1 2018-02-13 Jakub Jelinek <jakub@redhat.com>
3 PR tree-optimization/84339
4 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
5 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
9 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
10 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
11 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
13 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
15 * omp-simd-clone.h: New file.
16 * omp-simd-clone.c: Include omp-simd-clone.h.
17 (expand_simd_clones): No longer static.
18 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
19 cgraph.h and omp-simd-clone.h.
20 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
21 (vect_recog_widen_shift_pattern): Formatting fix.
22 (vect_pattern_recog_1): Don't check optab for calls.
25 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
26 operands[2] into a REG before using gen_lowpart on it.
28 2018-02-12 Jeff Law <law@redhat.com>
31 * config/sh/sh.c (find_barrier): Consider a sibling call
34 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
35 successfully back substituting a reg.
37 2018-02-12 Richard Biener <rguenther@suse.de>
39 PR tree-optimization/84037
40 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
41 parameter, move visited init to caller.
42 (vect_slp_analyze_operations): Separate cost from validity
43 check, initialize visited once for all instances.
44 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
46 * tree-vect-stmts.c (vect_model_simple_cost): Make early
48 (vect_model_promotion_demotion_cost): Likewise.
49 (vectorizable_bswap): Guard cost modeling with !slp_node
50 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
52 (vectorizable_call): Likewise.
53 (vectorizable_conversion): Likewise.
54 (vectorizable_assignment): Likewise.
55 (vectorizable_shift): Likewise.
56 (vectorizable_operation): Likewise.
57 (vectorizable_store): Likewise.
58 (vectorizable_load): Likewise.
59 (vectorizable_condition): Likewise.
60 (vectorizable_comparison): Likewise.
62 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
65 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
66 (ASAN_MARK): Fix fnspec to account for return value, change pointer
67 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
69 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
72 * params.def (inline-min-speedup): Increase from 8 to 15.
73 (max-inline-insns-auto): Decrease from 40 to 30.
74 * ipa-split.c (consider_split): Add some buffer for function to
75 be considered inlining candidate.
76 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
79 2018-02-12 Richard Biener <rguenther@suse.de>
81 PR tree-optimization/84037
82 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
83 matched stmts if we cannot swap the non-matched ones.
85 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
87 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
88 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
89 _mm_maskz_scalef_round_ss): New intrinsics.
90 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
91 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
92 __builtin_ia32_scalefss_round): Remove.
93 (__builtin_ia32_scalefsd_mask_round,
94 __builtin_ia32_scalefss_mask_round): New intrinsics.
95 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
96 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
97 ((match_operand:VF_128 2 "<round_nimm_predicate>"
98 "<round_constraint>")): Changed to ...
99 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
100 "<round_scalar_constraint>")): ... this.
101 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
102 %0, %1, %2<round_op3>}"): Changed to ...
103 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
104 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
105 %2<round_scalar_mask_op3>}"): ... this.
106 * config/i386/subst.md (round_scalar_nimm_predicate): New.
108 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
110 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
111 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
112 (_mm_maskz_sqrt_round_ss): New intrinsics.
113 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
114 (__builtin_ia32_sqrtsd_mask_round)
115 (__builtin_ia32_sqrtss_mask_round): New builtins.
116 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
117 (__builtin_ia32_sqrtss_round): Remove.
118 (__builtin_ia32_sqrtsd_mask_round)
119 (__builtin_ia32_sqrtss_mask_round): New builtins.
120 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
121 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
122 ((match_operand:VF_128 1 "vector_operand"
123 "xBm,<round_constraint>")): Changed to ...
124 ((match_operand:VF_128 1 "vector_operand"
125 "xBm,<round_scalar_constraint>")): ... this.
126 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
127 %0, %2, %<iptr>1<round_op3>}): Changed to ...
128 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
129 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
130 %<iptr>1<round_scalar_mask_op3>}): ... this.
131 ((set_attr "prefix" "<round_prefix>")): Changed to ...
132 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
134 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
137 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
138 Cast vec_cmpeq result to correct type.
139 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
140 Cast vec_cmpgt result to correct type.
142 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
144 * final.c (final_scan_insn_1): Renamed from...
145 (final_scan_insn): ... this. New wrapper, to recover
146 seen from the outermost call in recursive ones.
147 * config/sparc/sparc.c (output_return): Drop seen from call.
148 (output_sibcall): Likewise.
149 * config/visium/visium.c (output_branch): Likewise.
151 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
153 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
156 2018-02-10 Alan Modra <amodra@gmail.com>
159 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
160 Specify LR as an input.
162 2018-02-10 Jakub Jelinek <jakub@redhat.com>
165 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
166 remove_member_access_dummy_vars): New functions.
167 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
168 lower_omp_1, execute_lower_omp): Use them.
170 PR rtl-optimization/84308
171 * shrink-wrap.c (spread_components): Release todo vector.
173 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
175 PR rtl-optimization/57193
176 * ira-color.c (struct allocno_color_data): Add member
177 conflict_allocno_hard_prefs.
178 (update_conflict_allocno_hard_prefs): New.
179 (bucket_allocno_compare_func): Add a preference based on
180 conflict_allocno_hard_prefs.
181 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
182 (color_allocnos): Remove a dead code. Initiate
183 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
185 2018-02-09 Jakub Jelinek <jakub@redhat.com>
188 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
189 constraint from =wa to wa. Avoid a subreg on the output operand,
190 instead use a pseudo and subreg it in a move.
191 (p9_xxbrd_<mode>): Changed to ...
192 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
193 (p9_xxbrd_v2df): New expander.
194 (p9_xxbrw_<mode>): Changed to ...
195 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
196 (p9_xxbrw_v4sf): New expander.
198 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
200 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
202 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
205 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
206 multiply in 32-bit mode.
207 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
208 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
211 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
213 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
214 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
215 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
216 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
218 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
221 * dwarf2out.c (is_trivial_indirect_ref): New function.
222 (dwarf2out_late_global_decl): Do not generate a location
223 attribute for variables that have a non-trivial DECL_VALUE_EXPR
224 and that are not defined in the current unit.
226 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
228 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
229 instead of a libcall for UNORDERED.
231 2018-02-09 Tamar Christina <tamar.christina@arm.com>
234 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
235 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
237 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
240 * config/s390/s390.c (s390_set_current_function): Invoke
241 s390_indirect_branch_settings also if fndecl didn't change.
243 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
245 * config/rs6000/rs6000.md (blockage): Set length to zero.
247 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
249 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
251 2018-02-09 Jakub Jelinek <jakub@redhat.com>
254 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
255 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
259 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
260 PARALLEL incoming that failed vt_get_decl_and_offset check.
263 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
264 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
266 (get_variable_section): For decls in named .bss* sections pass true as
267 second argument to bss_initializer_p.
269 2018-02-09 Marek Polacek <polacek@redhat.com>
270 Jakub Jelinek <jakub@redhat.com>
273 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
274 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
275 Sync some changes from cxx_fold_indirect_ref.
277 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
279 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
281 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
282 (BLOCK_INLINE_ENTRY_LABEL): New.
283 (dwarf2out_var_location): Disregard inline entry markers.
284 (inline_entry_data): New struct.
285 (inline_entry_data_hasher): New hashtable type.
286 (inline_entry_data_hasher::hash): New.
287 (inline_entry_data_hasher::equal): New.
288 (inline_entry_data_table): New variable.
289 (add_high_low_attributes): Add DW_AT_entry_pc and
290 DW_AT_GNU_entry_view attributes if a pending entry is found
291 in inline_entry_data_table. Add old entry_pc attribute only
292 if debug nonbinding markers are disabled.
293 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
295 (block_within_block_p, dwarf2out_inline_entry): New.
296 (dwarf2out_finish): Check that no entries remained in
297 inline_entry_data_table.
298 * final.c (reemit_insn_block_notes): Handle inline entry notes.
299 (final_scan_insn, notice_source_line): Likewise.
300 (rest_of_clean_state): Skip inline entry markers.
301 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
303 * gimple.c (gimple_build_debug_inline_entry): New.
304 * gimple.h (enum gimple_debug_subcode): Add
305 GIMPLE_DEBUG_INLINE_ENTRY.
306 (gimple_build_debug_inline_entry): Declare.
307 (gimple_debug_inline_entry_p): New.
308 (gimple_debug_nonbind_marker_p): Adjust.
309 * insn-notes.def (INLINE_ENTRY): New.
310 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
311 inline entry marker notes.
312 (print_insn): Likewise.
313 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
314 (INSN_DEBUG_MARKER_KIND): Likewise.
315 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
316 * tree-inline.c (expand_call_inline): Build and insert
317 debug_inline_entry stmt.
318 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
319 inline entry blocks early, if nonbind markers are enabled.
320 (dump_scope_block): Dump fragment info.
321 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
322 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
323 (gimple_build_debug_inline_entry): New.
324 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
325 Enable/disable inline entry points too.
326 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
327 (DEBUG_INSN): Describe inline entry markers.
329 * common.opt (gvariable-location-views): New.
330 (gvariable-location-views=incompat5): New.
331 * config.in: Rebuilt.
332 * configure: Rebuilt.
333 * configure.ac: Test assembler for view support.
334 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
335 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
336 * dwarf2out.c (var_loc_view): New typedef.
337 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
338 (dwarf2out_locviews_in_attribute): New.
339 (dwarf2out_locviews_in_loclist): New.
340 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
341 (enum dw_line_info_opcode): Add LI_adv_address.
342 (struct dw_line_info_table): Add view.
343 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
344 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
345 (zero_view_p): New variable.
346 (ZERO_VIEW_P): New macro.
347 (output_asm_line_debug_info): New.
348 (struct var_loc_node): Add view.
349 (add_AT_view_list, AT_loc_list): New.
350 (add_var_loc_to_decl): Add view param. Test it against last.
351 (new_loc_list): Add view params. Record them.
352 (AT_loc_list_ptr): Handle loc and view lists.
353 (view_list_to_loc_list_val_node): New.
354 (print_dw_val): Handle dw_val_class_view_list.
355 (size_of_die): Likewise.
356 (value_format): Likewise.
357 (loc_list_has_views): New.
358 (gen_llsym): Set vl_symbol too.
359 (maybe_gen_llsym, skip_loc_list_entry): New.
360 (dwarf2out_maybe_output_loclist_view_pair): New.
361 (output_loc_list): Output view list or entries too.
362 (output_view_list_offset): New.
363 (output_die): Handle dw_val_class_view_list.
364 (output_dwarf_version): New.
365 (output_compilation_unit_header): Use it.
366 (output_skeleton_debug_sections): Likewise.
367 (output_rnglists, output_line_info): Likewise.
368 (output_pubnames, output_aranges): Update version comments.
369 (output_one_line_info_table): Output view numbers in asm comments.
370 (dw_loc_list): Determine current endview, pass it to new_loc_list.
371 Call maybe_gen_llsym.
372 (loc_list_from_tree_1): Adjust.
373 (add_AT_location_description): Create view list attribute if
374 needed, check it's absent otherwise.
375 (convert_cfa_to_fb_loc_list): Adjust.
376 (maybe_emit_file): Call output_asm_line_debug_info for test.
377 (dwarf2out_var_location): Reset views as needed. Precompute
378 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
380 (new_line_info_table): Reset next view.
381 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
382 (dwarf2out_source_line): Likewise. Output view resets and labels to
383 the assembler, or select appropriate line info opcodes.
384 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
385 (optimize_string_length): Catch it. Adjust.
386 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
387 dw_val_class_view_list, and remove it if no longer needed.
388 (hash_loc_list): Hash view numbers.
389 (loc_list_hasher::equal): Compare them.
390 (optimize_location_lists): Check whether a view list symbol is
391 needed, and whether the locview attribute is present, and
392 whether they match. Remove the locview attribute if no longer
394 (index_location_lists): Call skip_loc_list_entry for test.
395 (dwarf2out_finish): Call output_asm_line_debug_info for test.
396 Use output_dwarf_version.
397 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
398 (struct dw_val_node): Add val_view_list.
399 * final.c (SEEN_NEXT_VIEW): New.
400 (set_next_view_needed): New.
401 (clear_next_view_needed): New.
402 (maybe_output_next_view): New.
403 (final_start_function): Rename to...
404 (final_start_function_1): ... this. Take pointer to FIRST,
405 add SEEN parameter. Emit param bindings in the initial view.
406 (final_start_function): Reintroduce SEEN-less interface.
407 (final): Rename to...
408 (final_1): ... this. Take SEEN parameter. Output final pending
409 next view at the end.
410 (final): Reintroduce seen-less interface.
411 (final_scan_insn): Output pending next view before switching
412 sections or ending a block. Mark the next view as needed when
413 outputting variable locations. Notify debug backend of section
414 changes, and of location view changes.
415 (rest_of_handle_final): Adjust.
416 * toplev.c (process_options): Autodetect value for debug variable
417 location views option. Warn on incompat5 without -gdwarf-5.
418 * doc/invoke.texi (gvariable-location-views): New.
419 (gvariable-location-views=incompat5): New.
420 (gno-variable-location-views): New.
422 2018-02-08 David Malcolm <dmalcolm@redhat.com>
424 PR tree-optimization/84136
425 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
426 that the result of find_edge is non-NULL.
428 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
431 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
432 storing integer register in SImode. Fix cost of 256 and 512
433 byte aligned SSE register store.
435 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
437 * config/i386/i386.c (ix86_multiplication_cost): Fix
438 multiplication cost for TARGET_AVX512DQ.
440 2018-02-08 Marek Polacek <polacek@redhat.com>
442 PR tree-optimization/84238
443 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
446 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
448 PR tree-optimization/84265
449 * tree-vect-stmts.c (vectorizable_store): Don't treat
450 VMAT_CONTIGUOUS accesses as grouped.
451 (vectorizable_load): Likewise.
453 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
455 PR tree-optimization/81635
456 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
457 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
458 (test_round_for_mask): New functions.
459 (wide_int_cc_tests): Call test_round_for_mask.
460 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
461 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
462 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
463 range returned by get_range_info.
465 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
468 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
469 * symtab.c: Include builtins.h
470 (symtab_node::output_to_lto_symbol_table_p): Move here
471 from lto-streamer-out.c:output_symbol_p.
472 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
473 (output_symbol_p): Move all logic to symtab.c
474 (produce_symtab): Update.
476 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
478 * config/s390/s390-opts.h (enum indirect_branch): Define.
479 * config/s390/s390-protos.h (s390_return_addr_from_memory)
480 (s390_indirect_branch_via_thunk)
481 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
482 (enum s390_indirect_branch_type): Define.
483 * config/s390/s390.c (struct s390_frame_layout, struct
484 machine_function): Remove.
485 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
486 (indirect_branch_table_label_no, indirect_branch_table_name):
488 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
489 (enum s390_indirect_branch_option): Define.
490 (s390_return_addr_from_memory): New function.
491 (s390_handle_string_attribute): New function.
492 (s390_attribute_table): Add new attribute handler.
493 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
494 (s390_indirect_branch_via_thunk): New function.
495 (s390_indirect_branch_via_inline_thunk): New function.
496 (s390_function_ok_for_sibcall): When jumping via thunk disallow
497 sibling call optimization for non z10 compiles.
498 (s390_emit_call): Force indirect branch target to be a single
499 register. Add r1 clobber for non-z10 compiles.
500 (s390_emit_epilogue): Emit return jump via return_use expander.
501 (s390_reorg): Handle JUMP_INSNs as execute targets.
502 (s390_option_override_internal): Perform validity checks for the
503 new command line options.
504 (s390_indirect_branch_attrvalue): New function.
505 (s390_indirect_branch_settings): New function.
506 (s390_set_current_function): Invoke s390_indirect_branch_settings.
507 (s390_output_indirect_thunk_function): New function.
508 (s390_code_end): Implement target hook.
509 (s390_case_values_threshold): Implement target hook.
510 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
512 * config/s390/s390.h (struct s390_frame_layout)
513 (struct machine_function): Move here from s390.c.
514 (TARGET_INDIRECT_BRANCH_NOBP_RET)
515 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
516 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
517 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
518 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
519 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
520 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
521 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
522 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
523 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
524 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
525 (mnemonic attribute): Add values which aren't recognized
527 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
528 pattern for branch conversion. Fix mnemonic attribute.
529 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
530 indirect branch via thunk if requested.
531 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
532 ("*indirect_jump"): Disable for branch conversion using out of
534 ("indirect_jump_via_thunk<mode>_z10")
535 ("indirect_jump_via_thunk<mode>")
536 ("indirect_jump_via_inlinethunk<mode>_z10")
537 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
538 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
539 ("casesi_jump_via_inlinethunk<mode>_z10")
540 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
541 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
542 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
543 ("*indirect2_jump"): Disable for branch conversion.
544 ("casesi_jump"): Turn into expander and expand patterns for branch
546 ("return_use"): New expander.
547 ("*return"): Emit return via thunk and rename it to ...
548 ("*return<mode>"): ... this one.
549 * config/s390/s390.opt: Add new options and and enum for the
552 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
554 * lra-constraints.c (match_reload): Unconditionally use
555 gen_lowpart_SUBREG, rather than selecting between that
556 and equivalent gen_rtx_SUBREG code.
558 2018-02-08 Richard Biener <rguenther@suse.de>
560 PR tree-optimization/84233
561 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
562 changed flag instead of boguously re-using phi_inserted.
564 2018-02-08 Martin Jambor <mjambor@suse.cz>
566 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
567 static local variables.
569 2018-02-08 Richard Biener <rguenther@suse.de>
571 PR tree-optimization/84278
572 * tree-vect-stmts.c (vectorizable_store): When looking for
573 smaller vector types to perform grouped strided loads/stores
574 make sure the mode is supported by the target.
575 (vectorizable_load): Likewise.
577 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
579 * config/aarch64/aarch64.c (aarch64_components_for_bb):
580 Increase LDP/STP opportunities by adding adjacent callee-saves.
582 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
584 PR rtl-optimization/84068
585 PR rtl-optimization/83459
586 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
588 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
590 PR tree-optimization/84224
591 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
592 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
595 2018-02-07 Iain Sandoe <iain@codesourcery.com>
598 * config/rs6000/altivec.md (*restore_world): Remove LR use.
599 * config/rs6000/predicates.md (restore_world_operation): Adjust op
600 count, remove one USE.
602 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
604 * doc/install.texi (Configuration): Document the
605 --with-long-double-format={ibm,ieee} PowerPC configuration
609 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
610 Convert from define_expand to be define_insn_and_split. Rework
611 float/double/_Float128 conversions to QI/HI/SImode to work with
612 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
613 conversions to QI/HImode types did a store and then a load to
614 truncate the value. For conversions to VSX registers, don't split
615 the insn, instead emit the code directly. Use the code iterator
616 any_fix to combine signed and unsigned conversions.
617 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
618 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
619 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
620 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
621 (fix_<mode>di2_hw): Likewise.
622 (fixuns_<mode>di2_hw): Likewise.
623 (fix_<mode>si2_hw): Likewise.
624 (fixuns_<mode>si2_hw): Likewise.
625 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
626 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
627 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
628 fix<uns>_trunc<SFDF:mode>si2_p8.
629 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
631 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
632 (fix<uns>_<mode>_mem): Likewise.
633 (fctiw<u>z_<mode>_mem): Likewise.
634 (fix<uns>_<mode>_mem): Likewise.
635 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
636 the register allocator from doing a direct move to the GPRs to do
637 a store, and instead use the ISA 3.0 store byte/half-word from
638 vector register instruction. For IEEE 128-bit floating point,
639 also optimize stores of 32-bit ints.
640 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
642 2018-02-07 Alan Hayward <alan.hayward@arm.com>
644 * genextract.c (push_pathstr_operand): New function to support
646 (walk_rtx): Call push_pathstr_operand.
647 (print_path): Support [a-zA-Z].
649 2018-02-07 Richard Biener <rguenther@suse.de>
651 PR tree-optimization/84037
652 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
653 (cse_and_gimplify_to_preheader): Declare.
654 (vect_get_place_in_interleaving_chain): Likewise.
655 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
657 (_loop_vec_info::~_loop_vec_info): Delete it.
658 (cse_and_gimplify_to_preheader): New function.
659 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
660 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
661 (vectorizable_load): Likewise. For grouped stores always base
662 the IV on the first element.
663 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
664 condition before gimplifying.
666 2018-02-07 Jakub Jelinek <jakub@redhat.com>
668 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
669 *DIV_EXPR and *MOD_EXPR.
671 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
674 * config/i386/i386.c (ix86_option_override_internal): Mask out
675 the CF_SET bit when checking -fcf-protection.
677 2018-02-07 Tom de Vries <tom@codesourcery.com>
680 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
683 2018-02-07 Richard Biener <rguenther@suse.de>
685 PR tree-optimization/84204
686 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
689 PR tree-optimization/84205
690 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
691 special-case isl_ast_op_zdiv_r.
693 PR tree-optimization/84223
694 * graphite-scop-detection.c (gather_bbs::before_dom_children):
695 Only add conditions from within the region.
696 (gather_bbs::after_dom_children): Adjust.
698 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
701 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
702 * config/avr/avr.md: Only post-reload split REG-REG moves if
703 either register is REGERAL_REG_P.
705 2018-02-07 Jakub Jelinek <jakub@redhat.com>
707 PR tree-optimization/84235
708 * tree-ssa-scopedtables.c
709 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
710 if the subtraction is performed in floating point type where NaNs are
711 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
712 build 1. Formatting fix.
714 2018-02-06 Jakub Jelinek <jakub@redhat.com>
717 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
718 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
719 and skip it regardless of bb boundaries. Use CALL_P macro,
720 don't test INSN_P (insn) together with CALL_P or JUMP_P check
721 unnecessarily, formatting fix.
723 2018-02-06 Michael Collison <michael.collison@arm.com>
725 * config/arm/thumb2.md:
726 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
727 (*thumb_mov_notscc): Ditto.
729 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
732 * config/rs6000/rs6000.md (su code attribute): Use "u" for
733 unsigned_fix, not "s".
735 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
737 * configure.ac (gcc_fn_eh_frame_ro): New function.
738 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
739 correct .eh_frame permissions.
740 * configure: Regenerate.
742 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
744 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
747 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
749 * config/rs6000/rs6000.c (rs6000_option_override_internal):
750 Display warning message for -mno-speculate-indirect-jumps.
752 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
754 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
756 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
758 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
760 PR tree-optimization/84225
761 * tree-eh.c (find_trapping_overflow): Only call
762 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
764 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
767 * config/i386/i386.c: Reimplement the check of possible options
768 -mibt/-mshstk conbination. Change error messages.
769 * doc/invoke.texi: Fix a typo: remove extra '='.
771 2018-02-06 Marek Polacek <polacek@redhat.com>
773 PR tree-optimization/84228
774 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
776 2018-02-06 Tamar Christina <tamar.christina@arm.com>
779 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
780 emitted arch directives.
781 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
782 __ARM_FEATURE_COPROC before changing architectures.
784 2018-02-06 Richard Biener <rguenther@suse.de>
786 * config/i386/i386.c (print_reg): Fix typo.
787 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
789 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
791 * configure: Regenerate.
793 2018-02-05 Martin Sebor <msebor@redhat.com>
795 PR tree-optimization/83369
796 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
799 2018-02-05 Martin Liska <mliska@suse.cz>
801 * doc/invoke.texi: Cherry-pick upstream r323995.
803 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
805 * ira.c (ira_init_register_move_cost): Adjust comment.
807 2018-02-05 Martin Liska <mliska@suse.cz>
809 PR gcov-profile/84137
810 * doc/gcov.texi: Fix typo in documentation.
812 2018-02-05 Martin Liska <mliska@suse.cz>
814 PR gcov-profile/83879
815 * doc/gcov.texi: Document necessity of --dynamic-list-data when
816 using dlopen functionality.
818 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
820 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
821 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
822 _mm_maskz_range_ss, _mm_mask_range_round_ss,
823 _mm_maskz_range_round_ss): New intrinsics.
824 (__builtin_ia32_rangesd128_round)
825 (__builtin_ia32_rangess128_round): Remove.
826 (__builtin_ia32_rangesd128_mask_round,
827 __builtin_ia32_rangess128_mask_round): New builtins.
828 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
829 __builtin_ia32_rangess128_round): Remove.
830 (__builtin_ia32_rangesd128_mask_round,
831 __builtin_ia32_rangess128_mask_round): New builtins.
