1 2018-02-13 Jeff Law <law@redhat.com>
3 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
6 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
8 * common.opt (gas-loc-support, gas-locview-support): New.
9 (ginline-points, ginternal-reset-location-views): New.
10 * doc/invoke.texi: Document them. Use @itemx where intended.
11 (gvariable-location-views): Adjust.
12 * target.def (reset_location_view): New.
13 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
14 (TARGET_RESET_LOCATION_VIEW): New.
15 * doc/tm.texi: Rebuilt.
16 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
17 (dwarf2out_default_as_locview_support): New.
18 (output_asm_line_debug_info): Use option variables.
19 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
20 (output_loc_list): Likewise.
21 (add_high_low_attributes): Check option variables.
22 Don't output entry view attribute in strict mode.
23 (gen_inlined_subroutine_die): Check option variables.
24 (dwarf2out_inline_entry): Likewise.
25 (init_sections_and_labels): Likewise.
26 (dwarf2out_early_finish): Likewise.
27 (maybe_reset_location_view): New, from...
28 (dwarf2out_var_location): ... here. Call it.
29 * debug.h (dwarf2out_default_as_loc_support): Declare.
30 (dwarf2out_default_as_locview_support): Declare.
31 * hooks.c (hook_int_rtx_insn_0): New.
32 * hooks.h (hook_int_rtx_insn_0): Declare.
33 * toplev.c (process_options): Take -gas-loc-support and
34 -gas-locview-support from dwarf2out. Enable
35 -gvariable-location-views by default only with locview
36 assembler support. Enable -ginternal-reset-location-views by
37 default only if the target defines the corresponding hook.
38 Enable -ginline-points by default if location views are
39 enabled; force it disabled if statement frontiers are
41 * tree-inline.c (expand_call_inline): Check option variables.
42 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
44 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
46 PR tree-optimization/84321
47 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
48 handling. Also check whether the anti-range contains any values
49 that satisfy the mask; switch to a VR_RANGE if not.
51 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
54 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
56 2018-02-13 Martin Jambor <mjambor@suse.cz>
59 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
60 of call statements, also set location of a load to a temporary.
62 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
64 * config/rl78/rl78.c (add_vector_labels): New function.
65 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
66 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
67 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
68 which checks that no arguments are passed.
69 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
70 * doc/extend.texi: Documentation for the new attribute.
72 2018-02-13 Andreas Schwab <schwab@suse.de>
74 * config/riscv/linux.h (CPP_SPEC): Define.
76 2018-02-13 Jakub Jelinek <jakub@redhat.com>
79 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
80 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
81 OPTION_MASK_ISA_AES as first argument to def_builtin_const
82 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
83 instead of OPTION_MASK_ISA_PCLMUL as first argument to
84 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
85 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
86 temporarily for AES and PCLMUL builtins.
88 PR tree-optimization/84339
89 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
90 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
94 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
95 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
96 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
98 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
100 * omp-simd-clone.h: New file.
101 * omp-simd-clone.c: Include omp-simd-clone.h.
102 (expand_simd_clones): No longer static.
103 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
104 cgraph.h and omp-simd-clone.h.
105 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
106 (vect_recog_widen_shift_pattern): Formatting fix.
107 (vect_pattern_recog_1): Don't check optab for calls.
110 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
111 operands[2] into a REG before using gen_lowpart on it.
113 2018-02-12 Jeff Law <law@redhat.com>
116 * config/sh/sh.c (find_barrier): Consider a sibling call
119 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
120 successfully back substituting a reg.
122 2018-02-12 Richard Biener <rguenther@suse.de>
124 PR tree-optimization/84037
125 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
126 parameter, move visited init to caller.
127 (vect_slp_analyze_operations): Separate cost from validity
128 check, initialize visited once for all instances.
129 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
131 * tree-vect-stmts.c (vect_model_simple_cost): Make early
133 (vect_model_promotion_demotion_cost): Likewise.
134 (vectorizable_bswap): Guard cost modeling with !slp_node
135 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
137 (vectorizable_call): Likewise.
138 (vectorizable_conversion): Likewise.
139 (vectorizable_assignment): Likewise.
140 (vectorizable_shift): Likewise.
141 (vectorizable_operation): Likewise.
142 (vectorizable_store): Likewise.
143 (vectorizable_load): Likewise.
144 (vectorizable_condition): Likewise.
145 (vectorizable_comparison): Likewise.
147 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
150 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
151 (ASAN_MARK): Fix fnspec to account for return value, change pointer
152 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
154 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
157 * params.def (inline-min-speedup): Increase from 8 to 15.
158 (max-inline-insns-auto): Decrease from 40 to 30.
159 * ipa-split.c (consider_split): Add some buffer for function to
160 be considered inlining candidate.
161 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
164 2018-02-12 Richard Biener <rguenther@suse.de>
166 PR tree-optimization/84037
167 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
168 matched stmts if we cannot swap the non-matched ones.
170 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
172 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
173 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
174 _mm_maskz_scalef_round_ss): New intrinsics.
175 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
176 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
177 __builtin_ia32_scalefss_round): Remove.
178 (__builtin_ia32_scalefsd_mask_round,
179 __builtin_ia32_scalefss_mask_round): New intrinsics.
180 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
181 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
182 ((match_operand:VF_128 2 "<round_nimm_predicate>"
183 "<round_constraint>")): Changed to ...
184 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
185 "<round_scalar_constraint>")): ... this.
186 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
187 %0, %1, %2<round_op3>}"): Changed to ...
188 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
189 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
190 %2<round_scalar_mask_op3>}"): ... this.
191 * config/i386/subst.md (round_scalar_nimm_predicate): New.
193 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
195 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
196 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
197 (_mm_maskz_sqrt_round_ss): New intrinsics.
198 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
199 (__builtin_ia32_sqrtsd_mask_round)
200 (__builtin_ia32_sqrtss_mask_round): New builtins.
201 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
202 (__builtin_ia32_sqrtss_round): Remove.
203 (__builtin_ia32_sqrtsd_mask_round)
204 (__builtin_ia32_sqrtss_mask_round): New builtins.
205 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
206 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
207 ((match_operand:VF_128 1 "vector_operand"
208 "xBm,<round_constraint>")): Changed to ...
209 ((match_operand:VF_128 1 "vector_operand"
210 "xBm,<round_scalar_constraint>")): ... this.
211 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
212 %0, %2, %<iptr>1<round_op3>}): Changed to ...
213 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
214 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
215 %<iptr>1<round_scalar_mask_op3>}): ... this.
216 ((set_attr "prefix" "<round_prefix>")): Changed to ...
217 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
219 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
222 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
223 Cast vec_cmpeq result to correct type.
224 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
225 Cast vec_cmpgt result to correct type.
227 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
229 * final.c (final_scan_insn_1): Renamed from...
230 (final_scan_insn): ... this. New wrapper, to recover
231 seen from the outermost call in recursive ones.
232 * config/sparc/sparc.c (output_return): Drop seen from call.
233 (output_sibcall): Likewise.
234 * config/visium/visium.c (output_branch): Likewise.
236 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
238 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
241 2018-02-10 Alan Modra <amodra@gmail.com>
244 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
245 Specify LR as an input.
247 2018-02-10 Jakub Jelinek <jakub@redhat.com>
250 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
251 remove_member_access_dummy_vars): New functions.
252 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
253 lower_omp_1, execute_lower_omp): Use them.
255 PR rtl-optimization/84308
256 * shrink-wrap.c (spread_components): Release todo vector.
258 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
260 PR rtl-optimization/57193
261 * ira-color.c (struct allocno_color_data): Add member
262 conflict_allocno_hard_prefs.
263 (update_conflict_allocno_hard_prefs): New.
264 (bucket_allocno_compare_func): Add a preference based on
265 conflict_allocno_hard_prefs.
266 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
267 (color_allocnos): Remove a dead code. Initiate
268 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
270 2018-02-09 Jakub Jelinek <jakub@redhat.com>
273 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
274 constraint from =wa to wa. Avoid a subreg on the output operand,
275 instead use a pseudo and subreg it in a move.
276 (p9_xxbrd_<mode>): Changed to ...
277 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
278 (p9_xxbrd_v2df): New expander.
279 (p9_xxbrw_<mode>): Changed to ...
280 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
281 (p9_xxbrw_v4sf): New expander.
283 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
285 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
287 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
290 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
291 multiply in 32-bit mode.
292 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
293 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
296 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
298 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
299 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
300 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
301 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
303 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
306 * dwarf2out.c (is_trivial_indirect_ref): New function.
307 (dwarf2out_late_global_decl): Do not generate a location
308 attribute for variables that have a non-trivial DECL_VALUE_EXPR
309 and that are not defined in the current unit.
311 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
313 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
314 instead of a libcall for UNORDERED.
316 2018-02-09 Tamar Christina <tamar.christina@arm.com>
319 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
320 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
322 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
325 * config/s390/s390.c (s390_set_current_function): Invoke
326 s390_indirect_branch_settings also if fndecl didn't change.
328 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
330 * config/rs6000/rs6000.md (blockage): Set length to zero.
332 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
334 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
336 2018-02-09 Jakub Jelinek <jakub@redhat.com>
339 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
340 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
344 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
345 PARALLEL incoming that failed vt_get_decl_and_offset check.
348 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
349 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
351 (get_variable_section): For decls in named .bss* sections pass true as
352 second argument to bss_initializer_p.
354 2018-02-09 Marek Polacek <polacek@redhat.com>
355 Jakub Jelinek <jakub@redhat.com>
358 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
359 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
360 Sync some changes from cxx_fold_indirect_ref.
362 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
364 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
366 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
367 (BLOCK_INLINE_ENTRY_LABEL): New.
368 (dwarf2out_var_location): Disregard inline entry markers.
369 (inline_entry_data): New struct.
370 (inline_entry_data_hasher): New hashtable type.
371 (inline_entry_data_hasher::hash): New.
372 (inline_entry_data_hasher::equal): New.
373 (inline_entry_data_table): New variable.
374 (add_high_low_attributes): Add DW_AT_entry_pc and
375 DW_AT_GNU_entry_view attributes if a pending entry is found
376 in inline_entry_data_table. Add old entry_pc attribute only
377 if debug nonbinding markers are disabled.
378 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
380 (block_within_block_p, dwarf2out_inline_entry): New.
381 (dwarf2out_finish): Check that no entries remained in
382 inline_entry_data_table.
383 * final.c (reemit_insn_block_notes): Handle inline entry notes.
384 (final_scan_insn, notice_source_line): Likewise.
385 (rest_of_clean_state): Skip inline entry markers.
386 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
388 * gimple.c (gimple_build_debug_inline_entry): New.
389 * gimple.h (enum gimple_debug_subcode): Add
390 GIMPLE_DEBUG_INLINE_ENTRY.
391 (gimple_build_debug_inline_entry): Declare.
392 (gimple_debug_inline_entry_p): New.
393 (gimple_debug_nonbind_marker_p): Adjust.
394 * insn-notes.def (INLINE_ENTRY): New.
395 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
396 inline entry marker notes.
397 (print_insn): Likewise.
398 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
399 (INSN_DEBUG_MARKER_KIND): Likewise.
400 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
401 * tree-inline.c (expand_call_inline): Build and insert
402 debug_inline_entry stmt.
403 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
404 inline entry blocks early, if nonbind markers are enabled.
405 (dump_scope_block): Dump fragment info.
406 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
407 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
408 (gimple_build_debug_inline_entry): New.
409 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
410 Enable/disable inline entry points too.
411 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
412 (DEBUG_INSN): Describe inline entry markers.
414 * common.opt (gvariable-location-views): New.
415 (gvariable-location-views=incompat5): New.
416 * config.in: Rebuilt.
417 * configure: Rebuilt.
418 * configure.ac: Test assembler for view support.
419 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
420 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
421 * dwarf2out.c (var_loc_view): New typedef.
422 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
423 (dwarf2out_locviews_in_attribute): New.
424 (dwarf2out_locviews_in_loclist): New.
425 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
426 (enum dw_line_info_opcode): Add LI_adv_address.
427 (struct dw_line_info_table): Add view.
428 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
429 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
430 (zero_view_p): New variable.
431 (ZERO_VIEW_P): New macro.
432 (output_asm_line_debug_info): New.
433 (struct var_loc_node): Add view.
434 (add_AT_view_list, AT_loc_list): New.
435 (add_var_loc_to_decl): Add view param. Test it against last.
436 (new_loc_list): Add view params. Record them.
437 (AT_loc_list_ptr): Handle loc and view lists.
438 (view_list_to_loc_list_val_node): New.
439 (print_dw_val): Handle dw_val_class_view_list.
440 (size_of_die): Likewise.
441 (value_format): Likewise.
442 (loc_list_has_views): New.
443 (gen_llsym): Set vl_symbol too.
444 (maybe_gen_llsym, skip_loc_list_entry): New.
445 (dwarf2out_maybe_output_loclist_view_pair): New.
446 (output_loc_list): Output view list or entries too.
447 (output_view_list_offset): New.
448 (output_die): Handle dw_val_class_view_list.
449 (output_dwarf_version): New.
450 (output_compilation_unit_header): Use it.
451 (output_skeleton_debug_sections): Likewise.
452 (output_rnglists, output_line_info): Likewise.
453 (output_pubnames, output_aranges): Update version comments.
454 (output_one_line_info_table): Output view numbers in asm comments.
455 (dw_loc_list): Determine current endview, pass it to new_loc_list.
456 Call maybe_gen_llsym.
457 (loc_list_from_tree_1): Adjust.
458 (add_AT_location_description): Create view list attribute if
459 needed, check it's absent otherwise.
460 (convert_cfa_to_fb_loc_list): Adjust.
461 (maybe_emit_file): Call output_asm_line_debug_info for test.
462 (dwarf2out_var_location): Reset views as needed. Precompute
463 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
465 (new_line_info_table): Reset next view.
466 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
467 (dwarf2out_source_line): Likewise. Output view resets and labels to
468 the assembler, or select appropriate line info opcodes.
469 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
470 (optimize_string_length): Catch it. Adjust.
471 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
472 dw_val_class_view_list, and remove it if no longer needed.
473 (hash_loc_list): Hash view numbers.
474 (loc_list_hasher::equal): Compare them.
475 (optimize_location_lists): Check whether a view list symbol is
476 needed, and whether the locview attribute is present, and
477 whether they match. Remove the locview attribute if no longer
479 (index_location_lists): Call skip_loc_list_entry for test.
480 (dwarf2out_finish): Call output_asm_line_debug_info for test.
481 Use output_dwarf_version.
482 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
483 (struct dw_val_node): Add val_view_list.
484 * final.c (SEEN_NEXT_VIEW): New.
485 (set_next_view_needed): New.
486 (clear_next_view_needed): New.
487 (maybe_output_next_view): New.
488 (final_start_function): Rename to...
489 (final_start_function_1): ... this. Take pointer to FIRST,
490 add SEEN parameter. Emit param bindings in the initial view.
491 (final_start_function): Reintroduce SEEN-less interface.
492 (final): Rename to...
493 (final_1): ... this. Take SEEN parameter. Output final pending
494 next view at the end.
495 (final): Reintroduce seen-less interface.
496 (final_scan_insn): Output pending next view before switching
497 sections or ending a block. Mark the next view as needed when
498 outputting variable locations. Notify debug backend of section
499 changes, and of location view changes.
