce vectors by combining low and high
3859 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3860 (get_vectype_for_scalar_type_and_size): Export.
3861 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3862 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3863 * doc/tm.texi: Regenerate.
3864 * config/i386/i386.c (ix86_split_reduction): Implement
3865 TARGET_VECTORIZE_SPLIT_REDUCTION.
3867 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3870 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3871 in PIC mode except for TARGET_VXWORKS_RTP.
3872 * config/sparc/sparc.c: Include cfgrtl.h.
3873 (TARGET_INIT_PIC_REG): Define.
3874 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3875 (sparc_pic_register_p): New predicate.
3876 (sparc_legitimate_address_p): Use it.
3877 (sparc_legitimize_pic_address): Likewise.
3878 (sparc_delegitimize_address): Likewise.
3879 (sparc_mode_dependent_address_p): Likewise.
3880 (gen_load_pcrel_sym): Remove 4th parameter.
3881 (load_got_register): Adjust call to above. Remove obsolete stuff.
3882 (sparc_expand_prologue): Do not call load_got_register here.
3883 (sparc_flat_expand_prologue): Likewise.
3884 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3885 (sparc_use_pseudo_pic_reg): New function.
3886 (sparc_init_pic_reg): Likewise.
3887 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3888 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3890 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3892 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3893 Add item for branch_cost.
3895 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3897 PR rtl-optimization/83565
3898 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3899 not extend the result to a larger mode for rotate operations.
3900 (num_sign_bit_copies1): Likewise.
3902 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3905 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3907 Use values-Xc.o for -pedantic.
3908 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3910 2018-01-12 Martin Liska <mliska@suse.cz>
3913 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3915 (possible_polymorphic_call_targets): Use it.
3916 (ipa_devirt): Likewise.
3918 2018-01-12 Martin Liska <mliska@suse.cz>
3920 * profile-count.h (enum profile_quality): Use 0 as invalid
3921 enum value of profile_quality.
3923 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3925 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3926 -mext-string options.
3928 2018-01-12 Richard Biener <rguenther@suse.de>
3930 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3931 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3932 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3934 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3936 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3938 * configure.ac (--with-long-double-format): Add support for the
3939 configuration option to change the default long double format on
3941 * config.gcc (powerpc*-linux*-*): Likewise.
3942 * configure: Regenerate.
3943 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3944 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3945 used without modification.
3947 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3949 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3950 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3951 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3952 MISC_BUILTIN_SPEC_BARRIER.
3953 (rs6000_init_builtins): Likewise.
3954 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3956 (speculation_barrier): New define_insn.
3957 * doc/extend.texi: Document __builtin_speculation_barrier.
3959 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3962 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3963 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3964 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3966 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3967 integral modes instead of "ss" and "sd".
3968 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3969 vectors with 32-bit and 64-bit elements.
3970 (vecdupssescalarmodesuffix): New mode attribute.
3971 (vec_dup<mode>): Use it.
3973 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3976 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3977 frame if argument is passed on stack.
3979 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3982 * ree.c (combine_reaching_defs): Optimize also
3983 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3984 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3986 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3989 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3991 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3994 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3995 after they are computed.
3997 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3999 PR tree-optimization/83695
4000 * gimple-loop-linterchange.cc
4001 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4002 reset cached scev information after interchange.
4003 (pass_linterchange::execute): Remove call to scev_reset_htab.
4005 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4007 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4008 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4009 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4010 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4011 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4012 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4013 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4014 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4015 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4016 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4017 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4018 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4019 (V_lane_reg): Likewise.
4020 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4022 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4023 (vfmal_lane_low<mode>_intrinsic,
4024 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4025 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4026 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4027 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4028 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4029 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4031 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4033 * config/arm/arm-cpus.in (fp16fml): New feature.
4034 (ALL_SIMD): Add fp16fml.
4035 (armv8.2-a): Add fp16fml as an option.
4036 (armv8.3-a): Likewise.
4037 (armv8.4-a): Add fp16fml as part of fp16.
4038 * config/arm/arm.h (TARGET_FP16FML): Define.
4039 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4041 * config/arm/arm-modes.def (V2HF): Define.
4042 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4043 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4044 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4045 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4046 vfmsl_low, vfmsl_high): New set of builtins.
4047 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4048 (vfml_op): New code attribute.
4049 (VFMLHALVES): New int iterator.
4050 (VFML, VFMLSEL): New mode attributes.
4051 (V_reg): Define mapping for V2HF.
4052 (V_hi, V_lo): New mode attributes.
4053 (VF_constraint): Likewise.
4054 (vfml_half, vfml_half_selector): New int attributes.
4055 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4057 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4058 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4060 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4061 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4062 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4063 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4065 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4066 Document new effective target and option set.
4068 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4070 * config/arm/arm-cpus.in (armv8_4): New feature.
4071 (ARMv8_4a): New fgroup.
4072 (armv8.4-a): New arch.
4073 * config/arm/arm-tables.opt: Regenerate.
4074 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4075 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4076 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4077 Add matching rules for -march=armv8.4-a and extensions.
4078 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4080 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4083 * config/rx/rx.md (BW): New mode attribute.
4084 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4086 2018-01-11 Richard Biener <rguenther@suse.de>
4088 PR tree-optimization/83435
4089 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4090 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4091 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4093 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4094 Alan Hayward <alan.hayward@arm.com>
4095 David Sherwood <david.sherwood@arm.com>
4097 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4099 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4100 (aarch64_print_address_internal): Use it to check for a zero offset.
4102 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4103 Alan Hayward <alan.hayward@arm.com>
4104 David Sherwood <david.sherwood@arm.com>
4106 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4107 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4108 Return a poly_int64 rather than a HOST_WIDE_INT.
4109 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4110 rather than a HOST_WIDE_INT.
4111 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4112 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4113 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4114 final_offset from HOST_WIDE_INT to poly_int64.
4115 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4116 to_constant when getting the number of units in an Advanced SIMD
4118 (aarch64_builtin_vectorized_function): Check for a constant number
4120 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4122 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4123 attribute instead of GET_MODE_NUNITS.
4124 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4125 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4126 GET_MODE_SIZE for fixed-size registers.
4127 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4128 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4129 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4130 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4131 (aarch64_print_operand, aarch64_print_address_internal)
4132 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4133 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4134 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4135 Handle polynomial GET_MODE_SIZE.
4136 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4137 wider than SImode without modification.
4138 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4139 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4140 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4141 passing and returning SVE modes.
4142 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4143 rather than GEN_INT.
4144 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4145 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4146 (aarch64_allocate_and_probe_stack_space): Likewise.
4147 (aarch64_layout_frame): Cope with polynomial offsets.
4148 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4149 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4151 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4152 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4153 poly_int64 rather than a HOST_WIDE_INT.
4154 (aarch64_get_separate_components, aarch64_process_components)
4155 (aarch64_expand_prologue, aarch64_expand_epilogue)
4156 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4157 (aarch64_anchor_offset): New function, split out from...
4158 (aarch64_legitimize_address): ...here.
4159 (aarch64_builtin_vectorization_cost): Handle polynomial
4160 TYPE_VECTOR_SUBPARTS.
4161 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4163 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4164 number of elements from the PARALLEL rather than the mode.
4165 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4166 rather than GET_MODE_BITSIZE.
4167 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4168 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4169 (aarch64_expand_vec_perm_const_1): Handle polynomial
4170 d->perm.length () and d->perm elements.
4171 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4172 Apply to_constant to d->perm elements.
4173 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4174 polynomial CONST_VECTOR_NUNITS.
4175 (aarch64_move_pointer): Take amount as a poly_int64 rather
4177 (aarch64_progress_pointer): Avoid temporary variable.
4178 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4179 the mode attribute instead of GET_MODE.
4181 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4182 Alan Hayward <alan.hayward@arm.com>
4183 David Sherwood <david.sherwood@arm.com>
4185 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4186 x exists before using it.
4187 (aarch64_add_constant_internal): Rename to...
4188 (aarch64_add_offset_1): ...this. Replace regnum with separate
4189 src and dest rtxes. Handle the case in which they're different,
4190 including when the offset is zero. Replace scratchreg with an rtx.
4191 Use 2 additions if there is no spare register into which we can
4192 move a 16-bit constant.
4193 (aarch64_add_constant): Delete.
4194 (aarch64_add_offset): Replace reg with separate src and dest
4195 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4196 Use aarch64_add_offset_1.
4197 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4198 an rtx rather than an int. Take the delta as a poly_int64
4199 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4200 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4201 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4202 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4203 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4205 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4206 aarch64_add_constant.
4208 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4210 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4211 Use scalar_float_mode.
4213 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4215 * config/aarch64/aarch64-simd.md
4216 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4217 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4218 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4219 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4220 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4221 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4222 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4223 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4224 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4225 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4227 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4230 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4231 targ_options->x_arm_arch_string is non NULL.
4233 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4235 * config/aarch64/aarch64.h
4236 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4238 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4241 * expmed.c (emit_store_flag_force): Swap if const op0
4242 and change VOIDmode to mode of op0.
4244 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4246 PR rtl-optimization/83761
4247 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4248 than bytes to mode_for_size.
4250 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4253 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4254 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4257 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4260 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4261 when in layout mode.
4262 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4263 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4266 2018-01-10 Michael Collison <michael.collison@arm.com>
4268 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4269 * config/aarch64/aarch64-option-extension.def: Add
4270 AARCH64_OPT_EXTENSION of 'fp16fml'.
4271 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4272 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4273 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4274 * config/aarch64/constraints.md (Ui7): New constraint.
4275 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4276 (VFMLA_SEL_W): Ditto.
4279 (VFMLA16_LOW): New int iterator.
4280 (VFMLA16_HIGH): Ditto.
4281 (UNSPEC_FMLAL): New unspec.
4282 (UNSPEC_FMLSL): Ditto.
4283 (UNSPEC_FMLAL2): Ditto.
4284 (UNSPEC_FMLSL2): Ditto.
4285 (f16mac): New code attribute.
4286 * config/aarch64/aarch64-simd-builtins.def
4287 (aarch64_fmlal_lowv2sf): Ditto.
4288 (aarch64_fmlsl_lowv2sf): Ditto.
4289 (aarch64_fmlalq_lowv4sf): Ditto.
4290 (aarch64_fmlslq_lowv4sf): Ditto.
4291 (aarch64_fmlal_highv2sf): Ditto.
4292 (aarch64_fmlsl_highv2sf): Ditto.
4293 (aarch64_fmlalq_highv4sf): Ditto.
4294 (aarch64_fmlslq_highv4sf): Ditto.
4295 (aarch64_fmlal_lane_lowv2sf): Ditto.