832 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
833 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
834 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
835 "<round_saeonly_constraint>")): Changed to ...
836 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
837 "<round_saeonly_scalar_constraint>")): ... this.
838 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
839 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
840 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
841 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
842 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
844 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
846 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
848 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
849 Remove all values except native, 8540 and 8548.
851 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
853 * config/i386/i386.c (ix86_output_function_return): Pass
854 INVALID_REGNUM, instead of -1, as invalid register number to
855 indirect_thunk_name and output_indirect_thunk.
857 2018-02-02 Julia Koval <julia.koval@intel.com>
859 * config.gcc: Add -march=icelake.
860 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
861 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
862 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
863 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
864 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
865 (processor_target_table): Add icelake.
866 (ix86_option_override_internal): Handle new PTAs.
867 (get_builtin_code_for_version): Handle icelake.
868 (M_INTEL_COREI7_ICELAKE): New.
869 (fold_builtin_cpu): Handle icelake.
870 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
871 * doc/invoke.texi: Add -march=icelake.
873 2018-02-02 Julia Koval <julia.koval@intel.com>
875 * config/i386/i386.c (ix86_option_override_internal): Change flags type
877 * wide-int-bitmask.h: New.
879 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
882 * config/i386/i386.md: Replace Pmode with word_mode in
883 builtin_setjmp_setup and builtin_longjmp to support x32.
885 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
889 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
891 (rs6000_supported_cpu_names): New static variable.
892 (linux_cpu_translation_table): Likewise.
893 (elf_platform) <cpu>: Define new static variable and use it.
894 Translate kernel AT_PLATFORM name to canonical name if needed.
895 Error if platform name is unknown.
897 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
900 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
902 2018-02-01 Jeff Law <law@redhat.com>
905 * config/i386/i386.c (release_scratch_register_on_entry): Add new
906 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
907 the scratch if RELEASE_VIA_POP is false.
908 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
909 If we have to save a temporary register, decrement SIZE appropriately.
910 Pass new arguments to release_scratch_register_on_entry.
911 (ix86_adjust_stack_and_probe): Likewise.
912 (ix86_emit_probe_stack_range): Pass new arguments to
913 release_scratch_register_on_entry.
915 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
917 PR rtl-optimization/84157
918 * combine.c (change_zero_ext): Use REG_P predicate in
919 front of HARD_REGISTER_P predicate.
921 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
923 * config/avr/avr.c (avr_option_override): Move disabling of
924 -fdelete-null-pointer-checks to...
925 * common/config/avr/avr-common.c (avr_option_optimization_table):
928 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
930 PR tree-optimization/81635
931 * tree-data-ref.c (split_constant_offset_1): For types that
932 wrap on overflow, try to use range info to prove that wrapping
935 2018-02-01 Renlin Li <renlin.li@arm.com>
938 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
940 (aarch64_register_move_cost): Likewise.
941 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
943 (REG_CLASS_NAMES): Likewise.
944 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
945 TAILCALL_ADDR_REGS. Remove IP registers.
946 * config/aarch64/aarch64.md (Ucs): Update register constraint.
948 2018-02-01 Richard Biener <rguenther@suse.de>
950 * domwalk.h (dom_walker::dom_walker): Add additional constructor
951 for specifying RPO order and allow NULL for that.
952 * domwalk.c (dom_walker::dom_walker): Likewise.
953 (dom_walker::walk): Handle NULL RPO order.
954 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
956 (rewrite_update_dom_walker): Likewise.
957 (mark_def_dom_walker): Likewise.
959 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
961 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
962 (aarch64_maybe_expand_sve_subreg_move): Declare.
963 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
964 * config/aarch64/predicates.md (aarch64_any_register_operand): New
966 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
967 that are semantically a reverse operation.
968 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
969 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
970 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
972 (aarch64_can_change_mode_class): For big-endian, forbid changes
973 between two SVE modes if they have different element sizes.
975 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
977 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
978 the TImode handling for big-endian targets.
980 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
982 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
983 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
985 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
986 Remove BSWAP handing for big-endian targets and use the form of
987 LD1RQ appropariate for the mode.
989 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
991 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
992 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
995 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
998 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
999 check for operands that need to go through aarch64_sve_reload_be.
1001 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1003 PR tree-optimization/81661
1004 PR tree-optimization/84117
1005 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1006 * tree-eh.c: Include gimplify.h.
1007 (find_trapping_overflow, replace_trapping_overflow,
1008 rewrite_to_non_trapping_overflow): New functions.
1009 * tree-vect-loop.c: Include tree-eh.h.
1010 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1011 * tree-data-ref.c: Include tree-eh.h.
1012 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1014 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1016 PR rtl-optimization/84123
1017 * combine.c (change_zero_ext): Check if hard register satisfies
1018 can_change_dest_mode before calling gen_lowpart_SUBREG.
1020 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1023 * ira.c (ira_init_register_move_cost): Remove assert.
1025 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1027 PR rtl-optimization/84071
1028 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1029 * doc/tm.texi: Regenerate.
1031 2018-01-31 Richard Biener <rguenther@suse.de>
1033 PR tree-optimization/84132
1034 * tree-data-ref.c (analyze_miv_subscript): Properly
1035 check whether evolution_function_is_affine_multivariate_p
1036 before calling gcd_of_steps_may_divide_p.
1038 2018-01-31 Julia Koval <julia.koval@intel.com>
1041 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1042 * config/i386/i386.md (rdpid_rex64) New.
1043 (rdpid): Make 32bit only.
1045 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1048 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1049 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1051 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1054 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1056 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1058 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1060 PR rtl-optimization/84071
1061 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1062 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1064 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1066 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1067 (arc_attribute_table): Add 'aux' attribute.
1068 (arc_in_small_data_p): Consider aux like variables.
1069 (arc_is_aux_reg_p): New function.
1070 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1071 (arc_get_aux_arg): New function.
1072 (prepare_move_operands): Handle aux-register access.
1073 (arc_handle_aux_attribute): New function.
1074 * doc/extend.texi (ARC Variable attributes): Add subsection.
1076 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1078 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1079 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1080 (arc_attribute_table): Add 'uncached' attribute.
1081 (arc_print_operand): Print '.di' flag for uncached memory
1083 (arc_in_small_data_p): Do not consider for small data the uncached
1085 (arc_is_uncached_mem_p): New function.
1086 * config/arc/predicates.md (compact_store_memory_operand): Check
1087 for uncached memory accesses.
1088 (nonvol_nonimm_operand): Likewise.
1089 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1091 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1094 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1095 falign-loops=): Add Optimization flag.
1097 2018-01-30 Jeff Law <law@redhat.com>
1100 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1101 INT_REGISTERS_SAVED. Check it prior to calling
1102 get_scratch_register_on_entry.
1103 (ix86_adjust_stack_and_probe): Similarly.
1104 (ix86_emit_probe_stack_range): Similarly.
1105 (ix86_expand_prologue): Corresponding changes.
1107 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1110 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1111 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1113 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1116 * lra-constraints.c (curr_insn_transform): Process AND in the
1119 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1121 PR rtl-optimization/83986
1122 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1123 dependence against last_pending_memory_flush in addition to
1126 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1128 PR tree-optimization/81611
1129 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1132 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1134 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1137 2018-01-30 Richard Biener <rguenther@suse.de>
1138 Jakub Jelinek <jakub@redhat.com>
1140 PR tree-optimization/84111
1141 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1142 inner loops added during recursion, as they don't have up-to-date
1145 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1148 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1149 (can_inline_edge_by_limits_p): ... here.
1150 (can_early_inline_edge_p, check_callers,
1151 update_caller_keys, update_callee_keys, recursive_inlining,
1152 add_new_edges_to_heap, speculation_useful_p,
1153 inline_small_functions,
1154 inline_small_functions, flatten_function,
1155 inline_to_all_callers_1): Update.
1157 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1159 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1162 2018-01-30 Richard Biener <rguenther@suse.de>
1164 PR tree-optimization/83008
1165 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1166 invariant and constant vector uses in stmts when they need
1169 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1172 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1173 * configure: Regenerate.
1175 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1177 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1179 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1180 Use gen_rtx_REG rather than gen_lowpart.
1182 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1184 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1185 rather than 0 when creating partial subregs.
1187 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1189 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1192 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1195 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1196 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1197 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1198 flags. This restores the settings used before the 2017-07-24.
1199 Turning off pre increment/decrement/modify allows IVOPTS to
1200 optimize DF/SF loops where the index is an int.
1202 2018-01-29 Richard Biener <rguenther@suse.de>
1203 Kelvin Nilsen <kelvin@gcc.gnu.org>
1206 * tree-vect-stmts.c (vectorizable_call): Don't call
1207 targetm.vectorize_builtin_md_vectorized_function if callee is
1210 2018-01-22 Carl Love <cel@us.ibm.com>
1212 * doc/extend.tex: Fix typo in second arg in
1213 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1215 2018-01-29 Richard Biener <rguenther@suse.de>
1217 PR tree-optimization/84086
1218 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1219 (flush_ssaname_freelist): When SSA names were released reset
1220 the SCEV hash table.
1222 2018-01-29 Richard Biener <rguenther@suse.de>
1224 PR tree-optimization/84057
1225 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1226 removed paths when removing edges.
1228 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1230 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1231 -mfunction-return=@var{choice}.
1233 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1236 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1238 (layout::print_source_line): Likewise.
1239 (test_get_line_width_without_trailing_whitespace): Add test cases.
1241 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1244 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1247 2018-01-26 Jim Wilson <jimw@sifive.com>
1249 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1251 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1254 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1256 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1257 and CMP + SUB-immediate -> SUBS.
1259 2018-01-26 Martin Sebor <msebor@redhat.com>
1261 PR tree-optimization/83896
1262 * tree-ssa-strlen.c (get_string_len): Rename...
1263 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1264 Avoid assuming length is constant.
1265 (handle_char_store): Use HOST_WIDE_INT for string length.
1267 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1270 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1271 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1273 2018-01-26 Richard Biener <rguenther@suse.de>
1275 PR rtl-optimization/84003
1276 * dse.c (record_store): Only record redundant stores when
1277 the earlier store aliases at least all accesses the later one does.
1279 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1281 PR rtl-optimization/83985
1282 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1283 REG_CFA_RESTORE insns.
1284 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1285 REG_CFA_RESTORE insns here.
1288 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1289 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1291 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1293 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1294 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1295 (arc_init): Likewise.
1296 (arc_override_options): Likewise.
1297 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1299 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1301 * config/arc/arc.h (TARGET_DBNZ): Define.
1302 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1303 properly set the tune attribute.
1304 (dbnz): Use TARGET_DBNZ guard.
1305 * config/arc/arc.opt (mtune): Add core3 option.
1307 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1309 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1310 recognize new pic like addresses.
1311 (arc_delegitimize_address): Clean up.
1313 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1315 * config/arc/arc-arches.def: Option mrf16 valid for all
1317 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1318 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1319 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1320 * config/arc/arc-tables.opt: Regenerate.
1321 * config/arc/arc.c (arc_conditional_register_usage): Handle
1322 reduced register file case.
1323 (arc_file_start): Set must have build attributes.
1324 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1326 * config/arc/arc.opt (mrf16): Add new option.
1327 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1328 * config/arc/genmultilib.awk: Handle new mrf16 option.
1329 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1330 * config/arc/t-multilib: Regenerate.
1331 * doc/invoke.texi (ARC Options): Document mrf16 option.
1333 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1335 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1336 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1337 (arc_attribute_table): Add 'secure_call' attribute.
1338 (arc_print_operand): Print secure call operand.
1339 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1341 (arc_is_secure_call_p): New function. * config/arc/arc.md
1342 (call_i): Add support for sjli instruction.
1343 (call_value_i): Likewise.
1344 * config/arc/constraints.md (Csc): New constraint.
1346 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1347 John Eric Martin <John.Martin@emmicro-us.com>
1349 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1350 * config/arc/arc.c (_arc_jli_section): New struct.
1351 (arc_jli_section): New type.
1352 (rc_jli_sections): New static variable.
1353 (arc_handle_jli_attribute): New function.
1354 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1355 (arc_file_end): New function.
1356 (TARGET_ASM_FILE_END): Define.
1357 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1358 (arc_add_jli_section): New function.
1359 (jli_call_scan): Likewise.
1360 (arc_reorg): Call jli_call_scan.
1361 (arc_output_addsi): Remove 'S' from printing asm operand.
1362 (arc_is_jli_call_p): New function.
1363 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1365 (movhi_insn): Likewise.
1366 (movsi_insn): Likewise.
1367 (movsi_set_cc_insn): Likewise.
1368 (loadqi_update): Likewise.
1369 (load_zeroextendqisi_update): Likewise.
1370 (load_signextendqisi_update): Likewise.
1371 (loadhi_update): Likewise.
1372 (load_zeroextendhisi_update): Likewise.
1373 (load_signextendhisi_update): Likewise.
1374 (loadsi_update): Likewise.
1375 (loadsf_update): Likewise.
1376 (movsicc_insn): Likewise.
1377 (bset_insn): Likewise.
1378 (bxor_insn): Likewise.
1379 (bclr_insn): Likewise.
1380 (bmsk_insn): Likewise.
1381 (bicsi3_insn): Likewise.
1382 (cmpsi_cc_c_insn): Likewise.
1383 (movsi_ne): Likewise.
1384 (movsi_cond_exec): Likewise.
1385 (clrsbsi2): Likewise.
1396 (call_i): Remove 'S' asm letter, add jli instruction.
1397 (call_value_i): Likewise.
1398 * config/arc/arc.op (mjli-always): New option.
1399 * config/arc/constraints.md (Cji): New constraint.
1400 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1402 (subsf3_fpx): Likewise.
1403 (mulsf3_fpx): Likewise.
1404 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1406 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1408 * doc/invoke.texi (ARC): Document mjli-always option.
1410 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1412 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1413 avoid addition with 0 and use incw and decw where possible.
1415 2018-01-26 Richard Biener <rguenther@suse.de>
1417 PR tree-optimization/81082
1418 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1419 association if it requires casting to unsigned.
1420 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1421 from fold_plusminus_mult_expr to catch important cases late when
1422 range info is available.
1424 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1426 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1427 * configure.ac (hidden_linkonce): New test.
1428 * configure: Regenerate.
1429 * config.in: Regenerate.
1431 2018-01-26 Julia Koval <julia.koval@intel.com>
1433 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1434 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1435 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1436 _mm_mask_bitshuffle_epi64_mask): Fix type.
1437 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1438 USI_FTYPE_V4DI_V4DI_USI): Remove.
1439 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1440 __builtin_ia32_vpshufbitqmb256_mask,
1441 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1442 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1443 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1445 2018-01-26 Alan Modra <amodra@gmail.com>
1448 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1449 UNSPEC_VBPERMQ. Sort other unspecs.
1451 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1453 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1455 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1458 * predict.c (drop_profile): Do not push/pop cfun; update also
1460 (handle_missing_profiles): Fix logic looking for zero profiles.
1462 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1465 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1466 on functions with #pragma omp declare simd or functions with simd
1468 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1469 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1470 Remove trailing \n from warning_at calls.
1472 2018-01-25 Tom de Vries <tom@codesourcery.com>
1475 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1476 for neutered workers.
1478 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1481 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1482 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1484 2018-01-24 Jeff Law <law@redhat.com>
1487 * i386.c (get_probe_interval): Move to earlier point.
1488 (ix86_compute_frame_layout): If -fstack-clash-protection and
1489 the frame is larger than the probe interval, then use pushes
1490 to save registers rather than reg->mem moves.
1491 (ix86_expand_prologue): Remove conditional for int_registers_saved
1494 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1497 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1498 min/max for never referenced object.
1500 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1503 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1505 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1506 attributes from DECL_ATTRIBUTES (decl) without affecting
1507 DECL_ATTRIBUTES (current_function_decl).
1508 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1509 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1511 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1513 PR tree-optimization/83979
1514 * fold-const.c (fold_comparison): Use constant_boolean_node
1515 instead of boolean_{true,false}_node.
1517 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1519 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1522 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1524 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1525 Simplify the clause that sets the length attribute.
1526 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1527 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1528 clause that sets the length attribute.
1529 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1531 2018-01-24 Tom de Vries <tom@codesourcery.com>
1534 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1535 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1536 Add strict parameter.
1537 (prevent_branch_around_nothing): Insert dummy insn between branch to
1538 label and label with no ptx insn inbetween.
1539 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1541 2018-01-24 Tom de Vries <tom@codesourcery.com>
1544 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1545 for neutered threads in warp.
1546 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1548 2018-01-24 Richard Biener <rguenther@suse.de>
1550 PR tree-optimization/83176
1551 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1554 2018-01-24 Richard Biener <rguenther@suse.de>
1556 PR tree-optimization/82819
1557 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1558 code generating pluses that are no-ops in the target precision.
1560 2018-01-24 Richard Biener <rguenther@suse.de>
1563 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1565 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1567 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1568 to merge probabilities.
1569 * predict.c (probably_never_executed): Also mark as cold functions
1570 with global 0 profile and guessed local profile.
1571 * profile-count.c (profile_probability::combine_with_count): New
1573 * profile-count.h (profile_probability::operator*,
1574 profile_probability::operator*=, profile_probability::operator/,
1575 profile_probability::operator/=): Reduce precision to adjusted
1576 and set value to guessed on contradictory divisions.
1577 (profile_probability::combine_with_freq): Remove.
1578 (profile_probability::combine_wiht_count): Declare.
1579 (profile_count::force_nonzero):: Set to adjusted.
1580 (profile_count::probability_in):: Set quality to adjusted.
1581 * tree-ssa-tail-merge.c (replace_block_by): Use
1584 2018-01-23 Andrew Waterman <andrew@sifive.com>
1585 Jim Wilson <jimw@sifive.com>
1587 * config/riscv/riscv.c (riscv_stack_boundary): New.
1588 (riscv_option_override): Set riscv_stack_boundary. Handle
1589 riscv_preferred_stack_boundary_arg.
1590 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1591 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1592 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1593 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1594 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1595 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1597 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1600 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1601 of struct ix86_frame.
1602 (ix86_expand_epilogue): Likewise. Add a local variable for
1603 the reg_save_offset field in struct ix86_frame.
1605 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1607 PR tree-optimization/82604
1608 * tree-loop-distribution.c (enum partition_kind): New enum item
1609 PKIND_PARTIAL_MEMSET.
1610 (partition_builtin_p): Support above new enum item.
1611 (generate_code_for_partition): Ditto.
1612 (compute_access_range): Differentiate cases that equality can be
1613 proven at all loops, the innermost loops or no loops.
1614 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1615 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1616 (finalize_partitions, distribute_loop): Don't fuse partition of
1617 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1618 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1621 2018-01-23 Martin Liska <mliska@suse.cz>
1623 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1624 order to ignore the predictor.
1625 (PRED_POLYMORPHIC_CALL): Likewise.
1626 (PRED_RECURSIVE_CALL): Likewise.
1628 2018-01-23 Martin Liska <mliska@suse.cz>
1630 * tree-profile.c (tree_profiling): Print function header to
1631 aware reader which function we are working on.
1632 * value-prof.c (gimple_find_values_to_profile): Do not print
1633 not interesting value histograms.
1635 2018-01-23 Martin Liska <mliska@suse.cz>
1637 * profile-count.h (enum profile_quality): Add
1638 profile_uninitialized as the first value. Do not number values
1639 as they are zero based.
1640 (profile_count::verify): Update sanity check.
1641 (profile_probability::verify): Likewise.
1643 2018-01-23 Nathan Sidwell <nathan@acm.org>
1645 * doc/invoke.texi (ffor-scope): Deprecate.
1647 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1649 PR tree-optimization/83510
1650 * domwalk.c (set_all_edges_as_executable): New function.
1651 (dom_walker::dom_walker): Convert bool param
1652 "skip_unreachable_blocks" to enum reachability. Move setup of
1653 edge flags to set_all_edges_as_executable and only do it when
1654 reachability is REACHABLE_BLOCKS.
1655 * domwalk.h (enum dom_walker::reachability): New enum.
1656 (dom_walker::dom_walker): Convert bool param
1657 "skip_unreachable_blocks" to enum reachability.
1658 (set_all_edges_as_executable): New decl.
1659 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1660 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1662 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1663 but converting true to REACHABLE_BLOCKS.
1664 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1666 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1667 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1668 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1670 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1671 if check_all_array_refs will be called.
1673 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1675 * tree.c (selftest::test_location_wrappers): Add more test
1678 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1680 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1681 (selftest::test_bit_in_range): Likewise.
1683 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1686 * doc/sourcebuild.texi (vect_float): Say that the selector
1687 only describes the situation when -funsafe-math-optimizations is on.
1688 (vect_float_strict): Document.
1690 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1692 PR tree-optimization/83965
1693 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1694 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1695 instead of checking only for a reduction.
1696 (vect_recog_widen_sum_pattern): Likewise.
1698 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1700 * predict.c (probably_never_executed): Only use precise profile info.
1701 (compute_function_frequency): Skip after inlining hack since we now
1702 have quality checking.
1704 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1706 * profile-count.h (profile_probability::very_unlikely,
1707 profile_probability::unlikely, profile_probability::even): Set
1708 precision to guessed.
1710 2018-01-23 Richard Biener <rguenther@suse.de>
1712 PR tree-optimization/83963
1713 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1714 Properly terminate dominator walk when crossing the exit edge not
1715 when visiting its source block.
1717 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1720 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1721 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1723 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1725 PR tree-optimization/83957
1726 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1727 semicolon after for body surrounded by braces.
1729 PR tree-optimization/83081
1730 * profile-count.h (profile_probability::split): New method.
1731 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1732 Use profile_probability::split.
1733 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1734 when splitting a single conditional jump into 2.
1736 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1738 PR tree-optimization/69452
1739 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1742 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1744 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1745 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1746 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1748 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1750 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1751 * config/rl78/rl78.md (movdi): New define_expand.
1752 * config/rl78/rl78.c (rl78_split_movdi): New function.
1754 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1757 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1759 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1760 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1761 128-bit to produce an UNSPEC move to get the double word with the
1762 signbit and then a shift directly to do signbit.
1763 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1764 implementation with a new version that just does either a direct
1765 move or a regular move. Move memory interface to separate insns.
1766 Move insns so they are next to the expander.
1767 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1768 with signbit move. Split big and little endian case.
1769 (signbit<mode>2_dm_mem_le): Likewise.
1770 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1771 (signbit<mode>2_dm2): Likewise.
1773 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1775 * config/rl78/rl78.md (anddi3): New define_expand.
1777 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1779 * config/rl78/rl78.md (umindi3): New define_expand.
1781 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1783 * config/rl78/rl78.md (smindi3): New define_expand.
1785 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1787 * config/rl78/rl78.md (smaxdi3): New define_expand.
1789 2018-01-22 Carl Love <cel@us.ibm.com>
1791 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1792 LVX_V1TI): Add macro expansion.
1793 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1794 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1795 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1796 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1797 Change check to determine if the instruction is a byte reversing
1798 entry. Fix typo in comment.
1799 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1800 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1801 Add def_builtin calls for new builtins.
1802 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1803 Add define_insn expansion.
1805 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1807 * config/rl78/rl78.md (umaxdi3): New define_expand.
1809 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1811 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
1812 for non-QImode registers.
1814 2018-01-22 Richard Biener <rguenther@suse.de>
1816 PR tree-optimization/83963
1817 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1818 including the loop exit block.
1819 (scop_detection::merge_sese): Likewise.
1820 (scop_detection::add_scop): Do it here instead.
1822 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1824 * doc/sourcebuild.texi (arm_softfloat): Document.
1826 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1829 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1830 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1831 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1833 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1834 David Edelsohn <dje.gcc@gmail.com>
1837 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1838 Change "crset eq" to "crset 2".
1839 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1840 (*call_indirect_aix<mode>_nospec): Likewise.
1841 (*call_value_indirect_aix<mode>_nospec): Likewise.
1842 (*call_indirect_elfv2<mode>_nospec): Likewise.
1843 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1844 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1845 change assembly output from . to $.
1846 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1847 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1848 (*tablejump<mode>_internal1_nospec): Likewise.
1850 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1853 * config/sh/sh_optimize_sett_clrt.cc:
1854 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1856 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1858 PR tree-optimization/83940
1859 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1860 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1861 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1862 use it to pass back the definition type.