500 (rest_of_handle_final): Adjust.
501 * toplev.c (process_options): Autodetect value for debug variable
502 location views option. Warn on incompat5 without -gdwarf-5.
503 * doc/invoke.texi (gvariable-location-views): New.
504 (gvariable-location-views=incompat5): New.
505 (gno-variable-location-views): New.
507 2018-02-08 David Malcolm <dmalcolm@redhat.com>
509 PR tree-optimization/84136
510 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
511 that the result of find_edge is non-NULL.
513 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
516 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
517 storing integer register in SImode. Fix cost of 256 and 512
518 byte aligned SSE register store.
520 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
522 * config/i386/i386.c (ix86_multiplication_cost): Fix
523 multiplication cost for TARGET_AVX512DQ.
525 2018-02-08 Marek Polacek <polacek@redhat.com>
527 PR tree-optimization/84238
528 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
531 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
533 PR tree-optimization/84265
534 * tree-vect-stmts.c (vectorizable_store): Don't treat
535 VMAT_CONTIGUOUS accesses as grouped.
536 (vectorizable_load): Likewise.
538 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
540 PR tree-optimization/81635
541 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
542 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
543 (test_round_for_mask): New functions.
544 (wide_int_cc_tests): Call test_round_for_mask.
545 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
546 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
547 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
548 range returned by get_range_info.
550 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
553 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
554 * symtab.c: Include builtins.h
555 (symtab_node::output_to_lto_symbol_table_p): Move here
556 from lto-streamer-out.c:output_symbol_p.
557 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
558 (output_symbol_p): Move all logic to symtab.c
559 (produce_symtab): Update.
561 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
563 * config/s390/s390-opts.h (enum indirect_branch): Define.
564 * config/s390/s390-protos.h (s390_return_addr_from_memory)
565 (s390_indirect_branch_via_thunk)
566 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
567 (enum s390_indirect_branch_type): Define.
568 * config/s390/s390.c (struct s390_frame_layout, struct
569 machine_function): Remove.
570 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
571 (indirect_branch_table_label_no, indirect_branch_table_name):
573 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
574 (enum s390_indirect_branch_option): Define.
575 (s390_return_addr_from_memory): New function.
576 (s390_handle_string_attribute): New function.
577 (s390_attribute_table): Add new attribute handler.
578 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
579 (s390_indirect_branch_via_thunk): New function.
580 (s390_indirect_branch_via_inline_thunk): New function.
581 (s390_function_ok_for_sibcall): When jumping via thunk disallow
582 sibling call optimization for non z10 compiles.
583 (s390_emit_call): Force indirect branch target to be a single
584 register. Add r1 clobber for non-z10 compiles.
585 (s390_emit_epilogue): Emit return jump via return_use expander.
586 (s390_reorg): Handle JUMP_INSNs as execute targets.
587 (s390_option_override_internal): Perform validity checks for the
588 new command line options.
589 (s390_indirect_branch_attrvalue): New function.
590 (s390_indirect_branch_settings): New function.
591 (s390_set_current_function): Invoke s390_indirect_branch_settings.
592 (s390_output_indirect_thunk_function): New function.
593 (s390_code_end): Implement target hook.
594 (s390_case_values_threshold): Implement target hook.
595 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
597 * config/s390/s390.h (struct s390_frame_layout)
598 (struct machine_function): Move here from s390.c.
599 (TARGET_INDIRECT_BRANCH_NOBP_RET)
600 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
601 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
602 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
603 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
604 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
605 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
606 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
607 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
608 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
609 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
610 (mnemonic attribute): Add values which aren't recognized
612 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
613 pattern for branch conversion. Fix mnemonic attribute.
614 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
615 indirect branch via thunk if requested.
616 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
617 ("*indirect_jump"): Disable for branch conversion using out of
619 ("indirect_jump_via_thunk<mode>_z10")
620 ("indirect_jump_via_thunk<mode>")
621 ("indirect_jump_via_inlinethunk<mode>_z10")
622 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
623 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
624 ("casesi_jump_via_inlinethunk<mode>_z10")
625 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
626 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
627 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
628 ("*indirect2_jump"): Disable for branch conversion.
629 ("casesi_jump"): Turn into expander and expand patterns for branch
631 ("return_use"): New expander.
632 ("*return"): Emit return via thunk and rename it to ...
633 ("*return<mode>"): ... this one.
634 * config/s390/s390.opt: Add new options and and enum for the
637 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
639 * lra-constraints.c (match_reload): Unconditionally use
640 gen_lowpart_SUBREG, rather than selecting between that
641 and equivalent gen_rtx_SUBREG code.
643 2018-02-08 Richard Biener <rguenther@suse.de>
645 PR tree-optimization/84233
646 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
647 changed flag instead of boguously re-using phi_inserted.
649 2018-02-08 Martin Jambor <mjambor@suse.cz>
651 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
652 static local variables.
654 2018-02-08 Richard Biener <rguenther@suse.de>
656 PR tree-optimization/84278
657 * tree-vect-stmts.c (vectorizable_store): When looking for
658 smaller vector types to perform grouped strided loads/stores
659 make sure the mode is supported by the target.
660 (vectorizable_load): Likewise.
662 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
664 * config/aarch64/aarch64.c (aarch64_components_for_bb):
665 Increase LDP/STP opportunities by adding adjacent callee-saves.
667 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
669 PR rtl-optimization/84068
670 PR rtl-optimization/83459
671 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
673 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
675 PR tree-optimization/84224
676 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
677 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
680 2018-02-07 Iain Sandoe <iain@codesourcery.com>
683 * config/rs6000/altivec.md (*restore_world): Remove LR use.
684 * config/rs6000/predicates.md (restore_world_operation): Adjust op
685 count, remove one USE.
687 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
689 * doc/install.texi (Configuration): Document the
690 --with-long-double-format={ibm,ieee} PowerPC configuration
694 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
695 Convert from define_expand to be define_insn_and_split. Rework
696 float/double/_Float128 conversions to QI/HI/SImode to work with
697 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
698 conversions to QI/HImode types did a store and then a load to
699 truncate the value. For conversions to VSX registers, don't split
700 the insn, instead emit the code directly. Use the code iterator
701 any_fix to combine signed and unsigned conversions.
702 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
703 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
704 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
705 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
706 (fix_<mode>di2_hw): Likewise.
707 (fixuns_<mode>di2_hw): Likewise.
708 (fix_<mode>si2_hw): Likewise.
709 (fixuns_<mode>si2_hw): Likewise.
710 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
711 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
712 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
713 fix<uns>_trunc<SFDF:mode>si2_p8.
714 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
716 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
717 (fix<uns>_<mode>_mem): Likewise.
718 (fctiw<u>z_<mode>_mem): Likewise.
719 (fix<uns>_<mode>_mem): Likewise.
720 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
721 the register allocator from doing a direct move to the GPRs to do
722 a store, and instead use the ISA 3.0 store byte/half-word from
723 vector register instruction. For IEEE 128-bit floating point,
724 also optimize stores of 32-bit ints.
725 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
727 2018-02-07 Alan Hayward <alan.hayward@arm.com>
729 * genextract.c (push_pathstr_operand): New function to support
731 (walk_rtx): Call push_pathstr_operand.
732 (print_path): Support [a-zA-Z].
734 2018-02-07 Richard Biener <rguenther@suse.de>
736 PR tree-optimization/84037
737 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
738 (cse_and_gimplify_to_preheader): Declare.
739 (vect_get_place_in_interleaving_chain): Likewise.
740 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
742 (_loop_vec_info::~_loop_vec_info): Delete it.
743 (cse_and_gimplify_to_preheader): New function.
744 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
745 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
746 (vectorizable_load): Likewise. For grouped stores always base
747 the IV on the first element.
748 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
749 condition before gimplifying.
751 2018-02-07 Jakub Jelinek <jakub@redhat.com>
753 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
754 *DIV_EXPR and *MOD_EXPR.
756 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
759 * config/i386/i386.c (ix86_option_override_internal): Mask out
760 the CF_SET bit when checking -fcf-protection.
762 2018-02-07 Tom de Vries <tom@codesourcery.com>
765 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
768 2018-02-07 Richard Biener <rguenther@suse.de>
770 PR tree-optimization/84204
771 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
774 PR tree-optimization/84205
775 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
776 special-case isl_ast_op_zdiv_r.
778 PR tree-optimization/84223
779 * graphite-scop-detection.c (gather_bbs::before_dom_children):
780 Only add conditions from within the region.
781 (gather_bbs::after_dom_children): Adjust.
783 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
786 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
787 * config/avr/avr.md: Only post-reload split REG-REG moves if
788 either register is REGERAL_REG_P.
790 2018-02-07 Jakub Jelinek <jakub@redhat.com>
792 PR tree-optimization/84235
793 * tree-ssa-scopedtables.c
794 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
795 if the subtraction is performed in floating point type where NaNs are
796 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
797 build 1. Formatting fix.
799 2018-02-06 Jakub Jelinek <jakub@redhat.com>
802 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
803 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
804 and skip it regardless of bb boundaries. Use CALL_P macro,
805 don't test INSN_P (insn) together with CALL_P or JUMP_P check
806 unnecessarily, formatting fix.
808 2018-02-06 Michael Collison <michael.collison@arm.com>
810 * config/arm/thumb2.md:
811 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
812 (*thumb_mov_notscc): Ditto.
814 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
817 * config/rs6000/rs6000.md (su code attribute): Use "u" for
818 unsigned_fix, not "s".
820 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
822 * configure.ac (gcc_fn_eh_frame_ro): New function.
823 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
824 correct .eh_frame permissions.
825 * configure: Regenerate.
827 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
829 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
832 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
834 * config/rs6000/rs6000.c (rs6000_option_override_internal):
835 Display warning message for -mno-speculate-indirect-jumps.
837 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
839 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
841 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
843 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
845 PR tree-optimization/84225
846 * tree-eh.c (find_trapping_overflow): Only call
847 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
849 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
852 * config/i386/i386.c: Reimplement the check of possible options
853 -mibt/-mshstk conbination. Change error messages.
854 * doc/invoke.texi: Fix a typo: remove extra '='.
856 2018-02-06 Marek Polacek <polacek@redhat.com>
858 PR tree-optimization/84228
859 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
861 2018-02-06 Tamar Christina <tamar.christina@arm.com>
864 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
865 emitted arch directives.
866 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
867 __ARM_FEATURE_COPROC before changing architectures.
869 2018-02-06 Richard Biener <rguenther@suse.de>
871 * config/i386/i386.c (print_reg): Fix typo.
872 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
874 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
876 * configure: Regenerate.
878 2018-02-05 Martin Sebor <msebor@redhat.com>
880 PR tree-optimization/83369
881 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
884 2018-02-05 Martin Liska <mliska@suse.cz>
886 * doc/invoke.texi: Cherry-pick upstream r323995.
888 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
890 * ira.c (ira_init_register_move_cost): Adjust comment.
892 2018-02-05 Martin Liska <mliska@suse.cz>
894 PR gcov-profile/84137
895 * doc/gcov.texi: Fix typo in documentation.
897 2018-02-05 Martin Liska <mliska@suse.cz>
899 PR gcov-profile/83879
900 * doc/gcov.texi: Document necessity of --dynamic-list-data when
901 using dlopen functionality.
903 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
905 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
906 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
907 _mm_maskz_range_ss, _mm_mask_range_round_ss,
908 _mm_maskz_range_round_ss): New intrinsics.
909 (__builtin_ia32_rangesd128_round)
910 (__builtin_ia32_rangess128_round): Remove.
911 (__builtin_ia32_rangesd128_mask_round,
912 __builtin_ia32_rangess128_mask_round): New builtins.
913 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
914 __builtin_ia32_rangess128_round): Remove.
915 (__builtin_ia32_rangesd128_mask_round,
916 __builtin_ia32_rangess128_mask_round): New builtins.
917 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
918 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
919 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
920 "<round_saeonly_constraint>")): Changed to ...
921 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
922 "<round_saeonly_scalar_constraint>")): ... this.
923 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
924 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
925 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
926 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
927 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
929 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
931 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
933 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
934 Remove all values except native, 8540 and 8548.
936 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
938 * config/i386/i386.c (ix86_output_function_return): Pass
939 INVALID_REGNUM, instead of -1, as invalid register number to
940 indirect_thunk_name and output_indirect_thunk.
942 2018-02-02 Julia Koval <julia.koval@intel.com>
944 * config.gcc: Add -march=icelake.
945 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
946 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
947 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
948 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
949 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
950 (processor_target_table): Add icelake.
951 (ix86_option_override_internal): Handle new PTAs.
952 (get_builtin_code_for_version): Handle icelake.
953 (M_INTEL_COREI7_ICELAKE): New.
954 (fold_builtin_cpu): Handle icelake.
955 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
956 * doc/invoke.texi: Add -march=icelake.
958 2018-02-02 Julia Koval <julia.koval@intel.com>
960 * config/i386/i386.c (ix86_option_override_internal): Change flags type
962 * wide-int-bitmask.h: New.
964 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
967 * config/i386/i386.md: Replace Pmode with word_mode in
968 builtin_setjmp_setup and builtin_longjmp to support x32.
970 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
974 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
976 (rs6000_supported_cpu_names): New static variable.
977 (linux_cpu_translation_table): Likewise.
978 (elf_platform) <cpu>: Define new static variable and use it.
979 Translate kernel AT_PLATFORM name to canonical name if needed.
980 Error if platform name is unknown.
982 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
985 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
987 2018-02-01 Jeff Law <law@redhat.com>
990 * config/i386/i386.c (release_scratch_register_on_entry): Add new
991 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
992 the scratch if RELEASE_VIA_POP is false.
993 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
994 If we have to save a temporary register, decrement SIZE appropriately.
995 Pass new arguments to release_scratch_register_on_entry.
996 (ix86_adjust_stack_and_probe): Likewise.
997 (ix86_emit_probe_stack_range): Pass new arguments to
998 release_scratch_register_on_entry.
1000 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1002 PR rtl-optimization/84157
1003 * combine.c (change_zero_ext): Use REG_P predicate in
1004 front of HARD_REGISTER_P predicate.
1006 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1008 * config/avr/avr.c (avr_option_override): Move disabling of
1009 -fdelete-null-pointer-checks to...
1010 * common/config/avr/avr-common.c (avr_option_optimization_table):
1013 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1015 PR tree-optimization/81635
1016 * tree-data-ref.c (split_constant_offset_1): For types that
1017 wrap on overflow, try to use range info to prove that wrapping
1020 2018-02-01 Renlin Li <renlin.li@arm.com>
1023 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1025 (aarch64_register_move_cost): Likewise.
1026 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1028 (REG_CLASS_NAMES): Likewise.
1029 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1030 TAILCALL_ADDR_REGS. Remove IP registers.
1031 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1033 2018-02-01 Richard Biener <rguenther@suse.de>
1035 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1036 for specifying RPO order and allow NULL for that.
1037 * domwalk.c (dom_walker::dom_walker): Likewise.
1038 (dom_walker::walk): Handle NULL RPO order.
1039 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1041 (rewrite_update_dom_walker): Likewise.
1042 (mark_def_dom_walker): Likewise.