4296 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4297 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4298 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4299 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4300 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4301 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4302 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4303 (aarch64_fmlal_lane_highv2sf): Ditto.
4304 (aarch64_fmlsl_lane_highv2sf): Ditto.
4305 (aarch64_fmlal_laneq_highv2sf): Ditto.
4306 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4307 (aarch64_fmlalq_lane_highv4sf): Ditto.
4308 (aarch64_fmlsl_lane_highv4sf): Ditto.
4309 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4310 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4311 * config/aarch64/aarch64-simd.md:
4312 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4313 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4314 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4315 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4316 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4317 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4318 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4319 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4320 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4321 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4322 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4323 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4324 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4325 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4326 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4327 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4328 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4329 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4330 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4331 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4332 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4333 (vfmlsl_low_u32): Ditto.
4334 (vfmlalq_low_u32): Ditto.
4335 (vfmlslq_low_u32): Ditto.
4336 (vfmlal_high_u32): Ditto.
4337 (vfmlsl_high_u32): Ditto.
4338 (vfmlalq_high_u32): Ditto.
4339 (vfmlslq_high_u32): Ditto.
4340 (vfmlal_lane_low_u32): Ditto.
4341 (vfmlsl_lane_low_u32): Ditto.
4342 (vfmlal_laneq_low_u32): Ditto.
4343 (vfmlsl_laneq_low_u32): Ditto.
4344 (vfmlalq_lane_low_u32): Ditto.
4345 (vfmlslq_lane_low_u32): Ditto.
4346 (vfmlalq_laneq_low_u32): Ditto.
4347 (vfmlslq_laneq_low_u32): Ditto.
4348 (vfmlal_lane_high_u32): Ditto.
4349 (vfmlsl_lane_high_u32): Ditto.
4350 (vfmlal_laneq_high_u32): Ditto.
4351 (vfmlsl_laneq_high_u32): Ditto.
4352 (vfmlalq_lane_high_u32): Ditto.
4353 (vfmlslq_lane_high_u32): Ditto.
4354 (vfmlalq_laneq_high_u32): Ditto.
4355 (vfmlslq_laneq_high_u32): Ditto.
4356 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4357 (AARCH64_FL_FOR_ARCH8_4): New.
4358 (AARCH64_ISA_F16FML): New ISA flag.
4359 (TARGET_F16FML): New feature flag for fp16fml.
4360 (doc/invoke.texi): Document new fp16fml option.
4362 2018-01-10 Michael Collison <michael.collison@arm.com>
4364 * config/aarch64/aarch64-builtins.c:
4365 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4366 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4367 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4368 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4369 (AARCH64_ISA_SHA3): New ISA flag.
4370 (TARGET_SHA3): New feature flag for sha3.
4371 * config/aarch64/iterators.md (sha512_op): New int attribute.
4372 (CRYPTO_SHA512): New int iterator.
4373 (UNSPEC_SHA512H): New unspec.
4374 (UNSPEC_SHA512H2): Ditto.
4375 (UNSPEC_SHA512SU0): Ditto.
4376 (UNSPEC_SHA512SU1): Ditto.
4377 * config/aarch64/aarch64-simd-builtins.def
4378 (aarch64_crypto_sha512hqv2di): New builtin.
4379 (aarch64_crypto_sha512h2qv2di): Ditto.
4380 (aarch64_crypto_sha512su0qv2di): Ditto.
4381 (aarch64_crypto_sha512su1qv2di): Ditto.
4382 (aarch64_eor3qv8hi): Ditto.
4383 (aarch64_rax1qv2di): Ditto.
4384 (aarch64_xarqv2di): Ditto.
4385 (aarch64_bcaxqv8hi): Ditto.
4386 * config/aarch64/aarch64-simd.md:
4387 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4388 (aarch64_crypto_sha512su0qv2di): Ditto.
4389 (aarch64_crypto_sha512su1qv2di): Ditto.
4390 (aarch64_eor3qv8hi): Ditto.
4391 (aarch64_rax1qv2di): Ditto.
4392 (aarch64_xarqv2di): Ditto.
4393 (aarch64_bcaxqv8hi): Ditto.
4394 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4395 (vsha512h2q_u64): Ditto.
4396 (vsha512su0q_u64): Ditto.
4397 (vsha512su1q_u64): Ditto.
4398 (veor3q_u16): Ditto.
4399 (vrax1q_u64): Ditto.
4401 (vbcaxq_u16): Ditto.
4402 * config/arm/types.md (crypto_sha512): New type attribute.
4403 (crypto_sha3): Ditto.
4404 (doc/invoke.texi): Document new sha3 option.
4406 2018-01-10 Michael Collison <michael.collison@arm.com>
4408 * config/aarch64/aarch64-builtins.c:
4409 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4410 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4411 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4412 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4413 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4414 (AARCH64_ISA_SM4): New ISA flag.
4415 (TARGET_SM4): New feature flag for sm4.
4416 * config/aarch64/aarch64-simd-builtins.def
4417 (aarch64_sm3ss1qv4si): Ditto.
4418 (aarch64_sm3tt1aq4si): Ditto.
4419 (aarch64_sm3tt1bq4si): Ditto.
4420 (aarch64_sm3tt2aq4si): Ditto.
4421 (aarch64_sm3tt2bq4si): Ditto.
4422 (aarch64_sm3partw1qv4si): Ditto.
4423 (aarch64_sm3partw2qv4si): Ditto.
4424 (aarch64_sm4eqv4si): Ditto.
4425 (aarch64_sm4ekeyqv4si): Ditto.
4426 * config/aarch64/aarch64-simd.md:
4427 (aarch64_sm3ss1qv4si): Ditto.
4428 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4429 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4430 (aarch64_sm4eqv4si): Ditto.
4431 (aarch64_sm4ekeyqv4si): Ditto.
4432 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4433 (sm3part_op): Ditto.
4434 (CRYPTO_SM3TT): Ditto.
4435 (CRYPTO_SM3PART): Ditto.
4436 (UNSPEC_SM3SS1): New unspec.
4437 (UNSPEC_SM3TT1A): Ditto.
4438 (UNSPEC_SM3TT1B): Ditto.
4439 (UNSPEC_SM3TT2A): Ditto.
4440 (UNSPEC_SM3TT2B): Ditto.
4441 (UNSPEC_SM3PARTW1): Ditto.
4442 (UNSPEC_SM3PARTW2): Ditto.
4443 (UNSPEC_SM4E): Ditto.
4444 (UNSPEC_SM4EKEY): Ditto.
4445 * config/aarch64/constraints.md (Ui2): New constraint.
4446 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4447 * config/arm/types.md (crypto_sm3): New type attribute.
4448 (crypto_sm4): Ditto.
4449 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4450 (vsm3tt1aq_u32): Ditto.
4451 (vsm3tt1bq_u32): Ditto.
4452 (vsm3tt2aq_u32): Ditto.
4453 (vsm3tt2bq_u32): Ditto.
4454 (vsm3partw1q_u32): Ditto.
4455 (vsm3partw2q_u32): Ditto.
4456 (vsm4eq_u32): Ditto.
4457 (vsm4ekeyq_u32): Ditto.
4458 (doc/invoke.texi): Document new sm4 option.
4460 2018-01-10 Michael Collison <michael.collison@arm.com>
4462 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4463 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4464 (AARCH64_FL_FOR_ARCH8_4): New.
4465 (AARCH64_FL_V8_4): New flag.
4466 (doc/invoke.texi): Document new armv8.4-a option.
4468 2018-01-10 Michael Collison <michael.collison@arm.com>
4470 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4471 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4472 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4473 * config/aarch64/aarch64-option-extension.def: Add
4474 AARCH64_OPT_EXTENSION of 'sha2'.
4475 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4476 (crypto): Disable sha2 and aes if crypto disabled.
4477 (crypto): Enable aes and sha2 if enabled.
4478 (simd): Disable sha2 and aes if simd disabled.
4479 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4481 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4482 (TARGET_SHA2): New feature flag for sha2.
4483 (TARGET_AES): New feature flag for aes.
4484 * config/aarch64/aarch64-simd.md:
4485 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4486 conditional on TARGET_AES.
4487 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4488 (aarch64_crypto_sha1hsi): Make pattern conditional
4490 (aarch64_crypto_sha1hv4si): Ditto.
4491 (aarch64_be_crypto_sha1hv4si): Ditto.
4492 (aarch64_crypto_sha1su1v4si): Ditto.
4493 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4494 (aarch64_crypto_sha1su0v4si): Ditto.
4495 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4496 (aarch64_crypto_sha256su0v4si): Ditto.
4497 (aarch64_crypto_sha256su1v4si): Ditto.
4498 (doc/invoke.texi): Document new aes and sha2 options.
4500 2018-01-10 Martin Sebor <msebor@redhat.com>
4502 PR tree-optimization/83781
4503 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4506 2018-01-11 Martin Sebor <msebor@gmail.com>
4507 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4509 PR tree-optimization/83501
4510 PR tree-optimization/81703
4512 * tree-ssa-strlen.c (get_string_cst): Rename...
4513 (get_string_len): ...to this. Handle global constants.
4514 (handle_char_store): Adjust.
4516 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4517 Jim Wilson <jimw@sifive.com>
4519 * config/riscv/riscv-protos.h (riscv_output_return): New.
4520 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4521 (riscv_attribute_table, riscv_output_return),
4522 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4523 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4524 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4526 (riscv_expand_prologue): Add early return for naked function.
4527 (riscv_expand_epilogue): Likewise.
4528 (riscv_function_ok_for_sibcall): Return false for naked function.
4529 (riscv_set_current_function): New.
4530 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4531 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4532 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4533 * doc/extend.texi (RISC-V Function Attributes): New.
4535 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4537 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4538 check for 128-bit long double before checking TCmode.
4539 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4540 128-bit long doubles before checking TFmode or TCmode.
4541 (FLOAT128_IBM_P): Likewise.
4543 2018-01-10 Martin Sebor <msebor@redhat.com>
4545 PR tree-optimization/83671
4546 * builtins.c (c_strlen): Unconditionally return zero for the empty
4548 Use -Warray-bounds for warnings.
4549 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4550 for non-constant array indices with COMPONENT_REF, arrays of
4551 arrays, and pointers to arrays.
4552 (gimple_fold_builtin_strlen): Determine and set length range for
4553 non-constant character arrays.
4555 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4558 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4561 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4563 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4565 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4568 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4569 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4570 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4571 indexed_or_indirect_operand predicate.
4572 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4573 (*vsx_le_perm_load_v8hi): Likewise.
4574 (*vsx_le_perm_load_v16qi): Likewise.