1863 (vect_check_store_rhs): Likewise rhs_dt_out.
1864 (vect_build_gather_load_calls): Add a mask_dt argument and use
1865 it instead of a call to vect_is_simple_use.
1866 (vectorizable_store): Update calls to vect_check_load_store_mask
1867 and vect_check_store_rhs. Use the dt returned by the latter instead
1868 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1869 instead of calls to vect_is_simple_use. Pass the scalar rather
1870 than the vector operand to vect_is_simple_use when handling
1871 second and subsequent copies of an rhs value.
1872 (vectorizable_load): Update calls to vect_check_load_store_mask
1873 and vect_build_gather_load_calls. Use the cached mask_dt and
1874 gs_info.offset_dt instead of calls to vect_is_simple_use.
1876 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1879 * tree-emutls.c: Include gimplify.h.
1880 (lower_emutls_2): New function.
1881 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1882 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1883 it before further processing.
1886 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1887 UINTVAL (trueop1) instead of INTVAL (op1).
1889 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1893 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1894 INCOMING_FRAME_SP_OFFSET if not defined.
1895 (scan_trace): Add ENTRY argument. If true and
1896 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1897 emit a note to adjust the CFA offset.
1898 (create_cfi_notes): Adjust scan_trace callers.
1899 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1900 INCOMING_FRAME_SP_OFFSET in the CIE.
1901 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1902 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1904 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1905 * doc/tm.texi: Regenerated.
1907 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1909 PR rtl-optimization/83147
1910 * lra-constraints.c (remove_inheritance_pseudos): Use
1911 lra_substitute_pseudo_within_insn.
1913 2018-01-19 Tom de Vries <tom@codesourcery.com>
1914 Cesar Philippidis <cesar@codesourcery.com>
1917 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1919 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1922 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1923 spaces for function labels.
1925 2018-01-19 Martin Liska <mliska@suse.cz>
1927 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1928 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1929 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1930 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1931 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1932 (PRED_CONST_RETURN): Change from 69 to 65.
1933 (PRED_NULL_RETURN): Change from 91 to 71.
1934 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1935 (PRED_LOOP_GUARD): Change from 66 to 73.
1937 2018-01-19 Martin Liska <mliska@suse.cz>
1939 * predict.c (predict_insn_def): Add new assert.
1940 (struct branch_predictor): Change type to signed integer.
1941 (test_prediction_value_range): Amend test to cover
1943 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1944 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1945 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1946 (PRED_LOOP_IV_COMPARE): Likewise.
1947 * predict.h (PROB_UNINITIALIZED): Define new constant.
1949 2018-01-19 Martin Liska <mliska@suse.cz>
1951 * predict.c (dump_prediction): Add new format for
1952 analyze_brprob.py script which is enabled with -details
1954 * profile-count.h (precise_p): New function.
1956 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1958 PR tree-optimization/83922
1959 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1960 there are no statements that need masking.
1961 (vect_active_double_reduction_p): New function.
1962 (vect_analyze_loop_operations): Use it when handling phis that
1963 are not in the loop header.
1965 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1967 PR tree-optimization/83914
1968 * tree-vect-loop.c (vectorizable_induction): Don't convert
1969 init_expr or apply the peeling adjustment for inductions
1970 that are nested within the vectorized loop.
1972 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1974 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1977 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1981 * function.h (gimplify_parameters): Add gimple_seq * argument.
1982 * function.c: Include gimple.h and options.h.
1983 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1984 for the added local temporaries if needed.
1985 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1986 if there are any parameter cleanups, wrap whole body into a
1987 try/finally with the cleanups.
1989 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1992 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1993 Use GET_MODE_CLASS for scalar floating point.
1995 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1999 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2000 Fix call of call_cgraph_insertion_hooks.
2002 2018-01-18 Martin Sebor <msebor@redhat.com>
2004 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2006 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2009 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2012 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2015 * common.opt: (-ffile-prefix-map): New option.
2016 * opts.c (common_handle_option): Defer it.
2017 * opts-global.c (handle_common_deferred_options): Handle it.
2018 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2019 * file-prefix-map.h: New file.
2020 (remap_debug_filename, add_debug_prefix_map): ...here.
2021 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2022 * final.c (debug_prefix_map, add_debug_prefix_map
2023 remap_debug_filename): Move to...
2024 * file-prefix-map.c: New file.
2025 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2026 generalize, get rid of alloca(), use strrchr() instead of strchr().
2027 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2028 Implement in terms of add_prefix_map().
2029 (remap_macro_filename, remap_debug_filename): Implement in term of
2031 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2032 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2033 * dbxout.c: Include file-prefix-map.h.
2034 * varasm.c: Likewise.
2035 * vmsdbgout.c: Likewise.
2036 * xcoffout.c: Likewise.
2037 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2038 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2039 * doc/invoke.texi (-ffile-prefix-map): Document.
2040 (-fdebug-prefix-map): Update description.
2042 2018-01-18 Martin Liska <mliska@suse.cz>
2044 * config/i386/i386.c (indirect_thunk_name): Document that also
2046 (output_indirect_thunk): Document why both instructions
2047 (pause and lfence) are generated.
2049 2018-01-18 Richard Biener <rguenther@suse.de>
2051 PR tree-optimization/83887
2052 * graphite-scop-detection.c
2053 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2054 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2055 (scop_detection::merge_sese): Re-implement with a flood-fill
2056 algorithm that properly finds a SESE region if it exists.
2058 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2061 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2062 pointer_diff optimizations use view_convert instead of convert.
2064 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2066 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2067 Generate different code for -mno-speculate-indirect-jumps.
2068 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2069 (*call_indirect_aix<mode>): Disable for
2070 -mno-speculate-indirect-jumps.
2071 (*call_indirect_aix<mode>_nospec): New define_insn.
2072 (*call_value_indirect_aix<mode>): Disable for
2073 -mno-speculate-indirect-jumps.
2074 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2075 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2076 -mno-speculate-indirect-jumps.
2077 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2079 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2081 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2082 long double type, set the flags for noting the default long double
2083 type, even if we don't pass or return a long double type.
2085 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2088 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2091 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2093 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2094 support for merge[hl].
2095 (fold_mergehl_helper): New helper function.
2096 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2097 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2098 (altivec_vmrglw_direct): Add xxmrglw insn.
2100 2018-01-17 Andrew Waterman <andrew@sifive.com>
2102 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2103 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2105 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2108 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2109 call the lto_location_cache before reading the
2110 DECL_SOURCE_LOCATION of the types.
2112 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2113 Richard Sandiford <richard.sandiford@linaro.org>
2115 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2116 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2117 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2118 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2119 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2121 * config/aarch64/constraints.md (aarch64_movti_operand):
2123 * config/aarch64/predicates.md (Uti): Add new constraint.
2125 2018-01-17 Carl Love <cel@us.ibm.com>
2126 * config/rs6000/vsx.md (define_expand xl_len_r,
2127 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2128 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2130 (define_expand, define_insn): Move the shift left from the
2131 define_insn to the define_expand for lxvl and stxvl instructions.
2132 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2133 and XL_LEN_R definitions to PURE.
2135 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2137 * config/i386/i386.c (indirect_thunk_name): Declare regno
2138 as unsigned int. Compare regno with INVALID_REGNUM.
2139 (output_indirect_thunk): Ditto.
2140 (output_indirect_thunk_function): Ditto.
2141 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2142 in the call to output_indirect_thunk_function.
2144 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2147 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2148 rather than the size of inner_type to determine the stack slot size
2149 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2151 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2154 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2157 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2159 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2160 endian Linux systems to optionally enable multilibs for selecting
2161 the long double type if the user configured an explicit type.
2162 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2163 have no long double multilibs if not defined.
2164 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2165 warn if the user used -mabi={ieee,ibm}longdouble and we built
2166 multilibs for long double.
2167 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2168 appropriate multilib option.
2169 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2171 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2172 for building long double multilibs.
2173 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2175 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2177 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2180 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2182 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2185 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2188 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2191 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2193 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2194 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2196 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2198 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2199 different rtl trees depending on TARGET_64BIT.
2200 (rs6000_gen_lvx): Likewise.
2202 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2204 * config/visium/visium.md (nop): Tweak comment.
2205 (hazard_nop): Likewise.
2207 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2209 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2210 -mspeculate-indirect-jumps.
2211 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2212 for -mno-speculate-indirect-jumps.
2213 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2214 (*call_value_indirect_elfv2<mode>): Disable for
2215 -mno-speculate-indirect-jumps.
2216 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2217 (indirect_jump): Emit different RTL for
2218 -mno-speculate-indirect-jumps.
2219 (*indirect_jump<mode>): Disable for
2220 -mno-speculate-indirect-jumps.
2221 (*indirect_jump<mode>_nospec): New define_insn.
2222 (tablejump): Emit different RTL for
2223 -mno-speculate-indirect-jumps.
2224 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2225 (tablejumpsi_nospec): New define_expand.
2226 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2227 (tablejumpdi_nospec): New define_expand.
2228 (*tablejump<mode>_internal1): Disable for
2229 -mno-speculate-indirect-jumps.
2230 (*tablejump<mode>_internal1_nospec): New define_insn.
2231 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2234 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2236 * caller-save.c (insert_save): Drop unnecessary parameter. All
2239 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2240 Richard Biener <rguenth@suse.de>
2243 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2244 return early, inline manually is_gimple_sizepos. Make sure if we
2245 call gimplify_expr we don't end up with a gimple constant.
2246 * tree.c (variably_modified_type_p): Don't return true for
2247 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2248 * gimplify.h (is_gimple_sizepos): Remove.
2250 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2252 PR tree-optimization/83857
2253 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2254 vectorizable_live_operation for pure SLP statements.
2255 (vectorizable_live_operation): Handle PHIs.
2257 2018-01-16 Richard Biener <rguenther@suse.de>
2259 PR tree-optimization/83867
2260 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2261 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2263 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2266 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2267 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2268 If off is not INTEGER_CST, issue a may not be aligned warning
2269 rather than isn't aligned. Use isn%'t rather than isn't.
2270 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2272 <case MULT_EXPR>: Improve the case when bottom and one of the
2273 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2274 operand, in that case check if the other operand is multiple of
2275 bottom divided by the INTEGER_CST operand.
2277 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2280 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2281 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2282 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2283 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2284 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2285 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2286 (pa_function_arg_size): New function.
2288 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2290 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2291 in a separate statement.
2293 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2295 PR tree-optimization/83847
2296 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2297 group gathers and scatters.
2299 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2301 PR rtl-optimization/86620
2302 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2304 PR rtl-optimization/83213
2305 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2306 to last if both are JUMP_INSNs.
2308 PR tree-optimization/83843
2309 * gimple-ssa-store-merging.c
2310 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2311 store_immediate_info for bswap/nop orig_stores.
2313 2018-01-15 Andrew Waterman <andrew@sifive.com>
2315 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2317 <UDIV>: Increase cost if !TARGET_DIV.
2319 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2321 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2322 (define_attr "cr_logical_3op"): New.
2323 (cceq_ior_compare): Adjust.
2324 (cceq_ior_compare_complement): Adjust.
2325 (*cceq_rev_compare): Adjust.
2326 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2327 (is_cracked_insn): Adjust.
2328 (insn_must_be_first_in_group): Adjust.
2329 * config/rs6000/40x.md: Adjust.
2330 * config/rs6000/440.md: Adjust.
2331 * config/rs6000/476.md: Adjust.
2332 * config/rs6000/601.md: Adjust.
2333 * config/rs6000/603.md: Adjust.
2334 * config/rs6000/6xx.md: Adjust.
2335 * config/rs6000/7450.md: Adjust.
2336 * config/rs6000/7xx.md: Adjust.
2337 * config/rs6000/8540.md: Adjust.
2338 * config/rs6000/cell.md: Adjust.
2339 * config/rs6000/e300c2c3.md: Adjust.
2340 * config/rs6000/e500mc.md: Adjust.
2341 * config/rs6000/e500mc64.md: Adjust.
2342 * config/rs6000/e5500.md: Adjust.
2343 * config/rs6000/e6500.md: Adjust.
2344 * config/rs6000/mpc.md: Adjust.
2345 * config/rs6000/power4.md: Adjust.
2346 * config/rs6000/power5.md: Adjust.
2347 * config/rs6000/power6.md: Adjust.
2348 * config/rs6000/power7.md: Adjust.
2349 * config/rs6000/power8.md: Adjust.
2350 * config/rs6000/power9.md: Adjust.
2351 * config/rs6000/rs64.md: Adjust.
2352 * config/rs6000/titan.md: Adjust.
2354 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2356 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2357 ix86_indirect_branch_register logic.
2359 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2361 * config/i386/constraints.md (Bs): Update
2362 ix86_indirect_branch_register check. Don't check
2363 ix86_indirect_branch_register with GOT_memory_operand.
2365 * config/i386/predicates.md (GOT_memory_operand): Don't check
2366 ix86_indirect_branch_register here.
2367 (GOT32_symbol_operand): Likewise.
2369 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2371 * config/i386/predicates.md (constant_call_address_operand):
2372 Rewrite ix86_indirect_branch_register logic.
2373 (sibcall_insn_operand): Likewise.
2375 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2377 * config/i386/constraints.md (Bs): Replace
2378 ix86_indirect_branch_thunk_register with
2379 ix86_indirect_branch_register.
2381 * config/i386/i386.md (indirect_jump): Likewise.
2382 (tablejump): Likewise.
2383 (*sibcall_memory): Likewise.
2384 (*sibcall_value_memory): Likewise.
2385 Peepholes of indirect call and jump via memory: Likewise.
2386 * config/i386/i386.opt: Likewise.
2387 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2388 (GOT_memory_operand): Likewise.
2389 (call_insn_operand): Likewise.
2390 (sibcall_insn_operand): Likewise.
2391 (GOT32_symbol_operand): Likewise.
2393 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2396 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2397 type rather than type addr's type points to.
2398 (expand_omp_atomic_mutex): Likewise.
2399 (expand_omp_atomic): Likewise.
2401 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2404 * config/i386/i386.c (output_indirect_thunk_function): Use
2405 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2406 for __x86_return_thunk.
2408 2018-01-15 Richard Biener <rguenther@suse.de>
2411 * expmed.c (extract_bit_field_1): Fix typo.
2413 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2416 * config/arm/iterators.md (VF): New mode iterator.
2417 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2418 Remove integer-related logic from pattern.
2419 (neon_vabd<mode>_3): Likewise.
2421 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2424 * common.opt (fstrict-overflow): No longer an alias.
2425 (fwrapv-pointer): New option.
2426 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2427 also for pointer types based on flag_wrapv_pointer.
2428 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2429 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2430 opts->x_flag_wrapv got set.
2431 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2432 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2433 POINTER_TYPE_OVERFLOW_UNDEFINED.
2434 * match.pd: Likewise in address comparison pattern.
2435 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2437 2018-01-15 Richard Biener <rguenther@suse.de>
2440 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2441 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2442 Reset type names to their identifier if their TYPE_DECL doesn't
2443 have linkage (and thus is used for ODR and devirt).
2444 (save_debug_info_for_decl): Remove.
2445 (save_debug_info_for_type): Likewise.
2446 (add_tree_to_fld_list): Adjust.
2447 * tree-pretty-print.c (dump_generic_node): Make dumping of
2448 type names more robust.
2450 2018-01-15 Richard Biener <rguenther@suse.de>
2452 * BASE-VER: Bump to 8.0.1.
2454 2018-01-14 Martin Sebor <msebor@redhat.com>
2457 * builtins.c (check_access): Avoid warning when the no-warning bit
2460 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2462 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2463 * ira-color (allocno_hard_regs_compare): Likewise.
2465 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2468 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2469 Use .pushsection/.popsection.
2471 2018-01-14 Martin Sebor <msebor@redhat.com>
2474 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2476 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2478 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2479 entry from extra_headers.
2480 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2481 extra_headers, make the list bitwise identical to the i?86-*-* one.
2483 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2485 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2486 -mcmodel=large with -mindirect-branch=thunk,
2487 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2488 -mfunction-return=thunk-extern.
2489 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2490 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2491 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2493 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2495 * config/i386/i386.c (print_reg): Print the name of the full
2496 integer register without '%'.
2497 (ix86_print_operand): Handle 'V'.
2498 * doc/extend.texi: Document 'V' modifier.
2500 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2502 * config/i386/constraints.md (Bs): Disallow memory operand for
2503 -mindirect-branch-register.
2505 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2506 (GOT_memory_operand): Likewise.
2507 (call_insn_operand): Likewise.
2508 (sibcall_insn_operand): Likewise.
2509 (GOT32_symbol_operand): Likewise.
2510 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2511 for -mindirect-branch-register.
2512 (tablejump): Likewise.
2513 (*sibcall_memory): Likewise.
2514 (*sibcall_value_memory): Likewise.
2515 Disallow peepholes of indirect call and jump via memory for
2516 -mindirect-branch-register.
2517 (*call_pop): Replace m with Bw.
2518 (*call_value_pop): Likewise.
2519 (*sibcall_pop_memory): Replace m with Bs.
2520 * config/i386/i386.opt (mindirect-branch-register): New option.
2521 * doc/invoke.texi: Document -mindirect-branch-register option.
2523 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2525 * config/i386/i386-protos.h (ix86_output_function_return): New.
2526 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2527 set function_return_type.
2528 (indirect_thunk_name): Add ret_p to indicate thunk for function
2530 (output_indirect_thunk_function): Pass false to
2531 indirect_thunk_name.
2532 (ix86_output_indirect_branch_via_reg): Likewise.
2533 (ix86_output_indirect_branch_via_push): Likewise.
2534 (output_indirect_thunk_function): Create alias for function
2535 return thunk if regno < 0.
2536 (ix86_output_function_return): New function.
2537 (ix86_handle_fndecl_attribute): Handle function_return.
2538 (ix86_attribute_table): Add function_return.
2539 * config/i386/i386.h (machine_function): Add
2540 function_return_type.
2541 * config/i386/i386.md (simple_return_internal): Use
2542 ix86_output_function_return.
2543 (simple_return_internal_long): Likewise.
2544 * config/i386/i386.opt (mfunction-return=): New option.
2545 (indirect_branch): Mention -mfunction-return=.
2546 * doc/extend.texi: Document function_return function attribute.
2547 * doc/invoke.texi: Document -mfunction-return= option.
2549 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2551 * config/i386/i386-opts.h (indirect_branch): New.
2552 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2553 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2554 with local indirect jump when converting indirect call and jump.
2555 (ix86_set_indirect_branch_type): New.
2556 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2557 (indirectlabelno): New.
2558 (indirect_thunk_needed): Likewise.
2559 (indirect_thunk_bnd_needed): Likewise.
2560 (indirect_thunks_used): Likewise.
2561 (indirect_thunks_bnd_used): Likewise.
2562 (INDIRECT_LABEL): Likewise.
2563 (indirect_thunk_name): Likewise.
2564 (output_indirect_thunk): Likewise.
2565 (output_indirect_thunk_function): Likewise.
2566 (ix86_output_indirect_branch_via_reg): Likewise.
2567 (ix86_output_indirect_branch_via_push): Likewise.
2568 (ix86_output_indirect_branch): Likewise.
2569 (ix86_output_indirect_jmp): Likewise.
2570 (ix86_code_end): Call output_indirect_thunk_function if needed.
2571 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2573 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2574 (ix86_attribute_table): Add indirect_branch.
2575 * config/i386/i386.h (machine_function): Add indirect_branch_type
2576 and has_local_indirect_jump.
2577 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2579 (tablejump): Likewise.
2580 (*indirect_jump): Use ix86_output_indirect_jmp.
2581 (*tablejump_1): Likewise.
2582 (simple_return_indirect_internal): Likewise.
2583 * config/i386/i386.opt (mindirect-branch=): New option.
2584 (indirect_branch): New.
2587 (thunk-inline): Likewise.
2588 (thunk-extern): Likewise.
2589 * doc/extend.texi: Document indirect_branch function attribute.
2590 * doc/invoke.texi: Document -mindirect-branch= option.
2592 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2595 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2597 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2599 * ipa-inline.c (want_inline_small_function_p): Return false if
2600 inlining has already failed with CIF_FINAL_ERROR.
2601 (update_caller_keys): Call want_inline_small_function_p before
2603 (update_callee_keys): Likewise.
2605 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2607 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2609 (rs6000_quadword_masked_address_p): Likewise.
2610 (quad_aligned_load_p): Likewise.
2611 (quad_aligned_store_p): Likewise.
2612 (const_load_sequence_p): Add comment to describe the outer-most loop.
2613 (mimic_memory_attributes_and_flags): New function.
2614 (rs6000_gen_stvx): Likewise.
2615 (replace_swapped_aligned_store): Likewise.
2616 (rs6000_gen_lvx): Likewise.
2617 (replace_swapped_aligned_load): Likewise.
2618 (replace_swapped_load_constant): Capitalize argument name in
2619 comment describing this function.
2620 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2621 and stores that access quad-word aligned addresses and replace
2622 with stvx or lvx instructions when appropriate.
2623 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2624 New function prototype.
2625 (rs6000_quadword_masked_address_p): Likewise.
2626 (rs6000_gen_lvx): Likewise.
2627 (rs6000_gen_stvx): Likewise.
2628 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2629 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2630 when memory address is aligned.
2631 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2632 this split to select lvx instruction when memory address is aligned.
2633 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2634 instruction when memory address is aligned.
2635 (*vsx_le_perm_load_v16qi): Likewise.
2636 (four unnamed splitters): Modify to select the stvx instruction
2637 when memory is aligned.
2639 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2641 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2642 which appears in the queue multiple times.
2644 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2645 Alan Hayward <alan.hayward@arm.com>
2646 David Sherwood <david.sherwood@arm.com>
2648 * tree-vectorizer.h (vec_lower_bound): New structure.
2649 (_loop_vec_info): Add check_nonzero and lower_bounds.
2650 (LOOP_VINFO_CHECK_NONZERO): New macro.
2651 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2652 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2653 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2654 fields. Make seg_len the distance travelled, not including the
2656 (dr_direction_indicator): Declare.
2657 (dr_zero_step_indicator): Likewise.
2658 (dr_known_forward_stride_p): Likewise.
2659 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2661 (runtime_alias_check_p): Allow runtime alias checks with
2663 (operator ==): Compare access_size and align.
2664 (prune_runtime_alias_test_list): Rework for new distinction between
2665 the access_size and seg_len.
2666 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2668 (get_segment_min_max): New function.
2669 (create_intersect_range_checks): Use it.
2670 (dr_step_indicator): New function.
2671 (dr_direction_indicator): Likewise.
2672 (dr_zero_step_indicator): Likewise.
2673 (dr_known_forward_stride_p): Likewise.
2674 * tree-loop-distribution.c (data_ref_segment_size): Return
2675 DR_STEP * (niters - 1).
2676 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2678 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2679 (vect_preserves_scalar_order_p): New function, split out from...
2680 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2681 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2682 (vect_vfa_access_size): New function.
2683 (vect_vfa_align): Likewise.
2684 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2685 (dump_lower_bound): New function.
2686 (vect_check_lower_bound): Likewise.
2687 (vect_small_gap_p): Likewise.
2688 (vectorizable_with_step_bound_p): Likewise.
2689 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2690 depencies if the vectorization factor is 1. Convert the checks
2691 for nonzero steps into checks on the bounds of DR_STEP. Try using
2692 a bunds check for variable steps if the minimum required step is
2693 relatively small. Update calls to the dr_with_seg_len
2694 constructor and to vect_compile_time_alias.
2695 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2697 (vect_loop_versioning): Call it.
2698 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2700 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2702 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2703 Alan Hayward <alan.hayward@arm.com>
2704 David Sherwood <david.sherwood@arm.com>
2706 * doc/sourcebuild.texi (vect_scatter_store): Document.
2707 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2709 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2711 * genopinit.c (main): Add supports_vec_scatter_store and
2712 supports_vec_scatter_store_cached to target_optabs.
2713 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2714 IFN_MASK_SCATTER_STORE.
2715 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2717 * internal-fn.h (internal_store_fn_p): Declare.
2718 (internal_fn_stored_value_index): Likewise.
2719 * internal-fn.c (scatter_store_direct): New macro.
2720 (expand_scatter_store_optab_fn): New function.