1044 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1046 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1047 (aarch64_maybe_expand_sve_subreg_move): Declare.
1048 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1049 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1051 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1052 that are semantically a reverse operation.
1053 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1054 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1055 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1057 (aarch64_can_change_mode_class): For big-endian, forbid changes
1058 between two SVE modes if they have different element sizes.
1060 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1062 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1063 the TImode handling for big-endian targets.
1065 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1067 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1068 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1070 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1071 Remove BSWAP handing for big-endian targets and use the form of
1072 LD1RQ appropariate for the mode.
1074 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1076 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1077 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1080 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1083 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1084 check for operands that need to go through aarch64_sve_reload_be.
1086 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1088 PR tree-optimization/81661
1089 PR tree-optimization/84117
1090 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1091 * tree-eh.c: Include gimplify.h.
1092 (find_trapping_overflow, replace_trapping_overflow,
1093 rewrite_to_non_trapping_overflow): New functions.
1094 * tree-vect-loop.c: Include tree-eh.h.
1095 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1096 * tree-data-ref.c: Include tree-eh.h.
1097 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1099 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1101 PR rtl-optimization/84123
1102 * combine.c (change_zero_ext): Check if hard register satisfies
1103 can_change_dest_mode before calling gen_lowpart_SUBREG.
1105 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1108 * ira.c (ira_init_register_move_cost): Remove assert.
1110 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1112 PR rtl-optimization/84071
1113 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1114 * doc/tm.texi: Regenerate.
1116 2018-01-31 Richard Biener <rguenther@suse.de>
1118 PR tree-optimization/84132
1119 * tree-data-ref.c (analyze_miv_subscript): Properly
1120 check whether evolution_function_is_affine_multivariate_p
1121 before calling gcd_of_steps_may_divide_p.
1123 2018-01-31 Julia Koval <julia.koval@intel.com>
1126 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1127 * config/i386/i386.md (rdpid_rex64) New.
1128 (rdpid): Make 32bit only.
1130 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1133 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1134 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1136 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1139 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1141 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1143 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1145 PR rtl-optimization/84071
1146 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1147 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1149 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1151 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1152 (arc_attribute_table): Add 'aux' attribute.
1153 (arc_in_small_data_p): Consider aux like variables.
1154 (arc_is_aux_reg_p): New function.
1155 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1156 (arc_get_aux_arg): New function.
1157 (prepare_move_operands): Handle aux-register access.
1158 (arc_handle_aux_attribute): New function.
1159 * doc/extend.texi (ARC Variable attributes): Add subsection.
1161 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1163 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1164 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1165 (arc_attribute_table): Add 'uncached' attribute.
1166 (arc_print_operand): Print '.di' flag for uncached memory
1168 (arc_in_small_data_p): Do not consider for small data the uncached
1170 (arc_is_uncached_mem_p): New function.
1171 * config/arc/predicates.md (compact_store_memory_operand): Check
1172 for uncached memory accesses.
1173 (nonvol_nonimm_operand): Likewise.
1174 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1176 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1179 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1180 falign-loops=): Add Optimization flag.
1182 2018-01-30 Jeff Law <law@redhat.com>
1185 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1186 INT_REGISTERS_SAVED. Check it prior to calling
1187 get_scratch_register_on_entry.
1188 (ix86_adjust_stack_and_probe): Similarly.
1189 (ix86_emit_probe_stack_range): Similarly.
1190 (ix86_expand_prologue): Corresponding changes.
1192 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1195 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1196 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1198 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1201 * lra-constraints.c (curr_insn_transform): Process AND in the
1204 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1206 PR rtl-optimization/83986
1207 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1208 dependence against last_pending_memory_flush in addition to
1211 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1213 PR tree-optimization/81611
1214 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1217 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1219 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1222 2018-01-30 Richard Biener <rguenther@suse.de>
1223 Jakub Jelinek <jakub@redhat.com>
1225 PR tree-optimization/84111
1226 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1227 inner loops added during recursion, as they don't have up-to-date
1230 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1233 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1234 (can_inline_edge_by_limits_p): ... here.
1235 (can_early_inline_edge_p, check_callers,
1236 update_caller_keys, update_callee_keys, recursive_inlining,
1237 add_new_edges_to_heap, speculation_useful_p,
1238 inline_small_functions,
1239 inline_small_functions, flatten_function,
1240 inline_to_all_callers_1): Update.
1242 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1244 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1247 2018-01-30 Richard Biener <rguenther@suse.de>
1249 PR tree-optimization/83008
1250 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1251 invariant and constant vector uses in stmts when they need
1254 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1257 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1258 * configure: Regenerate.
1260 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1262 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1264 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1265 Use gen_rtx_REG rather than gen_lowpart.
1267 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1269 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1270 rather than 0 when creating partial subregs.
1272 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1274 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1277 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1280 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1281 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1282 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1283 flags. This restores the settings used before the 2017-07-24.
1284 Turning off pre increment/decrement/modify allows IVOPTS to
1285 optimize DF/SF loops where the index is an int.
1287 2018-01-29 Richard Biener <rguenther@suse.de>
1288 Kelvin Nilsen <kelvin@gcc.gnu.org>
1291 * tree-vect-stmts.c (vectorizable_call): Don't call
1292 targetm.vectorize_builtin_md_vectorized_function if callee is
1295 2018-01-22 Carl Love <cel@us.ibm.com>
1297 * doc/extend.tex: Fix typo in second arg in
1298 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1300 2018-01-29 Richard Biener <rguenther@suse.de>
1302 PR tree-optimization/84086
1303 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1304 (flush_ssaname_freelist): When SSA names were released reset
1305 the SCEV hash table.
1307 2018-01-29 Richard Biener <rguenther@suse.de>
1309 PR tree-optimization/84057
1310 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1311 removed paths when removing edges.
1313 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1315 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1316 -mfunction-return=@var{choice}.
1318 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1321 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1323 (layout::print_source_line): Likewise.
1324 (test_get_line_width_without_trailing_whitespace): Add test cases.
1326 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1329 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1332 2018-01-26 Jim Wilson <jimw@sifive.com>
1334 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1336 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1339 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1341 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1342 and CMP + SUB-immediate -> SUBS.
1344 2018-01-26 Martin Sebor <msebor@redhat.com>
1346 PR tree-optimization/83896
1347 * tree-ssa-strlen.c (get_string_len): Rename...
1348 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1349 Avoid assuming length is constant.
1350 (handle_char_store): Use HOST_WIDE_INT for string length.
1352 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1355 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1356 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1358 2018-01-26 Richard Biener <rguenther@suse.de>
1360 PR rtl-optimization/84003
1361 * dse.c (record_store): Only record redundant stores when
1362 the earlier store aliases at least all accesses the later one does.
1364 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1366 PR rtl-optimization/83985
1367 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1368 REG_CFA_RESTORE insns.
1369 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1370 REG_CFA_RESTORE insns here.
1373 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1374 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1376 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1378 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1379 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1380 (arc_init): Likewise.
1381 (arc_override_options): Likewise.
1382 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1384 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1386 * config/arc/arc.h (TARGET_DBNZ): Define.
1387 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1388 properly set the tune attribute.
1389 (dbnz): Use TARGET_DBNZ guard.
1390 * config/arc/arc.opt (mtune): Add core3 option.
1392 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1394 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1395 recognize new pic like addresses.
1396 (arc_delegitimize_address): Clean up.
1398 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1400 * config/arc/arc-arches.def: Option mrf16 valid for all
1402 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1403 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1404 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1405 * config/arc/arc-tables.opt: Regenerate.
1406 * config/arc/arc.c (arc_conditional_register_usage): Handle
1407 reduced register file case.
1408 (arc_file_start): Set must have build attributes.
1409 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1411 * config/arc/arc.opt (mrf16): Add new option.
1412 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1413 * config/arc/genmultilib.awk: Handle new mrf16 option.
1414 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1415 * config/arc/t-multilib: Regenerate.
1416 * doc/invoke.texi (ARC Options): Document mrf16 option.
1418 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1420 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1421 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1422 (arc_attribute_table): Add 'secure_call' attribute.
1423 (arc_print_operand): Print secure call operand.
1424 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1426 (arc_is_secure_call_p): New function. * config/arc/arc.md
1427 (call_i): Add support for sjli instruction.
1428 (call_value_i): Likewise.
1429 * config/arc/constraints.md (Csc): New constraint.
1431 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1432 John Eric Martin <John.Martin@emmicro-us.com>
1434 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1435 * config/arc/arc.c (_arc_jli_section): New struct.
1436 (arc_jli_section): New type.
1437 (rc_jli_sections): New static variable.
1438 (arc_handle_jli_attribute): New function.
1439 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1440 (arc_file_end): New function.
1441 (TARGET_ASM_FILE_END): Define.
1442 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1443 (arc_add_jli_section): New function.
1444 (jli_call_scan): Likewise.
1445 (arc_reorg): Call jli_call_scan.
1446 (arc_output_addsi): Remove 'S' from printing asm operand.
1447 (arc_is_jli_call_p): New function.
1448 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1450 (movhi_insn): Likewise.
1451 (movsi_insn): Likewise.
1452 (movsi_set_cc_insn): Likewise.
1453 (loadqi_update): Likewise.
1454 (load_zeroextendqisi_update): Likewise.
1455 (load_signextendqisi_update): Likewise.
1456 (loadhi_update): Likewise.
1457 (load_zeroextendhisi_update): Likewise.
1458 (load_signextendhisi_update): Likewise.
1459 (loadsi_update): Likewise.
1460 (loadsf_update): Likewise.
1461 (movsicc_insn): Likewise.
1462 (bset_insn): Likewise.
1463 (bxor_insn): Likewise.
1464 (bclr_insn): Likewise.
1465 (bmsk_insn): Likewise.
1466 (bicsi3_insn): Likewise.
1467 (cmpsi_cc_c_insn): Likewise.
1468 (movsi_ne): Likewise.
1469 (movsi_cond_exec): Likewise.
1470 (clrsbsi2): Likewise.
1481 (call_i): Remove 'S' asm letter, add jli instruction.
1482 (call_value_i): Likewise.
1483 * config/arc/arc.op (mjli-always): New option.
1484 * config/arc/constraints.md (Cji): New constraint.
1485 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1487 (subsf3_fpx): Likewise.
1488 (mulsf3_fpx): Likewise.
1489 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1491 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1493 * doc/invoke.texi (ARC): Document mjli-always option.
1495 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1497 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1498 avoid addition with 0 and use incw and decw where possible.
1500 2018-01-26 Richard Biener <rguenther@suse.de>
1502 PR tree-optimization/81082
1503 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1504 association if it requires casting to unsigned.
1505 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1506 from fold_plusminus_mult_expr to catch important cases late when
1507 range info is available.
1509 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1511 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1512 * configure.ac (hidden_linkonce): New test.
1513 * configure: Regenerate.
1514 * config.in: Regenerate.
1516 2018-01-26 Julia Koval <julia.koval@intel.com>
1518 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1519 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1520 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1521 _mm_mask_bitshuffle_epi64_mask): Fix type.
1522 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1523 USI_FTYPE_V4DI_V4DI_USI): Remove.
1524 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1525 __builtin_ia32_vpshufbitqmb256_mask,
1526 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1527 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1528 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1530 2018-01-26 Alan Modra <amodra@gmail.com>
1533 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1534 UNSPEC_VBPERMQ. Sort other unspecs.
1536 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1538 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1540 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1543 * predict.c (drop_profile): Do not push/pop cfun; update also
1545 (handle_missing_profiles): Fix logic looking for zero profiles.
1547 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1550 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1551 on functions with #pragma omp declare simd or functions with simd
1553 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1554 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1555 Remove trailing \n from warning_at calls.
1557 2018-01-25 Tom de Vries <tom@codesourcery.com>
1560 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1561 for neutered workers.
1563 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1566 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1567 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1569 2018-01-24 Jeff Law <law@redhat.com>
1572 * i386.c (get_probe_interval): Move to earlier point.
1573 (ix86_compute_frame_layout): If -fstack-clash-protection and
1574 the frame is larger than the probe interval, then use pushes
1575 to save registers rather than reg->mem moves.
1576 (ix86_expand_prologue): Remove conditional for int_registers_saved
1579 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1582 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1583 min/max for never referenced object.
1585 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1588 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1590 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1591 attributes from DECL_ATTRIBUTES (decl) without affecting
1592 DECL_ATTRIBUTES (current_function_decl).
1593 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1594 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1596 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1598 PR tree-optimization/83979
1599 * fold-const.c (fold_comparison): Use constant_boolean_node
1600 instead of boolean_{true,false}_node.
1602 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1604 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1607 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1609 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1610 Simplify the clause that sets the length attribute.
1611 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1612 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1613 clause that sets the length attribute.
1614 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1616 2018-01-24 Tom de Vries <tom@codesourcery.com>
1619 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1620 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1621 Add strict parameter.
1622 (prevent_branch_around_nothing): Insert dummy insn between branch to
1623 label and label with no ptx insn inbetween.
1624 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1626 2018-01-24 Tom de Vries <tom@codesourcery.com>
1629 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1630 for neutered threads in warp.
1631 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1633 2018-01-24 Richard Biener <rguenther@suse.de>
1635 PR tree-optimization/83176
1636 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1639 2018-01-24 Richard Biener <rguenther@suse.de>
1641 PR tree-optimization/82819
1642 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1643 code generating pluses that are no-ops in the target precision.
1645 2018-01-24 Richard Biener <rguenther@suse.de>
1648 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1650 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1652 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1653 to merge probabilities.
1654 * predict.c (probably_never_executed): Also mark as cold functions
1655 with global 0 profile and guessed local profile.
1656 * profile-count.c (profile_probability::combine_with_count): New
1658 * profile-count.h (profile_probability::operator*,
1659 profile_probability::operator*=, profile_probability::operator/,
1660 profile_probability::operator/=): Reduce precision to adjusted
1661 and set value to guessed on contradictory divisions.
1662 (profile_probability::combine_with_freq): Remove.
1663 (profile_probability::combine_wiht_count): Declare.
1664 (profile_count::force_nonzero):: Set to adjusted.
1665 (profile_count::probability_in):: Set quality to adjusted.
1666 * tree-ssa-tail-merge.c (replace_block_by): Use
1669 2018-01-23 Andrew Waterman <andrew@sifive.com>
1670 Jim Wilson <jimw@sifive.com>
1672 * config/riscv/riscv.c (riscv_stack_boundary): New.
1673 (riscv_option_override): Set riscv_stack_boundary. Handle
1674 riscv_preferred_stack_boundary_arg.
1675 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1676 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1677 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1678 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1679 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1680 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1682 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1685 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1686 of struct ix86_frame.
1687 (ix86_expand_epilogue): Likewise. Add a local variable for
1688 the reg_save_offset field in struct ix86_frame.
1690 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1692 PR tree-optimization/82604
1693 * tree-loop-distribution.c (enum partition_kind): New enum item
1694 PKIND_PARTIAL_MEMSET.
1695 (partition_builtin_p): Support above new enum item.
1696 (generate_code_for_partition): Ditto.
1697 (compute_access_range): Differentiate cases that equality can be
1698 proven at all loops, the innermost loops or no loops.
1699 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1700 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1701 (finalize_partitions, distribute_loop): Don't fuse partition of
1702 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1703 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1706 2018-01-23 Martin Liska <mliska@suse.cz>
1708 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1709 order to ignore the predictor.