4575 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4576 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4577 (*vsx_le_perm_store_v8hi): Likewise.
4578 (*vsx_le_perm_store_v16qi): Likewise.
4579 (eight unnamed splitters): Likewise.
4581 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4583 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4584 * config/rs6000/emmintrin.h: Likewise.
4585 * config/rs6000/mmintrin.h: Likewise.
4586 * config/rs6000/xmmintrin.h: Likewise.
4588 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4591 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4593 * tree.c (tree_nop_conversion): Return true for location wrapper
4595 (maybe_wrap_with_location): New function.
4596 (selftest::check_strip_nops): New function.
4597 (selftest::test_location_wrappers): New function.
4598 (selftest::tree_c_tests): Call it.
4599 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4600 (maybe_wrap_with_location): New decl.
4601 (EXPR_LOCATION_WRAPPER_P): New macro.
4602 (location_wrapper_p): New inline function.
4603 (tree_strip_any_location_wrapper): New inline function.
4605 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4608 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4609 stack_realign_offset for the largest alignment of stack slot
4611 (ix86_find_max_used_stack_alignment): New function.
4612 (ix86_finalize_stack_frame_flags): Use it. Set
4613 max_used_stack_alignment if we don't realign stack.
4614 * config/i386/i386.h (machine_function): Add
4615 max_used_stack_alignment.
4617 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4619 * config/arm/arm.opt (-mbranch-cost): New option.
4620 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4623 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4626 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4627 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4629 2018-01-10 Richard Biener <rguenther@suse.de>
4632 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4633 early out so it also covers the case where we have a non-NULL
4636 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4638 PR tree-optimization/83753
4639 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4640 for non-strided grouped accesses if the number of elements is 1.
4642 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4645 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4646 * i386.h (TARGET_USE_GATHER): Define.
4647 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4649 2018-01-10 Martin Liska <mliska@suse.cz>
4652 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4653 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4655 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4656 CLEANUP_NO_PARTITIONING is not set.
4658 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4660 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4661 for vectors, as a partial revert of r254296.
4662 * rtl.h (const_vec_p): Delete.
4663 (const_vec_duplicate_p): Don't test for vector CONSTs.
4664 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4665 * expmed.c (make_tree): Likewise.
4668 * common.md (E, F): Use CONSTANT_P instead of checking for
4670 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4671 checking for CONST_VECTOR.
4673 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4676 * predict.c (force_edge_cold): Handle in more sane way edges
4679 2018-01-09 Carl Love <cel@us.ibm.com>
4681 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4683 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4684 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4685 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4686 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4687 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4688 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4689 * config/rs6000/rs6000-protos.h: Add extern defition for
4690 rs6000_generate_float2_double_code.
4691 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4693 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4694 (float2_v2df): Add define_expand.
4696 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4699 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4700 op_mode in the force_to_mode call.
4702 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4704 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4705 instead of checking each element individually.
4706 (aarch64_evpc_uzp): Likewise.
4707 (aarch64_evpc_zip): Likewise.
4708 (aarch64_evpc_ext): Likewise.
4709 (aarch64_evpc_rev): Likewise.
4710 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4711 instead of checking each element individually. Return true without
4713 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4714 whether all selected elements come from the same input, instead of
4715 checking each element individually. Remove calls to gen_rtx_REG,
4716 start_sequence and end_sequence and instead assert that no rtl is
4719 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4721 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4722 order of HIGH and CONST checks.
4724 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4726 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4727 if the destination isn't an SSA_NAME.
4729 2018-01-09 Richard Biener <rguenther@suse.de>
4731 PR tree-optimization/83668
4732 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4734 (canonicalize_loop_form): ... here, renamed from ...
4735 (canonicalize_loop_closed_ssa_form): ... this and amended to
4736 swap successor edges for loop exit blocks to make us use
4737 the RPO order we need for initial schedule generation.
4739 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4741 PR tree-optimization/64811
4742 * match.pd: When optimizing comparisons with Inf, avoid
4743 introducing or losing exceptions from comparisons with NaN.
4745 2018-01-09 Martin Liska <mliska@suse.cz>
4748 * asan.c (shadow_mem_size): Add gcc_assert.
4750 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4752 Don't save registers in main().
4755 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4756 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4757 * config/avr/avr.c (avr_set_current_function): Don't error if
4758 naked, OS_task or OS_main are specified at the same time.
4759 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4761 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4763 * common/config/avr/avr-common.c (avr_option_optimization_table):
4764 Switch on -mmain-is-OS_task for optimizing compilations.
4766 2018-01-09 Richard Biener <rguenther@suse.de>
4768 PR tree-optimization/83572
4769 * graphite.c: Include cfganal.h.
4770 (graphite_transform_loops): Connect infinite loops to exit
4771 and remove fake edges at the end.
4773 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4775 * ipa-inline.c (edge_badness): Revert accidental checkin.
4777 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4780 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4781 symbols; not inline clones.
4783 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4786 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4787 hard registers. Formatting fixes.
4789 PR preprocessor/83722
4790 * gcc.c (try_generate_repro): Pass
4791 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4792 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4795 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4796 Kito Cheng <kito.cheng@gmail.com>
4798 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4799 (riscv_leaf_function_p): Delete.
4800 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4802 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4804 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4806 (do_ifelse): New function.
4807 (do_isel): New function.
4808 (do_sub3): New function.
4809 (do_add3): New function.
4810 (do_load_mask_compare): New function.
4811 (do_overlap_load_compare): New function.
4812 (expand_compare_loop): New function.
4813 (expand_block_compare): Call expand_compare_loop() when appropriate.
4814 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4816 (-mblock-compare-inline-loop-limit): New option.
4818 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4821 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4822 Reverse order of second and third operands in first alternative.
4823 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4824 of first and second elements in UNSPEC_VPERMR vector.
4825 (altivec_expand_vec_perm_le): Likewise.
4827 2018-01-08 Jeff Law <law@redhat.com>
4829 PR rtl-optimizatin/81308
4830 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4831 (process_switch): If group_case_labels makes a change, then set
4833 (pass_convert_switch::execute): If a switch is converted, then
4834 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4836 PR rtl-optimization/81308
4837 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4840 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4842 PR target/83663 - Revert r255946
4843 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4844 generation for cases where splatting a value is not useful.
4845 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4846 across a vec_duplicate and a paradoxical subreg forming a vector
4847 mode to a vec_concat.
4849 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4851 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4852 -march=armv8.3-a variants.
4853 * config/arm/t-multilib: Likewise.
4854 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4856 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4858 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4860 (cceq_ior_compare_complement): Give it a name so I can use it, and
4861 change boolean_or_operator predicate to boolean_operator so it can
4862 be used to generate a crand.
4863 (eqne): New code iterator.
4864 (bd/bd_neg): New code_attrs.
4865 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4866 a single define_insn.
4867 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4868 decrement (bdnzt/bdnzf/bdzt/bdzf).
4869 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4870 with the new names of the branch decrement patterns, and added the
4871 names of the branch decrement conditional patterns.
4873 2018-01-08 Richard Biener <rguenther@suse.de>
4875 PR tree-optimization/83563
4876 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4879 2018-01-08 Richard Biener <rguenther@suse.de>
4882 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4884 2018-01-08 Richard Biener <rguenther@suse.de>
4886 PR tree-optimization/83685
4887 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4888 references to abnormals.
4890 2018-01-08 Richard Biener <rguenther@suse.de>
4893 * dwarf2out.c (output_indirect_strings): Handle empty
4894 skeleton_debug_str_hash.
4895 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4897 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4899 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4900 (emit_store_direct): Likewise.
4901 (arc_trampoline_adjust_address): Likewise.
4902 (arc_asm_trampoline_template): New function.
4903 (arc_initialize_trampoline): Use asm_trampoline_template.
4904 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4905 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4906 * config/arc/arc.md (flush_icache): Delete pattern.
4908 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4910 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4911 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4914 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4917 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4918 by not USED_FOR_TARGET.
4919 (make_pass_resolve_sw_modes): Likewise.
4921 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4923 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4926 2018-01-08 Richard Biener <rguenther@suse.de>
4929 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4931 2018-01-08 Richard Biener <rguenther@suse.de>
4934 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4936 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4939 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4940 basic blocks with a small number of successors.
4941 (convert_control_dep_chain_into_preds): Improve handling of
4943 (dump_predicates): Split apart into...
4944 (dump_pred_chain): ...here...
4945 (dump_pred_info): ...and here.
4946 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4947 (can_chain_union_be_invalidated_p): Improve check for invalidation
4949 (uninit_uses_cannot_happen): Avoid unnecessary if
4950 convert_control_dep_chain_into_preds yielded nothing.
4952 2018-01-06 Martin Sebor <msebor@redhat.com>
4954 PR tree-optimization/83640
4955 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4956 subtracting negative offset from size.
4957 (builtin_access::overlap): Adjust offset bounds of the access to fall
4958 within the size of the object if possible.
4960 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4962 PR rtl-optimization/83699
4963 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4964 extract_bit_field_as_subreg to cases in which the extracted
4965 value is also a vector.
4967 * lra-constraints.c (process_alt_operands): Test for the equivalence
4968 substitutions when detecting a possible reload cycle.
4970 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4973 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4974 by default if flag_selective_schedling{,2}. Formatting fixes.
4976 PR rtl-optimization/83682
4977 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4978 if it has non-VECTOR_MODE element mode.
4979 (vec_duplicate_p): Likewise.
4982 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4983 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4985 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4988 * config/i386/i386-builtin.def
4989 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4990 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4991 Require also OPTION_MASK_ISA_AVX512F in addition to
4992 OPTION_MASK_ISA_GFNI.
4993 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4994 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4995 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4996 to OPTION_MASK_ISA_GFNI.
4997 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4998 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4999 OPTION_MASK_ISA_AVX512BW.
5000 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5001 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5002 addition to OPTION_MASK_ISA_GFNI.
5003 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5004 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5005 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5006 to OPTION_MASK_ISA_GFNI.
5007 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5008 a requirement for all ISAs rather than any of them with a few
5010 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5012 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5013 bitmasks to be enabled with 3 exceptions, instead of requiring any
5014 enabled ISA with lots of exceptions.
5015 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5016 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5017 Change avx512bw in isa attribute to avx512f.
5018 * config/i386/sgxintrin.h: Add license boilerplate.
5019 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5020 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5021 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5022 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5024 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5025 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5026 temporarily sse2 rather than sse if not enabled already.
5029 * config/i386/sse.md (VI248_VLBW): Rename to ...
5030 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5031 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5032 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5033 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5034 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5035 mode iterator instead of VI248_VLBW.
5037 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5039 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5040 (record_modified): Skip clobbers; add debug output.