2721 (direct_scatter_store_optab_supported_p): New macro.
2722 (internal_store_fn_p): New function.
2723 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2724 IFN_MASK_SCATTER_STORE.
2725 (internal_fn_mask_index): Likewise.
2726 (internal_fn_stored_value_index): New function.
2727 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2729 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2730 * optabs-query.c (supports_vec_scatter_store_p): New function.
2731 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2732 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2733 true for scatter stores.
2734 (vect_gather_scatter_fn_p): Handle scatter stores too.
2735 (vect_check_gather_scatter): Consider using scatter stores if
2736 supports_vec_scatter_store_p.
2737 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2739 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2740 internal_fn_stored_value_index.
2741 (check_load_store_masking): Handle scatter stores too.
2742 (vect_get_store_rhs): Make public.
2743 (vectorizable_call): Use internal_store_fn_p.
2744 (vectorizable_store): Handle scatter store internal functions.
2745 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2746 when deciding whether the end of the group has been reached.
2747 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2748 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2749 (mask_scatter_store<mode>): New insns.
2751 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2752 Alan Hayward <alan.hayward@arm.com>
2753 David Sherwood <david.sherwood@arm.com>
2755 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2756 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2757 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2759 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2760 Use vect_truncate_gather_scatter_offset if we can't treat the
2761 operation as a normal gather load or scatter store.
2762 (get_group_load_store_type): Take the gather_scatter_info
2763 as argument. Try using a gather load or scatter store for
2764 single-element groups.
2765 (get_load_store_type): Update calls to get_group_load_store_type
2766 and vect_use_strided_gather_scatters_p.
2768 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2769 Alan Hayward <alan.hayward@arm.com>
2770 David Sherwood <david.sherwood@arm.com>
2772 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2773 optional tree argument.
2774 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2776 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2777 but continue to use the current value as a fallback.
2778 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2779 to compare the updates.
2780 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2781 (get_load_store_type): Use it when handling a strided access.
2782 (vect_get_strided_load_store_ops): New function.
2783 (vect_get_data_ptr_increment): Likewise.
2784 (vectorizable_load): Handle strided gather loads. Always pass
2785 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2787 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2788 Alan Hayward <alan.hayward@arm.com>
2789 David Sherwood <david.sherwood@arm.com>
2791 * doc/md.texi (gather_load@var{m}): Document.
2792 (mask_gather_load@var{m}): Likewise.
2793 * genopinit.c (main): Add supports_vec_gather_load and
2794 supports_vec_gather_load_cached to target_optabs.
2795 * optabs-tree.c (init_tree_optimization_optabs): Use
2796 ggc_cleared_alloc to allocate target_optabs.
2797 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2798 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2800 * internal-fn.h (internal_load_fn_p): Declare.
2801 (internal_gather_scatter_fn_p): Likewise.
2802 (internal_fn_mask_index): Likewise.
2803 (internal_gather_scatter_fn_supported_p): Likewise.
2804 * internal-fn.c (gather_load_direct): New macro.
2805 (expand_gather_load_optab_fn): New function.
2806 (direct_gather_load_optab_supported_p): New macro.
2807 (direct_internal_fn_optab): New function.
2808 (internal_load_fn_p): Likewise.
2809 (internal_gather_scatter_fn_p): Likewise.
2810 (internal_fn_mask_index): Likewise.
2811 (internal_gather_scatter_fn_supported_p): Likewise.
2812 * optabs-query.c (supports_at_least_one_mode_p): New function.
2813 (supports_vec_gather_load_p): Likewise.
2814 * optabs-query.h (supports_vec_gather_load_p): Declare.
2815 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2816 and memory_type field.
2817 (NUM_PATTERNS): Bump to 15.
2818 * tree-vect-data-refs.c: Include internal-fn.h.
2819 (vect_gather_scatter_fn_p): New function.
2820 (vect_describe_gather_scatter_call): Likewise.
2821 (vect_check_gather_scatter): Try using internal functions for
2822 gather loads. Recognize existing calls to a gather load function.
2823 (vect_analyze_data_refs): Consider using gather loads if
2824 supports_vec_gather_load_p.
2825 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2826 (vect_get_gather_scatter_offset_type): Likewise.
2827 (vect_convert_mask_for_vectype): Likewise.
2828 (vect_add_conversion_to_patterm): Likewise.
2829 (vect_try_gather_scatter_pattern): Likewise.
2830 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2831 (vect_vect_recog_func_ptrs): Add it.
2832 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2833 internal_fn_mask_index and internal_gather_scatter_fn_p.
2834 (check_load_store_masking): Take the gather_scatter_info as an
2835 argument and handle gather loads.
2836 (vect_get_gather_scatter_ops): New function.
2837 (vectorizable_call): Check internal_load_fn_p.
2838 (vectorizable_load): Likewise. Handle gather load internal
2840 (vectorizable_store): Update call to check_load_store_masking.
2841 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2842 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2843 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2844 (aarch64_gather_scale_operand_d): New predicates.
2845 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2846 (mask_gather_load<mode>): New insns.
2848 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2849 Alan Hayward <alan.hayward@arm.com>
2850 David Sherwood <david.sherwood@arm.com>
2852 * optabs.def (fold_left_plus_optab): New optab.
2853 * doc/md.texi (fold_left_plus_@var{m}): Document.
2854 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2855 * internal-fn.c (fold_left_direct): Define.
2856 (expand_fold_left_optab_fn): Likewise.
2857 (direct_fold_left_optab_supported_p): Likewise.
2858 * fold-const-call.c (fold_const_fold_left): New function.
2859 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2860 * tree-parloops.c (valid_reduction_p): New function.
2861 (gather_scalar_reductions): Use it.
2862 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2863 (vect_finish_replace_stmt): Declare.
2864 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2865 (needs_fold_left_reduction_p): New function, split out from...
2866 (vect_is_simple_reduction): ...here. Accept reductions that
2867 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2868 (vect_force_simple_reduction): Also store the reduction type in
2869 the assignment's STMT_VINFO_REDUC_TYPE.
2870 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2871 (merge_with_identity): New function.
2872 (vect_expand_fold_left): Likewise.
2873 (vectorize_fold_left_reduction): Likewise.
2874 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2875 scalar phi in place for it. Check for target support and reject
2876 cases that would reassociate the operation. Defer the transform
2877 phase to vectorize_fold_left_reduction.
2878 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2879 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2880 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2882 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2884 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2885 call to ifc_temp_var.
2887 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2888 Alan Hayward <alan.hayward@arm.com>
2889 David Sherwood <david.sherwood@arm.com>
2891 * target.def (legitimize_address_displacement): Take the original
2892 offset as a poly_int.
2893 * targhooks.h (default_legitimize_address_displacement): Update
2895 * targhooks.c (default_legitimize_address_displacement): Likewise.
2896 * doc/tm.texi: Regenerate.
2897 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2898 as an argument, moving assert of ad->disp == ad->disp_term to...
2899 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2900 Try calling targetm.legitimize_address_displacement before expanding
2901 the address rather than afterwards, and adjust for the new interface.
2902 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2903 Match the new hook interface. Handle SVE addresses.
2904 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2907 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2909 * Makefile.in (OBJS): Add early-remat.o.
2910 * target.def (select_early_remat_modes): New hook.
2911 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2912 * doc/tm.texi: Regenerate.
2913 * targhooks.h (default_select_early_remat_modes): Declare.
2914 * targhooks.c (default_select_early_remat_modes): New function.
2915 * timevar.def (TV_EARLY_REMAT): New timevar.
2916 * passes.def (pass_early_remat): New pass.
2917 * tree-pass.h (make_pass_early_remat): Declare.
2918 * early-remat.c: New file.
2919 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2921 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2923 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2924 Alan Hayward <alan.hayward@arm.com>
2925 David Sherwood <david.sherwood@arm.com>
2927 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2928 vfm1 with a bound_epilog parameter.
2929 (vect_do_peeling): Update calls accordingly, and move the prologue
2930 call earlier in the function. Treat the base bound_epilog as 0 for
2931 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2932 this base when peeling for gaps.
2933 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2934 with fully-masked loops.
2935 (vect_estimate_min_profitable_iters): Handle the single peeled
2936 iteration in that case.
2938 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2939 Alan Hayward <alan.hayward@arm.com>
2940 David Sherwood <david.sherwood@arm.com>
2942 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2943 single-element interleaving even if the size is not a power of 2.
2944 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2945 accesses for single-element interleaving if the group size is
2948 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2949 Alan Hayward <alan.hayward@arm.com>
2950 David Sherwood <david.sherwood@arm.com>
2952 * doc/md.texi (fold_extract_last_@var{m}): Document.
2953 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2954 * optabs.def (fold_extract_last_optab): New optab.
2955 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2956 * internal-fn.c (fold_extract_direct): New macro.
2957 (expand_fold_extract_optab_fn): Likewise.
2958 (direct_fold_extract_optab_supported_p): Likewise.
2959 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2960 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2961 EXTRACT_LAST_REDUCTION.
2962 (get_initial_def_for_reduction): Do not create an initial vector
2963 for EXTRACT_LAST_REDUCTION reductions.
2964 (vectorizable_reduction): Leave the scalar phi in place for
2965 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2966 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2967 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2968 transform phase to vectorizable_condition.
2969 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2971 (vect_finish_stmt_generation): ...here.
2972 (vect_finish_replace_stmt): New function.
2973 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2974 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2976 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2978 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2979 Alan Hayward <alan.hayward@arm.com>
2980 David Sherwood <david.sherwood@arm.com>
2982 * doc/md.texi (extract_last_@var{m}): Document.
2983 * optabs.def (extract_last_optab): New optab.
2984 * internal-fn.def (EXTRACT_LAST): New internal function.
2985 * internal-fn.c (cond_unary_direct): New macro.
2986 (expand_cond_unary_optab_fn): Likewise.
2987 (direct_cond_unary_optab_supported_p): Likewise.
2988 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2989 loops using EXTRACT_LAST.
2990 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2991 (extract_last_<mode>): ...this optab.
2992 (vec_extract<mode><Vel>): Update accordingly.
2994 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2995 Alan Hayward <alan.hayward@arm.com>
2996 David Sherwood <david.sherwood@arm.com>
2998 * target.def (empty_mask_is_expensive): New hook.
2999 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3000 * doc/tm.texi: Regenerate.
3001 * targhooks.h (default_empty_mask_is_expensive): Declare.
3002 * targhooks.c (default_empty_mask_is_expensive): New function.
3003 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3004 if the target says that empty masks are expensive.
3005 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3007 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3009 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3010 Alan Hayward <alan.hayward@arm.com>
3011 David Sherwood <david.sherwood@arm.com>
3013 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3014 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3015 (vect_use_loop_mask_for_alignment_p): New function.
3016 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3017 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3018 niters_skip argument. Make sure that the first niters_skip elements
3019 of the first iteration are inactive.
3020 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3021 Update call to vect_set_loop_masks_directly.
3022 (get_misalign_in_elems): New function, split out from...
3023 (vect_gen_prolog_loop_niters): ...here.
3024 (vect_update_init_of_dr): Take a code argument that specifies whether
3025 the adjustment should be added or subtracted.
3026 (vect_update_init_of_drs): Likewise.
3027 (vect_prepare_for_masked_peels): New function.
3028 (vect_do_peeling): Skip prologue peeling if we're using a mask
3029 instead. Update call to vect_update_inits_of_drs.
3030 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3032 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3033 alignment. Do not include the number of peeled iterations in
3034 the minimum threshold in that case.
3035 (vectorizable_induction): Adjust the start value down by
3036 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3037 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3038 Take the number of skipped iterations into account when calculating
3040 * tree-vect-stmts.c (vect_gen_while_not): New function.
3042 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3043 Alan Hayward <alan.hayward@arm.com>
3044 David Sherwood <david.sherwood@arm.com>
3046 * doc/sourcebuild.texi (vect_fully_masked): Document.
3047 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3049 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3051 (vect_analyze_loop_2): ...here. Don't check the vectorization
3052 factor against the number of loop iterations if the loop is
3055 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3056 Alan Hayward <alan.hayward@arm.com>
3057 David Sherwood <david.sherwood@arm.com>
3059 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3060 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3061 (dump_groups): Update accordingly.
3062 (iv_use::mem_type): New member variable.
3063 (address_p): New function.
3064 (record_use): Add a mem_type argument and initialize the new
3066 (record_group_use): Add a mem_type argument. Use address_p.
3067 Remove obsolete null checks of base_object. Update call to record_use.
3068 (find_interesting_uses_op): Update call to record_group_use.
3069 (find_interesting_uses_cond): Likewise.
3070 (find_interesting_uses_address): Likewise.
3071 (get_mem_type_for_internal_fn): New function.
3072 (find_address_like_use): Likewise.
3073 (find_interesting_uses_stmt): Try find_address_like_use before
3074 calling find_interesting_uses_op.
3075 (addr_offset_valid_p): Use the iv mem_type field as the type
3076 of the addressed memory.
3077 (add_autoinc_candidates): Likewise.
3078 (get_address_cost): Likewise.
3079 (split_small_address_groups_p): Use address_p.
3080 (split_address_groups): Likewise.
3081 (add_iv_candidate_for_use): Likewise.
3082 (autoinc_possible_for_pair): Likewise.
3083 (rewrite_groups): Likewise.
3084 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3085 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3086 (get_alias_ptr_type_for_ptr_address): New function.
3087 (rewrite_use_address): Rewrite address uses in calls that were
3088 identified by find_address_like_use.
3090 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3091 Alan Hayward <alan.hayward@arm.com>
3092 David Sherwood <david.sherwood@arm.com>
3094 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3096 * gimple-expr.h (is_gimple_addressable: Likewise.
3097 * gimple-expr.c (is_gimple_address): Likewise.
3098 * internal-fn.c (expand_call_mem_ref): New function.
3099 (expand_mask_load_optab_fn): Use it.
3100 (expand_mask_store_optab_fn): Likewise.
3102 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3103 Alan Hayward <alan.hayward@arm.com>
3104 David Sherwood <david.sherwood@arm.com>
3106 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3107 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3108 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3109 (cond_umax@var{mode}): Document.
3110 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3111 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3112 (cond_umin_optab, cond_umax_optab): New optabs.
3113 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3114 (COND_IOR, COND_XOR): New internal functions.
3115 * internal-fn.h (get_conditional_internal_fn): Declare.
3116 * internal-fn.c (cond_binary_direct): New macro.
3117 (expand_cond_binary_optab_fn): Likewise.
3118 (direct_cond_binary_optab_supported_p): Likewise.
3119 (get_conditional_internal_fn): New function.
3120 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3121 Cope with reduction statements that are vectorized as calls rather
3123 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3124 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3125 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3126 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3127 (UNSPEC_COND_EOR): New unspecs.
3128 (optab): Add mappings for them.
3129 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3130 (sve_int_op, sve_fp_op): New int attributes.
3132 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3133 Alan Hayward <alan.hayward@arm.com>
3134 David Sherwood <david.sherwood@arm.com>
3136 * optabs.def (while_ult_optab): New optab.
3137 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3138 * internal-fn.def (WHILE_ULT): New internal function.
3139 * internal-fn.h (direct_internal_fn_supported_p): New override
3140 that takes two types as argument.
3141 * internal-fn.c (while_direct): New macro.
3142 (expand_while_optab_fn): New function.
3143 (convert_optab_supported_p): Likewise.
3144 (direct_while_optab_supported_p): New macro.
3145 * wide-int.h (wi::udiv_ceil): New function.
3146 * tree-vectorizer.h (rgroup_masks): New structure.
3147 (vec_loop_masks): New typedef.
3148 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3150 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3151 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3152 (vect_max_vf): New function.
3153 (slpeel_make_loop_iterate_ntimes): Delete.
3154 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3155 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3156 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3157 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3158 internal-fn.h, stor-layout.h and optabs-query.h.
3159 (vect_set_loop_mask): New function.
3160 (add_preheader_seq): Likewise.
3161 (add_header_seq): Likewise.
3162 (interleave_supported_p): Likewise.
3163 (vect_maybe_permute_loop_masks): Likewise.
3164 (vect_set_loop_masks_directly): Likewise.
3165 (vect_set_loop_condition_masked): Likewise.
3166 (vect_set_loop_condition_unmasked): New function, split out from
3167 slpeel_make_loop_iterate_ntimes.
3168 (slpeel_make_loop_iterate_ntimes): Rename to..
3169 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3170 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3171 (vect_do_peeling): Update call accordingly.
3172 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3174 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3175 mask_compare_type, can_fully_mask_p and fully_masked_p.
3176 (release_vec_loop_masks): New function.
3177 (_loop_vec_info): Use it to free the loop masks.
3178 (can_produce_all_loop_masks_p): New function.
3179 (vect_get_max_nscalars_per_iter): Likewise.
3180 (vect_verify_full_masking): Likewise.
3181 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3182 retries, and free the mask rgroups before retrying. Check loop-wide
3183 reasons for disallowing fully-masked loops. Make the final decision
3184 about whether use a fully-masked loop or not.
3185 (vect_estimate_min_profitable_iters): Do not assume that peeling
3186 for the number of iterations will be needed for fully-masked loops.
3187 (vectorizable_reduction): Disable fully-masked loops.
3188 (vectorizable_live_operation): Likewise.
3189 (vect_halve_mask_nunits): New function.
3190 (vect_double_mask_nunits): Likewise.
3191 (vect_record_loop_mask): Likewise.
3192 (vect_get_loop_mask): Likewise.
3193 (vect_transform_loop): Handle the case in which the final loop
3194 iteration might handle a partial vector. Call vect_set_loop_condition
3195 instead of slpeel_make_loop_iterate_ntimes.
3196 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3197 (check_load_store_masking): New function.
3198 (prepare_load_store_mask): Likewise.
3199 (vectorizable_store): Handle fully-masked loops.
3200 (vectorizable_load): Likewise.
3201 (supportable_widening_operation): Use vect_halve_mask_nunits for
3203 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3204 (vect_gen_while): New function.
3205 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3206 (aarch64_uqdec<mode>): New insn.
3208 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3209 Alan Hayward <alan.hayward@arm.com>
3210 David Sherwood <david.sherwood@arm.com>
3212 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3213 (reduc_xor_scal_optab): New optabs.
3214 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3215 (reduc_xor_scal_@var{m}): Document.
3216 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3217 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3219 * fold-const-call.c (fold_const_call): Handle them.
3220 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3221 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3222 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3223 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3224 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3225 (UNSPEC_XORV): New unspecs.
3226 (optab): Add entries for them.
3227 (BITWISEV): New int iterator.
3228 (bit_reduc_op): New int attributes.
3230 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3231 Alan Hayward <alan.hayward@arm.com>
3232 David Sherwood <david.sherwood@arm.com>
3234 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3235 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3236 * optabs.def (vec_shl_insert_optab): New optab.
3237 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3238 (duplicate_and_interleave): Likewise.
3239 * tree-vect-loop.c: Include internal-fn.h.
3240 (neutral_op_for_slp_reduction): New function, split out from
3241 get_initial_defs_for_reduction.
3242 (get_initial_def_for_reduction): Handle option 2 for variable-length
3243 vectors by loading the neutral value into a vector and then shifting
3244 the initial value into element 0.
3245 (get_initial_defs_for_reduction): Replace the code argument with
3246 the neutral value calculated by neutral_op_for_slp_reduction.
3247 Use gimple_build_vector for constant-length vectors.
3248 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3249 but the first group_size elements have a neutral value.
3250 Use duplicate_and_interleave otherwise.
3251 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3252 Update call to get_initial_defs_for_reduction. Handle SLP
3253 reductions for variable-length vectors by creating one vector
3254 result for each scalar result, with the elements associated
3255 with other scalar results stubbed out with the neutral value.
3256 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3257 Require IFN_VEC_SHL_INSERT for double reductions on
3258 variable-length vectors, or SLP reductions that have
3259 a neutral value. Require can_duplicate_and_interleave_p
3260 support for variable-length unchained SLP reductions if there
3261 is no neutral value, such as for MIN/MAX reductions. Also require
3262 the number of vector elements to be a multiple of the number of
3263 SLP statements when doing variable-length unchained SLP reductions.
3264 Update call to vect_create_epilog_for_reduction.
3265 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3266 and remove initial values.
3267 (duplicate_and_interleave): Make public.
3268 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3269 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3271 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3272 Alan Hayward <alan.hayward@arm.com>
3273 David Sherwood <david.sherwood@arm.com>
3275 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3276 (can_duplicate_and_interleave_p): New function.
3277 (vect_get_and_check_slp_defs): Take the vector of statements
3278 rather than just the current one. Remove excess parentheses.
3279 Restriction rejectinon of vect_constant_def and vect_external_def
3280 for variable-length vectors to boolean types, or types for which
3281 can_duplicate_and_interleave_p is false.
3282 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3283 (duplicate_and_interleave): New function.
3284 (vect_get_constant_vectors): Use gimple_build_vector for
3285 constant-length vectors and suitable variable-length constant
3286 vectors. Use duplicate_and_interleave for other variable-length
3287 vectors. Don't defer the update when inserting new statements.
3289 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3290 Alan Hayward <alan.hayward@arm.com>
3291 David Sherwood <david.sherwood@arm.com>
3293 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3294 min_profitable_iters doesn't go negative.
3296 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3297 Alan Hayward <alan.hayward@arm.com>
3298 David Sherwood <david.sherwood@arm.com>
3300 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3301 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3302 * optabs.def (vec_mask_load_lanes_optab): New optab.
3303 (vec_mask_store_lanes_optab): Likewise.
3304 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3305 (MASK_STORE_LANES): Likewise.
3306 * internal-fn.c (mask_load_lanes_direct): New macro.
3307 (mask_store_lanes_direct): Likewise.
3308 (expand_mask_load_optab_fn): Handle masked operations.
3309 (expand_mask_load_lanes_optab_fn): New macro.
3310 (expand_mask_store_optab_fn): Handle masked operations.
3311 (expand_mask_store_lanes_optab_fn): New macro.
3312 (direct_mask_load_lanes_optab_supported_p): Likewise.
3313 (direct_mask_store_lanes_optab_supported_p): Likewise.
3314 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3316 (vect_load_lanes_supported): Likewise.
3317 * tree-vect-data-refs.c (strip_conversion): New function.
3318 (can_group_stmts_p): Likewise.
3319 (vect_analyze_data_ref_accesses): Use it instead of checking
3320 for a pair of assignments.
3321 (vect_store_lanes_supported): Take a masked_p parameter.
3322 (vect_load_lanes_supported): Likewise.
3323 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3324 vect_store_lanes_supported and vect_load_lanes_supported.
3325 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3326 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3327 parameter. Don't allow gaps for masked accesses.
3328 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3329 and vect_load_lanes_supported.
3330 (get_load_store_type): Take a masked_p parameter and update
3331 call to get_group_load_store_type.
3332 (vectorizable_store): Update call to get_load_store_type.
3333 Handle IFN_MASK_STORE_LANES.
3334 (vectorizable_load): Update call to get_load_store_type.
3335 Handle IFN_MASK_LOAD_LANES.
3337 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3338 Alan Hayward <alan.hayward@arm.com>
3339 David Sherwood <david.sherwood@arm.com>
3341 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3343 * config/aarch64/aarch64-protos.h
3344 (aarch64_sve_struct_memory_operand_p): Declare.
3345 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3346 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3347 (VPRED, vpred): Handle SVE structure modes.
3348 * config/aarch64/constraints.md (Utx): New constraint.
3349 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3350 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3351 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3352 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3353 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3354 structure modes. Split into pieces after RA.
3355 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3356 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3358 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3359 SVE structure modes.
3360 (aarch64_classify_address): Likewise.
3361 (sizetochar): Move earlier in file.
3362 (aarch64_print_operand): Handle SVE register lists.
3363 (aarch64_array_mode): New function.
3364 (aarch64_sve_struct_memory_operand_p): Likewise.
3365 (TARGET_ARRAY_MODE): Redefine.
3367 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3368 Alan Hayward <alan.hayward@arm.com>
3369 David Sherwood <david.sherwood@arm.com>
3371 * target.def (array_mode): New target hook.
3372 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3373 * doc/tm.texi: Regenerate.
3374 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3375 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3376 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3378 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3381 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3382 Alan Hayward <alan.hayward@arm.com>
3383 David Sherwood <david.sherwood@arm.com>
3385 * fold-const.c (fold_binary_loc): Check the argument types
3386 rather than the result type when testing for a vector operation.