1710 (PRED_POLYMORPHIC_CALL): Likewise.
1711 (PRED_RECURSIVE_CALL): Likewise.
1713 2018-01-23 Martin Liska <mliska@suse.cz>
1715 * tree-profile.c (tree_profiling): Print function header to
1716 aware reader which function we are working on.
1717 * value-prof.c (gimple_find_values_to_profile): Do not print
1718 not interesting value histograms.
1720 2018-01-23 Martin Liska <mliska@suse.cz>
1722 * profile-count.h (enum profile_quality): Add
1723 profile_uninitialized as the first value. Do not number values
1724 as they are zero based.
1725 (profile_count::verify): Update sanity check.
1726 (profile_probability::verify): Likewise.
1728 2018-01-23 Nathan Sidwell <nathan@acm.org>
1730 * doc/invoke.texi (ffor-scope): Deprecate.
1732 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1734 PR tree-optimization/83510
1735 * domwalk.c (set_all_edges_as_executable): New function.
1736 (dom_walker::dom_walker): Convert bool param
1737 "skip_unreachable_blocks" to enum reachability. Move setup of
1738 edge flags to set_all_edges_as_executable and only do it when
1739 reachability is REACHABLE_BLOCKS.
1740 * domwalk.h (enum dom_walker::reachability): New enum.
1741 (dom_walker::dom_walker): Convert bool param
1742 "skip_unreachable_blocks" to enum reachability.
1743 (set_all_edges_as_executable): New decl.
1744 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1745 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1747 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1748 but converting true to REACHABLE_BLOCKS.
1749 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1751 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1752 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1753 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1755 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1756 if check_all_array_refs will be called.
1758 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1760 * tree.c (selftest::test_location_wrappers): Add more test
1763 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1765 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1766 (selftest::test_bit_in_range): Likewise.
1768 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1771 * doc/sourcebuild.texi (vect_float): Say that the selector
1772 only describes the situation when -funsafe-math-optimizations is on.
1773 (vect_float_strict): Document.
1775 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1777 PR tree-optimization/83965
1778 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1779 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1780 instead of checking only for a reduction.
1781 (vect_recog_widen_sum_pattern): Likewise.
1783 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1785 * predict.c (probably_never_executed): Only use precise profile info.
1786 (compute_function_frequency): Skip after inlining hack since we now
1787 have quality checking.
1789 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1791 * profile-count.h (profile_probability::very_unlikely,
1792 profile_probability::unlikely, profile_probability::even): Set
1793 precision to guessed.
1795 2018-01-23 Richard Biener <rguenther@suse.de>
1797 PR tree-optimization/83963
1798 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1799 Properly terminate dominator walk when crossing the exit edge not
1800 when visiting its source block.
1802 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1805 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1806 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1808 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1810 PR tree-optimization/83957
1811 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1812 semicolon after for body surrounded by braces.
1814 PR tree-optimization/83081
1815 * profile-count.h (profile_probability::split): New method.
1816 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1817 Use profile_probability::split.
1818 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1819 when splitting a single conditional jump into 2.
1821 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1823 PR tree-optimization/69452
1824 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1827 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1829 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1830 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1831 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1833 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1835 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1836 * config/rl78/rl78.md (movdi): New define_expand.
1837 * config/rl78/rl78.c (rl78_split_movdi): New function.
1839 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1842 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1844 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1845 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1846 128-bit to produce an UNSPEC move to get the double word with the
1847 signbit and then a shift directly to do signbit.
1848 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1849 implementation with a new version that just does either a direct
1850 move or a regular move. Move memory interface to separate insns.
1851 Move insns so they are next to the expander.
1852 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1853 with signbit move. Split big and little endian case.
1854 (signbit<mode>2_dm_mem_le): Likewise.
1855 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1856 (signbit<mode>2_dm2): Likewise.
1858 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1860 * config/rl78/rl78.md (anddi3): New define_expand.
1862 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1864 * config/rl78/rl78.md (umindi3): New define_expand.
1866 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1868 * config/rl78/rl78.md (smindi3): New define_expand.
1870 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1872 * config/rl78/rl78.md (smaxdi3): New define_expand.
1874 2018-01-22 Carl Love <cel@us.ibm.com>
1876 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1877 LVX_V1TI): Add macro expansion.
1878 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1879 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1880 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1881 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1882 Change check to determine if the instruction is a byte reversing
1883 entry. Fix typo in comment.
1884 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1885 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1886 Add def_builtin calls for new builtins.
1887 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1888 Add define_insn expansion.
1890 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1892 * config/rl78/rl78.md (umaxdi3): New define_expand.
1894 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1896 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
1897 for non-QImode registers.
1899 2018-01-22 Richard Biener <rguenther@suse.de>
1901 PR tree-optimization/83963
1902 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1903 including the loop exit block.
1904 (scop_detection::merge_sese): Likewise.
1905 (scop_detection::add_scop): Do it here instead.
1907 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1909 * doc/sourcebuild.texi (arm_softfloat): Document.
1911 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1914 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1915 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1916 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1918 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1919 David Edelsohn <dje.gcc@gmail.com>
1922 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1923 Change "crset eq" to "crset 2".
1924 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1925 (*call_indirect_aix<mode>_nospec): Likewise.
1926 (*call_value_indirect_aix<mode>_nospec): Likewise.
1927 (*call_indirect_elfv2<mode>_nospec): Likewise.
1928 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1929 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1930 change assembly output from . to $.
1931 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1932 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1933 (*tablejump<mode>_internal1_nospec): Likewise.
1935 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1938 * config/sh/sh_optimize_sett_clrt.cc:
1939 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1941 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1943 PR tree-optimization/83940
1944 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1945 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1946 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1947 use it to pass back the definition type.
1948 (vect_check_store_rhs): Likewise rhs_dt_out.
1949 (vect_build_gather_load_calls): Add a mask_dt argument and use
1950 it instead of a call to vect_is_simple_use.
1951 (vectorizable_store): Update calls to vect_check_load_store_mask
1952 and vect_check_store_rhs. Use the dt returned by the latter instead
1953 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1954 instead of calls to vect_is_simple_use. Pass the scalar rather
1955 than the vector operand to vect_is_simple_use when handling
1956 second and subsequent copies of an rhs value.
1957 (vectorizable_load): Update calls to vect_check_load_store_mask
1958 and vect_build_gather_load_calls. Use the cached mask_dt and
1959 gs_info.offset_dt instead of calls to vect_is_simple_use.
1961 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1964 * tree-emutls.c: Include gimplify.h.
1965 (lower_emutls_2): New function.
1966 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1967 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1968 it before further processing.
1971 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1972 UINTVAL (trueop1) instead of INTVAL (op1).
1974 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1978 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1979 INCOMING_FRAME_SP_OFFSET if not defined.
1980 (scan_trace): Add ENTRY argument. If true and
1981 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1982 emit a note to adjust the CFA offset.
1983 (create_cfi_notes): Adjust scan_trace callers.
1984 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1985 INCOMING_FRAME_SP_OFFSET in the CIE.
1986 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1987 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1989 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1990 * doc/tm.texi: Regenerated.
1992 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1994 PR rtl-optimization/83147
1995 * lra-constraints.c (remove_inheritance_pseudos): Use
1996 lra_substitute_pseudo_within_insn.
1998 2018-01-19 Tom de Vries <tom@codesourcery.com>
1999 Cesar Philippidis <cesar@codesourcery.com>
2002 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2004 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2007 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2008 spaces for function labels.
2010 2018-01-19 Martin Liska <mliska@suse.cz>
2012 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2013 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2014 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2015 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2016 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2017 (PRED_CONST_RETURN): Change from 69 to 65.
2018 (PRED_NULL_RETURN): Change from 91 to 71.
2019 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2020 (PRED_LOOP_GUARD): Change from 66 to 73.
2022 2018-01-19 Martin Liska <mliska@suse.cz>
2024 * predict.c (predict_insn_def): Add new assert.
2025 (struct branch_predictor): Change type to signed integer.
2026 (test_prediction_value_range): Amend test to cover
2028 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2029 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2030 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2031 (PRED_LOOP_IV_COMPARE): Likewise.
2032 * predict.h (PROB_UNINITIALIZED): Define new constant.
2034 2018-01-19 Martin Liska <mliska@suse.cz>
2036 * predict.c (dump_prediction): Add new format for
2037 analyze_brprob.py script which is enabled with -details
2039 * profile-count.h (precise_p): New function.
2041 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2043 PR tree-optimization/83922
2044 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2045 there are no statements that need masking.
2046 (vect_active_double_reduction_p): New function.
2047 (vect_analyze_loop_operations): Use it when handling phis that
2048 are not in the loop header.
2050 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2052 PR tree-optimization/83914
2053 * tree-vect-loop.c (vectorizable_induction): Don't convert
2054 init_expr or apply the peeling adjustment for inductions
2055 that are nested within the vectorized loop.
2057 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2059 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2062 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2066 * function.h (gimplify_parameters): Add gimple_seq * argument.
2067 * function.c: Include gimple.h and options.h.
2068 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2069 for the added local temporaries if needed.
2070 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2071 if there are any parameter cleanups, wrap whole body into a
2072 try/finally with the cleanups.
2074 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2077 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2078 Use GET_MODE_CLASS for scalar floating point.
2080 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2084 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2085 Fix call of call_cgraph_insertion_hooks.
2087 2018-01-18 Martin Sebor <msebor@redhat.com>
2089 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2091 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2094 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2097 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2100 * common.opt: (-ffile-prefix-map): New option.
2101 * opts.c (common_handle_option): Defer it.
2102 * opts-global.c (handle_common_deferred_options): Handle it.
2103 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2104 * file-prefix-map.h: New file.
2105 (remap_debug_filename, add_debug_prefix_map): ...here.
2106 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2107 * final.c (debug_prefix_map, add_debug_prefix_map
2108 remap_debug_filename): Move to...
2109 * file-prefix-map.c: New file.
2110 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2111 generalize, get rid of alloca(), use strrchr() instead of strchr().
2112 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2113 Implement in terms of add_prefix_map().
2114 (remap_macro_filename, remap_debug_filename): Implement in term of
2116 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2117 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2118 * dbxout.c: Include file-prefix-map.h.
2119 * varasm.c: Likewise.
2120 * vmsdbgout.c: Likewise.
2121 * xcoffout.c: Likewise.
2122 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2123 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2124 * doc/invoke.texi (-ffile-prefix-map): Document.
2125 (-fdebug-prefix-map): Update description.
2127 2018-01-18 Martin Liska <mliska@suse.cz>
2129 * config/i386/i386.c (indirect_thunk_name): Document that also
2131 (output_indirect_thunk): Document why both instructions
2132 (pause and lfence) are generated.
2134 2018-01-18 Richard Biener <rguenther@suse.de>
2136 PR tree-optimization/83887
2137 * graphite-scop-detection.c
2138 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2139 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2140 (scop_detection::merge_sese): Re-implement with a flood-fill
2141 algorithm that properly finds a SESE region if it exists.
2143 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2146 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2147 pointer_diff optimizations use view_convert instead of convert.
2149 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2151 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2152 Generate different code for -mno-speculate-indirect-jumps.
2153 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2154 (*call_indirect_aix<mode>): Disable for
2155 -mno-speculate-indirect-jumps.
2156 (*call_indirect_aix<mode>_nospec): New define_insn.
2157 (*call_value_indirect_aix<mode>): Disable for
2158 -mno-speculate-indirect-jumps.
2159 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2160 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2161 -mno-speculate-indirect-jumps.
2162 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2164 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2166 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2167 long double type, set the flags for noting the default long double
2168 type, even if we don't pass or return a long double type.
2170 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2173 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2176 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2178 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2179 support for merge[hl].
2180 (fold_mergehl_helper): New helper function.
2181 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2182 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2183 (altivec_vmrglw_direct): Add xxmrglw insn.
2185 2018-01-17 Andrew Waterman <andrew@sifive.com>
2187 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2188 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2190 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2193 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2194 call the lto_location_cache before reading the
2195 DECL_SOURCE_LOCATION of the types.
2197 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2198 Richard Sandiford <richard.sandiford@linaro.org>
2200 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2201 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2202 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2203 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2204 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2206 * config/aarch64/constraints.md (aarch64_movti_operand):
2208 * config/aarch64/predicates.md (Uti): Add new constraint.
2210 2018-01-17 Carl Love <cel@us.ibm.com>
2211 * config/rs6000/vsx.md (define_expand xl_len_r,
2212 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2213 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2215 (define_expand, define_insn): Move the shift left from the
2216 define_insn to the define_expand for lxvl and stxvl instructions.
2217 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2218 and XL_LEN_R definitions to PURE.
2220 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2222 * config/i386/i386.c (indirect_thunk_name): Declare regno
2223 as unsigned int. Compare regno with INVALID_REGNUM.
2224 (output_indirect_thunk): Ditto.
2225 (output_indirect_thunk_function): Ditto.
2226 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2227 in the call to output_indirect_thunk_function.
2229 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2232 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2233 rather than the size of inner_type to determine the stack slot size
2234 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2236 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2239 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2242 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2244 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2245 endian Linux systems to optionally enable multilibs for selecting
2246 the long double type if the user configured an explicit type.
2247 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2248 have no long double multilibs if not defined.
2249 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2250 warn if the user used -mabi={ieee,ibm}longdouble and we built
2251 multilibs for long double.
2252 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2253 appropriate multilib option.
2254 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2256 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2257 for building long double multilibs.
2258 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2260 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2262 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2265 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2267 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2270 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2273 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2276 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2278 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2279 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2281 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2283 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2284 different rtl trees depending on TARGET_64BIT.
2285 (rs6000_gen_lvx): Likewise.
2287 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2289 * config/visium/visium.md (nop): Tweak comment.
2290 (hazard_nop): Likewise.
2292 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2294 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2295 -mspeculate-indirect-jumps.
2296 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2297 for -mno-speculate-indirect-jumps.
2298 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2299 (*call_value_indirect_elfv2<mode>): Disable for
2300 -mno-speculate-indirect-jumps.
2301 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2302 (indirect_jump): Emit different RTL for
2303 -mno-speculate-indirect-jumps.
2304 (*indirect_jump<mode>): Disable for
2305 -mno-speculate-indirect-jumps.
2306 (*indirect_jump<mode>_nospec): New define_insn.
2307 (tablejump): Emit different RTL for
2308 -mno-speculate-indirect-jumps.
2309 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2310 (tablejumpsi_nospec): New define_expand.
2311 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2312 (tablejumpdi_nospec): New define_expand.
2313 (*tablejump<mode>_internal1): Disable for
2314 -mno-speculate-indirect-jumps.
2315 (*tablejump<mode>_internal1_nospec): New define_insn.
2316 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2319 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2321 * caller-save.c (insert_save): Drop unnecessary parameter. All
2324 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2325 Richard Biener <rguenth@suse.de>
2328 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2329 return early, inline manually is_gimple_sizepos. Make sure if we
2330 call gimplify_expr we don't end up with a gimple constant.
2331 * tree.c (variably_modified_type_p): Don't return true for
2332 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2333 * gimplify.h (is_gimple_sizepos): Remove.
2335 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2337 PR tree-optimization/83857
2338 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2339 vectorizable_live_operation for pure SLP statements.