5041 (param_change_prob): Use sreal frequencies.
5043 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5045 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5046 punt for user-aligned variables.
5048 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5050 * tree-chrec.c (chrec_contains_symbols): Return true for
5053 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5056 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5057 of (x|y) == x for BICS pattern.
5059 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5061 PR tree-optimization/83605
5062 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5063 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5066 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5068 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5069 * config/epiphany/rtems.h: New file.
5071 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5072 Uros Bizjak <ubizjak@gmail.com>
5075 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5076 QIreg_operand instead of register_operand predicate.
5077 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5078 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5079 comments instead of -fmitigate[-_]rop.
5081 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5084 * cgraphunit.c (symbol_table::compile): Switch to text_section
5085 before calling assembly_start debug hook.
5086 * run-rtl-passes.c (run_rtl_passes): Likewise.
5089 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5091 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5092 range_int_cst_p rather than !symbolic_range_p before calling
5093 extract_range_from_multiplicative_op_1.
5095 2018-01-04 Jeff Law <law@redhat.com>
5097 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5098 redundant test in assertion.
5100 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5102 * doc/rtl.texi: Document machine_mode wrapper classes.
5104 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5106 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5109 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5111 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5112 the VEC_PERM_EXPR fold to fail.
5114 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5117 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5118 to switched_sections.
5120 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5123 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5126 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5129 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5130 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5132 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5135 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5136 is BLKmode and bitpos not zero or mode change is needed.
5138 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5141 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5144 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5147 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5148 instead of MULT rtx. Update all corresponding splitters.
5150 (*ssub<modesuffix>): Ditto.
5152 (*cmp_sadd_di): Update split patterns.
5153 (*cmp_sadd_si): Ditto.
5154 (*cmp_sadd_sidi): Ditto.
5155 (*cmp_ssub_di): Ditto.
5156 (*cmp_ssub_si): Ditto.
5157 (*cmp_ssub_sidi): Ditto.
5158 * config/alpha/predicates.md (const23_operand): New predicate.
5159 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5160 Look for ASHIFT, not MULT inner operand.
5161 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5163 2018-01-04 Martin Liska <mliska@suse.cz>
5165 PR gcov-profile/83669
5166 * gcov.c (output_intermediate_file): Add version to intermediate
5168 * doc/gcov.texi: Document new field 'version' in intermediate
5169 file format. Fix location of '-k' option of gcov command.
5171 2018-01-04 Martin Liska <mliska@suse.cz>
5174 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5176 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5178 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5180 2018-01-03 Martin Sebor <msebor@redhat.com>
5182 PR tree-optimization/83655
5183 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5184 checking calls with invalid arguments.
5186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5188 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5189 (vectorizable_mask_load_store): Delete.
5190 (vectorizable_call): Return false for masked loads and stores.
5191 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5192 instead of gimple_assign_rhs1.
5193 (vectorizable_load): Handle IFN_MASK_LOAD.
5194 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5196 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5198 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5200 (vectorizable_mask_load_store): ...here.
5201 (vectorizable_load): ...and here.
5203 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5205 * tree-vect-stmts.c (vect_build_all_ones_mask)
5206 (vect_build_zero_merge_argument): New functions, split out from...
5207 (vectorizable_load): ...here.
5209 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5211 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5213 (vectorizable_mask_load_store): ...here.
5214 (vectorizable_store): ...and here.
5216 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5218 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5220 (vectorizable_mask_load_store): ...here.
5222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5224 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5225 (vect_model_store_cost): Take a vec_load_store_type instead of a
5227 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5228 (vect_model_store_cost): Take a vec_load_store_type instead of a
5230 (vectorizable_mask_load_store): Update accordingly.
5231 (vectorizable_store): Likewise.
5232 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5236 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5237 IFN_MASK_LOAD calls here rather than...
5238 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5241 Alan Hayward <alan.hayward@arm.com>
5242 David Sherwood <david.sherwood@arm.com>
5244 * expmed.c (extract_bit_field_1): For vector extracts,
5245 fall back to extract_bit_field_as_subreg if vec_extract
5248 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5249 Alan Hayward <alan.hayward@arm.com>
5250 David Sherwood <david.sherwood@arm.com>
5252 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5253 they are variable or constant sized.
5254 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5255 slots for constant-sized data.
5257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5258 Alan Hayward <alan.hayward@arm.com>
5259 David Sherwood <david.sherwood@arm.com>
5261 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5262 handling COND_EXPRs with boolean comparisons, try to find a better
5263 basis for the mask type than the boolean itself.
5265 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5267 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5268 is calculated and how it can be overridden.
5269 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5270 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5272 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5275 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5276 Alan Hayward <alan.hayward@arm.com>
5277 David Sherwood <david.sherwood@arm.com>
5279 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5280 Remove the mode argument.
5281 (aarch64_simd_valid_immediate): Remove the mode and inverse
5283 * config/aarch64/iterators.md (bitsize): New iterator.
5284 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5285 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5286 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5287 aarch64_simd_valid_immediate.
5288 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5289 (aarch64_reg_or_bic_imm): Likewise.
5290 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5291 with an insn_type enum and msl with a modifier_type enum.
5292 Replace element_width with a scalar_mode. Change the shift
5293 to unsigned int. Add constructors for scalar_float_mode and
5294 scalar_int_mode elements.
5295 (aarch64_vect_float_const_representable_p): Delete.
5296 (aarch64_can_const_movi_rtx_p)
5297 (aarch64_simd_scalar_immediate_valid_for_move)
5298 (aarch64_simd_make_constant): Update call to
5299 aarch64_simd_valid_immediate.
5300 (aarch64_advsimd_valid_immediate_hs): New function.
5301 (aarch64_advsimd_valid_immediate): Likewise.
5302 (aarch64_simd_valid_immediate): Remove mode and inverse
5303 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5304 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5305 and aarch64_float_const_representable_p on the result.
5306 (aarch64_output_simd_mov_immediate): Remove mode argument.
5307 Update call to aarch64_simd_valid_immediate and use of
5308 simd_immediate_info.
5309 (aarch64_output_scalar_simd_mov_immediate): Update call
5312 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5313 Alan Hayward <alan.hayward@arm.com>
5314 David Sherwood <david.sherwood@arm.com>
5316 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5317 (mode_nunits): Likewise CONST_MODE_NUNITS.
5318 * machmode.def (ADJUST_NUNITS): Document.
5319 * genmodes.c (mode_data::need_nunits_adj): New field.
5320 (blank_mode): Update accordingly.
5321 (adj_nunits): New variable.
5322 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5324 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5325 listed in adj_nunits.
5326 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5327 listed in adj_nunits. Don't emit case statements for such modes.
5328 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5329 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5330 nothing if adj_nunits is nonnull.
5331 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5332 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5333 (emit_mode_fbit): Update use of print_maybe_const_decl.
5334 (emit_move_size): Likewise. Treat the array as non-const
5336 (emit_mode_adjustments): Handle adj_nunits.
5338 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5340 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5341 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5342 (VECTOR_MODES): Use it.
5343 (make_vector_modes): Take the prefix as an argument.
5345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5346 Alan Hayward <alan.hayward@arm.com>
5347 David Sherwood <david.sherwood@arm.com>
5349 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5350 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5351 for MODE_VECTOR_BOOL.
5352 * machmode.def (VECTOR_BOOL_MODE): Document.
5353 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5354 (make_vector_bool_mode): New function.
5355 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5357 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5358 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5360 * stor-layout.c (int_mode_for_mode): Likewise.
5361 * tree.c (build_vector_type_for_mode): Likewise.
5362 * varasm.c (output_constant_pool_2): Likewise.
5363 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5364 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5365 for MODE_VECTOR_BOOL.
5366 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5367 of mode class checks.
5368 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5369 instead of a list of mode class checks.
5370 (expand_vector_scalar_condition): Likewise.
5371 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5374 Alan Hayward <alan.hayward@arm.com>
5375 David Sherwood <david.sherwood@arm.com>
5377 * machmode.h (mode_size): Change from unsigned short to
5379 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5380 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5381 or if measurement_type is not polynomial.
5382 (fixed_size_mode::includes_p): Check for constant-sized modes.
5383 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5384 return a poly_uint16 rather than an unsigned short.
5385 (emit_mode_size): Change the type of mode_size from unsigned short
5386 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5387 (emit_mode_adjustments): Cope with polynomial vector sizes.
5388 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5390 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5392 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5393 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5394 * caller-save.c (setup_save_areas): Likewise.
5395 (replace_reg_with_saved_mem): Likewise.
5396 * calls.c (emit_library_call_value_1): Likewise.
5397 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5398 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5399 (gen_lowpart_for_combine): Likewise.
5400 * convert.c (convert_to_integer_1): Likewise.
5401 * cse.c (equiv_constant, cse_insn): Likewise.
5402 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5403 (cselib_subst_to_values): Likewise.
5404 * dce.c (word_dce_process_block): Likewise.
5405 * df-problems.c (df_word_lr_mark_ref): Likewise.
5406 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5407 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5408 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5409 (rtl_for_decl_location): Likewise.
5410 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5411 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5412 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5413 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5414 (expand_expr_real_1): Likewise.
5415 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5416 (pad_below): Likewise.
5417 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5418 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5419 * ira.c (get_subreg_tracking_sizes): Likewise.
5420 * ira-build.c (ira_create_allocno_objects): Likewise.
5421 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5422 (ira_sort_regnos_for_alter_reg): Likewise.
5423 * ira-costs.c (record_operand_costs): Likewise.
5424 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5425 (resolve_simple_move): Likewise.
5426 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5427 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5428 (lra_constraints): Likewise.
5429 (CONST_POOL_OK_P): Reject variable-sized modes.
5430 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5431 (add_pseudo_to_slot, lra_spill): Likewise.
5432 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5433 * optabs-query.c (get_best_extraction_insn): Likewise.
5434 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5435 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5436 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5437 * recog.c (offsettable_address_addr_space_p): Likewise.
5438 * regcprop.c (maybe_mode_change): Likewise.
5439 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5440 * regrename.c (build_def_use): Likewise.
5441 * regstat.c (dump_reg_info): Likewise.
5442 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5443 (find_reloads, find_reloads_subreg_address): Likewise.
5444 * reload1.c (eliminate_regs_1): Likewise.
5445 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5446 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5447 (simplify_binary_operation_1, simplify_subreg): Likewise.
5448 * targhooks.c (default_function_arg_padding): Likewise.
5449 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5450 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5451 (verify_gimple_assign_ternary): Likewise.
5452 * tree-inline.c (estimate_move_cost): Likewise.