3388 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3390 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3391 * doc/tm.texi: Regenerate.
3393 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3394 Alan Hayward <alan.hayward@arm.com>
3395 David Sherwood <david.sherwood@arm.com>
3397 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3398 (sve): Document new AArch64 extension.
3399 * doc/md.texi (w): Extend the description of the AArch64
3400 constraint to include SVE vectors.
3401 (Upl, Upa): Document new AArch64 predicate constraints.
3402 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3404 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3405 (msve-vector-bits=): New option.
3406 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3407 SVE when these are disabled.
3408 (sve): New extension.
3409 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3410 modes. Adjust their number of units based on aarch64_sve_vg.
3411 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3412 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3413 aarch64_addr_query_type.
3414 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3415 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3416 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3417 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3418 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3419 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3420 (aarch64_simd_imm_zero_p): Delete.
3421 (aarch64_check_zero_based_sve_index_immediate): Declare.
3422 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3423 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3424 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3425 (aarch64_sve_float_mul_immediate_p): Likewise.
3426 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3428 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3429 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3430 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3431 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3432 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3433 (aarch64_regmode_natural_size): Likewise.
3434 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3435 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3437 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3438 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3439 for VG and the SVE predicate registers.
3440 (V_ALIASES): Add a "z"-prefixed alias.
3441 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3442 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3443 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3444 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3445 (REG_CLASS_NAMES): Add entries for them.
3446 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3447 and the predicate registers.
3448 (aarch64_sve_vg): Declare.
3449 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3450 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3451 (REGMODE_NATURAL_SIZE): Define.
3452 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3454 * config/aarch64/aarch64.c: Include cfgrtl.h.
3455 (simd_immediate_info): Add a constructor for series vectors,
3456 and an associated step field.
3457 (aarch64_sve_vg): New variable.
3458 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3459 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3460 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3461 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3462 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3463 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3464 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3465 (aarch64_get_mask_mode): New functions.
3466 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3467 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3468 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3469 predicate modes and predicate registers. Explicitly restrict
3470 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3471 to store a vector mode if it is recognized by
3472 aarch64_classify_vector_mode.
3473 (aarch64_regmode_natural_size): New function.
3474 (aarch64_hard_regno_caller_save_mode): Return the original mode
3476 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3477 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3478 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3479 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3481 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3482 does not overlap dest if the function is frame-related. Handle
3484 (aarch64_split_add_offset): New function.
3485 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3486 them aarch64_add_offset.
3487 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3488 and update call to aarch64_sub_sp.
3489 (aarch64_add_cfa_expression): New function.
3490 (aarch64_expand_prologue): Pass extra temporary registers to the
3491 functions above. Handle the case in which we need to emit new
3492 DW_CFA_expressions for registers that were originally saved
3493 relative to the stack pointer, but now have to be expressed
3494 relative to the frame pointer.
3495 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3497 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3498 IP0 and IP1 values for SVE frames.
3499 (aarch64_expand_vec_series): New function.
3500 (aarch64_expand_sve_widened_duplicate): Likewise.
3501 (aarch64_expand_sve_const_vector): Likewise.
3502 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3503 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3504 into the register, rather than emitting a SET directly.
3505 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3506 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3507 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3508 (offset_9bit_signed_scaled_p): New functions.
3509 (aarch64_replicate_bitmask_imm): New function.
3510 (aarch64_bitmask_imm): Use it.
3511 (aarch64_cannot_force_const_mem): Reject expressions involving
3512 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3513 (aarch64_classify_index): Handle SVE indices, by requiring
3514 a plain register index with a scale that matches the element size.
3515 (aarch64_classify_address): Handle SVE addresses. Assert that
3516 the mode of the address is VOIDmode or an integer mode.
3517 Update call to aarch64_classify_symbol.
3518 (aarch64_classify_symbolic_expression): Update call to
3519 aarch64_classify_symbol.
3520 (aarch64_const_vec_all_in_range_p): New function.
3521 (aarch64_print_vector_float_operand): Likewise.
3522 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3523 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3524 and the FP immediates 1.0 and 0.5.
3525 (aarch64_print_address_internal): Handle SVE addresses.
3526 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3527 (aarch64_regno_regclass): Handle predicate registers.
3528 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3530 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3531 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3532 (aarch64_convert_sve_vector_bits): New function.
3533 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3534 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3536 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3537 Handle SVE vector and predicate modes. Accept VL-based constants
3538 that need only one temporary register, and VL offsets that require
3539 no temporary registers.
3540 (aarch64_conditional_register_usage): Mark the predicate registers
3541 as fixed if SVE isn't available.
3542 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3543 Return true for SVE vector and predicate modes.
3544 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3545 rather than an unsigned int. Handle SVE modes.
3546 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3548 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3550 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3551 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3552 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3553 (aarch64_sve_float_mul_immediate_p): New functions.
3554 (aarch64_sve_valid_immediate): New function.
3555 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3556 Explicitly reject structure modes. Check for INDEX constants.
3557 Handle PTRUE and PFALSE constants.
3558 (aarch64_check_zero_based_sve_index_immediate): New function.
3559 (aarch64_simd_imm_zero_p): Delete.
3560 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3561 vector modes. Accept constants in the range of CNT[BHWD].
3562 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3563 ask for an Advanced SIMD mode.
3564 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3565 (aarch64_simd_vector_alignment): Handle SVE predicates.
3566 (aarch64_vectorize_preferred_vector_alignment): New function.
3567 (aarch64_simd_vector_alignment_reachable): Use it instead of
3569 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3570 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3572 (MAX_VECT_LEN): Delete.
3573 (expand_vec_perm_d): Add a vec_flags field.
3574 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3575 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3576 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3578 (aarch64_evpc_rev): Rename to...
3579 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3580 (aarch64_evpc_rev_global): New function.
3581 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3582 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3584 (aarch64_evpc_sve_tbl): New function.
3585 (aarch64_expand_vec_perm_const_1): Update after rename of
3586 aarch64_evpc_rev. Handle SVE permutes too, trying
3587 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3588 than aarch64_evpc_tbl.
3589 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3590 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3591 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3592 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3593 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3594 (aarch64_expand_sve_vcond): New functions.
3595 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3596 of aarch64_vector_mode_p.
3597 (aarch64_dwarf_poly_indeterminate_value): New function.
3598 (aarch64_compute_pressure_classes): Likewise.
3599 (aarch64_can_change_mode_class): Likewise.
3600 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3601 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3602 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3603 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3604 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3605 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3606 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3607 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3609 (Dn, Dl, Dr): Accept const as well as const_vector.
3610 (Dz): Likewise. Compare against CONST0_RTX.
3611 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3612 of "vector" where appropriate.
3613 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3614 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3615 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3616 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3617 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3618 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3619 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3620 (v_int_equiv): Extend to SVE modes.
3621 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3623 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3624 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3625 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3626 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3627 (SVE_COND_FP_CMP): New int iterators.
3628 (perm_hilo): Handle the new unpack unspecs.
3629 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3631 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3632 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3633 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3634 (aarch64_equality_operator, aarch64_constant_vector_operand)
3635 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3636 (aarch64_sve_nonimmediate_operand): Likewise.
3637 (aarch64_sve_general_operand): Likewise.
3638 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3639 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3640 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3641 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3642 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3643 (aarch64_sve_float_arith_immediate): Likewise.
3644 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3645 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3646 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3647 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3648 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3649 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3650 (aarch64_sve_float_arith_operand): Likewise.
3651 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3652 (aarch64_sve_float_mul_operand): Likewise.
3653 (aarch64_sve_vec_perm_operand): Likewise.
3654 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3655 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3656 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3657 as well as const_vector.
3658 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3659 in file. Use CONST0_RTX and CONSTM1_RTX.
3660 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3661 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3662 Use aarch64_simd_imm_zero.
3663 * config/aarch64/aarch64-sve.md: New file.
3664 * config/aarch64/aarch64.md: Include it.
3665 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3666 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3667 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3668 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3669 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3670 (sve): New attribute.
3671 (enabled): Disable instructions with the sve attribute unless
3673 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3674 aarch64_expand_mov_immediate.
3675 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3676 CNT[BHSD] immediates.
3677 (movti): Split CONST_POLY_INT moves into two halves.
3678 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3679 Split additions that need a temporary here if the destination
3680 is the stack pointer.
3681 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3682 (*add<mode>3_poly_1): New instruction.
3683 (set_clobber_cc): New expander.
3685 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3687 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3688 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3689 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3690 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3691 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3692 Change innermode from fixed_mode_size to machine_mode.
3693 (simplify_subreg): Update call accordingly. Handle a constant-sized
3694 subreg of a variable-length CONST_VECTOR.
3696 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3697 Alan Hayward <alan.hayward@arm.com>
3698 David Sherwood <david.sherwood@arm.com>
3700 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3701 (add_offset_to_base): New function, split out from...
3702 (create_mem_ref): ...here. When handling a scale other than 1,
3703 check first whether the address is valid without the offset.
3704 Add it into the base if so, leaving the index and scale as-is.
3706 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3709 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3710 fold_for_warn before checking if arg2 is INTEGER_CST.
3712 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3714 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3715 (store_multiple_operation): Delete.
3716 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3717 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3718 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3719 guarded by TARGET_STRING.
3720 (rs6000_output_load_multiple): Delete.
3721 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3722 OPTION_MASK_STRING / TARGET_STRING handling.
3723 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3724 (const rs6000_opt_masks) <"string">: Change mask to 0.
3725 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3726 (MASK_STRING): Delete.
3727 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3729 (load_multiple): Delete.
3736 (store_multiple): Delete.
3743 (movmemsi_8reg): Delete.
3744 (corresponding unnamed define_insn): Delete.
3745 (movmemsi_6reg): Delete.
3746 (corresponding unnamed define_insn): Delete.
3747 (movmemsi_4reg): Delete.
3748 (corresponding unnamed define_insn): Delete.
3749 (movmemsi_2reg): Delete.
3750 (corresponding unnamed define_insn): Delete.
3751 (movmemsi_1reg): Delete.
3752 (corresponding unnamed define_insn): Delete.
3753 * config/rs6000/rs6000.opt (mno-string): New.
3754 (mstring): Replace by deprecation warning stub.
3755 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3757 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3759 * regrename.c (regrename_do_replace): If replacing the same
3760 reg multiple times, try to reuse last created gen_raw_REG.
3763 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3764 main to workaround a bug in GDB.
3766 2018-01-12 Tom de Vries <tom@codesourcery.com>
3769 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3771 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3773 PR rtl-optimization/80481
3774 * ira-color.c (get_cap_member): New function.
3775 (allocnos_conflict_by_live_ranges_p): Use it.
3776 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3777 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3779 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3782 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3783 (*saddl_se_1): Ditto.
3785 (*ssubl_se_1): Ditto.
3787 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3789 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3790 rather than wi::to_widest for DR_INITs.
3791 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3792 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3793 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3795 (vect_analyze_group_access_1): Note that here.
3797 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3799 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3800 polynomial type sizes.
3802 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3804 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3805 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3806 (gimple_add_tmp_var): Likewise.
3808 2018-01-12 Martin Liska <mliska@suse.cz>
3810 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3811 (gimple_alloc_sizes): Likewise.
3812 (dump_gimple_statistics): Use PRIu64 in printf format.
3813 * gimple.h: Change uint64_t to int.
3815 2018-01-12 Martin Liska <mliska@suse.cz>
3817 * tree-core.h: Use uint64_t instead of int.
3818 * tree.c (tree_node_counts): Likewise.
3819 (tree_node_sizes): Likewise.
3820 (dump_tree_statistics): Use PRIu64 in printf format.
3822 2018-01-12 Martin Liska <mliska@suse.cz>
3824 * Makefile.in: As qsort_chk is implemented in vec.c, add
3825 vec.o to linkage of gencfn-macros.
3826 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3827 passing the info to record_node_allocation_statistics.
3828 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3830 * ggc-common.c (struct ggc_usage): Add operator== and use
3831 it in operator< and compare function.
3832 * mem-stats.h (struct mem_usage): Likewise.
3833 * vec.c (struct vec_usage): Remove operator< and compare
3834 function. Can be simply inherited.
3836 2018-01-12 Martin Jambor <mjambor@suse.cz>
3839 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3840 * tree-ssa-math-opts.c: Include domwalk.h.
3841 (convert_mult_to_fma_1): New function.
3842 (fma_transformation_info): New type.
3843 (fma_deferring_state): Likewise.
3844 (cancel_fma_deferring): New function.
3845 (result_of_phi): Likewise.
3846 (last_fma_candidate_feeds_initial_phi): Likewise.
3847 (convert_mult_to_fma): Added deferring logic, split actual
3848 transformation to convert_mult_to_fma_1.
3849 (math_opts_dom_walker): New type.
3850 (math_opts_dom_walker::after_dom_children): New method, body moved
3851 here from pass_optimize_widening_mul::execute, added deferring logic
3853 (pass_optimize_widening_mul::execute): Moved most of code to
3854 math_opts_dom_walker::after_dom_children.
3855 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3856 * config/i386/i386.c (ix86_option_override_internal): Added
3857 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3859 2018-01-12 Richard Biener <rguenther@suse.de>
3862 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3863 inline instance vars.
3865 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3868 * config/rx/rx.c (rx_is_restricted_memory_address):
3871 2018-01-12 Richard Biener <rguenther@suse.de>
3873 PR tree-optimization/80846
3874 * target.def (split_reduction): New target hook.
3875 * targhooks.c (default_split_reduction): New function.
3876 * targhooks.h (default_split_reduction): Declare.
3877 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3878 target requests first reduce vectors by combining low and high
3880 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3881 (get_vectype_for_scalar_type_and_size): Export.
3882 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3883 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3884 * doc/tm.texi: Regenerate.
3885 * config/i386/i386.c (ix86_split_reduction): Implement
3886 TARGET_VECTORIZE_SPLIT_REDUCTION.
3888 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3891 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3892 in PIC mode except for TARGET_VXWORKS_RTP.
3893 * config/sparc/sparc.c: Include cfgrtl.h.
3894 (TARGET_INIT_PIC_REG): Define.
3895 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3896 (sparc_pic_register_p): New predicate.
3897 (sparc_legitimate_address_p): Use it.
3898 (sparc_legitimize_pic_address): Likewise.
3899 (sparc_delegitimize_address): Likewise.
3900 (sparc_mode_dependent_address_p): Likewise.
3901 (gen_load_pcrel_sym): Remove 4th parameter.
3902 (load_got_register): Adjust call to above. Remove obsolete stuff.
3903 (sparc_expand_prologue): Do not call load_got_register here.
3904 (sparc_flat_expand_prologue): Likewise.
3905 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3906 (sparc_use_pseudo_pic_reg): New function.
3907 (sparc_init_pic_reg): Likewise.
3908 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3909 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3911 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3913 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3914 Add item for branch_cost.
3916 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3918 PR rtl-optimization/83565
3919 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3920 not extend the result to a larger mode for rotate operations.
3921 (num_sign_bit_copies1): Likewise.
3923 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3926 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3928 Use values-Xc.o for -pedantic.
3929 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3931 2018-01-12 Martin Liska <mliska@suse.cz>
3934 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3936 (possible_polymorphic_call_targets): Use it.
3937 (ipa_devirt): Likewise.
3939 2018-01-12 Martin Liska <mliska@suse.cz>
3941 * profile-count.h (enum profile_quality): Use 0 as invalid
3942 enum value of profile_quality.
3944 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3946 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3947 -mext-string options.
3949 2018-01-12 Richard Biener <rguenther@suse.de>
3951 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3952 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3953 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3955 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3957 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3959 * configure.ac (--with-long-double-format): Add support for the
3960 configuration option to change the default long double format on
3962 * config.gcc (powerpc*-linux*-*): Likewise.
3963 * configure: Regenerate.
3964 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3965 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3966 used without modification.
3968 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3970 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3971 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3972 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3973 MISC_BUILTIN_SPEC_BARRIER.
3974 (rs6000_init_builtins): Likewise.
3975 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3977 (speculation_barrier): New define_insn.
3978 * doc/extend.texi: Document __builtin_speculation_barrier.
3980 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3983 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3984 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3985 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3987 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3988 integral modes instead of "ss" and "sd".
3989 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3990 vectors with 32-bit and 64-bit elements.
3991 (vecdupssescalarmodesuffix): New mode attribute.
3992 (vec_dup<mode>): Use it.
3994 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3997 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3998 frame if argument is passed on stack.
4000 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4003 * ree.c (combine_reaching_defs): Optimize also
4004 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4005 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4007 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4010 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4012 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4015 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4016 after they are computed.
4018 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4020 PR tree-optimization/83695
4021 * gimple-loop-linterchange.cc
4022 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4023 reset cached scev information after interchange.
4024 (pass_linterchange::execute): Remove call to scev_reset_htab.
4026 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4028 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4029 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4030 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4031 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4032 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4033 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4034 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4035 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4036 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4037 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4038 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4039 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4040 (V_lane_reg): Likewise.
4041 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4043 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4044 (vfmal_lane_low<mode>_intrinsic,
4045 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4046 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4047 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4048 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4049 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4050 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4052 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4054 * config/arm/arm-cpus.in (fp16fml): New feature.
4055 (ALL_SIMD): Add fp16fml.
4056 (armv8.2-a): Add fp16fml as an option.
4057 (armv8.3-a): Likewise.
4058 (armv8.4-a): Add fp16fml as part of fp16.
4059 * config/arm/arm.h (TARGET_FP16FML): Define.
4060 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4062 * config/arm/arm-modes.def (V2HF): Define.
4063 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4064 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4065 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4066 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4067 vfmsl_low, vfmsl_high): New set of builtins.
4068 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4069 (vfml_op): New code attribute.
4070 (VFMLHALVES): New int iterator.
4071 (VFML, VFMLSEL): New mode attributes.
4072 (V_reg): Define mapping for V2HF.
4073 (V_hi, V_lo): New mode attributes.
4074 (VF_constraint): Likewise.
4075 (vfml_half, vfml_half_selector): New int attributes.
4076 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4078 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4079 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4081 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4082 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4083 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4084 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4086 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4087 Document new effective target and option set.
4089 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4091 * config/arm/arm-cpus.in (armv8_4): New feature.
4092 (ARMv8_4a): New fgroup.
4093 (armv8.4-a): New arch.
4094 * config/arm/arm-tables.opt: Regenerate.
4095 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4096 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4097 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4098 Add matching rules for -march=armv8.4-a and extensions.
4099 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4101 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4104 * config/rx/rx.md (BW): New mode attribute.
4105 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4107 2018-01-11 Richard Biener <rguenther@suse.de>
4109 PR tree-optimization/83435
4110 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4111 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4112 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4114 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4115 Alan Hayward <alan.hayward@arm.com>
4116 David Sherwood <david.sherwood@arm.com>
4118 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4120 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4121 (aarch64_print_address_internal): Use it to check for a zero offset.
4123 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4124 Alan Hayward <alan.hayward@arm.com>
4125 David Sherwood <david.sherwood@arm.com>
4127 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4128 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4129 Return a poly_int64 rather than a HOST_WIDE_INT.
4130 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4131 rather than a HOST_WIDE_INT.
4132 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4133 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4134 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4135 final_offset from HOST_WIDE_INT to poly_int64.
4136 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4137 to_constant when getting the number of units in an Advanced SIMD
4139 (aarch64_builtin_vectorized_function): Check for a constant number
4141 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4143 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4144 attribute instead of GET_MODE_NUNITS.
4145 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4146 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4147 GET_MODE_SIZE for fixed-size registers.
4148 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4149 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4150 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4151 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4152 (aarch64_print_operand, aarch64_print_address_internal)
4153 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4154 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4155 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4156 Handle polynomial GET_MODE_SIZE.
4157 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4158 wider than SImode without modification.
4159 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4160 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4161 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4162 passing and returning SVE modes.
4163 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4164 rather than GEN_INT.
4165 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4166 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4167 (aarch64_allocate_and_probe_stack_space): Likewise.
4168 (aarch64_layout_frame): Cope with polynomial offsets.
4169 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4170 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4172 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4173 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4174 poly_int64 rather than a HOST_WIDE_INT.
4175 (aarch64_get_separate_components, aarch64_process_components)
4176 (aarch64_expand_prologue, aarch64_expand_epilogue)
4177 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4178 (aarch64_anchor_offset): New function, split out from...
4179 (aarch64_legitimize_address): ...here.
4180 (aarch64_builtin_vectorization_cost): Handle polynomial
4181 TYPE_VECTOR_SUBPARTS.
4182 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4184 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4185 number of elements from the PARALLEL rather than the mode.
4186 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4187 rather than GET_MODE_BITSIZE.
4188 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4189 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4190 (aarch64_expand_vec_perm_const_1): Handle polynomial
4191 d->perm.length () and d->perm elements.
4192 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4193 Apply to_constant to d->perm elements.
4194 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4195 polynomial CONST_VECTOR_NUNITS.
4196 (aarch64_move_pointer): Take amount as a poly_int64 rather
4198 (aarch64_progress_pointer): Avoid temporary variable.
4199 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4200 the mode attribute instead of GET_MODE.
4202 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4203 Alan Hayward <alan.hayward@arm.com>
4204 David Sherwood <david.sherwood@arm.com>
4206 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4207 x exists before using it.
4208 (aarch64_add_constant_internal): Rename to...
4209 (aarch64_add_offset_1): ...this. Replace regnum with separate
4210 src and dest rtxes. Handle the case in which they're different,
4211 including when the offset is zero. Replace scratchreg with an rtx.
4212 Use 2 additions if there is no spare register into which we can
4213 move a 16-bit constant.
4214 (aarch64_add_constant): Delete.
4215 (aarch64_add_offset): Replace reg with separate src and dest
4216 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4217 Use aarch64_add_offset_1.
4218 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4219 an rtx rather than an int. Take the delta as a poly_int64
4220 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4221 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4222 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4223 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4224 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4226 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4227 aarch64_add_constant.
4229 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4231 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4232 Use scalar_float_mode.
4234 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4236 * config/aarch64/aarch64-simd.md
4237 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4238 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4239 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4240 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4241 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4242 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4243 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4244 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4245 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4246 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4248 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4251 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4252 targ_options->x_arm_arch_string is non NULL.
4254 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4256 * config/aarch64/aarch64.h
4257 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4259 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4262 * expmed.c (emit_store_flag_force): Swap if const op0
4263 and change VOIDmode to mode of op0.
4265 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4267 PR rtl-optimization/83761
4268 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4269 than bytes to mode_for_size.
4271 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4274 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4275 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4278 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4281 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4282 when in layout mode.
4283 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4284 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4287 2018-01-10 Michael Collison <michael.collison@arm.com>
4289 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4290 * config/aarch64/aarch64-option-extension.def: Add
4291 AARCH64_OPT_EXTENSION of 'fp16fml'.
4292 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4293 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4294 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4295 * config/aarch64/constraints.md (Ui7): New constraint.
4296 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4297 (VFMLA_SEL_W): Ditto.
4300 (VFMLA16_LOW): New int iterator.
4301 (VFMLA16_HIGH): Ditto.
4302 (UNSPEC_FMLAL): New unspec.
4303 (UNSPEC_FMLSL): Ditto.
4304 (UNSPEC_FMLAL2): Ditto.
4305 (UNSPEC_FMLSL2): Ditto.
4306 (f16mac): New code attribute.
4307 * config/aarch64/aarch64-simd-builtins.def
4308 (aarch64_fmlal_lowv2sf): Ditto.
4309 (aarch64_fmlsl_lowv2sf): Ditto.
4310 (aarch64_fmlalq_lowv4sf): Ditto.
4311 (aarch64_fmlslq_lowv4sf): Ditto.
4312 (aarch64_fmlal_highv2sf): Ditto.
4313 (aarch64_fmlsl_highv2sf): Ditto.
4314 (aarch64_fmlalq_highv4sf): Ditto.
4315 (aarch64_fmlslq_highv4sf): Ditto.
4316 (aarch64_fmlal_lane_lowv2sf): Ditto.
4317 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4318 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4319 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4320 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4321 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4322 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4323 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4324 (aarch64_fmlal_lane_highv2sf): Ditto.
4325 (aarch64_fmlsl_lane_highv2sf): Ditto.