2340 (vectorizable_live_operation): Handle PHIs.
2342 2018-01-16 Richard Biener <rguenther@suse.de>
2344 PR tree-optimization/83867
2345 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2346 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2348 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2351 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2352 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2353 If off is not INTEGER_CST, issue a may not be aligned warning
2354 rather than isn't aligned. Use isn%'t rather than isn't.
2355 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2357 <case MULT_EXPR>: Improve the case when bottom and one of the
2358 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2359 operand, in that case check if the other operand is multiple of
2360 bottom divided by the INTEGER_CST operand.
2362 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2365 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2366 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2367 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2368 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2369 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2370 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2371 (pa_function_arg_size): New function.
2373 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2375 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2376 in a separate statement.
2378 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2380 PR tree-optimization/83847
2381 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2382 group gathers and scatters.
2384 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2386 PR rtl-optimization/86620
2387 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2389 PR rtl-optimization/83213
2390 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2391 to last if both are JUMP_INSNs.
2393 PR tree-optimization/83843
2394 * gimple-ssa-store-merging.c
2395 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2396 store_immediate_info for bswap/nop orig_stores.
2398 2018-01-15 Andrew Waterman <andrew@sifive.com>
2400 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2402 <UDIV>: Increase cost if !TARGET_DIV.
2404 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2406 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2407 (define_attr "cr_logical_3op"): New.
2408 (cceq_ior_compare): Adjust.
2409 (cceq_ior_compare_complement): Adjust.
2410 (*cceq_rev_compare): Adjust.
2411 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2412 (is_cracked_insn): Adjust.
2413 (insn_must_be_first_in_group): Adjust.
2414 * config/rs6000/40x.md: Adjust.
2415 * config/rs6000/440.md: Adjust.
2416 * config/rs6000/476.md: Adjust.
2417 * config/rs6000/601.md: Adjust.
2418 * config/rs6000/603.md: Adjust.
2419 * config/rs6000/6xx.md: Adjust.
2420 * config/rs6000/7450.md: Adjust.
2421 * config/rs6000/7xx.md: Adjust.
2422 * config/rs6000/8540.md: Adjust.
2423 * config/rs6000/cell.md: Adjust.
2424 * config/rs6000/e300c2c3.md: Adjust.
2425 * config/rs6000/e500mc.md: Adjust.
2426 * config/rs6000/e500mc64.md: Adjust.
2427 * config/rs6000/e5500.md: Adjust.
2428 * config/rs6000/e6500.md: Adjust.
2429 * config/rs6000/mpc.md: Adjust.
2430 * config/rs6000/power4.md: Adjust.
2431 * config/rs6000/power5.md: Adjust.
2432 * config/rs6000/power6.md: Adjust.
2433 * config/rs6000/power7.md: Adjust.
2434 * config/rs6000/power8.md: Adjust.
2435 * config/rs6000/power9.md: Adjust.
2436 * config/rs6000/rs64.md: Adjust.
2437 * config/rs6000/titan.md: Adjust.
2439 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2441 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2442 ix86_indirect_branch_register logic.
2444 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2446 * config/i386/constraints.md (Bs): Update
2447 ix86_indirect_branch_register check. Don't check
2448 ix86_indirect_branch_register with GOT_memory_operand.
2450 * config/i386/predicates.md (GOT_memory_operand): Don't check
2451 ix86_indirect_branch_register here.
2452 (GOT32_symbol_operand): Likewise.
2454 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2456 * config/i386/predicates.md (constant_call_address_operand):
2457 Rewrite ix86_indirect_branch_register logic.
2458 (sibcall_insn_operand): Likewise.
2460 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2462 * config/i386/constraints.md (Bs): Replace
2463 ix86_indirect_branch_thunk_register with
2464 ix86_indirect_branch_register.
2466 * config/i386/i386.md (indirect_jump): Likewise.
2467 (tablejump): Likewise.
2468 (*sibcall_memory): Likewise.
2469 (*sibcall_value_memory): Likewise.
2470 Peepholes of indirect call and jump via memory: Likewise.
2471 * config/i386/i386.opt: Likewise.
2472 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2473 (GOT_memory_operand): Likewise.
2474 (call_insn_operand): Likewise.
2475 (sibcall_insn_operand): Likewise.
2476 (GOT32_symbol_operand): Likewise.
2478 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2481 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2482 type rather than type addr's type points to.
2483 (expand_omp_atomic_mutex): Likewise.
2484 (expand_omp_atomic): Likewise.
2486 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2489 * config/i386/i386.c (output_indirect_thunk_function): Use
2490 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2491 for __x86_return_thunk.
2493 2018-01-15 Richard Biener <rguenther@suse.de>
2496 * expmed.c (extract_bit_field_1): Fix typo.
2498 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2501 * config/arm/iterators.md (VF): New mode iterator.
2502 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2503 Remove integer-related logic from pattern.
2504 (neon_vabd<mode>_3): Likewise.
2506 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2509 * common.opt (fstrict-overflow): No longer an alias.
2510 (fwrapv-pointer): New option.
2511 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2512 also for pointer types based on flag_wrapv_pointer.
2513 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2514 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2515 opts->x_flag_wrapv got set.
2516 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2517 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2518 POINTER_TYPE_OVERFLOW_UNDEFINED.
2519 * match.pd: Likewise in address comparison pattern.
2520 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2522 2018-01-15 Richard Biener <rguenther@suse.de>
2525 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2526 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2527 Reset type names to their identifier if their TYPE_DECL doesn't
2528 have linkage (and thus is used for ODR and devirt).
2529 (save_debug_info_for_decl): Remove.
2530 (save_debug_info_for_type): Likewise.
2531 (add_tree_to_fld_list): Adjust.
2532 * tree-pretty-print.c (dump_generic_node): Make dumping of
2533 type names more robust.
2535 2018-01-15 Richard Biener <rguenther@suse.de>
2537 * BASE-VER: Bump to 8.0.1.
2539 2018-01-14 Martin Sebor <msebor@redhat.com>
2542 * builtins.c (check_access): Avoid warning when the no-warning bit
2545 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2547 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2548 * ira-color (allocno_hard_regs_compare): Likewise.
2550 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2553 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2554 Use .pushsection/.popsection.
2556 2018-01-14 Martin Sebor <msebor@redhat.com>
2559 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2561 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2563 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2564 entry from extra_headers.
2565 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2566 extra_headers, make the list bitwise identical to the i?86-*-* one.
2568 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2570 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2571 -mcmodel=large with -mindirect-branch=thunk,
2572 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2573 -mfunction-return=thunk-extern.
2574 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2575 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2576 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2578 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2580 * config/i386/i386.c (print_reg): Print the name of the full
2581 integer register without '%'.
2582 (ix86_print_operand): Handle 'V'.
2583 * doc/extend.texi: Document 'V' modifier.
2585 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2587 * config/i386/constraints.md (Bs): Disallow memory operand for
2588 -mindirect-branch-register.
2590 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2591 (GOT_memory_operand): Likewise.
2592 (call_insn_operand): Likewise.
2593 (sibcall_insn_operand): Likewise.
2594 (GOT32_symbol_operand): Likewise.
2595 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2596 for -mindirect-branch-register.
2597 (tablejump): Likewise.
2598 (*sibcall_memory): Likewise.
2599 (*sibcall_value_memory): Likewise.
2600 Disallow peepholes of indirect call and jump via memory for
2601 -mindirect-branch-register.
2602 (*call_pop): Replace m with Bw.
2603 (*call_value_pop): Likewise.
2604 (*sibcall_pop_memory): Replace m with Bs.
2605 * config/i386/i386.opt (mindirect-branch-register): New option.
2606 * doc/invoke.texi: Document -mindirect-branch-register option.
2608 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2610 * config/i386/i386-protos.h (ix86_output_function_return): New.
2611 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2612 set function_return_type.
2613 (indirect_thunk_name): Add ret_p to indicate thunk for function
2615 (output_indirect_thunk_function): Pass false to
2616 indirect_thunk_name.
2617 (ix86_output_indirect_branch_via_reg): Likewise.
2618 (ix86_output_indirect_branch_via_push): Likewise.
2619 (output_indirect_thunk_function): Create alias for function
2620 return thunk if regno < 0.
2621 (ix86_output_function_return): New function.
2622 (ix86_handle_fndecl_attribute): Handle function_return.
2623 (ix86_attribute_table): Add function_return.
2624 * config/i386/i386.h (machine_function): Add
2625 function_return_type.
2626 * config/i386/i386.md (simple_return_internal): Use
2627 ix86_output_function_return.
2628 (simple_return_internal_long): Likewise.
2629 * config/i386/i386.opt (mfunction-return=): New option.
2630 (indirect_branch): Mention -mfunction-return=.
2631 * doc/extend.texi: Document function_return function attribute.
2632 * doc/invoke.texi: Document -mfunction-return= option.
2634 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2636 * config/i386/i386-opts.h (indirect_branch): New.
2637 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2638 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2639 with local indirect jump when converting indirect call and jump.
2640 (ix86_set_indirect_branch_type): New.
2641 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2642 (indirectlabelno): New.
2643 (indirect_thunk_needed): Likewise.
2644 (indirect_thunk_bnd_needed): Likewise.
2645 (indirect_thunks_used): Likewise.
2646 (indirect_thunks_bnd_used): Likewise.
2647 (INDIRECT_LABEL): Likewise.
2648 (indirect_thunk_name): Likewise.
2649 (output_indirect_thunk): Likewise.
2650 (output_indirect_thunk_function): Likewise.
2651 (ix86_output_indirect_branch_via_reg): Likewise.
2652 (ix86_output_indirect_branch_via_push): Likewise.
2653 (ix86_output_indirect_branch): Likewise.
2654 (ix86_output_indirect_jmp): Likewise.
2655 (ix86_code_end): Call output_indirect_thunk_function if needed.
2656 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2658 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2659 (ix86_attribute_table): Add indirect_branch.
2660 * config/i386/i386.h (machine_function): Add indirect_branch_type
2661 and has_local_indirect_jump.
2662 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2664 (tablejump): Likewise.
2665 (*indirect_jump): Use ix86_output_indirect_jmp.
2666 (*tablejump_1): Likewise.
2667 (simple_return_indirect_internal): Likewise.
2668 * config/i386/i386.opt (mindirect-branch=): New option.
2669 (indirect_branch): New.
2672 (thunk-inline): Likewise.
2673 (thunk-extern): Likewise.
2674 * doc/extend.texi: Document indirect_branch function attribute.
2675 * doc/invoke.texi: Document -mindirect-branch= option.
2677 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2680 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2682 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2684 * ipa-inline.c (want_inline_small_function_p): Return false if
2685 inlining has already failed with CIF_FINAL_ERROR.
2686 (update_caller_keys): Call want_inline_small_function_p before
2688 (update_callee_keys): Likewise.
2690 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2692 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2694 (rs6000_quadword_masked_address_p): Likewise.
2695 (quad_aligned_load_p): Likewise.
2696 (quad_aligned_store_p): Likewise.
2697 (const_load_sequence_p): Add comment to describe the outer-most loop.
2698 (mimic_memory_attributes_and_flags): New function.
2699 (rs6000_gen_stvx): Likewise.
2700 (replace_swapped_aligned_store): Likewise.
2701 (rs6000_gen_lvx): Likewise.
2702 (replace_swapped_aligned_load): Likewise.
2703 (replace_swapped_load_constant): Capitalize argument name in
2704 comment describing this function.
2705 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2706 and stores that access quad-word aligned addresses and replace
2707 with stvx or lvx instructions when appropriate.
2708 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2709 New function prototype.
2710 (rs6000_quadword_masked_address_p): Likewise.
2711 (rs6000_gen_lvx): Likewise.
2712 (rs6000_gen_stvx): Likewise.
2713 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2714 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2715 when memory address is aligned.
2716 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2717 this split to select lvx instruction when memory address is aligned.
2718 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2719 instruction when memory address is aligned.
2720 (*vsx_le_perm_load_v16qi): Likewise.
2721 (four unnamed splitters): Modify to select the stvx instruction
2722 when memory is aligned.
2724 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2726 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2727 which appears in the queue multiple times.
2729 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2730 Alan Hayward <alan.hayward@arm.com>
2731 David Sherwood <david.sherwood@arm.com>
2733 * tree-vectorizer.h (vec_lower_bound): New structure.
2734 (_loop_vec_info): Add check_nonzero and lower_bounds.
2735 (LOOP_VINFO_CHECK_NONZERO): New macro.
2736 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2737 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2738 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2739 fields. Make seg_len the distance travelled, not including the
2741 (dr_direction_indicator): Declare.
2742 (dr_zero_step_indicator): Likewise.
2743 (dr_known_forward_stride_p): Likewise.
2744 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2746 (runtime_alias_check_p): Allow runtime alias checks with
2748 (operator ==): Compare access_size and align.
2749 (prune_runtime_alias_test_list): Rework for new distinction between
2750 the access_size and seg_len.
2751 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2753 (get_segment_min_max): New function.
2754 (create_intersect_range_checks): Use it.
2755 (dr_step_indicator): New function.
2756 (dr_direction_indicator): Likewise.
2757 (dr_zero_step_indicator): Likewise.
2758 (dr_known_forward_stride_p): Likewise.
2759 * tree-loop-distribution.c (data_ref_segment_size): Return
2760 DR_STEP * (niters - 1).
2761 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2763 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2764 (vect_preserves_scalar_order_p): New function, split out from...
2765 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2766 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2767 (vect_vfa_access_size): New function.
2768 (vect_vfa_align): Likewise.
2769 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2770 (dump_lower_bound): New function.
2771 (vect_check_lower_bound): Likewise.
2772 (vect_small_gap_p): Likewise.
2773 (vectorizable_with_step_bound_p): Likewise.
2774 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2775 depencies if the vectorization factor is 1. Convert the checks
2776 for nonzero steps into checks on the bounds of DR_STEP. Try using
2777 a bunds check for variable steps if the minimum required step is
2778 relatively small. Update calls to the dr_with_seg_len
2779 constructor and to vect_compile_time_alias.
2780 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2782 (vect_loop_versioning): Call it.
2783 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2785 (vect_estimate_min_profitable_iters): Account for any bounds checks.
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/sourcebuild.texi (vect_scatter_store): Document.
2792 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2794 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2796 * genopinit.c (main): Add supports_vec_scatter_store and
2797 supports_vec_scatter_store_cached to target_optabs.
2798 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2799 IFN_MASK_SCATTER_STORE.
2800 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2802 * internal-fn.h (internal_store_fn_p): Declare.
2803 (internal_fn_stored_value_index): Likewise.
2804 * internal-fn.c (scatter_store_direct): New macro.
2805 (expand_scatter_store_optab_fn): New function.
2806 (direct_scatter_store_optab_supported_p): New macro.
2807 (internal_store_fn_p): New function.
2808 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2809 IFN_MASK_SCATTER_STORE.
2810 (internal_fn_mask_index): Likewise.
2811 (internal_fn_stored_value_index): New function.
2812 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2814 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2815 * optabs-query.c (supports_vec_scatter_store_p): New function.
2816 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2817 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2818 true for scatter stores.
2819 (vect_gather_scatter_fn_p): Handle scatter stores too.