5453 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5454 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5455 (get_address_cost_ainc): Likewise.
5456 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5457 (vect_supportable_dr_alignment): Likewise.
5458 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5459 (vectorizable_reduction): Likewise.
5460 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5461 (vectorizable_operation, vectorizable_load): Likewise.
5462 * tree.c (build_same_sized_truth_vector_type): Likewise.
5463 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5464 * var-tracking.c (emit_note_insn_var_location): Likewise.
5465 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5466 (ADDR_VEC_ALIGN): Likewise.
5468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5469 Alan Hayward <alan.hayward@arm.com>
5470 David Sherwood <david.sherwood@arm.com>
5472 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5474 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5475 or if measurement_type is polynomial.
5476 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5477 * combine.c (make_extraction): Likewise.
5478 * dse.c (find_shift_sequence): Likewise.
5479 * dwarf2out.c (mem_loc_descriptor): Likewise.
5480 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5481 (extract_bit_field, extract_low_bits): Likewise.
5482 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5483 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5484 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5485 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5486 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5487 * reload.c (find_reloads): Likewise.
5488 * reload1.c (alter_reg): Likewise.
5489 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5490 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5491 * tree-if-conv.c (predicate_mem_writes): Likewise.
5492 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5493 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5494 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5495 * valtrack.c (dead_debug_insert_temp): Likewise.
5496 * varasm.c (mergeable_constant_section): Likewise.
5497 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5499 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5500 Alan Hayward <alan.hayward@arm.com>
5501 David Sherwood <david.sherwood@arm.com>
5503 * expr.c (expand_assignment): Cope with polynomial mode sizes
5504 when assigning to a CONCAT.
5506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5507 Alan Hayward <alan.hayward@arm.com>
5508 David Sherwood <david.sherwood@arm.com>
5510 * machmode.h (mode_precision): Change from unsigned short to
5512 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5514 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5515 or if measurement_type is not polynomial.
5516 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5517 in which the mode is already known to be a scalar_int_mode.
5518 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5519 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5521 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5522 for GET_MODE_PRECISION.
5523 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5524 for GET_MODE_PRECISION.
5525 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5527 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5528 (expand_field_assignment, make_extraction): Likewise.
5529 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5530 (get_last_value): Likewise.
5531 * convert.c (convert_to_integer_1): Likewise.
5532 * cse.c (cse_insn): Likewise.
5533 * expr.c (expand_expr_real_1): Likewise.
5534 * lra-constraints.c (simplify_operand_subreg): Likewise.
5535 * optabs-query.c (can_atomic_load_p): Likewise.
5536 * optabs.c (expand_atomic_load): Likewise.
5537 (expand_atomic_store): Likewise.
5538 * ree.c (combine_reaching_defs): Likewise.
5539 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5540 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5541 * tree.h (type_has_mode_precision_p): Likewise.
5542 * ubsan.c (instrument_si_overflow): Likewise.
5544 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5545 Alan Hayward <alan.hayward@arm.com>
5546 David Sherwood <david.sherwood@arm.com>
5548 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5549 polynomial numbers of units.
5550 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5551 (valid_vector_subparts_p): New function.
5552 (build_vector_type): Remove temporary shim and take the number
5553 of units as a poly_uint64 rather than an int.
5554 (build_opaque_vector_type): Take the number of units as a
5555 poly_uint64 rather than an int.
5556 * tree.c (build_vector_from_ctor): Handle polynomial
5557 TYPE_VECTOR_SUBPARTS.
5558 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5559 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5560 (build_vector_from_val): If the number of units is variable,
5561 use build_vec_duplicate_cst for constant operands and
5562 VEC_DUPLICATE_EXPR otherwise.
5563 (make_vector_type): Remove temporary is_constant ().
5564 (build_vector_type, build_opaque_vector_type): Take the number of
5565 units as a poly_uint64 rather than an int.
5566 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5568 * cfgexpand.c (expand_debug_expr): Likewise.
5569 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5570 (store_constructor, expand_expr_real_1): Likewise.
5571 (const_scalar_mask_from_tree): Likewise.
5572 * fold-const-call.c (fold_const_reduction): Likewise.
5573 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5574 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5575 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5576 (fold_relational_const): Likewise.
5577 (native_interpret_vector): Likewise. Change the size from an
5578 int to an unsigned int.
5579 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5580 TYPE_VECTOR_SUBPARTS.
5581 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5582 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5583 duplicating a non-constant operand into a variable-length vector.
5584 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5585 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5586 * ipa-icf.c (sem_variable::equals): Likewise.
5587 * match.pd: Likewise.
5588 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5589 * print-tree.c (print_node): Likewise.
5590 * stor-layout.c (layout_type): Likewise.
5591 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5592 * tree-cfg.c (verify_gimple_comparison): Likewise.
5593 (verify_gimple_assign_binary): Likewise.
5594 (verify_gimple_assign_ternary): Likewise.
5595 (verify_gimple_assign_single): Likewise.
5596 * tree-pretty-print.c (dump_generic_node): Likewise.
5597 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5598 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5599 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5600 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5601 (vect_shift_permute_load_chain): Likewise.
5602 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5603 (expand_vector_condition, optimize_vector_constructor): Likewise.
5604 (lower_vec_perm, get_compute_type): Likewise.
5605 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5606 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5607 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5608 (vect_recog_mask_conversion_pattern): Likewise.
5609 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5610 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5611 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5612 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5613 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5614 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5615 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5616 (supportable_widening_operation): Likewise.
5617 (supportable_narrowing_operation): Likewise.
5618 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5620 * varasm.c (output_constant): Likewise.
5622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5623 Alan Hayward <alan.hayward@arm.com>
5624 David Sherwood <david.sherwood@arm.com>
5626 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5627 so that both the length == 3 and length != 3 cases set up their
5628 own permute vectors. Add comments explaining why we know the
5629 number of elements is constant.
5630 (vect_permute_load_chain): Likewise.
5632 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5633 Alan Hayward <alan.hayward@arm.com>
5634 David Sherwood <david.sherwood@arm.com>
5636 * machmode.h (mode_nunits): Change from unsigned char to
5638 (ONLY_FIXED_SIZE_MODES): New macro.
5639 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5640 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5641 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5643 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5644 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5645 or if measurement_type is not polynomial.
5646 * genmodes.c (ZERO_COEFFS): New macro.
5647 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5649 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5650 Use ZERO_COEFFS when emitting initializers.
5651 * data-streamer.h (bp_pack_poly_value): New function.
5652 (bp_unpack_poly_value): Likewise.
5653 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5654 for GET_MODE_NUNITS.
5655 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5656 for GET_MODE_NUNITS.
5657 * tree.c (make_vector_type): Remove temporary shim and make
5658 the real function take the number of units as a poly_uint64
5660 (build_vector_type_for_mode): Handle polynomial nunits.
5661 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5662 * emit-rtl.c (const_vec_series_p_1): Likewise.
5663 (gen_rtx_CONST_VECTOR): Likewise.
5664 * fold-const.c (test_vec_duplicate_folding): Likewise.
5665 * genrecog.c (validate_pattern): Likewise.
5666 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5667 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5668 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5669 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5670 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5671 * rtlanal.c (subreg_get_info): Likewise.
5672 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5673 (vect_grouped_load_supported): Likewise.
5674 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5675 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5676 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5677 (simplify_const_unary_operation, simplify_binary_operation_1)
5678 (simplify_const_binary_operation, simplify_ternary_operation)
5679 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5680 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5681 instead of CONST_VECTOR_NUNITS.
5682 * varasm.c (output_constant_pool_2): Likewise.
5683 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5684 explicit-encoded elements in the XVEC for variable-length vectors.
5686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5688 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5691 Alan Hayward <alan.hayward@arm.com>
5692 David Sherwood <david.sherwood@arm.com>
5694 * coretypes.h (fixed_size_mode): Declare.
5695 (fixed_size_mode_pod): New typedef.
5696 * builtins.h (target_builtins::x_apply_args_mode)
5697 (target_builtins::x_apply_result_mode): Change type to
5698 fixed_size_mode_pod.
5699 * builtins.c (apply_args_size, apply_result_size, result_vector)
5700 (expand_builtin_apply_args_1, expand_builtin_apply)
5701 (expand_builtin_return): Update accordingly.
5703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5705 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5706 * cselib.c (cselib_hash_rtx): Likewise.
5707 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5708 CONST_VECTOR encoding.
5710 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5711 Jeff Law <law@redhat.com>
5714 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5715 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5716 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5717 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5720 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5721 explicitly probe *sp in a noreturn function if there were any callee
5722 register saves or frame pointer is needed.
5724 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5727 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5728 BLKmode for ternary, binary or unary expressions.
5731 * var-tracking.c (delete_vta_debug_insn): New inline function.
5732 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5733 insns from get_insns () to NULL instead of each bb separately.
5734 Use delete_vta_debug_insn. No longer static.
5735 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5736 delete_vta_debug_insns callers.
5737 * rtl.h (delete_vta_debug_insns): Declare.
5738 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5739 instead of variable_tracking_main.
5741 2018-01-03 Martin Sebor <msebor@redhat.com>
5743 PR tree-optimization/83603
5744 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5745 arguments past the endof the argument list in functions declared
5746 without a prototype.
5747 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5748 Avoid checking when arguments are null.
5750 2018-01-03 Martin Sebor <msebor@redhat.com>
5753 * doc/extend.texi (attribute const): Fix a typo.
5754 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5755 issuing -Wsuggest-attribute for void functions.
5757 2018-01-03 Martin Sebor <msebor@redhat.com>
5759 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5760 offset_int::from instead of wide_int::to_shwi.
5761 (maybe_diag_overlap): Remove assertion.
5762 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5763 * gimple-ssa-sprintf.c (format_directive): Same.
5764 (parse_directive): Same.
5765 (sprintf_dom_walker::compute_format_length): Same.
5766 (try_substitute_return_value): Same.
5768 2018-01-03 Jeff Law <law@redhat.com>
5771 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5772 non-constant residual for zero at runtime and avoid probing in
5773 that case. Reorganize code for trailing problem to mirror handling
5776 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5778 PR tree-optimization/83501
5779 * tree-ssa-strlen.c (get_string_cst): New.
5780 (handle_char_store): Call get_string_cst.
5782 2018-01-03 Martin Liska <mliska@suse.cz>
5784 PR tree-optimization/83593
5785 * tree-ssa-strlen.c: Include tree-cfg.h.
5786 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5787 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5788 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5790 (strlen_dom_walker::before_dom_children): Call
5791 gimple_purge_dead_eh_edges. Dump tranformation with details
5793 (strlen_dom_walker::before_dom_children): Update call by adding
5794 new argument cleanup_eh.