4326 (aarch64_fmlal_laneq_highv2sf): Ditto.
4327 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4328 (aarch64_fmlalq_lane_highv4sf): Ditto.
4329 (aarch64_fmlsl_lane_highv4sf): Ditto.
4330 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4331 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4332 * config/aarch64/aarch64-simd.md:
4333 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4334 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4335 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4336 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4337 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4338 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4339 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4340 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4341 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4342 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4343 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4344 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4345 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4346 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4347 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4348 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4349 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4350 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4351 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4352 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4353 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4354 (vfmlsl_low_u32): Ditto.
4355 (vfmlalq_low_u32): Ditto.
4356 (vfmlslq_low_u32): Ditto.
4357 (vfmlal_high_u32): Ditto.
4358 (vfmlsl_high_u32): Ditto.
4359 (vfmlalq_high_u32): Ditto.
4360 (vfmlslq_high_u32): Ditto.
4361 (vfmlal_lane_low_u32): Ditto.
4362 (vfmlsl_lane_low_u32): Ditto.
4363 (vfmlal_laneq_low_u32): Ditto.
4364 (vfmlsl_laneq_low_u32): Ditto.
4365 (vfmlalq_lane_low_u32): Ditto.
4366 (vfmlslq_lane_low_u32): Ditto.
4367 (vfmlalq_laneq_low_u32): Ditto.
4368 (vfmlslq_laneq_low_u32): Ditto.
4369 (vfmlal_lane_high_u32): Ditto.
4370 (vfmlsl_lane_high_u32): Ditto.
4371 (vfmlal_laneq_high_u32): Ditto.
4372 (vfmlsl_laneq_high_u32): Ditto.
4373 (vfmlalq_lane_high_u32): Ditto.
4374 (vfmlslq_lane_high_u32): Ditto.
4375 (vfmlalq_laneq_high_u32): Ditto.
4376 (vfmlslq_laneq_high_u32): Ditto.
4377 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4378 (AARCH64_FL_FOR_ARCH8_4): New.
4379 (AARCH64_ISA_F16FML): New ISA flag.
4380 (TARGET_F16FML): New feature flag for fp16fml.
4381 (doc/invoke.texi): Document new fp16fml option.
4383 2018-01-10 Michael Collison <michael.collison@arm.com>
4385 * config/aarch64/aarch64-builtins.c:
4386 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4387 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4388 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4389 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4390 (AARCH64_ISA_SHA3): New ISA flag.
4391 (TARGET_SHA3): New feature flag for sha3.
4392 * config/aarch64/iterators.md (sha512_op): New int attribute.
4393 (CRYPTO_SHA512): New int iterator.
4394 (UNSPEC_SHA512H): New unspec.
4395 (UNSPEC_SHA512H2): Ditto.
4396 (UNSPEC_SHA512SU0): Ditto.
4397 (UNSPEC_SHA512SU1): Ditto.
4398 * config/aarch64/aarch64-simd-builtins.def
4399 (aarch64_crypto_sha512hqv2di): New builtin.
4400 (aarch64_crypto_sha512h2qv2di): Ditto.
4401 (aarch64_crypto_sha512su0qv2di): Ditto.
4402 (aarch64_crypto_sha512su1qv2di): Ditto.
4403 (aarch64_eor3qv8hi): Ditto.
4404 (aarch64_rax1qv2di): Ditto.
4405 (aarch64_xarqv2di): Ditto.
4406 (aarch64_bcaxqv8hi): Ditto.
4407 * config/aarch64/aarch64-simd.md:
4408 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4409 (aarch64_crypto_sha512su0qv2di): Ditto.
4410 (aarch64_crypto_sha512su1qv2di): Ditto.
4411 (aarch64_eor3qv8hi): Ditto.
4412 (aarch64_rax1qv2di): Ditto.
4413 (aarch64_xarqv2di): Ditto.
4414 (aarch64_bcaxqv8hi): Ditto.
4415 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4416 (vsha512h2q_u64): Ditto.
4417 (vsha512su0q_u64): Ditto.
4418 (vsha512su1q_u64): Ditto.
4419 (veor3q_u16): Ditto.
4420 (vrax1q_u64): Ditto.
4422 (vbcaxq_u16): Ditto.
4423 * config/arm/types.md (crypto_sha512): New type attribute.
4424 (crypto_sha3): Ditto.
4425 (doc/invoke.texi): Document new sha3 option.
4427 2018-01-10 Michael Collison <michael.collison@arm.com>
4429 * config/aarch64/aarch64-builtins.c:
4430 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4431 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4432 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4433 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4434 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4435 (AARCH64_ISA_SM4): New ISA flag.
4436 (TARGET_SM4): New feature flag for sm4.
4437 * config/aarch64/aarch64-simd-builtins.def
4438 (aarch64_sm3ss1qv4si): Ditto.
4439 (aarch64_sm3tt1aq4si): Ditto.
4440 (aarch64_sm3tt1bq4si): Ditto.
4441 (aarch64_sm3tt2aq4si): Ditto.
4442 (aarch64_sm3tt2bq4si): Ditto.
4443 (aarch64_sm3partw1qv4si): Ditto.
4444 (aarch64_sm3partw2qv4si): Ditto.
4445 (aarch64_sm4eqv4si): Ditto.
4446 (aarch64_sm4ekeyqv4si): Ditto.
4447 * config/aarch64/aarch64-simd.md:
4448 (aarch64_sm3ss1qv4si): Ditto.
4449 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4450 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4451 (aarch64_sm4eqv4si): Ditto.
4452 (aarch64_sm4ekeyqv4si): Ditto.
4453 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4454 (sm3part_op): Ditto.
4455 (CRYPTO_SM3TT): Ditto.
4456 (CRYPTO_SM3PART): Ditto.
4457 (UNSPEC_SM3SS1): New unspec.
4458 (UNSPEC_SM3TT1A): Ditto.
4459 (UNSPEC_SM3TT1B): Ditto.
4460 (UNSPEC_SM3TT2A): Ditto.
4461 (UNSPEC_SM3TT2B): Ditto.
4462 (UNSPEC_SM3PARTW1): Ditto.
4463 (UNSPEC_SM3PARTW2): Ditto.
4464 (UNSPEC_SM4E): Ditto.
4465 (UNSPEC_SM4EKEY): Ditto.
4466 * config/aarch64/constraints.md (Ui2): New constraint.
4467 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4468 * config/arm/types.md (crypto_sm3): New type attribute.
4469 (crypto_sm4): Ditto.
4470 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4471 (vsm3tt1aq_u32): Ditto.
4472 (vsm3tt1bq_u32): Ditto.
4473 (vsm3tt2aq_u32): Ditto.
4474 (vsm3tt2bq_u32): Ditto.
4475 (vsm3partw1q_u32): Ditto.
4476 (vsm3partw2q_u32): Ditto.
4477 (vsm4eq_u32): Ditto.
4478 (vsm4ekeyq_u32): Ditto.
4479 (doc/invoke.texi): Document new sm4 option.
4481 2018-01-10 Michael Collison <michael.collison@arm.com>
4483 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4484 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4485 (AARCH64_FL_FOR_ARCH8_4): New.
4486 (AARCH64_FL_V8_4): New flag.
4487 (doc/invoke.texi): Document new armv8.4-a option.
4489 2018-01-10 Michael Collison <michael.collison@arm.com>
4491 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4492 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4493 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4494 * config/aarch64/aarch64-option-extension.def: Add
4495 AARCH64_OPT_EXTENSION of 'sha2'.
4496 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4497 (crypto): Disable sha2 and aes if crypto disabled.
4498 (crypto): Enable aes and sha2 if enabled.
4499 (simd): Disable sha2 and aes if simd disabled.
4500 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4502 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4503 (TARGET_SHA2): New feature flag for sha2.
4504 (TARGET_AES): New feature flag for aes.
4505 * config/aarch64/aarch64-simd.md:
4506 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4507 conditional on TARGET_AES.
4508 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4509 (aarch64_crypto_sha1hsi): Make pattern conditional
4511 (aarch64_crypto_sha1hv4si): Ditto.
4512 (aarch64_be_crypto_sha1hv4si): Ditto.
4513 (aarch64_crypto_sha1su1v4si): Ditto.
4514 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4515 (aarch64_crypto_sha1su0v4si): Ditto.
4516 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4517 (aarch64_crypto_sha256su0v4si): Ditto.
4518 (aarch64_crypto_sha256su1v4si): Ditto.
4519 (doc/invoke.texi): Document new aes and sha2 options.
4521 2018-01-10 Martin Sebor <msebor@redhat.com>
4523 PR tree-optimization/83781
4524 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4527 2018-01-11 Martin Sebor <msebor@gmail.com>
4528 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4530 PR tree-optimization/83501
4531 PR tree-optimization/81703
4533 * tree-ssa-strlen.c (get_string_cst): Rename...
4534 (get_string_len): ...to this. Handle global constants.
4535 (handle_char_store): Adjust.
4537 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4538 Jim Wilson <jimw@sifive.com>
4540 * config/riscv/riscv-protos.h (riscv_output_return): New.
4541 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4542 (riscv_attribute_table, riscv_output_return),
4543 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4544 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4545 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4547 (riscv_expand_prologue): Add early return for naked function.
4548 (riscv_expand_epilogue): Likewise.
4549 (riscv_function_ok_for_sibcall): Return false for naked function.
4550 (riscv_set_current_function): New.
4551 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4552 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4553 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4554 * doc/extend.texi (RISC-V Function Attributes): New.
4556 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4558 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4559 check for 128-bit long double before checking TCmode.
4560 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4561 128-bit long doubles before checking TFmode or TCmode.
4562 (FLOAT128_IBM_P): Likewise.
4564 2018-01-10 Martin Sebor <msebor@redhat.com>
4566 PR tree-optimization/83671
4567 * builtins.c (c_strlen): Unconditionally return zero for the empty
4569 Use -Warray-bounds for warnings.
4570 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4571 for non-constant array indices with COMPONENT_REF, arrays of
4572 arrays, and pointers to arrays.
4573 (gimple_fold_builtin_strlen): Determine and set length range for
4574 non-constant character arrays.
4576 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4579 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4582 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4584 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4586 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4589 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4590 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4591 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4592 indexed_or_indirect_operand predicate.
4593 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4594 (*vsx_le_perm_load_v8hi): Likewise.
4595 (*vsx_le_perm_load_v16qi): Likewise.
4596 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4597 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4598 (*vsx_le_perm_store_v8hi): Likewise.
4599 (*vsx_le_perm_store_v16qi): Likewise.
4600 (eight unnamed splitters): Likewise.
4602 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4604 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4605 * config/rs6000/emmintrin.h: Likewise.
4606 * config/rs6000/mmintrin.h: Likewise.
4607 * config/rs6000/xmmintrin.h: Likewise.
4609 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4612 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4614 * tree.c (tree_nop_conversion): Return true for location wrapper
4616 (maybe_wrap_with_location): New function.
4617 (selftest::check_strip_nops): New function.
4618 (selftest::test_location_wrappers): New function.
4619 (selftest::tree_c_tests): Call it.
4620 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4621 (maybe_wrap_with_location): New decl.
4622 (EXPR_LOCATION_WRAPPER_P): New macro.
4623 (location_wrapper_p): New inline function.
4624 (tree_strip_any_location_wrapper): New inline function.
4626 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4629 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4630 stack_realign_offset for the largest alignment of stack slot
4632 (ix86_find_max_used_stack_alignment): New function.
4633 (ix86_finalize_stack_frame_flags): Use it. Set
4634 max_used_stack_alignment if we don't realign stack.
4635 * config/i386/i386.h (machine_function): Add
4636 max_used_stack_alignment.
4638 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4640 * config/arm/arm.opt (-mbranch-cost): New option.
4641 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4644 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4647 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4648 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4650 2018-01-10 Richard Biener <rguenther@suse.de>
4653 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4654 early out so it also covers the case where we have a non-NULL
4657 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4659 PR tree-optimization/83753
4660 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4661 for non-strided grouped accesses if the number of elements is 1.
4663 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4666 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4667 * i386.h (TARGET_USE_GATHER): Define.
4668 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4670 2018-01-10 Martin Liska <mliska@suse.cz>
4673 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4674 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4676 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4677 CLEANUP_NO_PARTITIONING is not set.
4679 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4681 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4682 for vectors, as a partial revert of r254296.
4683 * rtl.h (const_vec_p): Delete.
4684 (const_vec_duplicate_p): Don't test for vector CONSTs.
4685 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4686 * expmed.c (make_tree): Likewise.
4689 * common.md (E, F): Use CONSTANT_P instead of checking for
4691 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4692 checking for CONST_VECTOR.
4694 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4697 * predict.c (force_edge_cold): Handle in more sane way edges
4700 2018-01-09 Carl Love <cel@us.ibm.com>
4702 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4704 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4705 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4706 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4707 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4708 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4709 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4710 * config/rs6000/rs6000-protos.h: Add extern defition for
4711 rs6000_generate_float2_double_code.
4712 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4714 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4715 (float2_v2df): Add define_expand.
4717 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4720 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4721 op_mode in the force_to_mode call.
4723 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4725 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4726 instead of checking each element individually.
4727 (aarch64_evpc_uzp): Likewise.
4728 (aarch64_evpc_zip): Likewise.
4729 (aarch64_evpc_ext): Likewise.
4730 (aarch64_evpc_rev): Likewise.
4731 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4732 instead of checking each element individually. Return true without
4734 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4735 whether all selected elements come from the same input, instead of
4736 checking each element individually. Remove calls to gen_rtx_REG,
4737 start_sequence and end_sequence and instead assert that no rtl is
4740 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4742 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4743 order of HIGH and CONST checks.
4745 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4747 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4748 if the destination isn't an SSA_NAME.
4750 2018-01-09 Richard Biener <rguenther@suse.de>
4752 PR tree-optimization/83668
4753 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4755 (canonicalize_loop_form): ... here, renamed from ...
4756 (canonicalize_loop_closed_ssa_form): ... this and amended to
4757 swap successor edges for loop exit blocks to make us use
4758 the RPO order we need for initial schedule generation.
4760 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4762 PR tree-optimization/64811
4763 * match.pd: When optimizing comparisons with Inf, avoid
4764 introducing or losing exceptions from comparisons with NaN.
4766 2018-01-09 Martin Liska <mliska@suse.cz>
4769 * asan.c (shadow_mem_size): Add gcc_assert.
4771 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4773 Don't save registers in main().
4776 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4777 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4778 * config/avr/avr.c (avr_set_current_function): Don't error if
4779 naked, OS_task or OS_main are specified at the same time.
4780 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4782 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4784 * common/config/avr/avr-common.c (avr_option_optimization_table):
4785 Switch on -mmain-is-OS_task for optimizing compilations.
4787 2018-01-09 Richard Biener <rguenther@suse.de>
4789 PR tree-optimization/83572
4790 * graphite.c: Include cfganal.h.
4791 (graphite_transform_loops): Connect infinite loops to exit
4792 and remove fake edges at the end.
4794 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4796 * ipa-inline.c (edge_badness): Revert accidental checkin.
4798 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4801 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4802 symbols; not inline clones.
4804 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4807 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4808 hard registers. Formatting fixes.
4810 PR preprocessor/83722
4811 * gcc.c (try_generate_repro): Pass
4812 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4813 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4816 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4817 Kito Cheng <kito.cheng@gmail.com>
4819 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4820 (riscv_leaf_function_p): Delete.
4821 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4823 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4825 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4827 (do_ifelse): New function.
4828 (do_isel): New function.
4829 (do_sub3): New function.
4830 (do_add3): New function.
4831 (do_load_mask_compare): New function.
4832 (do_overlap_load_compare): New function.
4833 (expand_compare_loop): New function.
4834 (expand_block_compare): Call expand_compare_loop() when appropriate.
4835 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4837 (-mblock-compare-inline-loop-limit): New option.
4839 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4842 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4843 Reverse order of second and third operands in first alternative.
4844 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4845 of first and second elements in UNSPEC_VPERMR vector.
4846 (altivec_expand_vec_perm_le): Likewise.
4848 2018-01-08 Jeff Law <law@redhat.com>
4850 PR rtl-optimizatin/81308
4851 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4852 (process_switch): If group_case_labels makes a change, then set
4854 (pass_convert_switch::execute): If a switch is converted, then
4855 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4857 PR rtl-optimization/81308
4858 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4861 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4863 PR target/83663 - Revert r255946
4864 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4865 generation for cases where splatting a value is not useful.
4866 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4867 across a vec_duplicate and a paradoxical subreg forming a vector
4868 mode to a vec_concat.
4870 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4872 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4873 -march=armv8.3-a variants.
4874 * config/arm/t-multilib: Likewise.
4875 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4877 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4879 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4881 (cceq_ior_compare_complement): Give it a name so I can use it, and
4882 change boolean_or_operator predicate to boolean_operator so it can
4883 be used to generate a crand.
4884 (eqne): New code iterator.
4885 (bd/bd_neg): New code_attrs.
4886 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4887 a single define_insn.
4888 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4889 decrement (bdnzt/bdnzf/bdzt/bdzf).
4890 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4891 with the new names of the branch decrement patterns, and added the
4892 names of the branch decrement conditional patterns.
4894 2018-01-08 Richard Biener <rguenther@suse.de>
4896 PR tree-optimization/83563
4897 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4900 2018-01-08 Richard Biener <rguenther@suse.de>
4903 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4905 2018-01-08 Richard Biener <rguenther@suse.de>
4907 PR tree-optimization/83685
4908 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4909 references to abnormals.
4911 2018-01-08 Richard Biener <rguenther@suse.de>
4914 * dwarf2out.c (output_indirect_strings): Handle empty
4915 skeleton_debug_str_hash.
4916 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4918 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4920 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4921 (emit_store_direct): Likewise.
4922 (arc_trampoline_adjust_address): Likewise.
4923 (arc_asm_trampoline_template): New function.
4924 (arc_initialize_trampoline): Use asm_trampoline_template.
4925 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4926 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4927 * config/arc/arc.md (flush_icache): Delete pattern.
4929 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4931 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4932 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4935 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4938 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4939 by not USED_FOR_TARGET.
4940 (make_pass_resolve_sw_modes): Likewise.
4942 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4944 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4947 2018-01-08 Richard Biener <rguenther@suse.de>
4950 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4952 2018-01-08 Richard Biener <rguenther@suse.de>
4955 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4957 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4960 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4961 basic blocks with a small number of successors.
4962 (convert_control_dep_chain_into_preds): Improve handling of
4964 (dump_predicates): Split apart into...
4965 (dump_pred_chain): ...here...
4966 (dump_pred_info): ...and here.
4967 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4968 (can_chain_union_be_invalidated_p): Improve check for invalidation
4970 (uninit_uses_cannot_happen): Avoid unnecessary if
4971 convert_control_dep_chain_into_preds yielded nothing.
4973 2018-01-06 Martin Sebor <msebor@redhat.com>
4975 PR tree-optimization/83640
4976 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4977 subtracting negative offset from size.
4978 (builtin_access::overlap): Adjust offset bounds of the access to fall
4979 within the size of the object if possible.
4981 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4983 PR rtl-optimization/83699
4984 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4985 extract_bit_field_as_subreg to cases in which the extracted
4986 value is also a vector.
4988 * lra-constraints.c (process_alt_operands): Test for the equivalence
4989 substitutions when detecting a possible reload cycle.
4991 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4994 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4995 by default if flag_selective_schedling{,2}. Formatting fixes.
4997 PR rtl-optimization/83682
4998 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4999 if it has non-VECTOR_MODE element mode.
5000 (vec_duplicate_p): Likewise.
5003 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5004 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5006 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5009 * config/i386/i386-builtin.def
5010 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5011 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5012 Require also OPTION_MASK_ISA_AVX512F in addition to
5013 OPTION_MASK_ISA_GFNI.
5014 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5015 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5016 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5017 to OPTION_MASK_ISA_GFNI.
5018 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5019 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5020 OPTION_MASK_ISA_AVX512BW.
5021 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5022 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5023 addition to OPTION_MASK_ISA_GFNI.
5024 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5025 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5026 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5027 to OPTION_MASK_ISA_GFNI.
5028 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5029 a requirement for all ISAs rather than any of them with a few
5031 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5033 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5034 bitmasks to be enabled with 3 exceptions, instead of requiring any
5035 enabled ISA with lots of exceptions.
5036 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5037 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5038 Change avx512bw in isa attribute to avx512f.
5039 * config/i386/sgxintrin.h: Add license boilerplate.
5040 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5041 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5042 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5043 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5045 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5046 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5047 temporarily sse2 rather than sse if not enabled already.
5050 * config/i386/sse.md (VI248_VLBW): Rename to ...
5051 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5052 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5053 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5054 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5055 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5056 mode iterator instead of VI248_VLBW.
5058 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5060 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5061 (record_modified): Skip clobbers; add debug output.
5062 (param_change_prob): Use sreal frequencies.
5064 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5066 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5067 punt for user-aligned variables.
5069 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5071 * tree-chrec.c (chrec_contains_symbols): Return true for
5074 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5077 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5078 of (x|y) == x for BICS pattern.
5080 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5082 PR tree-optimization/83605
5083 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5084 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5087 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5089 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5090 * config/epiphany/rtems.h: New file.
5092 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5093 Uros Bizjak <ubizjak@gmail.com>
5096 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5097 QIreg_operand instead of register_operand predicate.
5098 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5099 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5100 comments instead of -fmitigate[-_]rop.
5102 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5105 * cgraphunit.c (symbol_table::compile): Switch to text_section
5106 before calling assembly_start debug hook.
5107 * run-rtl-passes.c (run_rtl_passes): Likewise.
5110 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5112 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5113 range_int_cst_p rather than !symbolic_range_p before calling
5114 extract_range_from_multiplicative_op_1.
5116 2018-01-04 Jeff Law <law@redhat.com>
5118 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5119 redundant test in assertion.
5121 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5123 * doc/rtl.texi: Document machine_mode wrapper classes.
5125 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5127 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5130 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5132 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5133 the VEC_PERM_EXPR fold to fail.
5135 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5138 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5139 to switched_sections.
5141 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5144 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5147 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5150 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5151 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5153 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5156 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5157 is BLKmode and bitpos not zero or mode change is needed.
5159 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5162 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5165 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5168 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5169 instead of MULT rtx. Update all corresponding splitters.
5171 (*ssub<modesuffix>): Ditto.
5173 (*cmp_sadd_di): Update split patterns.
5174 (*cmp_sadd_si): Ditto.
5175 (*cmp_sadd_sidi): Ditto.
5176 (*cmp_ssub_di): Ditto.
5177 (*cmp_ssub_si): Ditto.
5178 (*cmp_ssub_sidi): Ditto.
5179 * config/alpha/predicates.md (const23_operand): New predicate.
5180 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5181 Look for ASHIFT, not MULT inner operand.
5182 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5184 2018-01-04 Martin Liska <mliska@suse.cz>
5186 PR gcov-profile/83669
5187 * gcov.c (output_intermediate_file): Add version to intermediate
5189 * doc/gcov.texi: Document new field 'version' in intermediate
5190 file format. Fix location of '-k' option of gcov command.
5192 2018-01-04 Martin Liska <mliska@suse.cz>
5195 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5197 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5199 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5201 2018-01-03 Martin Sebor <msebor@redhat.com>
5203 PR tree-optimization/83655
5204 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5205 checking calls with invalid arguments.
5207 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5209 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5210 (vectorizable_mask_load_store): Delete.
5211 (vectorizable_call): Return false for masked loads and stores.
5212 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5213 instead of gimple_assign_rhs1.
5214 (vectorizable_load): Handle IFN_MASK_LOAD.
5215 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5219 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5221 (vectorizable_mask_load_store): ...here.
5222 (vectorizable_load): ...and here.
5224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5226 * tree-vect-stmts.c (vect_build_all_ones_mask)
5227 (vect_build_zero_merge_argument): New functions, split out from...
5228 (vectorizable_load): ...here.
5230 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5232 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5234 (vectorizable_mask_load_store): ...here.
5235 (vectorizable_store): ...and here.
5237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5239 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5241 (vectorizable_mask_load_store): ...here.
5243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5245 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5246 (vect_model_store_cost): Take a vec_load_store_type instead of a
5248 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5249 (vect_model_store_cost): Take a vec_load_store_type instead of a
5251 (vectorizable_mask_load_store): Update accordingly.