2820 (vect_check_gather_scatter): Consider using scatter stores if
2821 supports_vec_scatter_store_p.
2822 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2824 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2825 internal_fn_stored_value_index.
2826 (check_load_store_masking): Handle scatter stores too.
2827 (vect_get_store_rhs): Make public.
2828 (vectorizable_call): Use internal_store_fn_p.
2829 (vectorizable_store): Handle scatter store internal functions.
2830 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2831 when deciding whether the end of the group has been reached.
2832 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2833 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2834 (mask_scatter_store<mode>): New insns.
2836 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2837 Alan Hayward <alan.hayward@arm.com>
2838 David Sherwood <david.sherwood@arm.com>
2840 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2841 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2842 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2844 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2845 Use vect_truncate_gather_scatter_offset if we can't treat the
2846 operation as a normal gather load or scatter store.
2847 (get_group_load_store_type): Take the gather_scatter_info
2848 as argument. Try using a gather load or scatter store for
2849 single-element groups.
2850 (get_load_store_type): Update calls to get_group_load_store_type
2851 and vect_use_strided_gather_scatters_p.
2853 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2854 Alan Hayward <alan.hayward@arm.com>
2855 David Sherwood <david.sherwood@arm.com>
2857 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2858 optional tree argument.
2859 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2861 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2862 but continue to use the current value as a fallback.
2863 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2864 to compare the updates.
2865 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2866 (get_load_store_type): Use it when handling a strided access.
2867 (vect_get_strided_load_store_ops): New function.
2868 (vect_get_data_ptr_increment): Likewise.
2869 (vectorizable_load): Handle strided gather loads. Always pass
2870 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2872 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2873 Alan Hayward <alan.hayward@arm.com>
2874 David Sherwood <david.sherwood@arm.com>
2876 * doc/md.texi (gather_load@var{m}): Document.
2877 (mask_gather_load@var{m}): Likewise.
2878 * genopinit.c (main): Add supports_vec_gather_load and
2879 supports_vec_gather_load_cached to target_optabs.
2880 * optabs-tree.c (init_tree_optimization_optabs): Use
2881 ggc_cleared_alloc to allocate target_optabs.
2882 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2883 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2885 * internal-fn.h (internal_load_fn_p): Declare.
2886 (internal_gather_scatter_fn_p): Likewise.
2887 (internal_fn_mask_index): Likewise.
2888 (internal_gather_scatter_fn_supported_p): Likewise.
2889 * internal-fn.c (gather_load_direct): New macro.
2890 (expand_gather_load_optab_fn): New function.
2891 (direct_gather_load_optab_supported_p): New macro.
2892 (direct_internal_fn_optab): New function.
2893 (internal_load_fn_p): Likewise.
2894 (internal_gather_scatter_fn_p): Likewise.
2895 (internal_fn_mask_index): Likewise.
2896 (internal_gather_scatter_fn_supported_p): Likewise.
2897 * optabs-query.c (supports_at_least_one_mode_p): New function.
2898 (supports_vec_gather_load_p): Likewise.
2899 * optabs-query.h (supports_vec_gather_load_p): Declare.
2900 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2901 and memory_type field.
2902 (NUM_PATTERNS): Bump to 15.
2903 * tree-vect-data-refs.c: Include internal-fn.h.
2904 (vect_gather_scatter_fn_p): New function.
2905 (vect_describe_gather_scatter_call): Likewise.
2906 (vect_check_gather_scatter): Try using internal functions for
2907 gather loads. Recognize existing calls to a gather load function.
2908 (vect_analyze_data_refs): Consider using gather loads if
2909 supports_vec_gather_load_p.
2910 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2911 (vect_get_gather_scatter_offset_type): Likewise.
2912 (vect_convert_mask_for_vectype): Likewise.
2913 (vect_add_conversion_to_patterm): Likewise.
2914 (vect_try_gather_scatter_pattern): Likewise.
2915 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2916 (vect_vect_recog_func_ptrs): Add it.
2917 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2918 internal_fn_mask_index and internal_gather_scatter_fn_p.
2919 (check_load_store_masking): Take the gather_scatter_info as an
2920 argument and handle gather loads.
2921 (vect_get_gather_scatter_ops): New function.
2922 (vectorizable_call): Check internal_load_fn_p.
2923 (vectorizable_load): Likewise. Handle gather load internal
2925 (vectorizable_store): Update call to check_load_store_masking.
2926 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2927 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2928 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2929 (aarch64_gather_scale_operand_d): New predicates.
2930 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2931 (mask_gather_load<mode>): New insns.
2933 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2934 Alan Hayward <alan.hayward@arm.com>
2935 David Sherwood <david.sherwood@arm.com>
2937 * optabs.def (fold_left_plus_optab): New optab.
2938 * doc/md.texi (fold_left_plus_@var{m}): Document.
2939 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2940 * internal-fn.c (fold_left_direct): Define.
2941 (expand_fold_left_optab_fn): Likewise.
2942 (direct_fold_left_optab_supported_p): Likewise.
2943 * fold-const-call.c (fold_const_fold_left): New function.
2944 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2945 * tree-parloops.c (valid_reduction_p): New function.
2946 (gather_scalar_reductions): Use it.
2947 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2948 (vect_finish_replace_stmt): Declare.
2949 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2950 (needs_fold_left_reduction_p): New function, split out from...
2951 (vect_is_simple_reduction): ...here. Accept reductions that
2952 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2953 (vect_force_simple_reduction): Also store the reduction type in
2954 the assignment's STMT_VINFO_REDUC_TYPE.
2955 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2956 (merge_with_identity): New function.
2957 (vect_expand_fold_left): Likewise.
2958 (vectorize_fold_left_reduction): Likewise.
2959 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2960 scalar phi in place for it. Check for target support and reject
2961 cases that would reassociate the operation. Defer the transform
2962 phase to vectorize_fold_left_reduction.
2963 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2964 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2965 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2967 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2969 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2970 call to ifc_temp_var.
2972 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2973 Alan Hayward <alan.hayward@arm.com>
2974 David Sherwood <david.sherwood@arm.com>
2976 * target.def (legitimize_address_displacement): Take the original
2977 offset as a poly_int.
2978 * targhooks.h (default_legitimize_address_displacement): Update
2980 * targhooks.c (default_legitimize_address_displacement): Likewise.
2981 * doc/tm.texi: Regenerate.
2982 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2983 as an argument, moving assert of ad->disp == ad->disp_term to...
2984 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2985 Try calling targetm.legitimize_address_displacement before expanding
2986 the address rather than afterwards, and adjust for the new interface.
2987 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2988 Match the new hook interface. Handle SVE addresses.
2989 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2992 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2994 * Makefile.in (OBJS): Add early-remat.o.
2995 * target.def (select_early_remat_modes): New hook.
2996 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2997 * doc/tm.texi: Regenerate.
2998 * targhooks.h (default_select_early_remat_modes): Declare.
2999 * targhooks.c (default_select_early_remat_modes): New function.
3000 * timevar.def (TV_EARLY_REMAT): New timevar.
3001 * passes.def (pass_early_remat): New pass.
3002 * tree-pass.h (make_pass_early_remat): Declare.
3003 * early-remat.c: New file.
3004 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3006 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3008 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3009 Alan Hayward <alan.hayward@arm.com>
3010 David Sherwood <david.sherwood@arm.com>
3012 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3013 vfm1 with a bound_epilog parameter.
3014 (vect_do_peeling): Update calls accordingly, and move the prologue
3015 call earlier in the function. Treat the base bound_epilog as 0 for
3016 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3017 this base when peeling for gaps.
3018 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3019 with fully-masked loops.
3020 (vect_estimate_min_profitable_iters): Handle the single peeled
3021 iteration in that case.
3023 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3024 Alan Hayward <alan.hayward@arm.com>
3025 David Sherwood <david.sherwood@arm.com>
3027 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3028 single-element interleaving even if the size is not a power of 2.
3029 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3030 accesses for single-element interleaving if the group size is
3033 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3034 Alan Hayward <alan.hayward@arm.com>
3035 David Sherwood <david.sherwood@arm.com>
3037 * doc/md.texi (fold_extract_last_@var{m}): Document.
3038 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3039 * optabs.def (fold_extract_last_optab): New optab.
3040 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3041 * internal-fn.c (fold_extract_direct): New macro.
3042 (expand_fold_extract_optab_fn): Likewise.
3043 (direct_fold_extract_optab_supported_p): Likewise.
3044 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3045 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3046 EXTRACT_LAST_REDUCTION.
3047 (get_initial_def_for_reduction): Do not create an initial vector
3048 for EXTRACT_LAST_REDUCTION reductions.
3049 (vectorizable_reduction): Leave the scalar phi in place for
3050 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3051 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3052 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3053 transform phase to vectorizable_condition.
3054 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3056 (vect_finish_stmt_generation): ...here.
3057 (vect_finish_replace_stmt): New function.
3058 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3059 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3061 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3063 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3064 Alan Hayward <alan.hayward@arm.com>
3065 David Sherwood <david.sherwood@arm.com>
3067 * doc/md.texi (extract_last_@var{m}): Document.
3068 * optabs.def (extract_last_optab): New optab.
3069 * internal-fn.def (EXTRACT_LAST): New internal function.
3070 * internal-fn.c (cond_unary_direct): New macro.
3071 (expand_cond_unary_optab_fn): Likewise.
3072 (direct_cond_unary_optab_supported_p): Likewise.
3073 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3074 loops using EXTRACT_LAST.
3075 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3076 (extract_last_<mode>): ...this optab.
3077 (vec_extract<mode><Vel>): Update accordingly.
3079 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3080 Alan Hayward <alan.hayward@arm.com>
3081 David Sherwood <david.sherwood@arm.com>
3083 * target.def (empty_mask_is_expensive): New hook.
3084 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3085 * doc/tm.texi: Regenerate.
3086 * targhooks.h (default_empty_mask_is_expensive): Declare.
3087 * targhooks.c (default_empty_mask_is_expensive): New function.
3088 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3089 if the target says that empty masks are expensive.
3090 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3092 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3094 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3095 Alan Hayward <alan.hayward@arm.com>
3096 David Sherwood <david.sherwood@arm.com>
3098 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3099 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3100 (vect_use_loop_mask_for_alignment_p): New function.
3101 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3102 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3103 niters_skip argument. Make sure that the first niters_skip elements
3104 of the first iteration are inactive.
3105 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3106 Update call to vect_set_loop_masks_directly.
3107 (get_misalign_in_elems): New function, split out from...
3108 (vect_gen_prolog_loop_niters): ...here.
3109 (vect_update_init_of_dr): Take a code argument that specifies whether
3110 the adjustment should be added or subtracted.
3111 (vect_update_init_of_drs): Likewise.
3112 (vect_prepare_for_masked_peels): New function.
3113 (vect_do_peeling): Skip prologue peeling if we're using a mask
3114 instead. Update call to vect_update_inits_of_drs.
3115 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3117 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3118 alignment. Do not include the number of peeled iterations in
3119 the minimum threshold in that case.
3120 (vectorizable_induction): Adjust the start value down by
3121 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3122 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3123 Take the number of skipped iterations into account when calculating
3125 * tree-vect-stmts.c (vect_gen_while_not): New function.
3127 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3128 Alan Hayward <alan.hayward@arm.com>
3129 David Sherwood <david.sherwood@arm.com>
3131 * doc/sourcebuild.texi (vect_fully_masked): Document.
3132 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3134 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3136 (vect_analyze_loop_2): ...here. Don't check the vectorization
3137 factor against the number of loop iterations if the loop is
3140 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3141 Alan Hayward <alan.hayward@arm.com>
3142 David Sherwood <david.sherwood@arm.com>
3144 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3145 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3146 (dump_groups): Update accordingly.
3147 (iv_use::mem_type): New member variable.
3148 (address_p): New function.
3149 (record_use): Add a mem_type argument and initialize the new
3151 (record_group_use): Add a mem_type argument. Use address_p.
3152 Remove obsolete null checks of base_object. Update call to record_use.
3153 (find_interesting_uses_op): Update call to record_group_use.
3154 (find_interesting_uses_cond): Likewise.
3155 (find_interesting_uses_address): Likewise.
3156 (get_mem_type_for_internal_fn): New function.
3157 (find_address_like_use): Likewise.
3158 (find_interesting_uses_stmt): Try find_address_like_use before
3159 calling find_interesting_uses_op.
3160 (addr_offset_valid_p): Use the iv mem_type field as the type
3161 of the addressed memory.
3162 (add_autoinc_candidates): Likewise.
3163 (get_address_cost): Likewise.
3164 (split_small_address_groups_p): Use address_p.
3165 (split_address_groups): Likewise.
3166 (add_iv_candidate_for_use): Likewise.
3167 (autoinc_possible_for_pair): Likewise.
3168 (rewrite_groups): Likewise.
3169 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3170 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3171 (get_alias_ptr_type_for_ptr_address): New function.
3172 (rewrite_use_address): Rewrite address uses in calls that were
3173 identified by find_address_like_use.
3175 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3176 Alan Hayward <alan.hayward@arm.com>
3177 David Sherwood <david.sherwood@arm.com>
3179 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3181 * gimple-expr.h (is_gimple_addressable: Likewise.
3182 * gimple-expr.c (is_gimple_address): Likewise.
3183 * internal-fn.c (expand_call_mem_ref): New function.
3184 (expand_mask_load_optab_fn): Use it.
3185 (expand_mask_store_optab_fn): Likewise.
3187 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3188 Alan Hayward <alan.hayward@arm.com>
3189 David Sherwood <david.sherwood@arm.com>
3191 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3192 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3193 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3194 (cond_umax@var{mode}): Document.
3195 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3196 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3197 (cond_umin_optab, cond_umax_optab): New optabs.
3198 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3199 (COND_IOR, COND_XOR): New internal functions.
3200 * internal-fn.h (get_conditional_internal_fn): Declare.
3201 * internal-fn.c (cond_binary_direct): New macro.
3202 (expand_cond_binary_optab_fn): Likewise.
3203 (direct_cond_binary_optab_supported_p): Likewise.
3204 (get_conditional_internal_fn): New function.
3205 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3206 Cope with reduction statements that are vectorized as calls rather
3208 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3209 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3210 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3211 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3212 (UNSPEC_COND_EOR): New unspecs.
3213 (optab): Add mappings for them.
3214 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3215 (sve_int_op, sve_fp_op): New int attributes.
3217 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3218 Alan Hayward <alan.hayward@arm.com>
3219 David Sherwood <david.sherwood@arm.com>
3221 * optabs.def (while_ult_optab): New optab.
3222 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3223 * internal-fn.def (WHILE_ULT): New internal function.
3224 * internal-fn.h (direct_internal_fn_supported_p): New override
3225 that takes two types as argument.
3226 * internal-fn.c (while_direct): New macro.
3227 (expand_while_optab_fn): New function.
3228 (convert_optab_supported_p): Likewise.
3229 (direct_while_optab_supported_p): New macro.
3230 * wide-int.h (wi::udiv_ceil): New function.
3231 * tree-vectorizer.h (rgroup_masks): New structure.
3232 (vec_loop_masks): New typedef.
3233 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3235 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3236 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3237 (vect_max_vf): New function.
3238 (slpeel_make_loop_iterate_ntimes): Delete.