5795 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5797 2018-01-03 Martin Liska <mliska@suse.cz>
5800 * cif-code.def (VARIADIC_THUNK): New enum value.
5801 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5804 2018-01-03 Jan Beulich <jbeulich@suse.com>
5806 * sse.md (mov<mode>_internal): Tighten condition for when to use
5807 vmovdqu<ssescalarsize> for TI and OI modes.
5809 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5811 Update copyright years.
5813 2018-01-03 Martin Liska <mliska@suse.cz>
5816 * ipa-visibility.c (function_and_variable_visibility): Skip
5817 functions with noipa attribure.
5819 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5821 * gcc.c (process_command): Update copyright notice dates.
5822 * gcov-dump.c (print_version): Ditto.
5823 * gcov.c (print_version): Ditto.
5824 * gcov-tool.c (print_version): Ditto.
5825 * gengtype.c (create_file): Ditto.
5826 * doc/cpp.texi: Bump @copying's copyright year.
5827 * doc/cppinternals.texi: Ditto.
5828 * doc/gcc.texi: Ditto.
5829 * doc/gccint.texi: Ditto.
5830 * doc/gcov.texi: Ditto.
5831 * doc/install.texi: Ditto.
5832 * doc/invoke.texi: Ditto.
5834 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5836 * vector-builder.h (vector_builder::m_full_nelts): Change from
5837 unsigned int to poly_uint64.
5838 (vector_builder::full_nelts): Update prototype accordingly.
5839 (vector_builder::new_vector): Likewise.
5840 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5841 (vector_builder::operator ==): Likewise.
5842 (vector_builder::finalize): Likewise.
5843 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5844 Take the number of elements as a poly_uint64 rather than an
5846 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5847 from unsigned int to poly_uint64.
5848 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5849 (vec_perm_indices::new_vector): Likewise.
5850 (vec_perm_indices::length): Likewise.
5851 (vec_perm_indices::nelts_per_input): Likewise.
5852 (vec_perm_indices::input_nelts): Likewise.
5853 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5854 number of elements per input as a poly_uint64 rather than an
5855 unsigned int. Use the original encoding for variable-length
5856 vectors, rather than clamping each individual element.
5857 For the second and subsequent elements in each pattern,
5858 clamp the step and base before clamping their sum.
5859 (vec_perm_indices::series_p): Handle polynomial element counts.
5860 (vec_perm_indices::all_in_range_p): Likewise.
5861 (vec_perm_indices_to_tree): Likewise.
5862 (vec_perm_indices_to_rtx): Likewise.
5863 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5864 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5865 (tree_vector_builder::new_binary_operation): Handle polynomial
5866 element counts. Return false if we need to know the number
5867 of elements at compile time.
5868 * fold-const.c (fold_vec_perm): Punt if the number of elements
5869 isn't known at compile time.
5871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5873 * vec-perm-indices.h (vec_perm_builder): Change element type
5874 from HOST_WIDE_INT to poly_int64.
5875 (vec_perm_indices::element_type): Update accordingly.
5876 (vec_perm_indices::clamp): Handle polynomial element_types.
5877 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5878 (vec_perm_indices::all_in_range_p): Likewise.
5879 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5881 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5882 polynomial vec_perm_indices element types.
5883 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5884 * fold-const.c (fold_vec_perm): Likewise.
5885 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5886 * tree-vect-generic.c (lower_vec_perm): Likewise.
5887 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5888 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5889 element type to HOST_WIDE_INT.
5891 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5892 Alan Hayward <alan.hayward@arm.com>
5893 David Sherwood <david.sherwood@arm.com>
5895 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5896 rather than an int. Use plus_constant.
5897 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5898 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5901 Alan Hayward <alan.hayward@arm.com>
5902 David Sherwood <david.sherwood@arm.com>
5904 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5905 a HOST_WIDE_INT to a poly_int64.
5907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5908 Alan Hayward <alan.hayward@arm.com>
5909 David Sherwood <david.sherwood@arm.com>
5911 * calls.c (load_register_parameters): Cope with polynomial
5912 mode sizes. Require a constant size for BLKmode parameters
5913 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5914 forces a parameter to be padded at the lsb end in order to
5915 fill a complete number of words, require the parameter size
5916 to be ordered wrt UNITS_PER_WORD.
5918 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5919 Alan Hayward <alan.hayward@arm.com>
5920 David Sherwood <david.sherwood@arm.com>
5922 * reload1.c (spill_stack_slot_width): Change element type
5923 from unsigned int to poly_uint64_pod.
5924 (alter_reg): Treat mode sizes as polynomial.
5926 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5927 Alan Hayward <alan.hayward@arm.com>
5928 David Sherwood <david.sherwood@arm.com>
5930 * reload.c (complex_word_subreg_p): New function.
5931 (reload_inner_reg_of_subreg, push_reload): Use it.
5933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5934 Alan Hayward <alan.hayward@arm.com>
5935 David Sherwood <david.sherwood@arm.com>
5937 * lra-constraints.c (process_alt_operands): Reject matched
5938 operands whose sizes aren't ordered.
5939 (match_reload): Refer to this check here.
5941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5942 Alan Hayward <alan.hayward@arm.com>
5943 David Sherwood <david.sherwood@arm.com>
5945 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5946 that the mode size is in the set {1, 2, 4, 8, 16}.
5948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5949 Alan Hayward <alan.hayward@arm.com>
5950 David Sherwood <david.sherwood@arm.com>
5952 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5953 Use plus_constant instead of gen_rtx_PLUS.
5955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5956 Alan Hayward <alan.hayward@arm.com>
5957 David Sherwood <david.sherwood@arm.com>
5959 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5960 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5961 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5962 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5963 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5964 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5965 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5966 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5967 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5968 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5970 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5971 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5972 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5973 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5974 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5975 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5976 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5977 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5978 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5979 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5981 * expr.c (emit_move_resolve_push): Treat the input and result
5982 of PUSH_ROUNDING as a poly_int64.
5983 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5984 (emit_push_insn): Likewise.
5985 * lra-eliminations.c (mark_not_eliminable): Likewise.
5986 * recog.c (push_operand): Likewise.
5987 * reload1.c (elimination_effects): Likewise.
5988 * rtlanal.c (nonzero_bits1): Likewise.
5989 * calls.c (store_one_arg): Likewise. Require the padding to be
5990 known at compile time.
5992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5993 Alan Hayward <alan.hayward@arm.com>
5994 David Sherwood <david.sherwood@arm.com>
5996 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5997 Use plus_constant instead of gen_rtx_PLUS.
5999 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6000 Alan Hayward <alan.hayward@arm.com>
6001 David Sherwood <david.sherwood@arm.com>
6003 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6006 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6007 Alan Hayward <alan.hayward@arm.com>
6008 David Sherwood <david.sherwood@arm.com>
6010 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6011 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6012 via stack temporaries. Treat the mode size as polynomial too.
6014 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6015 Alan Hayward <alan.hayward@arm.com>
6016 David Sherwood <david.sherwood@arm.com>
6018 * expr.c (expand_expr_real_2): When handling conversions involving
6019 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6020 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6021 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6030 Alan Hayward <alan.hayward@arm.com>
6031 David Sherwood <david.sherwood@arm.com>
6033 * combine.c (can_change_dest_mode): Handle polynomial
6034 REGMODE_NATURAL_SIZE.
6035 * expmed.c (store_bit_field_1): Likewise.
6036 * expr.c (store_constructor): Likewise.
6037 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6038 and polynomial REGMODE_NATURAL_SIZE.
6039 (gen_lowpart_common): Likewise.
6040 * reginfo.c (record_subregs_of_mode): Likewise.
6041 * rtlanal.c (read_modify_subreg_p): Likewise.
6043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6044 Alan Hayward <alan.hayward@arm.com>
6045 David Sherwood <david.sherwood@arm.com>
6047 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6048 numbers of 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 * match.pd: Cope with polynomial numbers of vector elements.
6056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6057 Alan Hayward <alan.hayward@arm.com>
6058 David Sherwood <david.sherwood@arm.com>
6060 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6061 in a POINTER_PLUS_EXPR.
6063 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6064 Alan Hayward <alan.hayward@arm.com>
6065 David Sherwood <david.sherwood@arm.com>
6067 * omp-simd-clone.c (simd_clone_subparts): New function.
6068 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6069 (ipa_simd_modify_function_body): Likewise.
6071 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6072 Alan Hayward <alan.hayward@arm.com>
6073 David Sherwood <david.sherwood@arm.com>
6075 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6076 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6077 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6078 (expand_vector_condition, vector_element): Likewise.
6079 (subparts_gt): New function.
6080 (get_compute_type): Use subparts_gt.
6081 (count_type_subparts): Delete.
6082 (expand_vector_operations_1): Use subparts_gt instead of
6083 count_type_subparts.
6085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6086 Alan Hayward <alan.hayward@arm.com>
6087 David Sherwood <david.sherwood@arm.com>
6089 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6090 (vect_compile_time_alias): ...this new function. Do the calculation
6091 on poly_ints rather than trees.
6092 (vect_prune_runtime_alias_test_list): Update call accordingly.
6094 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6095 Alan Hayward <alan.hayward@arm.com>
6096 David Sherwood <david.sherwood@arm.com>
6098 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6100 (vect_schedule_slp_instance): Likewise.
6102 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6103 Alan Hayward <alan.hayward@arm.com>
6104 David Sherwood <david.sherwood@arm.com>
6106 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6107 constant and extern definitions for variable-length vectors.
6108 (vect_get_constant_vectors): Note that the number of units
6109 is known to be constant.
6111 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6112 Alan Hayward <alan.hayward@arm.com>
6113 David Sherwood <david.sherwood@arm.com>
6115 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6116 of units as polynomial. Choose between WIDE and NARROW based
6119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6120 Alan Hayward <alan.hayward@arm.com>
6121 David Sherwood <david.sherwood@arm.com>
6123 * tree-vect-stmts.c (simd_clone_subparts): New function.
6124 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6127 Alan Hayward <alan.hayward@arm.com>
6128 David Sherwood <david.sherwood@arm.com>
6130 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6131 vectors as polynomial. Use build_index_vector for
6134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6135 Alan Hayward <alan.hayward@arm.com>
6136 David Sherwood <david.sherwood@arm.com>
6138 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6139 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6140 for variable-length vectors.
6141 (vectorizable_mask_load_store): Treat the number of units as
6142 polynomial, asserting that it is constant if the condition has
6143 already been enforced.
6144 (vectorizable_store, vectorizable_load): Likewise.