5252 (vectorizable_store): Likewise.
5253 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5255 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5257 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5258 IFN_MASK_LOAD calls here rather than...
5259 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5262 Alan Hayward <alan.hayward@arm.com>
5263 David Sherwood <david.sherwood@arm.com>
5265 * expmed.c (extract_bit_field_1): For vector extracts,
5266 fall back to extract_bit_field_as_subreg if vec_extract
5269 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5270 Alan Hayward <alan.hayward@arm.com>
5271 David Sherwood <david.sherwood@arm.com>
5273 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5274 they are variable or constant sized.
5275 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5276 slots for constant-sized data.
5278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5279 Alan Hayward <alan.hayward@arm.com>
5280 David Sherwood <david.sherwood@arm.com>
5282 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5283 handling COND_EXPRs with boolean comparisons, try to find a better
5284 basis for the mask type than the boolean itself.
5286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5288 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5289 is calculated and how it can be overridden.
5290 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5291 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5293 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5297 Alan Hayward <alan.hayward@arm.com>
5298 David Sherwood <david.sherwood@arm.com>
5300 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5301 Remove the mode argument.
5302 (aarch64_simd_valid_immediate): Remove the mode and inverse
5304 * config/aarch64/iterators.md (bitsize): New iterator.
5305 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5306 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5307 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5308 aarch64_simd_valid_immediate.
5309 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5310 (aarch64_reg_or_bic_imm): Likewise.
5311 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5312 with an insn_type enum and msl with a modifier_type enum.
5313 Replace element_width with a scalar_mode. Change the shift
5314 to unsigned int. Add constructors for scalar_float_mode and
5315 scalar_int_mode elements.
5316 (aarch64_vect_float_const_representable_p): Delete.
5317 (aarch64_can_const_movi_rtx_p)
5318 (aarch64_simd_scalar_immediate_valid_for_move)
5319 (aarch64_simd_make_constant): Update call to
5320 aarch64_simd_valid_immediate.
5321 (aarch64_advsimd_valid_immediate_hs): New function.
5322 (aarch64_advsimd_valid_immediate): Likewise.
5323 (aarch64_simd_valid_immediate): Remove mode and inverse
5324 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5325 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5326 and aarch64_float_const_representable_p on the result.
5327 (aarch64_output_simd_mov_immediate): Remove mode argument.
5328 Update call to aarch64_simd_valid_immediate and use of
5329 simd_immediate_info.
5330 (aarch64_output_scalar_simd_mov_immediate): Update call
5333 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5334 Alan Hayward <alan.hayward@arm.com>
5335 David Sherwood <david.sherwood@arm.com>
5337 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5338 (mode_nunits): Likewise CONST_MODE_NUNITS.
5339 * machmode.def (ADJUST_NUNITS): Document.
5340 * genmodes.c (mode_data::need_nunits_adj): New field.
5341 (blank_mode): Update accordingly.
5342 (adj_nunits): New variable.
5343 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5345 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5346 listed in adj_nunits.
5347 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5348 listed in adj_nunits. Don't emit case statements for such modes.
5349 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5350 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5351 nothing if adj_nunits is nonnull.
5352 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5353 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5354 (emit_mode_fbit): Update use of print_maybe_const_decl.
5355 (emit_move_size): Likewise. Treat the array as non-const
5357 (emit_mode_adjustments): Handle adj_nunits.
5359 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5361 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5362 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5363 (VECTOR_MODES): Use it.
5364 (make_vector_modes): Take the prefix as an argument.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5371 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5372 for MODE_VECTOR_BOOL.
5373 * machmode.def (VECTOR_BOOL_MODE): Document.
5374 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5375 (make_vector_bool_mode): New function.
5376 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5378 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5379 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5381 * stor-layout.c (int_mode_for_mode): Likewise.
5382 * tree.c (build_vector_type_for_mode): Likewise.
5383 * varasm.c (output_constant_pool_2): Likewise.
5384 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5385 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5386 for MODE_VECTOR_BOOL.
5387 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5388 of mode class checks.
5389 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5390 instead of a list of mode class checks.
5391 (expand_vector_scalar_condition): Likewise.
5392 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5394 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5395 Alan Hayward <alan.hayward@arm.com>
5396 David Sherwood <david.sherwood@arm.com>
5398 * machmode.h (mode_size): Change from unsigned short to
5400 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5401 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5402 or if measurement_type is not polynomial.
5403 (fixed_size_mode::includes_p): Check for constant-sized modes.
5404 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5405 return a poly_uint16 rather than an unsigned short.
5406 (emit_mode_size): Change the type of mode_size from unsigned short
5407 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5408 (emit_mode_adjustments): Cope with polynomial vector sizes.
5409 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5411 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5413 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5414 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5415 * caller-save.c (setup_save_areas): Likewise.
5416 (replace_reg_with_saved_mem): Likewise.
5417 * calls.c (emit_library_call_value_1): Likewise.
5418 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5419 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5420 (gen_lowpart_for_combine): Likewise.
5421 * convert.c (convert_to_integer_1): Likewise.
5422 * cse.c (equiv_constant, cse_insn): Likewise.
5423 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5424 (cselib_subst_to_values): Likewise.
5425 * dce.c (word_dce_process_block): Likewise.
5426 * df-problems.c (df_word_lr_mark_ref): Likewise.
5427 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5428 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5429 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5430 (rtl_for_decl_location): Likewise.
5431 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5432 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5433 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5434 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5435 (expand_expr_real_1): Likewise.
5436 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5437 (pad_below): Likewise.
5438 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5439 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5440 * ira.c (get_subreg_tracking_sizes): Likewise.
5441 * ira-build.c (ira_create_allocno_objects): Likewise.
5442 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5443 (ira_sort_regnos_for_alter_reg): Likewise.
5444 * ira-costs.c (record_operand_costs): Likewise.
5445 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5446 (resolve_simple_move): Likewise.
5447 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5448 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5449 (lra_constraints): Likewise.
5450 (CONST_POOL_OK_P): Reject variable-sized modes.
5451 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5452 (add_pseudo_to_slot, lra_spill): Likewise.
5453 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5454 * optabs-query.c (get_best_extraction_insn): Likewise.
5455 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5456 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5457 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5458 * recog.c (offsettable_address_addr_space_p): Likewise.
5459 * regcprop.c (maybe_mode_change): Likewise.
5460 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5461 * regrename.c (build_def_use): Likewise.
5462 * regstat.c (dump_reg_info): Likewise.
5463 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5464 (find_reloads, find_reloads_subreg_address): Likewise.
5465 * reload1.c (eliminate_regs_1): Likewise.
5466 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5467 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5468 (simplify_binary_operation_1, simplify_subreg): Likewise.
5469 * targhooks.c (default_function_arg_padding): Likewise.
5470 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5471 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5472 (verify_gimple_assign_ternary): Likewise.
5473 * tree-inline.c (estimate_move_cost): Likewise.
5474 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5475 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5476 (get_address_cost_ainc): Likewise.
5477 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5478 (vect_supportable_dr_alignment): Likewise.
5479 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5480 (vectorizable_reduction): Likewise.
5481 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5482 (vectorizable_operation, vectorizable_load): Likewise.
5483 * tree.c (build_same_sized_truth_vector_type): Likewise.
5484 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5485 * var-tracking.c (emit_note_insn_var_location): Likewise.
5486 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5487 (ADDR_VEC_ALIGN): Likewise.
5489 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5490 Alan Hayward <alan.hayward@arm.com>
5491 David Sherwood <david.sherwood@arm.com>
5493 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5495 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5496 or if measurement_type is polynomial.
5497 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5498 * combine.c (make_extraction): Likewise.
5499 * dse.c (find_shift_sequence): Likewise.
5500 * dwarf2out.c (mem_loc_descriptor): Likewise.
5501 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5502 (extract_bit_field, extract_low_bits): Likewise.
5503 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5504 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5505 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5506 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5507 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5508 * reload.c (find_reloads): Likewise.
5509 * reload1.c (alter_reg): Likewise.
5510 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5511 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5512 * tree-if-conv.c (predicate_mem_writes): Likewise.
5513 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5514 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5515 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5516 * valtrack.c (dead_debug_insert_temp): Likewise.
5517 * varasm.c (mergeable_constant_section): Likewise.
5518 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5520 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5521 Alan Hayward <alan.hayward@arm.com>
5522 David Sherwood <david.sherwood@arm.com>
5524 * expr.c (expand_assignment): Cope with polynomial mode sizes
5525 when assigning to a CONCAT.
5527 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5528 Alan Hayward <alan.hayward@arm.com>
5529 David Sherwood <david.sherwood@arm.com>
5531 * machmode.h (mode_precision): Change from unsigned short to
5533 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5535 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5536 or if measurement_type is not polynomial.
5537 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5538 in which the mode is already known to be a scalar_int_mode.
5539 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5540 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5542 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5543 for GET_MODE_PRECISION.
5544 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5545 for GET_MODE_PRECISION.
5546 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5548 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5549 (expand_field_assignment, make_extraction): Likewise.
5550 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5551 (get_last_value): Likewise.
5552 * convert.c (convert_to_integer_1): Likewise.
5553 * cse.c (cse_insn): Likewise.
5554 * expr.c (expand_expr_real_1): Likewise.
5555 * lra-constraints.c (simplify_operand_subreg): Likewise.
5556 * optabs-query.c (can_atomic_load_p): Likewise.
5557 * optabs.c (expand_atomic_load): Likewise.
5558 (expand_atomic_store): Likewise.
5559 * ree.c (combine_reaching_defs): Likewise.
5560 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5561 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5562 * tree.h (type_has_mode_precision_p): Likewise.
5563 * ubsan.c (instrument_si_overflow): Likewise.
5565 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5566 Alan Hayward <alan.hayward@arm.com>
5567 David Sherwood <david.sherwood@arm.com>
5569 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5570 polynomial numbers of units.
5571 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5572 (valid_vector_subparts_p): New function.
5573 (build_vector_type): Remove temporary shim and take the number
5574 of units as a poly_uint64 rather than an int.
5575 (build_opaque_vector_type): Take the number of units as a
5576 poly_uint64 rather than an int.
5577 * tree.c (build_vector_from_ctor): Handle polynomial
5578 TYPE_VECTOR_SUBPARTS.
5579 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5580 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5581 (build_vector_from_val): If the number of units is variable,
5582 use build_vec_duplicate_cst for constant operands and
5583 VEC_DUPLICATE_EXPR otherwise.
5584 (make_vector_type): Remove temporary is_constant ().
5585 (build_vector_type, build_opaque_vector_type): Take the number of
5586 units as a poly_uint64 rather than an int.
5587 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5589 * cfgexpand.c (expand_debug_expr): Likewise.
5590 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5591 (store_constructor, expand_expr_real_1): Likewise.
5592 (const_scalar_mask_from_tree): Likewise.
5593 * fold-const-call.c (fold_const_reduction): Likewise.
5594 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5595 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5596 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5597 (fold_relational_const): Likewise.
5598 (native_interpret_vector): Likewise. Change the size from an
5599 int to an unsigned int.
5600 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5601 TYPE_VECTOR_SUBPARTS.
5602 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5603 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5604 duplicating a non-constant operand into a variable-length vector.
5605 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5606 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5607 * ipa-icf.c (sem_variable::equals): Likewise.
5608 * match.pd: Likewise.
5609 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5610 * print-tree.c (print_node): Likewise.
5611 * stor-layout.c (layout_type): Likewise.
5612 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5613 * tree-cfg.c (verify_gimple_comparison): Likewise.
5614 (verify_gimple_assign_binary): Likewise.
5615 (verify_gimple_assign_ternary): Likewise.
5616 (verify_gimple_assign_single): Likewise.
5617 * tree-pretty-print.c (dump_generic_node): Likewise.
5618 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5619 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5620 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5621 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5622 (vect_shift_permute_load_chain): Likewise.
5623 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5624 (expand_vector_condition, optimize_vector_constructor): Likewise.
5625 (lower_vec_perm, get_compute_type): Likewise.
5626 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5627 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5628 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5629 (vect_recog_mask_conversion_pattern): Likewise.
5630 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5631 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5632 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5633 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5634 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5635 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5636 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5637 (supportable_widening_operation): Likewise.
5638 (supportable_narrowing_operation): Likewise.
5639 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5641 * varasm.c (output_constant): Likewise.
5643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5644 Alan Hayward <alan.hayward@arm.com>
5645 David Sherwood <david.sherwood@arm.com>
5647 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5648 so that both the length == 3 and length != 3 cases set up their
5649 own permute vectors. Add comments explaining why we know the
5650 number of elements is constant.
5651 (vect_permute_load_chain): Likewise.
5653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5654 Alan Hayward <alan.hayward@arm.com>
5655 David Sherwood <david.sherwood@arm.com>
5657 * machmode.h (mode_nunits): Change from unsigned char to
5659 (ONLY_FIXED_SIZE_MODES): New macro.
5660 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5661 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5662 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5664 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5665 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5666 or if measurement_type is not polynomial.
5667 * genmodes.c (ZERO_COEFFS): New macro.
5668 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5670 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5671 Use ZERO_COEFFS when emitting initializers.
5672 * data-streamer.h (bp_pack_poly_value): New function.
5673 (bp_unpack_poly_value): Likewise.
5674 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5675 for GET_MODE_NUNITS.
5676 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5677 for GET_MODE_NUNITS.
5678 * tree.c (make_vector_type): Remove temporary shim and make
5679 the real function take the number of units as a poly_uint64
5681 (build_vector_type_for_mode): Handle polynomial nunits.
5682 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5683 * emit-rtl.c (const_vec_series_p_1): Likewise.
5684 (gen_rtx_CONST_VECTOR): Likewise.
5685 * fold-const.c (test_vec_duplicate_folding): Likewise.
5686 * genrecog.c (validate_pattern): Likewise.
5687 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5688 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5689 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5690 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5691 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5692 * rtlanal.c (subreg_get_info): Likewise.
5693 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5694 (vect_grouped_load_supported): Likewise.
5695 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5696 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5697 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5698 (simplify_const_unary_operation, simplify_binary_operation_1)
5699 (simplify_const_binary_operation, simplify_ternary_operation)
5700 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5701 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5702 instead of CONST_VECTOR_NUNITS.
5703 * varasm.c (output_constant_pool_2): Likewise.
5704 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5705 explicit-encoded elements in the XVEC for variable-length vectors.
5707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5709 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5711 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5712 Alan Hayward <alan.hayward@arm.com>
5713 David Sherwood <david.sherwood@arm.com>
5715 * coretypes.h (fixed_size_mode): Declare.
5716 (fixed_size_mode_pod): New typedef.
5717 * builtins.h (target_builtins::x_apply_args_mode)
5718 (target_builtins::x_apply_result_mode): Change type to
5719 fixed_size_mode_pod.
5720 * builtins.c (apply_args_size, apply_result_size, result_vector)
5721 (expand_builtin_apply_args_1, expand_builtin_apply)
5722 (expand_builtin_return): Update accordingly.
5724 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5726 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5727 * cselib.c (cselib_hash_rtx): Likewise.
5728 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5729 CONST_VECTOR encoding.
5731 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5732 Jeff Law <law@redhat.com>
5735 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5736 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5737 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5738 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5741 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5742 explicitly probe *sp in a noreturn function if there were any callee
5743 register saves or frame pointer is needed.
5745 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5748 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5749 BLKmode for ternary, binary or unary expressions.
5752 * var-tracking.c (delete_vta_debug_insn): New inline function.
5753 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5754 insns from get_insns () to NULL instead of each bb separately.
5755 Use delete_vta_debug_insn. No longer static.
5756 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5757 delete_vta_debug_insns callers.
5758 * rtl.h (delete_vta_debug_insns): Declare.
5759 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5760 instead of variable_tracking_main.
5762 2018-01-03 Martin Sebor <msebor@redhat.com>
5764 PR tree-optimization/83603
5765 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5766 arguments past the endof the argument list in functions declared
5767 without a prototype.
5768 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5769 Avoid checking when arguments are null.
5771 2018-01-03 Martin Sebor <msebor@redhat.com>
5774 * doc/extend.texi (attribute const): Fix a typo.
5775 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5776 issuing -Wsuggest-attribute for void functions.
5778 2018-01-03 Martin Sebor <msebor@redhat.com>
5780 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5781 offset_int::from instead of wide_int::to_shwi.
5782 (maybe_diag_overlap): Remove assertion.
5783 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5784 * gimple-ssa-sprintf.c (format_directive): Same.
5785 (parse_directive): Same.
5786 (sprintf_dom_walker::compute_format_length): Same.
5787 (try_substitute_return_value): Same.
5789 2018-01-03 Jeff Law <law@redhat.com>
5792 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5793 non-constant residual for zero at runtime and avoid probing in
5794 that case. Reorganize code for trailing problem to mirror handling
5797 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5799 PR tree-optimization/83501
5800 * tree-ssa-strlen.c (get_string_cst): New.
5801 (handle_char_store): Call get_string_cst.
5803 2018-01-03 Martin Liska <mliska@suse.cz>
5805 PR tree-optimization/83593
5806 * tree-ssa-strlen.c: Include tree-cfg.h.
5807 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5808 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5809 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5811 (strlen_dom_walker::before_dom_children): Call
5812 gimple_purge_dead_eh_edges. Dump tranformation with details
5814 (strlen_dom_walker::before_dom_children): Update call by adding
5815 new argument cleanup_eh.
5816 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5818 2018-01-03 Martin Liska <mliska@suse.cz>
5821 * cif-code.def (VARIADIC_THUNK): New enum value.
5822 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5825 2018-01-03 Jan Beulich <jbeulich@suse.com>
5827 * sse.md (mov<mode>_internal): Tighten condition for when to use
5828 vmovdqu<ssescalarsize> for TI and OI modes.
5830 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5832 Update copyright years.
5834 2018-01-03 Martin Liska <mliska@suse.cz>
5837 * ipa-visibility.c (function_and_variable_visibility): Skip
5838 functions with noipa attribure.
5840 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5842 * gcc.c (process_command): Update copyright notice dates.
5843 * gcov-dump.c (print_version): Ditto.
5844 * gcov.c (print_version): Ditto.
5845 * gcov-tool.c (print_version): Ditto.
5846 * gengtype.c (create_file): Ditto.
5847 * doc/cpp.texi: Bump @copying's copyright year.
5848 * doc/cppinternals.texi: Ditto.
5849 * doc/gcc.texi: Ditto.
5850 * doc/gccint.texi: Ditto.
5851 * doc/gcov.texi: Ditto.
5852 * doc/install.texi: Ditto.
5853 * doc/invoke.texi: Ditto.
5855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5857 * vector-builder.h (vector_builder::m_full_nelts): Change from
5858 unsigned int to poly_uint64.
5859 (vector_builder::full_nelts): Update prototype accordingly.
5860 (vector_builder::new_vector): Likewise.
5861 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5862 (vector_builder::operator ==): Likewise.
5863 (vector_builder::finalize): Likewise.
5864 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5865 Take the number of elements as a poly_uint64 rather than an
5867 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5868 from unsigned int to poly_uint64.
5869 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5870 (vec_perm_indices::new_vector): Likewise.
5871 (vec_perm_indices::length): Likewise.
5872 (vec_perm_indices::nelts_per_input): Likewise.
5873 (vec_perm_indices::input_nelts): Likewise.
5874 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5875 number of elements per input as a poly_uint64 rather than an
5876 unsigned int. Use the original encoding for variable-length
5877 vectors, rather than clamping each individual element.
5878 For the second and subsequent elements in each pattern,
5879 clamp the step and base before clamping their sum.
5880 (vec_perm_indices::series_p): Handle polynomial element counts.
5881 (vec_perm_indices::all_in_range_p): Likewise.
5882 (vec_perm_indices_to_tree): Likewise.
5883 (vec_perm_indices_to_rtx): Likewise.
5884 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5885 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5886 (tree_vector_builder::new_binary_operation): Handle polynomial
5887 element counts. Return false if we need to know the number
5888 of elements at compile time.
5889 * fold-const.c (fold_vec_perm): Punt if the number of elements
5890 isn't known at compile time.
5892 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5894 * vec-perm-indices.h (vec_perm_builder): Change element type
5895 from HOST_WIDE_INT to poly_int64.
5896 (vec_perm_indices::element_type): Update accordingly.
5897 (vec_perm_indices::clamp): Handle polynomial element_types.
5898 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5899 (vec_perm_indices::all_in_range_p): Likewise.
5900 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5902 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5903 polynomial vec_perm_indices element types.
5904 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5905 * fold-const.c (fold_vec_perm): Likewise.
5906 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5907 * tree-vect-generic.c (lower_vec_perm): Likewise.
5908 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5909 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5910 element type to HOST_WIDE_INT.
5912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5913 Alan Hayward <alan.hayward@arm.com>
5914 David Sherwood <david.sherwood@arm.com>
5916 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5917 rather than an int. Use plus_constant.
5918 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5919 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5921 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5922 Alan Hayward <alan.hayward@arm.com>
5923 David Sherwood <david.sherwood@arm.com>
5925 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5926 a HOST_WIDE_INT to a poly_int64.
5928 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5929 Alan Hayward <alan.hayward@arm.com>
5930 David Sherwood <david.sherwood@arm.com>
5932 * calls.c (load_register_parameters): Cope with polynomial
5933 mode sizes. Require a constant size for BLKmode parameters
5934 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5935 forces a parameter to be padded at the lsb end in order to
5936 fill a complete number of words, require the parameter size
5937 to be ordered wrt UNITS_PER_WORD.
5939 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5940 Alan Hayward <alan.hayward@arm.com>
5941 David Sherwood <david.sherwood@arm.com>
5943 * reload1.c (spill_stack_slot_width): Change element type
5944 from unsigned int to poly_uint64_pod.
5945 (alter_reg): Treat mode sizes as polynomial.
5947 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5948 Alan Hayward <alan.hayward@arm.com>
5949 David Sherwood <david.sherwood@arm.com>
5951 * reload.c (complex_word_subreg_p): New function.
5952 (reload_inner_reg_of_subreg, push_reload): Use it.
5954 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5955 Alan Hayward <alan.hayward@arm.com>
5956 David Sherwood <david.sherwood@arm.com>
5958 * lra-constraints.c (process_alt_operands): Reject matched
5959 operands whose sizes aren't ordered.
5960 (match_reload): Refer to this check here.
5962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5963 Alan Hayward <alan.hayward@arm.com>
5964 David Sherwood <david.sherwood@arm.com>
5966 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5967 that the mode size is in the set {1, 2, 4, 8, 16}.
5969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5970 Alan Hayward <alan.hayward@arm.com>
5971 David Sherwood <david.sherwood@arm.com>
5973 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5974 Use plus_constant instead of gen_rtx_PLUS.
5976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5977 Alan Hayward <alan.hayward@arm.com>
5978 David Sherwood <david.sherwood@arm.com>
5980 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5981 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5982 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5983 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5984 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5985 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5986 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5987 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5988 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5989 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5991 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5992 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5993 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5994 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5995 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5996 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5997 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5998 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5999 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6000 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6002 * expr.c (emit_move_resolve_push): Treat the input and result
6003 of PUSH_ROUNDING as a poly_int64.
6004 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6005 (emit_push_insn): Likewise.
6006 * lra-eliminations.c (mark_not_eliminable): Likewise.
6007 * recog.c (push_operand): Likewise.
6008 * reload1.c (elimination_effects): Likewise.
6009 * rtlanal.c (nonzero_bits1): Likewise.
6010 * calls.c (store_one_arg): Likewise. Require the padding to be
6011 known at compile time.
6013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6014 Alan Hayward <alan.hayward@arm.com>
6015 David Sherwood <david.sherwood@arm.com>
6017 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6018 Use plus_constant instead of gen_rtx_PLUS.
6020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6021 Alan Hayward <alan.hayward@arm.com>
6022 David Sherwood <david.sherwood@arm.com>
6024 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6027 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6028 Alan Hayward <alan.hayward@arm.com>
6029 David Sherwood <david.sherwood@arm.com>
6031 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6032 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6033 via stack temporaries. Treat the mode size as polynomial too.
6035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6036 Alan Hayward <alan.hayward@arm.com>
6037 David Sherwood <david.sherwood@arm.com>
6039 * expr.c (expand_expr_real_2): When handling conversions involving
6040 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6041 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6042 as a poly_uint64 too.