3239 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3240 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3241 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3242 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3243 internal-fn.h, stor-layout.h and optabs-query.h.
3244 (vect_set_loop_mask): New function.
3245 (add_preheader_seq): Likewise.
3246 (add_header_seq): Likewise.
3247 (interleave_supported_p): Likewise.
3248 (vect_maybe_permute_loop_masks): Likewise.
3249 (vect_set_loop_masks_directly): Likewise.
3250 (vect_set_loop_condition_masked): Likewise.
3251 (vect_set_loop_condition_unmasked): New function, split out from
3252 slpeel_make_loop_iterate_ntimes.
3253 (slpeel_make_loop_iterate_ntimes): Rename to..
3254 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3255 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3256 (vect_do_peeling): Update call accordingly.
3257 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3259 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3260 mask_compare_type, can_fully_mask_p and fully_masked_p.
3261 (release_vec_loop_masks): New function.
3262 (_loop_vec_info): Use it to free the loop masks.
3263 (can_produce_all_loop_masks_p): New function.
3264 (vect_get_max_nscalars_per_iter): Likewise.
3265 (vect_verify_full_masking): Likewise.
3266 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3267 retries, and free the mask rgroups before retrying. Check loop-wide
3268 reasons for disallowing fully-masked loops. Make the final decision
3269 about whether use a fully-masked loop or not.
3270 (vect_estimate_min_profitable_iters): Do not assume that peeling
3271 for the number of iterations will be needed for fully-masked loops.
3272 (vectorizable_reduction): Disable fully-masked loops.
3273 (vectorizable_live_operation): Likewise.
3274 (vect_halve_mask_nunits): New function.
3275 (vect_double_mask_nunits): Likewise.
3276 (vect_record_loop_mask): Likewise.
3277 (vect_get_loop_mask): Likewise.
3278 (vect_transform_loop): Handle the case in which the final loop
3279 iteration might handle a partial vector. Call vect_set_loop_condition
3280 instead of slpeel_make_loop_iterate_ntimes.
3281 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3282 (check_load_store_masking): New function.
3283 (prepare_load_store_mask): Likewise.
3284 (vectorizable_store): Handle fully-masked loops.
3285 (vectorizable_load): Likewise.
3286 (supportable_widening_operation): Use vect_halve_mask_nunits for
3288 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3289 (vect_gen_while): New function.
3290 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3291 (aarch64_uqdec<mode>): New insn.
3293 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3294 Alan Hayward <alan.hayward@arm.com>
3295 David Sherwood <david.sherwood@arm.com>
3297 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3298 (reduc_xor_scal_optab): New optabs.
3299 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3300 (reduc_xor_scal_@var{m}): Document.
3301 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3302 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3304 * fold-const-call.c (fold_const_call): Handle them.
3305 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3306 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3307 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3308 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3309 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3310 (UNSPEC_XORV): New unspecs.
3311 (optab): Add entries for them.
3312 (BITWISEV): New int iterator.
3313 (bit_reduc_op): New int attributes.
3315 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3316 Alan Hayward <alan.hayward@arm.com>
3317 David Sherwood <david.sherwood@arm.com>
3319 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3320 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3321 * optabs.def (vec_shl_insert_optab): New optab.
3322 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3323 (duplicate_and_interleave): Likewise.
3324 * tree-vect-loop.c: Include internal-fn.h.
3325 (neutral_op_for_slp_reduction): New function, split out from
3326 get_initial_defs_for_reduction.
3327 (get_initial_def_for_reduction): Handle option 2 for variable-length
3328 vectors by loading the neutral value into a vector and then shifting
3329 the initial value into element 0.
3330 (get_initial_defs_for_reduction): Replace the code argument with
3331 the neutral value calculated by neutral_op_for_slp_reduction.
3332 Use gimple_build_vector for constant-length vectors.
3333 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3334 but the first group_size elements have a neutral value.
3335 Use duplicate_and_interleave otherwise.
3336 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3337 Update call to get_initial_defs_for_reduction. Handle SLP
3338 reductions for variable-length vectors by creating one vector
3339 result for each scalar result, with the elements associated
3340 with other scalar results stubbed out with the neutral value.
3341 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3342 Require IFN_VEC_SHL_INSERT for double reductions on
3343 variable-length vectors, or SLP reductions that have
3344 a neutral value. Require can_duplicate_and_interleave_p
3345 support for variable-length unchained SLP reductions if there
3346 is no neutral value, such as for MIN/MAX reductions. Also require
3347 the number of vector elements to be a multiple of the number of
3348 SLP statements when doing variable-length unchained SLP reductions.
3349 Update call to vect_create_epilog_for_reduction.
3350 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3351 and remove initial values.
3352 (duplicate_and_interleave): Make public.
3353 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3354 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3356 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3357 Alan Hayward <alan.hayward@arm.com>
3358 David Sherwood <david.sherwood@arm.com>
3360 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3361 (can_duplicate_and_interleave_p): New function.
3362 (vect_get_and_check_slp_defs): Take the vector of statements
3363 rather than just the current one. Remove excess parentheses.
3364 Restriction rejectinon of vect_constant_def and vect_external_def
3365 for variable-length vectors to boolean types, or types for which
3366 can_duplicate_and_interleave_p is false.
3367 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3368 (duplicate_and_interleave): New function.
3369 (vect_get_constant_vectors): Use gimple_build_vector for
3370 constant-length vectors and suitable variable-length constant
3371 vectors. Use duplicate_and_interleave for other variable-length
3372 vectors. Don't defer the update when inserting new statements.
3374 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3375 Alan Hayward <alan.hayward@arm.com>
3376 David Sherwood <david.sherwood@arm.com>
3378 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3379 min_profitable_iters doesn't go negative.
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 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3386 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3387 * optabs.def (vec_mask_load_lanes_optab): New optab.
3388 (vec_mask_store_lanes_optab): Likewise.
3389 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3390 (MASK_STORE_LANES): Likewise.
3391 * internal-fn.c (mask_load_lanes_direct): New macro.
3392 (mask_store_lanes_direct): Likewise.
3393 (expand_mask_load_optab_fn): Handle masked operations.
3394 (expand_mask_load_lanes_optab_fn): New macro.
3395 (expand_mask_store_optab_fn): Handle masked operations.
3396 (expand_mask_store_lanes_optab_fn): New macro.
3397 (direct_mask_load_lanes_optab_supported_p): Likewise.
3398 (direct_mask_store_lanes_optab_supported_p): Likewise.
3399 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3401 (vect_load_lanes_supported): Likewise.
3402 * tree-vect-data-refs.c (strip_conversion): New function.
3403 (can_group_stmts_p): Likewise.
3404 (vect_analyze_data_ref_accesses): Use it instead of checking
3405 for a pair of assignments.
3406 (vect_store_lanes_supported): Take a masked_p parameter.
3407 (vect_load_lanes_supported): Likewise.
3408 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3409 vect_store_lanes_supported and vect_load_lanes_supported.
3410 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3411 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3412 parameter. Don't allow gaps for masked accesses.
3413 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3414 and vect_load_lanes_supported.
3415 (get_load_store_type): Take a masked_p parameter and update
3416 call to get_group_load_store_type.
3417 (vectorizable_store): Update call to get_load_store_type.
3418 Handle IFN_MASK_STORE_LANES.
3419 (vectorizable_load): Update call to get_load_store_type.
3420 Handle IFN_MASK_LOAD_LANES.
3422 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3423 Alan Hayward <alan.hayward@arm.com>
3424 David Sherwood <david.sherwood@arm.com>
3426 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3428 * config/aarch64/aarch64-protos.h
3429 (aarch64_sve_struct_memory_operand_p): Declare.
3430 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3431 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3432 (VPRED, vpred): Handle SVE structure modes.
3433 * config/aarch64/constraints.md (Utx): New constraint.
3434 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3435 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3436 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3437 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3438 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3439 structure modes. Split into pieces after RA.
3440 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3441 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3443 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3444 SVE structure modes.
3445 (aarch64_classify_address): Likewise.
3446 (sizetochar): Move earlier in file.
3447 (aarch64_print_operand): Handle SVE register lists.
3448 (aarch64_array_mode): New function.
3449 (aarch64_sve_struct_memory_operand_p): Likewise.
3450 (TARGET_ARRAY_MODE): Redefine.
3452 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3453 Alan Hayward <alan.hayward@arm.com>
3454 David Sherwood <david.sherwood@arm.com>
3456 * target.def (array_mode): New target hook.
3457 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3458 * doc/tm.texi: Regenerate.
3459 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3460 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3461 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3463 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3466 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3467 Alan Hayward <alan.hayward@arm.com>
3468 David Sherwood <david.sherwood@arm.com>
3470 * fold-const.c (fold_binary_loc): Check the argument types
3471 rather than the result type when testing for a vector operation.
3473 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3475 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3476 * doc/tm.texi: Regenerate.
3478 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3479 Alan Hayward <alan.hayward@arm.com>
3480 David Sherwood <david.sherwood@arm.com>
3482 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3483 (sve): Document new AArch64 extension.
3484 * doc/md.texi (w): Extend the description of the AArch64
3485 constraint to include SVE vectors.
3486 (Upl, Upa): Document new AArch64 predicate constraints.
3487 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3489 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3490 (msve-vector-bits=): New option.
3491 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3492 SVE when these are disabled.
3493 (sve): New extension.
3494 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3495 modes. Adjust their number of units based on aarch64_sve_vg.
3496 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3497 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3498 aarch64_addr_query_type.
3499 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3500 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3501 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3502 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3503 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3504 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3505 (aarch64_simd_imm_zero_p): Delete.
3506 (aarch64_check_zero_based_sve_index_immediate): Declare.
3507 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3508 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3509 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3510 (aarch64_sve_float_mul_immediate_p): Likewise.
3511 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3513 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3514 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3515 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3516 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3517 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3518 (aarch64_regmode_natural_size): Likewise.
3519 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3520 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3522 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3523 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3524 for VG and the SVE predicate registers.
3525 (V_ALIASES): Add a "z"-prefixed alias.
3526 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3527 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3528 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3529 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3530 (REG_CLASS_NAMES): Add entries for them.
3531 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3532 and the predicate registers.
3533 (aarch64_sve_vg): Declare.
3534 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3535 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3536 (REGMODE_NATURAL_SIZE): Define.
3537 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3539 * config/aarch64/aarch64.c: Include cfgrtl.h.
3540 (simd_immediate_info): Add a constructor for series vectors,
3541 and an associated step field.
3542 (aarch64_sve_vg): New variable.
3543 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3544 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3545 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3546 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3547 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3548 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3549 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3550 (aarch64_get_mask_mode): New functions.
3551 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3552 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3553 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3554 predicate modes and predicate registers. Explicitly restrict
3555 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3556 to store a vector mode if it is recognized by
3557 aarch64_classify_vector_mode.
3558 (aarch64_regmode_natural_size): New function.
3559 (aarch64_hard_regno_caller_save_mode): Return the original mode
3561 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3562 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3563 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3564 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3566 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3567 does not overlap dest if the function is frame-related. Handle
3569 (aarch64_split_add_offset): New function.
3570 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3571 them aarch64_add_offset.
3572 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3573 and update call to aarch64_sub_sp.
3574 (aarch64_add_cfa_expression): New function.
3575 (aarch64_expand_prologue): Pass extra temporary registers to the
3576 functions above. Handle the case in which we need to emit new
3577 DW_CFA_expressions for registers that were originally saved
3578 relative to the stack pointer, but now have to be expressed
3579 relative to the frame pointer.
3580 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3582 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3583 IP0 and IP1 values for SVE frames.
3584 (aarch64_expand_vec_series): New function.
3585 (aarch64_expand_sve_widened_duplicate): Likewise.
3586 (aarch64_expand_sve_const_vector): Likewise.
3587 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3588 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3589 into the register, rather than emitting a SET directly.
3590 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3591 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3592 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3593 (offset_9bit_signed_scaled_p): New functions.
3594 (aarch64_replicate_bitmask_imm): New function.
3595 (aarch64_bitmask_imm): Use it.
3596 (aarch64_cannot_force_const_mem): Reject expressions involving
3597 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3598 (aarch64_classify_index): Handle SVE indices, by requiring
3599 a plain register index with a scale that matches the element size.
3600 (aarch64_classify_address): Handle SVE addresses. Assert that
3601 the mode of the address is VOIDmode or an integer mode.
3602 Update call to aarch64_classify_symbol.
3603 (aarch64_classify_symbolic_expression): Update call to
3604 aarch64_classify_symbol.
3605 (aarch64_const_vec_all_in_range_p): New function.
3606 (aarch64_print_vector_float_operand): Likewise.
3607 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3608 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3609 and the FP immediates 1.0 and 0.5.
3610 (aarch64_print_address_internal): Handle SVE addresses.
3611 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3612 (aarch64_regno_regclass): Handle predicate registers.
3613 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3615 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3616 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3617 (aarch64_convert_sve_vector_bits): New function.
3618 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3619 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3621 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3622 Handle SVE vector and predicate modes. Accept VL-based constants
3623 that need only one temporary register, and VL offsets that require
3624 no temporary registers.
3625 (aarch64_conditional_register_usage): Mark the predicate registers
3626 as fixed if SVE isn't available.
3627 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3628 Return true for SVE vector and predicate modes.
3629 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3630 rather than an unsigned int. Handle SVE modes.
3631 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3633 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3635 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3636 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3637 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3638 (aarch64_sve_float_mul_immediate_p): New functions.
3639 (aarch64_sve_valid_immediate): New function.
3640 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3641 Explicitly reject structure modes. Check for INDEX constants.
3642 Handle PTRUE and PFALSE constants.
3643 (aarch64_check_zero_based_sve_index_immediate): New function.
3644 (aarch64_simd_imm_zero_p): Delete.
3645 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3646 vector modes. Accept constants in the range of CNT[BHWD].
3647 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3648 ask for an Advanced SIMD mode.
3649 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3650 (aarch64_simd_vector_alignment): Handle SVE predicates.
3651 (aarch64_vectorize_preferred_vector_alignment): New function.
3652 (aarch64_simd_vector_alignment_reachable): Use it instead of
3654 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3655 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3657 (MAX_VECT_LEN): Delete.
3658 (expand_vec_perm_d): Add a vec_flags field.
3659 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3660 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3661 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3663 (aarch64_evpc_rev): Rename to...
3664 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3665 (aarch64_evpc_rev_global): New function.
3666 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3667 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3669 (aarch64_evpc_sve_tbl): New function.
3670 (aarch64_expand_vec_perm_const_1): Update after rename of
3671 aarch64_evpc_rev. Handle SVE permutes too, trying
3672 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3673 than aarch64_evpc_tbl.
3674 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3675 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3676 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3677 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3678 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3679 (aarch64_expand_sve_vcond): New functions.
3680 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3681 of aarch64_vector_mode_p.
3682 (aarch64_dwarf_poly_indeterminate_value): New function.
3683 (aarch64_compute_pressure_classes): Likewise.
3684 (aarch64_can_change_mode_class): Likewise.
3685 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3686 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3687 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3688 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3689 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3690 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3691 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3692 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3694 (Dn, Dl, Dr): Accept const as well as const_vector.
3695 (Dz): Likewise. Compare against CONST0_RTX.
3696 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3697 of "vector" where appropriate.