6146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6147 Alan Hayward <alan.hayward@arm.com>
6148 David Sherwood <david.sherwood@arm.com>
6150 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6151 of units as polynomial. Punt if we can't tell at compile time
6152 which vector contains the final result.
6154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6155 Alan Hayward <alan.hayward@arm.com>
6156 David Sherwood <david.sherwood@arm.com>
6158 * tree-vect-loop.c (vectorizable_induction): Treat the number
6159 of units as polynomial. Punt on SLP inductions. Use an integer
6160 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6161 cast of such a series for variable-length floating-point
6164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6165 Alan Hayward <alan.hayward@arm.com>
6166 David Sherwood <david.sherwood@arm.com>
6168 * tree.h (build_index_vector): Declare.
6169 * tree.c (build_index_vector): New function.
6170 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6171 of units as polynomial, forcibly converting it to a constant if
6172 vectorizable_reduction has already enforced the condition.
6173 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6174 to create a {1,2,3,...} vector.
6175 (vectorizable_reduction): Treat the number of units as polynomial.
6176 Choose vectype_in based on the largest scalar element size rather
6177 than the smallest number of units. Enforce the restrictions
6180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6181 Alan Hayward <alan.hayward@arm.com>
6182 David Sherwood <david.sherwood@arm.com>
6184 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6185 number of units as polynomial.
6187 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6188 Alan Hayward <alan.hayward@arm.com>
6189 David Sherwood <david.sherwood@arm.com>
6191 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6192 * target.def (autovectorize_vector_sizes): Return the vector sizes
6193 by pointer, using vector_sizes rather than a bitmask.
6194 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6195 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6196 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6198 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6199 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6200 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6201 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6202 * omp-general.c (omp_max_vf): Likewise.
6203 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6204 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6205 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6206 * tree-vect-slp.c (vect_slp_bb): Likewise.
6207 * doc/tm.texi: Regenerate.
6208 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6210 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6211 the vector size as a poly_uint64 rather than an unsigned int.
6212 (current_vector_size): Change from an unsigned int to a poly_uint64.
6213 (get_vectype_for_scalar_type): Update accordingly.
6214 * tree.h (build_truth_vector_type): Take the size and number of
6215 units as a poly_uint64 rather than an unsigned int.
6216 (build_vector_type): Add a temporary overload that takes
6217 the number of units as a poly_uint64 rather than an unsigned int.
6218 * tree.c (make_vector_type): Likewise.
6219 (build_truth_vector_type): Take the number of units as a poly_uint64
6220 rather than an unsigned int.
6222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6223 Alan Hayward <alan.hayward@arm.com>
6224 David Sherwood <david.sherwood@arm.com>
6226 * target.def (get_mask_mode): Take the number of units and length
6227 as poly_uint64s rather than unsigned ints.
6228 * targhooks.h (default_get_mask_mode): Update accordingly.
6229 * targhooks.c (default_get_mask_mode): Likewise.
6230 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6231 * doc/tm.texi: Regenerate.
6233 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6234 Alan Hayward <alan.hayward@arm.com>
6235 David Sherwood <david.sherwood@arm.com>
6237 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6238 * omp-general.c (omp_max_vf): Likewise.
6239 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6240 (expand_omp_simd): Handle polynomial safelen.
6241 * omp-low.c (omplow_simd_context): Add a default constructor.
6242 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6243 (lower_rec_simd_input_clauses): Update accordingly.
6244 (lower_rec_input_clauses): Likewise.
6246 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6247 Alan Hayward <alan.hayward@arm.com>
6248 David Sherwood <david.sherwood@arm.com>
6250 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6251 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6252 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6253 (vect_analyze_slp_cost): Likewise.
6254 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6255 (vect_model_load_cost): Likewise.
6257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6258 Alan Hayward <alan.hayward@arm.com>
6259 David Sherwood <david.sherwood@arm.com>
6261 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6262 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6263 from an unsigned int * to a poly_uint64_pod *.
6264 (calculate_unrolling_factor): New function.
6265 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6267 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6268 Alan Hayward <alan.hayward@arm.com>
6269 David Sherwood <david.sherwood@arm.com>
6271 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6272 from an unsigned int to a poly_uint64.
6273 (_loop_vec_info::slp_unrolling_factor): Likewise.
6274 (_loop_vec_info::vectorization_factor): Change from an int
6276 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6277 (vect_get_num_vectors): New function.
6278 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6279 (vect_get_num_copies): Use vect_get_num_vectors.
6280 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6281 to an unsigned int *.
6282 (vect_analyze_data_refs): Change min_vf from an int * to a
6284 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6285 than an unsigned HOST_WIDE_INT.
6286 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6287 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6288 to an unsigned int *.
6289 (vect_analyze_data_ref_dependences): Likewise.
6290 (vect_compute_data_ref_alignment): Handle polynomial vf.
6291 (vect_enhance_data_refs_alignment): Likewise.
6292 (vect_prune_runtime_alias_test_list): Likewise.
6293 (vect_shift_permute_load_chain): Likewise.
6294 (vect_supportable_dr_alignment): Likewise.
6295 (dependence_distance_ge_vf): Take the vectorization factor as a
6296 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6297 (vect_analyze_data_refs): Change min_vf from an int * to a
6299 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6300 vfm1 as a poly_uint64 rather than an int. Make the same change
6301 for the returned bound_scalar.
6302 (vect_gen_vector_loop_niters): Handle polynomial vf.
6303 (vect_do_peeling): Likewise. Update call to
6304 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6305 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6307 * tree-vect-loop.c (vect_determine_vectorization_factor)
6308 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6309 (vect_get_known_peeling_cost): Likewise.
6310 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6311 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6312 (vect_transform_loop): Likewise. Use the lowest possible VF when
6313 updating the upper bounds of the loop.
6314 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6316 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6317 polynomial unroll factors.
6318 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6319 (vect_make_slp_decision): Likewise.
6320 (vect_supported_load_permutation_p): Likewise, and polynomial
6322 (vect_analyze_slp_cost): Handle polynomial vf.
6323 (vect_slp_analyze_node_operations): Likewise.
6324 (vect_slp_analyze_bb_1): Likewise.
6325 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6326 than an unsigned HOST_WIDE_INT.
6327 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6328 (vectorizable_load): Handle polynomial vf.
6329 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6331 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6333 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6334 Alan Hayward <alan.hayward@arm.com>
6335 David Sherwood <david.sherwood@arm.com>
6337 * match.pd: Handle bit operations involving three constants
6338 and try to fold one pair.
6340 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6342 * tree-vect-loop-manip.c: Include gimple-fold.h.
6343 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6344 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6345 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6346 Add a path that uses a step of VF instead of 1, but disable it
6348 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6349 and niters_no_overflow parameters. Update calls to
6350 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6351 Create a new SSA name if the latter choses to use a ste other
6352 than zero, and return it via niters_vector_mult_vf_var.
6353 * tree-vect-loop.c (vect_transform_loop): Update calls to
6354 vect_do_peeling, vect_gen_vector_loop_niters and
6355 slpeel_make_loop_iterate_ntimes.
6356 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6357 (vect_gen_vector_loop_niters): Update declarations after above changes.
6359 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6361 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6362 128-bit round to integer instructions.
6363 (ceil<mode>2): Likewise.
6364 (btrunc<mode>2): Likewise.
6365 (round<mode>2): Likewise.
6367 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6369 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6370 unaligned VSX load/store on P8/P9.
6371 (expand_block_clear): Allow the use of unaligned VSX
6372 load/store on P8/P9.
6374 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6376 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6378 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6379 swap associated with both a load and a store.
6381 2018-01-02 Andrew Waterman <andrew@sifive.com>
6383 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6384 * config/riscv/riscv.md (clear_cache): Use it.
6386 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6388 * web.c: Remove out-of-date comment.
6390 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6392 * expr.c (fixup_args_size_notes): Check that any existing
6393 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6394 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6395 (emit_single_push_insn): ...here.
6397 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6399 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6400 (const_vector_encoded_nelts): New function.
6401 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6402 (const_vector_int_elt, const_vector_elt): Declare.
6403 * emit-rtl.c (const_vector_int_elt_1): New function.
6404 (const_vector_elt): Likewise.
6405 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6406 of CONST_VECTOR_ELT.
6408 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6410 * expr.c: Include rtx-vector-builder.h.
6411 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6412 directly on the tree encoding.
6413 (const_vector_from_tree): Likewise.
6414 * optabs.c: Include rtx-vector-builder.h.
6415 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6416 sequence of "u" values.
6417 * vec-perm-indices.c: Include rtx-vector-builder.h.
6418 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6419 directly on the vec_perm_indices encoding.
6421 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6423 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6424 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6425 * rtx-vector-builder.h: New file.
6426 * rtx-vector-builder.c: Likewise.
6427 * rtl.h (rtx_def::u2): Add a const_vector field.
6428 (CONST_VECTOR_NPATTERNS): New macro.
6429 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6430 (CONST_VECTOR_DUPLICATE_P): Likewise.
6431 (CONST_VECTOR_STEPPED_P): Likewise.
6432 (CONST_VECTOR_ENCODED_ELT): Likewise.
6433 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6434 (unwrap_const_vec_duplicate): Likewise.
6435 (const_vec_series_p): Check for a non-duplicated vector encoding.
6436 Say that the function only returns true for integer vectors.
6437 * emit-rtl.c: Include rtx-vector-builder.h.
6438 (gen_const_vec_duplicate_1): Delete.
6439 (gen_const_vector): Call gen_const_vec_duplicate instead of
6440 gen_const_vec_duplicate_1.
6441 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6442 (gen_const_vec_duplicate): Use rtx_vector_builder.
6443 (gen_const_vec_series): Likewise.
6444 (gen_rtx_CONST_VECTOR): Likewise.
6445 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6446 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6447 Build a new vector rather than modifying a CONST_VECTOR in-place.
6448 (handle_special_swappables): Update call accordingly.
6449 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6450 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6451 Build a new vector rather than modifying a CONST_VECTOR in-place.
6452 (handle_special_swappables): Update call accordingly.
6454 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6456 * simplify-rtx.c (simplify_const_binary_operation): Use
6457 CONST_VECTOR_ELT instead of XVECEXP.
6459 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6461 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6462 the selector elements to be different from the data elements
6463 if the selector is a VECTOR_CST.
6464 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6465 ssizetype for the selector.
6467 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6469 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6470 before testing each element individually.
6471 * tree-vect-generic.c (lower_vec_perm): Likewise.
6473 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6475 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6476 * selftest-run-tests.c (selftest::run_tests): Call it.
6477 * vector-builder.h (vector_builder::operator ==): New function.
6478 (vector_builder::operator !=): Likewise.
6479 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6480 (vec_perm_indices::all_from_input_p): New function.