6044 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6045 Alan Hayward <alan.hayward@arm.com>
6046 David Sherwood <david.sherwood@arm.com>
6048 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
6055 REGMODE_NATURAL_SIZE.
6056 * expmed.c (store_bit_field_1): Likewise.
6057 * expr.c (store_constructor): Likewise.
6058 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6059 and polynomial REGMODE_NATURAL_SIZE.
6060 (gen_lowpart_common): Likewise.
6061 * reginfo.c (record_subregs_of_mode): Likewise.
6062 * rtlanal.c (read_modify_subreg_p): Likewise.
6064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6065 Alan Hayward <alan.hayward@arm.com>
6066 David Sherwood <david.sherwood@arm.com>
6068 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6069 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
6077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6078 Alan Hayward <alan.hayward@arm.com>
6079 David Sherwood <david.sherwood@arm.com>
6081 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6082 in a POINTER_PLUS_EXPR.
6084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6085 Alan Hayward <alan.hayward@arm.com>
6086 David Sherwood <david.sherwood@arm.com>
6088 * omp-simd-clone.c (simd_clone_subparts): New function.
6089 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6090 (ipa_simd_modify_function_body): Likewise.
6092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6093 Alan Hayward <alan.hayward@arm.com>
6094 David Sherwood <david.sherwood@arm.com>
6096 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6097 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6098 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6099 (expand_vector_condition, vector_element): Likewise.
6100 (subparts_gt): New function.
6101 (get_compute_type): Use subparts_gt.
6102 (count_type_subparts): Delete.
6103 (expand_vector_operations_1): Use subparts_gt instead of
6104 count_type_subparts.
6106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6107 Alan Hayward <alan.hayward@arm.com>
6108 David Sherwood <david.sherwood@arm.com>
6110 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6111 (vect_compile_time_alias): ...this new function. Do the calculation
6112 on poly_ints rather than trees.
6113 (vect_prune_runtime_alias_test_list): Update call accordingly.
6115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6116 Alan Hayward <alan.hayward@arm.com>
6117 David Sherwood <david.sherwood@arm.com>
6119 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6121 (vect_schedule_slp_instance): Likewise.
6123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6124 Alan Hayward <alan.hayward@arm.com>
6125 David Sherwood <david.sherwood@arm.com>
6127 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6128 constant and extern definitions for variable-length vectors.
6129 (vect_get_constant_vectors): Note that the number of units
6130 is known to be constant.
6132 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6133 Alan Hayward <alan.hayward@arm.com>
6134 David Sherwood <david.sherwood@arm.com>
6136 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6137 of units as polynomial. Choose between WIDE and NARROW based
6140 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6141 Alan Hayward <alan.hayward@arm.com>
6142 David Sherwood <david.sherwood@arm.com>
6144 * tree-vect-stmts.c (simd_clone_subparts): New function.
6145 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6147 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6148 Alan Hayward <alan.hayward@arm.com>
6149 David Sherwood <david.sherwood@arm.com>
6151 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6152 vectors as polynomial. Use build_index_vector for
6155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6156 Alan Hayward <alan.hayward@arm.com>
6157 David Sherwood <david.sherwood@arm.com>
6159 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6160 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6161 for variable-length vectors.
6162 (vectorizable_mask_load_store): Treat the number of units as
6163 polynomial, asserting that it is constant if the condition has
6164 already been enforced.
6165 (vectorizable_store, vectorizable_load): Likewise.
6167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6168 Alan Hayward <alan.hayward@arm.com>
6169 David Sherwood <david.sherwood@arm.com>
6171 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6172 of units as polynomial. Punt if we can't tell at compile time
6173 which vector contains the final result.
6175 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6176 Alan Hayward <alan.hayward@arm.com>
6177 David Sherwood <david.sherwood@arm.com>
6179 * tree-vect-loop.c (vectorizable_induction): Treat the number
6180 of units as polynomial. Punt on SLP inductions. Use an integer
6181 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6182 cast of such a series for variable-length floating-point
6185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6186 Alan Hayward <alan.hayward@arm.com>
6187 David Sherwood <david.sherwood@arm.com>
6189 * tree.h (build_index_vector): Declare.
6190 * tree.c (build_index_vector): New function.
6191 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6192 of units as polynomial, forcibly converting it to a constant if
6193 vectorizable_reduction has already enforced the condition.
6194 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6195 to create a {1,2,3,...} vector.
6196 (vectorizable_reduction): Treat the number of units as polynomial.
6197 Choose vectype_in based on the largest scalar element size rather
6198 than the smallest number of units. Enforce the restrictions
6201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6202 Alan Hayward <alan.hayward@arm.com>
6203 David Sherwood <david.sherwood@arm.com>
6205 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6206 number of units as polynomial.
6208 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6209 Alan Hayward <alan.hayward@arm.com>
6210 David Sherwood <david.sherwood@arm.com>
6212 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6213 * target.def (autovectorize_vector_sizes): Return the vector sizes
6214 by pointer, using vector_sizes rather than a bitmask.
6215 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6216 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6217 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6219 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6220 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6221 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6222 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6223 * omp-general.c (omp_max_vf): Likewise.
6224 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6225 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6226 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6227 * tree-vect-slp.c (vect_slp_bb): Likewise.
6228 * doc/tm.texi: Regenerate.
6229 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6231 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6232 the vector size as a poly_uint64 rather than an unsigned int.
6233 (current_vector_size): Change from an unsigned int to a poly_uint64.
6234 (get_vectype_for_scalar_type): Update accordingly.
6235 * tree.h (build_truth_vector_type): Take the size and number of
6236 units as a poly_uint64 rather than an unsigned int.
6237 (build_vector_type): Add a temporary overload that takes
6238 the number of units as a poly_uint64 rather than an unsigned int.
6239 * tree.c (make_vector_type): Likewise.
6240 (build_truth_vector_type): Take the number of units as a poly_uint64
6241 rather than an unsigned int.
6243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6244 Alan Hayward <alan.hayward@arm.com>
6245 David Sherwood <david.sherwood@arm.com>
6247 * target.def (get_mask_mode): Take the number of units and length
6248 as poly_uint64s rather than unsigned ints.
6249 * targhooks.h (default_get_mask_mode): Update accordingly.
6250 * targhooks.c (default_get_mask_mode): Likewise.
6251 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6252 * doc/tm.texi: Regenerate.
6254 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6255 Alan Hayward <alan.hayward@arm.com>
6256 David Sherwood <david.sherwood@arm.com>
6258 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6259 * omp-general.c (omp_max_vf): Likewise.
6260 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6261 (expand_omp_simd): Handle polynomial safelen.
6262 * omp-low.c (omplow_simd_context): Add a default constructor.
6263 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6264 (lower_rec_simd_input_clauses): Update accordingly.
6265 (lower_rec_input_clauses): Likewise.
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 (vect_nunits_for_cost): New function.
6272 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6273 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6274 (vect_analyze_slp_cost): Likewise.
6275 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6276 (vect_model_load_cost): Likewise.
6278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6279 Alan Hayward <alan.hayward@arm.com>
6280 David Sherwood <david.sherwood@arm.com>
6282 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6283 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6284 from an unsigned int * to a poly_uint64_pod *.
6285 (calculate_unrolling_factor): New function.
6286 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6289 Alan Hayward <alan.hayward@arm.com>
6290 David Sherwood <david.sherwood@arm.com>
6292 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6293 from an unsigned int to a poly_uint64.
6294 (_loop_vec_info::slp_unrolling_factor): Likewise.
6295 (_loop_vec_info::vectorization_factor): Change from an int
6297 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6298 (vect_get_num_vectors): New function.
6299 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6300 (vect_get_num_copies): Use vect_get_num_vectors.
6301 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6302 to an unsigned int *.
6303 (vect_analyze_data_refs): Change min_vf from an int * to a
6305 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6306 than an unsigned HOST_WIDE_INT.
6307 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6308 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6309 to an unsigned int *.
6310 (vect_analyze_data_ref_dependences): Likewise.
6311 (vect_compute_data_ref_alignment): Handle polynomial vf.
6312 (vect_enhance_data_refs_alignment): Likewise.
6313 (vect_prune_runtime_alias_test_list): Likewise.
6314 (vect_shift_permute_load_chain): Likewise.
6315 (vect_supportable_dr_alignment): Likewise.
6316 (dependence_distance_ge_vf): Take the vectorization factor as a
6317 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6318 (vect_analyze_data_refs): Change min_vf from an int * to a
6320 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6321 vfm1 as a poly_uint64 rather than an int. Make the same change
6322 for the returned bound_scalar.
6323 (vect_gen_vector_loop_niters): Handle polynomial vf.
6324 (vect_do_peeling): Likewise. Update call to
6325 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6326 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6328 * tree-vect-loop.c (vect_determine_vectorization_factor)
6329 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6330 (vect_get_known_peeling_cost): Likewise.
6331 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6332 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6333 (vect_transform_loop): Likewise. Use the lowest possible VF when
6334 updating the upper bounds of the loop.
6335 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6337 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6338 polynomial unroll factors.
6339 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6340 (vect_make_slp_decision): Likewise.
6341 (vect_supported_load_permutation_p): Likewise, and polynomial
6343 (vect_analyze_slp_cost): Handle polynomial vf.
6344 (vect_slp_analyze_node_operations): Likewise.
6345 (vect_slp_analyze_bb_1): Likewise.
6346 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6347 than an unsigned HOST_WIDE_INT.
6348 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6349 (vectorizable_load): Handle polynomial vf.
6350 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6352 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6355 Alan Hayward <alan.hayward@arm.com>
6356 David Sherwood <david.sherwood@arm.com>
6358 * match.pd: Handle bit operations involving three constants
6359 and try to fold one pair.
6361 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6363 * tree-vect-loop-manip.c: Include gimple-fold.h.
6364 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6365 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6366 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6367 Add a path that uses a step of VF instead of 1, but disable it
6369 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6370 and niters_no_overflow parameters. Update calls to
6371 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6372 Create a new SSA name if the latter choses to use a ste other
6373 than zero, and return it via niters_vector_mult_vf_var.
6374 * tree-vect-loop.c (vect_transform_loop): Update calls to
6375 vect_do_peeling, vect_gen_vector_loop_niters and
6376 slpeel_make_loop_iterate_ntimes.
6377 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6378 (vect_gen_vector_loop_niters): Update declarations after above changes.
6380 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6382 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6383 128-bit round to integer instructions.
6384 (ceil<mode>2): Likewise.
6385 (btrunc<mode>2): Likewise.
6386 (round<mode>2): Likewise.
6388 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6390 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6391 unaligned VSX load/store on P8/P9.
6392 (expand_block_clear): Allow the use of unaligned VSX
6393 load/store on P8/P9.
6395 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6397 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6399 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6400 swap associated with both a load and a store.
6402 2018-01-02 Andrew Waterman <andrew@sifive.com>
6404 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6405 * config/riscv/riscv.md (clear_cache): Use it.
6407 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6409 * web.c: Remove out-of-date comment.
6411 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6413 * expr.c (fixup_args_size_notes): Check that any existing
6414 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6415 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6416 (emit_single_push_insn): ...here.
6418 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6420 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6421 (const_vector_encoded_nelts): New function.
6422 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6423 (const_vector_int_elt, const_vector_elt): Declare.
6424 * emit-rtl.c (const_vector_int_elt_1): New function.
6425 (const_vector_elt): Likewise.
6426 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6427 of CONST_VECTOR_ELT.
6429 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6431 * expr.c: Include rtx-vector-builder.h.
6432 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6433 directly on the tree encoding.
6434 (const_vector_from_tree): Likewise.
6435 * optabs.c: Include rtx-vector-builder.h.
6436 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6437 sequence of "u" values.
6438 * vec-perm-indices.c: Include rtx-vector-builder.h.
6439 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6440 directly on the vec_perm_indices encoding.
6442 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6444 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6445 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6446 * rtx-vector-builder.h: New file.
6447 * rtx-vector-builder.c: Likewise.
6448 * rtl.h (rtx_def::u2): Add a const_vector field.
6449 (CONST_VECTOR_NPATTERNS): New macro.
6450 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6451 (CONST_VECTOR_DUPLICATE_P): Likewise.
6452 (CONST_VECTOR_STEPPED_P): Likewise.
6453 (CONST_VECTOR_ENCODED_ELT): Likewise.
6454 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6455 (unwrap_const_vec_duplicate): Likewise.
6456 (const_vec_series_p): Check for a non-duplicated vector encoding.
6457 Say that the function only returns true for integer vectors.
6458 * emit-rtl.c: Include rtx-vector-builder.h.
6459 (gen_const_vec_duplicate_1): Delete.
6460 (gen_const_vector): Call gen_const_vec_duplicate instead of
6461 gen_const_vec_duplicate_1.
6462 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6463 (gen_const_vec_duplicate): Use rtx_vector_builder.
6464 (gen_const_vec_series): Likewise.
6465 (gen_rtx_CONST_VECTOR): Likewise.
6466 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6467 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6468 Build a new vector rather than modifying a CONST_VECTOR in-place.
6469 (handle_special_swappables): Update call accordingly.
6470 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6471 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6472 Build a new vector rather than modifying a CONST_VECTOR in-place.
6473 (handle_special_swappables): Update call accordingly.
6475 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6477 * simplify-rtx.c (simplify_const_binary_operation): Use
6478 CONST_VECTOR_ELT instead of XVECEXP.
6480 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6482 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6483 the selector elements to be different from the data elements
6484 if the selector is a VECTOR_CST.
6485 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6486 ssizetype for the selector.
6488 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6490 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6491 before testing each element individually.
6492 * tree-vect-generic.c (lower_vec_perm): Likewise.
6494 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6496 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6497 * selftest-run-tests.c (selftest::run_tests): Call it.
6498 * vector-builder.h (vector_builder::operator ==): New function.
6499 (vector_builder::operator !=): Likewise.
6500 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6501 (vec_perm_indices::all_from_input_p): New function.
6502 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6503 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6504 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6505 instead of reading the VECTOR_CST directly. Detect whether both
6506 vector inputs are the same before constructing the vec_perm_indices,
6507 and update the number of inputs argument accordingly. Use the
6508 utility functions added above. Only construct sel2 if we need to.
6510 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6512 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6513 the broadcast of the low byte.
6514 (expand_mult_highpart): Use an explicit encoding for the permutes.
6515 * optabs-query.c (can_mult_highpart_p): Likewise.
6516 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6517 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6518 (vectorizable_bswap): Likewise.
6519 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6520 explicit encoding for the power-of-2 permutes.
6521 (vect_permute_store_chain): Likewise.
6522 (vect_grouped_load_supported): Likewise.
6523 (vect_permute_load_chain): Likewise.
6525 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6527 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6528 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6529 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6530 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6531 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6532 (vect_gen_perm_mask_any): Likewise.
6534 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6536 * int-vector-builder.h: New file.
6537 * vec-perm-indices.h: Include int-vector-builder.h.
6538 (vec_perm_indices): Redefine as an int_vector_builder.
6539 (auto_vec_perm_indices): Delete.
6540 (vec_perm_builder): Redefine as a stand-alone class.
6541 (vec_perm_indices::vec_perm_indices): New function.
6542 (vec_perm_indices::clamp): Likewise.
6543 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6544 (vec_perm_indices::new_vector): New function.
6545 (vec_perm_indices::new_expanded_vector): Update for new
6546 vec_perm_indices class.
6547 (vec_perm_indices::rotate_inputs): New function.
6548 (vec_perm_indices::all_in_range_p): Operate directly on the
6549 encoded form, without computing elided elements.
6550 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6551 encoding. Update for new vec_perm_indices class.
6552 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6553 the given vec_perm_builder.
6554 (expand_vec_perm_var): Update vec_perm_builder constructor.
6555 (expand_mult_highpart): Use vec_perm_builder instead of
6556 auto_vec_perm_indices.
6557 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6558 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6559 or double series encoding as appropriate.
6560 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6561 vec_perm_indices instead of auto_vec_perm_indices.
6562 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6563 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6564 (vect_permute_store_chain): Likewise.
6565 (vect_grouped_load_supported): Likewise.
6566 (vect_permute_load_chain): Likewise.
6567 (vect_shift_permute_load_chain): Likewise.
6568 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6569 (vect_transform_slp_perm_load): Likewise.
6570 (vect_schedule_slp_instance): Likewise.
6571 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6572 (vectorizable_mask_load_store): Likewise.
6573 (vectorizable_bswap): Likewise.
6574 (vectorizable_store): Likewise.
6575 (vectorizable_load): Likewise.
6576 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6577 vec_perm_indices instead of auto_vec_perm_indices. Use
6578 tree_to_vec_perm_builder to read the vector from a tree.
6579 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6580 vec_perm_builder instead of a vec_perm_indices.
6581 (have_whole_vector_shift): Use vec_perm_builder and
6582 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6583 truncation to calc_vec_perm_mask_for_shift.
6584 (vect_create_epilog_for_reduction): Likewise.
6585 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6586 from auto_vec_perm_indices to vec_perm_indices.
6587 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6588 instead of changing individual elements.
6589 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6590 the vector in d.perm.
6591 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6592 from auto_vec_perm_indices to vec_perm_indices.
6593 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6594 instead of changing individual elements.
6595 (arm_vectorize_vec_perm_const): Use new_vector to install
6596 the vector in d.perm.
6597 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6598 Update vec_perm_builder constructor.
6599 (rs6000_expand_interleave): Likewise.
6600 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6601 (rs6000_expand_interleave): Likewise.
6603 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6605 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6606 to qimode could truncate the indices.
6607 * optabs.c (expand_vec_perm_var): Likewise.
6609 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6611 * Makefile.in (OBJS): Add vec-perm-indices.o.
6612 * vec-perm-indices.h: New file.
6613 * vec-perm-indices.c: Likewise.
6614 * target.h (vec_perm_indices): Replace with a forward class
6616 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6617 * optabs.h: Include vec-perm-indices.h.
6618 (expand_vec_perm): Delete.
6619 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6620 (expand_vec_perm_const): Declare.
6621 * target.def (vec_perm_const_ok): Replace with...
6622 (vec_perm_const): ...this new hook.
6623 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6624 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6625 * doc/tm.texi: Regenerate.
6626 * optabs.def (vec_perm_const): Delete.
6627 * doc/md.texi (vec_perm_const): Likewise.
6628 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6629 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6630 expand_vec_perm for constant permutation vectors. Assert that
6631 the mode of variable permutation vectors is the integer equivalent
6632 of the mode that is being permuted.
6633 * optabs-query.h (selector_fits_mode_p): Declare.
6634 * optabs-query.c: Include vec-perm-indices.h.
6635 (selector_fits_mode_p): New function.
6636 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6637 is defined, instead of checking whether the vec_perm_const_optab
6638 exists. Use targetm.vectorize.vec_perm_const instead of
6639 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6640 fit in the vector mode before using a variable permute.
6641 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6642 vec_perm_indices instead of an rtx.
6643 (expand_vec_perm): Replace with...
6644 (expand_vec_perm_const): ...this new function. Take the selector
6645 as a vec_perm_indices rather than an rtx. Also take the mode of
6646 the selector. Update call to shift_amt_for_vec_perm_mask.
6647 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6648 Use vec_perm_indices::new_expanded_vector to expand the original
6649 selector into bytes. Check whether the indices fit in the vector
6650 mode before using a variable permute.
6651 (expand_vec_perm_var): Make global.
6652 (expand_mult_highpart): Use expand_vec_perm_const.
6653 * fold-const.c: Includes vec-perm-indices.h.
6654 * tree-ssa-forwprop.c: Likewise.
6655 * tree-vect-data-refs.c: Likewise.
6656 * tree-vect-generic.c: Likewise.
6657 * tree-vect-loop.c: Likewise.
6658 * tree-vect-slp.c: Likewise.
6659 * tree-vect-stmts.c: Likewise.
6660 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6662 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6663 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6664 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6665 (aarch64_vectorize_vec_perm_const): ...this new function.
6666 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6667 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6668 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6669 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6670 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6671 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6672 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6674 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6675 check for NEON modes.
6676 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6677 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6678 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6679 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6681 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6682 the old VEC_PERM_CONST conditions.
6683 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6684 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6685 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6686 (ia64_vectorize_vec_perm_const_ok): Merge into...
6687 (ia64_vectorize_vec_perm_const): ...this new function.
6688 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6689 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6690 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6691 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6692 * config/mips/mips.c (mips_expand_vec_perm_const)
6693 (mips_vectorize_vec_perm_const_ok): Merge into...
6694 (mips_vectorize_vec_perm_const): ...this new function.
6695 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6696 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6697 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6698 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6699 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6700 (rs6000_expand_vec_perm_const): Delete.
6701 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6703 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6704 (altivec_expand_vec_perm_const_le): Take each operand individually.
6705 Operate on constant selectors rather than rtxes.
6706 (altivec_expand_vec_perm_const): Likewise. Update call to
6707 altivec_expand_vec_perm_const_le.
6708 (rs6000_expand_vec_perm_const): Delete.
6709 (rs6000_vectorize_vec_perm_const_ok): Delete.
6710 (rs6000_vectorize_vec_perm_const): New function.
6711 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6712 an element count and rtx array.
6713 (rs6000_expand_extract_even): Update call accordingly.
6714 (rs6000_expand_interleave): Likewise.
6715 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6716 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6717 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6718 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6719 (rs6000_expand_vec_perm_const): Delete.
6720 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6721 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6722 (altivec_expand_vec_perm_const_le): Take each operand individually.
6723 Operate on constant selectors rather than rtxes.
6724 (altivec_expand_vec_perm_const): Likewise. Update call to
6725 altivec_expand_vec_perm_const_le.
6726 (rs6000_expand_vec_perm_const): Delete.
6727 (rs6000_vectorize_vec_perm_const_ok): Delete.
6728 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6729 reference to the SPE evmerge intructions.
6730 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6731 an element count and rtx array.
6732 (rs6000_expand_extract_even): Update call accordingly.
6733 (rs6000_expand_interleave): Likewise.
6734 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6735 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6737 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6739 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6741 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6742 vector mode and that that mode matches the mode of the data
6744 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6745 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6746 directly using expand_vec_perm_1 when forcing selectors into
6748 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6750 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6752 * optabs-query.h (can_vec_perm_p): Delete.
6753 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6754 * optabs-query.c (can_vec_perm_p): Split into...
6755 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6756 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6757 particular selector is valid.
6758 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6759 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6760 (vect_grouped_load_supported): Likewise.
6761 (vect_shift_permute_load_chain): Likewise.
6762 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6763 (vect_transform_slp_perm_load): Likewise.
6764 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6765 (vectorizable_bswap): Likewise.
6766 (vect_gen_perm_mask_checked): Likewise.
6767 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6768 implementations of variable permutation vectors into account
6769 when deciding which selector to use.
6770 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6771 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6772 with a false third argument.
6773 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6774 to test whether the constant selector is valid and can_vec_perm_var_p
6775 to test whether a variable selector is valid.
6777 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6779 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6780 * optabs-query.c (can_vec_perm_p): Likewise.
6781 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6782 instead of vec_perm_indices.
6783 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6784 (vect_gen_perm_mask_checked): Likewise,
6785 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6786 (vect_gen_perm_mask_checked): Likewise,
6788 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6790 * optabs-query.h (qimode_for_vec_perm): Declare.
6791 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6792 (qimode_for_vec_perm): ...this new function.
6793 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6795 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6797 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6798 does not have a conditional at the top.
6800 2018-01-02 Richard Biener <rguenther@suse.de>
6802 * ipa-inline.c (big_speedup_p): Fix expression.
6804 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6807 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6810 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6814 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6815 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6816 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6817 cond_taken_branch_cost 3->4.
6819 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6821 PR tree-optimization/83581
6822 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6823 TODO_cleanup_cfg if any changes have been made.
6826 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6827 convert_modes if target mode has the right side, but different mode
6831 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6832 last argument when extracting from CONCAT. If either from_real or
6833 from_imag is NULL, use expansion through memory. If result is not
6834 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6835 the parts directly to inner mode, if even that fails, use expansion
6839 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6840 check for bswap in mode rather than HImode and use that in expand_unop
6843 Copyright (C) 2018 Free Software Foundation, Inc.
6845 Copying and distribution of this file, with or without modification,
6846 are permitted in any medium without royalty provided the copyright
6847 notice and this notice are preserved.