6481 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6482 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6483 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6484 instead of reading the VECTOR_CST directly. Detect whether both
6485 vector inputs are the same before constructing the vec_perm_indices,
6486 and update the number of inputs argument accordingly. Use the
6487 utility functions added above. Only construct sel2 if we need to.
6489 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6491 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6492 the broadcast of the low byte.
6493 (expand_mult_highpart): Use an explicit encoding for the permutes.
6494 * optabs-query.c (can_mult_highpart_p): Likewise.
6495 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6496 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6497 (vectorizable_bswap): Likewise.
6498 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6499 explicit encoding for the power-of-2 permutes.
6500 (vect_permute_store_chain): Likewise.
6501 (vect_grouped_load_supported): Likewise.
6502 (vect_permute_load_chain): Likewise.
6504 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6506 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6507 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6508 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6509 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6510 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6511 (vect_gen_perm_mask_any): Likewise.
6513 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6515 * int-vector-builder.h: New file.
6516 * vec-perm-indices.h: Include int-vector-builder.h.
6517 (vec_perm_indices): Redefine as an int_vector_builder.
6518 (auto_vec_perm_indices): Delete.
6519 (vec_perm_builder): Redefine as a stand-alone class.
6520 (vec_perm_indices::vec_perm_indices): New function.
6521 (vec_perm_indices::clamp): Likewise.
6522 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6523 (vec_perm_indices::new_vector): New function.
6524 (vec_perm_indices::new_expanded_vector): Update for new
6525 vec_perm_indices class.
6526 (vec_perm_indices::rotate_inputs): New function.
6527 (vec_perm_indices::all_in_range_p): Operate directly on the
6528 encoded form, without computing elided elements.
6529 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6530 encoding. Update for new vec_perm_indices class.
6531 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6532 the given vec_perm_builder.
6533 (expand_vec_perm_var): Update vec_perm_builder constructor.
6534 (expand_mult_highpart): Use vec_perm_builder instead of
6535 auto_vec_perm_indices.
6536 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6537 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6538 or double series encoding as appropriate.
6539 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6540 vec_perm_indices instead of auto_vec_perm_indices.
6541 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6542 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6543 (vect_permute_store_chain): Likewise.
6544 (vect_grouped_load_supported): Likewise.
6545 (vect_permute_load_chain): Likewise.
6546 (vect_shift_permute_load_chain): Likewise.
6547 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6548 (vect_transform_slp_perm_load): Likewise.
6549 (vect_schedule_slp_instance): Likewise.
6550 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6551 (vectorizable_mask_load_store): Likewise.
6552 (vectorizable_bswap): Likewise.
6553 (vectorizable_store): Likewise.
6554 (vectorizable_load): Likewise.
6555 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6556 vec_perm_indices instead of auto_vec_perm_indices. Use
6557 tree_to_vec_perm_builder to read the vector from a tree.
6558 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6559 vec_perm_builder instead of a vec_perm_indices.
6560 (have_whole_vector_shift): Use vec_perm_builder and
6561 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6562 truncation to calc_vec_perm_mask_for_shift.
6563 (vect_create_epilog_for_reduction): Likewise.
6564 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6565 from auto_vec_perm_indices to vec_perm_indices.
6566 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6567 instead of changing individual elements.
6568 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6569 the vector in d.perm.
6570 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6571 from auto_vec_perm_indices to vec_perm_indices.
6572 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6573 instead of changing individual elements.
6574 (arm_vectorize_vec_perm_const): Use new_vector to install
6575 the vector in d.perm.
6576 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6577 Update vec_perm_builder constructor.
6578 (rs6000_expand_interleave): Likewise.
6579 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6580 (rs6000_expand_interleave): Likewise.
6582 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6584 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6585 to qimode could truncate the indices.
6586 * optabs.c (expand_vec_perm_var): Likewise.
6588 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6590 * Makefile.in (OBJS): Add vec-perm-indices.o.
6591 * vec-perm-indices.h: New file.
6592 * vec-perm-indices.c: Likewise.
6593 * target.h (vec_perm_indices): Replace with a forward class
6595 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6596 * optabs.h: Include vec-perm-indices.h.
6597 (expand_vec_perm): Delete.
6598 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6599 (expand_vec_perm_const): Declare.
6600 * target.def (vec_perm_const_ok): Replace with...
6601 (vec_perm_const): ...this new hook.
6602 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6603 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6604 * doc/tm.texi: Regenerate.
6605 * optabs.def (vec_perm_const): Delete.
6606 * doc/md.texi (vec_perm_const): Likewise.
6607 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6608 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6609 expand_vec_perm for constant permutation vectors. Assert that
6610 the mode of variable permutation vectors is the integer equivalent
6611 of the mode that is being permuted.
6612 * optabs-query.h (selector_fits_mode_p): Declare.
6613 * optabs-query.c: Include vec-perm-indices.h.
6614 (selector_fits_mode_p): New function.
6615 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6616 is defined, instead of checking whether the vec_perm_const_optab
6617 exists. Use targetm.vectorize.vec_perm_const instead of
6618 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6619 fit in the vector mode before using a variable permute.
6620 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6621 vec_perm_indices instead of an rtx.
6622 (expand_vec_perm): Replace with...
6623 (expand_vec_perm_const): ...this new function. Take the selector
6624 as a vec_perm_indices rather than an rtx. Also take the mode of
6625 the selector. Update call to shift_amt_for_vec_perm_mask.
6626 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6627 Use vec_perm_indices::new_expanded_vector to expand the original
6628 selector into bytes. Check whether the indices fit in the vector
6629 mode before using a variable permute.
6630 (expand_vec_perm_var): Make global.
6631 (expand_mult_highpart): Use expand_vec_perm_const.
6632 * fold-const.c: Includes vec-perm-indices.h.
6633 * tree-ssa-forwprop.c: Likewise.
6634 * tree-vect-data-refs.c: Likewise.
6635 * tree-vect-generic.c: Likewise.
6636 * tree-vect-loop.c: Likewise.
6637 * tree-vect-slp.c: Likewise.
6638 * tree-vect-stmts.c: Likewise.
6639 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6641 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6642 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6643 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6644 (aarch64_vectorize_vec_perm_const): ...this new function.
6645 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6646 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6647 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6648 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6649 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6650 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6651 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6653 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6654 check for NEON modes.
6655 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6656 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6657 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6658 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6660 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6661 the old VEC_PERM_CONST conditions.
6662 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6663 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6664 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6665 (ia64_vectorize_vec_perm_const_ok): Merge into...
6666 (ia64_vectorize_vec_perm_const): ...this new function.
6667 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6668 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6669 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6670 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6671 * config/mips/mips.c (mips_expand_vec_perm_const)
6672 (mips_vectorize_vec_perm_const_ok): Merge into...
6673 (mips_vectorize_vec_perm_const): ...this new function.
6674 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6675 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6676 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6677 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6678 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6679 (rs6000_expand_vec_perm_const): Delete.
6680 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6682 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6683 (altivec_expand_vec_perm_const_le): Take each operand individually.
6684 Operate on constant selectors rather than rtxes.
6685 (altivec_expand_vec_perm_const): Likewise. Update call to
6686 altivec_expand_vec_perm_const_le.
6687 (rs6000_expand_vec_perm_const): Delete.
6688 (rs6000_vectorize_vec_perm_const_ok): Delete.
6689 (rs6000_vectorize_vec_perm_const): New function.
6690 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6691 an element count and rtx array.
6692 (rs6000_expand_extract_even): Update call accordingly.
6693 (rs6000_expand_interleave): Likewise.
6694 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6695 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6696 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6697 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6698 (rs6000_expand_vec_perm_const): Delete.
6699 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6700 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6701 (altivec_expand_vec_perm_const_le): Take each operand individually.
6702 Operate on constant selectors rather than rtxes.
6703 (altivec_expand_vec_perm_const): Likewise. Update call to
6704 altivec_expand_vec_perm_const_le.
6705 (rs6000_expand_vec_perm_const): Delete.
6706 (rs6000_vectorize_vec_perm_const_ok): Delete.
6707 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6708 reference to the SPE evmerge intructions.
6709 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6710 an element count and rtx array.
6711 (rs6000_expand_extract_even): Update call accordingly.
6712 (rs6000_expand_interleave): Likewise.
6713 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6714 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6716 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6718 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6720 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6721 vector mode and that that mode matches the mode of the data
6723 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6724 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6725 directly using expand_vec_perm_1 when forcing selectors into
6727 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6729 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6731 * optabs-query.h (can_vec_perm_p): Delete.
6732 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6733 * optabs-query.c (can_vec_perm_p): Split into...
6734 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6735 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6736 particular selector is valid.
6737 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6738 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6739 (vect_grouped_load_supported): Likewise.
6740 (vect_shift_permute_load_chain): Likewise.
6741 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6742 (vect_transform_slp_perm_load): Likewise.
6743 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6744 (vectorizable_bswap): Likewise.
6745 (vect_gen_perm_mask_checked): Likewise.
6746 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6747 implementations of variable permutation vectors into account
6748 when deciding which selector to use.
6749 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6750 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6751 with a false third argument.
6752 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6753 to test whether the constant selector is valid and can_vec_perm_var_p
6754 to test whether a variable selector is valid.
6756 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6758 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6759 * optabs-query.c (can_vec_perm_p): Likewise.
6760 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6761 instead of vec_perm_indices.
6762 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6763 (vect_gen_perm_mask_checked): Likewise,
6764 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6765 (vect_gen_perm_mask_checked): Likewise,
6767 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6769 * optabs-query.h (qimode_for_vec_perm): Declare.
6770 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6771 (qimode_for_vec_perm): ...this new function.
6772 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6774 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6776 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6777 does not have a conditional at the top.
6779 2018-01-02 Richard Biener <rguenther@suse.de>
6781 * ipa-inline.c (big_speedup_p): Fix expression.
6783 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6786 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6789 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6793 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6794 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6795 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6796 cond_taken_branch_cost 3->4.
6798 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6800 PR tree-optimization/83581
6801 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6802 TODO_cleanup_cfg if any changes have been made.
6805 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6806 convert_modes if target mode has the right side, but different mode
6810 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6811 last argument when extracting from CONCAT. If either from_real or
6812 from_imag is NULL, use expansion through memory. If result is not
6813 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6814 the parts directly to inner mode, if even that fails, use expansion
6818 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6819 check for bswap in mode rather than HImode and use that in expand_unop
6822 Copyright (C) 2018 Free Software Foundation, Inc.
6824 Copying and distribution of this file, with or without modification,
6825 are permitted in any medium without royalty provided the copyright
6826 notice and this notice are preserved.