3698 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3699 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3700 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3701 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3702 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3703 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3704 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3705 (v_int_equiv): Extend to SVE modes.
3706 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3708 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3709 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3710 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3711 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3712 (SVE_COND_FP_CMP): New int iterators.
3713 (perm_hilo): Handle the new unpack unspecs.
3714 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3716 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3717 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3718 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3719 (aarch64_equality_operator, aarch64_constant_vector_operand)
3720 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3721 (aarch64_sve_nonimmediate_operand): Likewise.
3722 (aarch64_sve_general_operand): Likewise.
3723 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3724 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3725 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3726 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3727 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3728 (aarch64_sve_float_arith_immediate): Likewise.
3729 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3730 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3731 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3732 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3733 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3734 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3735 (aarch64_sve_float_arith_operand): Likewise.
3736 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3737 (aarch64_sve_float_mul_operand): Likewise.
3738 (aarch64_sve_vec_perm_operand): Likewise.
3739 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3740 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3741 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3742 as well as const_vector.
3743 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3744 in file. Use CONST0_RTX and CONSTM1_RTX.
3745 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3746 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3747 Use aarch64_simd_imm_zero.
3748 * config/aarch64/aarch64-sve.md: New file.
3749 * config/aarch64/aarch64.md: Include it.
3750 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3751 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3752 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3753 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3754 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3755 (sve): New attribute.
3756 (enabled): Disable instructions with the sve attribute unless
3758 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3759 aarch64_expand_mov_immediate.
3760 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3761 CNT[BHSD] immediates.
3762 (movti): Split CONST_POLY_INT moves into two halves.
3763 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3764 Split additions that need a temporary here if the destination
3765 is the stack pointer.
3766 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3767 (*add<mode>3_poly_1): New instruction.
3768 (set_clobber_cc): New expander.
3770 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3772 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3773 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3774 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3775 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3776 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3777 Change innermode from fixed_mode_size to machine_mode.
3778 (simplify_subreg): Update call accordingly. Handle a constant-sized
3779 subreg of a variable-length CONST_VECTOR.
3781 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3782 Alan Hayward <alan.hayward@arm.com>
3783 David Sherwood <david.sherwood@arm.com>
3785 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3786 (add_offset_to_base): New function, split out from...
3787 (create_mem_ref): ...here. When handling a scale other than 1,
3788 check first whether the address is valid without the offset.
3789 Add it into the base if so, leaving the index and scale as-is.
3791 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3794 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3795 fold_for_warn before checking if arg2 is INTEGER_CST.
3797 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3799 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3800 (store_multiple_operation): Delete.
3801 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3802 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3803 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3804 guarded by TARGET_STRING.
3805 (rs6000_output_load_multiple): Delete.
3806 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3807 OPTION_MASK_STRING / TARGET_STRING handling.
3808 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3809 (const rs6000_opt_masks) <"string">: Change mask to 0.
3810 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3811 (MASK_STRING): Delete.
3812 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3814 (load_multiple): Delete.
3821 (store_multiple): Delete.
3828 (movmemsi_8reg): Delete.
3829 (corresponding unnamed define_insn): Delete.
3830 (movmemsi_6reg): Delete.
3831 (corresponding unnamed define_insn): Delete.
3832 (movmemsi_4reg): Delete.
3833 (corresponding unnamed define_insn): Delete.
3834 (movmemsi_2reg): Delete.
3835 (corresponding unnamed define_insn): Delete.
3836 (movmemsi_1reg): Delete.
3837 (corresponding unnamed define_insn): Delete.
3838 * config/rs6000/rs6000.opt (mno-string): New.
3839 (mstring): Replace by deprecation warning stub.
3840 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3842 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3844 * regrename.c (regrename_do_replace): If replacing the same
3845 reg multiple times, try to reuse last created gen_raw_REG.
3848 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3849 main to workaround a bug in GDB.
3851 2018-01-12 Tom de Vries <tom@codesourcery.com>
3854 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3856 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3858 PR rtl-optimization/80481
3859 * ira-color.c (get_cap_member): New function.
3860 (allocnos_conflict_by_live_ranges_p): Use it.
3861 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3862 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3864 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3867 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3868 (*saddl_se_1): Ditto.
3870 (*ssubl_se_1): Ditto.
3872 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3874 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3875 rather than wi::to_widest for DR_INITs.
3876 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3877 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3878 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3880 (vect_analyze_group_access_1): Note that here.
3882 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3884 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3885 polynomial type sizes.
3887 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3889 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3890 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3891 (gimple_add_tmp_var): Likewise.
3893 2018-01-12 Martin Liska <mliska@suse.cz>
3895 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3896 (gimple_alloc_sizes): Likewise.
3897 (dump_gimple_statistics): Use PRIu64 in printf format.
3898 * gimple.h: Change uint64_t to int.
3900 2018-01-12 Martin Liska <mliska@suse.cz>
3902 * tree-core.h: Use uint64_t instead of int.
3903 * tree.c (tree_node_counts): Likewise.
3904 (tree_node_sizes): Likewise.
3905 (dump_tree_statistics): Use PRIu64 in printf format.
3907 2018-01-12 Martin Liska <mliska@suse.cz>
3909 * Makefile.in: As qsort_chk is implemented in vec.c, add
3910 vec.o to linkage of gencfn-macros.
3911 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3912 passing the info to record_node_allocation_statistics.
3913 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3915 * ggc-common.c (struct ggc_usage): Add operator== and use
3916 it in operator< and compare function.
3917 * mem-stats.h (struct mem_usage): Likewise.
3918 * vec.c (struct vec_usage): Remove operator< and compare
3919 function. Can be simply inherited.
3921 2018-01-12 Martin Jambor <mjambor@suse.cz>
3924 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3925 * tree-ssa-math-opts.c: Include domwalk.h.
3926 (convert_mult_to_fma_1): New function.
3927 (fma_transformation_info): New type.
3928 (fma_deferring_state): Likewise.
3929 (cancel_fma_deferring): New function.
3930 (result_of_phi): Likewise.
3931 (last_fma_candidate_feeds_initial_phi): Likewise.
3932 (convert_mult_to_fma): Added deferring logic, split actual
3933 transformation to convert_mult_to_fma_1.
3934 (math_opts_dom_walker): New type.
3935 (math_opts_dom_walker::after_dom_children): New method, body moved
3936 here from pass_optimize_widening_mul::execute, added deferring logic
3938 (pass_optimize_widening_mul::execute): Moved most of code to
3939 math_opts_dom_walker::after_dom_children.
3940 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3941 * config/i386/i386.c (ix86_option_override_internal): Added
3942 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3944 2018-01-12 Richard Biener <rguenther@suse.de>
3947 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3948 inline instance vars.
3950 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3953 * config/rx/rx.c (rx_is_restricted_memory_address):
3956 2018-01-12 Richard Biener <rguenther@suse.de>
3958 PR tree-optimization/80846
3959 * target.def (split_reduction): New target hook.
3960 * targhooks.c (default_split_reduction): New function.
3961 * targhooks.h (default_split_reduction): Declare.
3962 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3963 target requests first reduce vectors by combining low and high
3965 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3966 (get_vectype_for_scalar_type_and_size): Export.
3967 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3968 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3969 * doc/tm.texi: Regenerate.
3970 * config/i386/i386.c (ix86_split_reduction): Implement
3971 TARGET_VECTORIZE_SPLIT_REDUCTION.
3973 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3976 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3977 in PIC mode except for TARGET_VXWORKS_RTP.
3978 * config/sparc/sparc.c: Include cfgrtl.h.
3979 (TARGET_INIT_PIC_REG): Define.
3980 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3981 (sparc_pic_register_p): New predicate.
3982 (sparc_legitimate_address_p): Use it.
3983 (sparc_legitimize_pic_address): Likewise.
3984 (sparc_delegitimize_address): Likewise.
3985 (sparc_mode_dependent_address_p): Likewise.
3986 (gen_load_pcrel_sym): Remove 4th parameter.
3987 (load_got_register): Adjust call to above. Remove obsolete stuff.
3988 (sparc_expand_prologue): Do not call load_got_register here.
3989 (sparc_flat_expand_prologue): Likewise.
3990 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3991 (sparc_use_pseudo_pic_reg): New function.
3992 (sparc_init_pic_reg): Likewise.
3993 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3994 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3996 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3998 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3999 Add item for branch_cost.
4001 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4003 PR rtl-optimization/83565
4004 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4005 not extend the result to a larger mode for rotate operations.
4006 (num_sign_bit_copies1): Likewise.
4008 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4011 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4013 Use values-Xc.o for -pedantic.
4014 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4016 2018-01-12 Martin Liska <mliska@suse.cz>
4019 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4021 (possible_polymorphic_call_targets): Use it.
4022 (ipa_devirt): Likewise.
4024 2018-01-12 Martin Liska <mliska@suse.cz>
4026 * profile-count.h (enum profile_quality): Use 0 as invalid
4027 enum value of profile_quality.
4029 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4031 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4032 -mext-string options.
4034 2018-01-12 Richard Biener <rguenther@suse.de>
4036 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4037 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4038 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4040 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4042 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4044 * configure.ac (--with-long-double-format): Add support for the
4045 configuration option to change the default long double format on
4047 * config.gcc (powerpc*-linux*-*): Likewise.
4048 * configure: Regenerate.
4049 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4050 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4051 used without modification.
4053 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4055 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4056 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4057 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4058 MISC_BUILTIN_SPEC_BARRIER.
4059 (rs6000_init_builtins): Likewise.
4060 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4062 (speculation_barrier): New define_insn.
4063 * doc/extend.texi: Document __builtin_speculation_barrier.
4065 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4068 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4069 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4070 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4072 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4073 integral modes instead of "ss" and "sd".
4074 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4075 vectors with 32-bit and 64-bit elements.
4076 (vecdupssescalarmodesuffix): New mode attribute.
4077 (vec_dup<mode>): Use it.
4079 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4082 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4083 frame if argument is passed on stack.
4085 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4088 * ree.c (combine_reaching_defs): Optimize also
4089 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4090 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4092 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4095 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4097 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4100 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4101 after they are computed.
4103 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4105 PR tree-optimization/83695
4106 * gimple-loop-linterchange.cc
4107 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4108 reset cached scev information after interchange.
4109 (pass_linterchange::execute): Remove call to scev_reset_htab.
4111 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4113 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4114 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4115 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4116 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4117 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4118 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4119 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4120 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4121 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4122 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4123 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4124 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4125 (V_lane_reg): Likewise.
4126 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4128 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4129 (vfmal_lane_low<mode>_intrinsic,
4130 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4131 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4132 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4133 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4134 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4135 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4137 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4139 * config/arm/arm-cpus.in (fp16fml): New feature.
4140 (ALL_SIMD): Add fp16fml.
4141 (armv8.2-a): Add fp16fml as an option.
4142 (armv8.3-a): Likewise.
4143 (armv8.4-a): Add fp16fml as part of fp16.
4144 * config/arm/arm.h (TARGET_FP16FML): Define.
4145 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4147 * config/arm/arm-modes.def (V2HF): Define.
4148 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4149 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4150 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4151 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4152 vfmsl_low, vfmsl_high): New set of builtins.
4153 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4154 (vfml_op): New code attribute.
4155 (VFMLHALVES): New int iterator.
4156 (VFML, VFMLSEL): New mode attributes.
4157 (V_reg): Define mapping for V2HF.
4158 (V_hi, V_lo): New mode attributes.
4159 (VF_constraint): Likewise.
4160 (vfml_half, vfml_half_selector): New int attributes.
4161 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4163 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4164 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4166 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4167 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4168 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4169 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4171 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4172 Document new effective target and option set.
4174 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4176 * config/arm/arm-cpus.in (armv8_4): New feature.
4177 (ARMv8_4a): New fgroup.
4178 (armv8.4-a): New arch.
4179 * config/arm/arm-tables.opt: Regenerate.
4180 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4181 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4182 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4183 Add matching rules for -march=armv8.4-a and extensions.
4184 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4186 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4189 * config/rx/rx.md (BW): New mode attribute.
4190 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4192 2018-01-11 Richard Biener <rguenther@suse.de>
4194 PR tree-optimization/83435
4195 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4196 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4197 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4199 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4200 Alan Hayward <alan.hayward@arm.com>
4201 David Sherwood <david.sherwood@arm.com>
4203 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4205 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4206 (aarch64_print_address_internal): Use it to check for a zero offset.
4208 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4209 Alan Hayward <alan.hayward@arm.com>
4210 David Sherwood <david.sherwood@arm.com>
4212 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4213 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4214 Return a poly_int64 rather than a HOST_WIDE_INT.
4215 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4216 rather than a HOST_WIDE_INT.
4217 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4218 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4219 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4220 final_offset from HOST_WIDE_INT to poly_int64.
4221 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4222 to_constant when getting the number of units in an Advanced SIMD
4224 (aarch64_builtin_vectorized_function): Check for a constant number
4226 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4228 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4229 attribute instead of GET_MODE_NUNITS.
4230 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4231 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4232 GET_MODE_SIZE for fixed-size registers.
4233 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4234 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4235 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4236 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4237 (aarch64_print_operand, aarch64_print_address_internal)
4238 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4239 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4240 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4241 Handle polynomial GET_MODE_SIZE.
4242 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4243 wider than SImode without modification.
4244 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4245 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4246 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4247 passing and returning SVE modes.
4248 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4249 rather than GEN_INT.
4250 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4251 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4252 (aarch64_allocate_and_probe_stack_space): Likewise.
4253 (aarch64_layout_frame): Cope with polynomial offsets.
4254 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4255 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4257 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4258 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4259 poly_int64 rather than a HOST_WIDE_INT.
4260 (aarch64_get_separate_components, aarch64_process_components)
4261 (aarch64_expand_prologue, aarch64_expand_epilogue)
4262 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4263 (aarch64_anchor_offset): New function, split out from...
4264 (aarch64_legitimize_address): ...here.
4265 (aarch64_builtin_vectorization_cost): Handle polynomial
4266 TYPE_VECTOR_SUBPARTS.
4267 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4269 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4270 number of elements from the PARALLEL rather than the mode.
4271 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4272 rather than GET_MODE_BITSIZE.
4273 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4274 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4275 (aarch64_expand_vec_perm_const_1): Handle polynomial
4276 d->perm.length () and d->perm elements.
4277 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4278 Apply to_constant to d->perm elements.
4279 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4280 polynomial CONST_VECTOR_NUNITS.
4281 (aarch64_move_pointer): Take amount as a poly_int64 rather
4283 (aarch64_progress_pointer): Avoid temporary variable.
4284 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4285 the mode attribute instead of GET_MODE.
4287 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4288 Alan Hayward <alan.hayward@arm.com>
4289 David Sherwood <david.sherwood@arm.com>
4291 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4292 x exists before using it.
4293 (aarch64_add_constant_internal): Rename to...
4294 (aarch64_add_offset_1): ...this. Replace regnum with separate
4295 src and dest rtxes. Handle the case in which they're different,
4296 including when the offset is zero. Replace scratchreg with an rtx.
4297 Use 2 additions if there is no spare register into which we can
4298 move a 16-bit constant.
4299 (aarch64_add_constant): Delete.
4300 (aarch64_add_offset): Replace reg with separate src and dest
4301 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4302 Use aarch64_add_offset_1.
4303 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4304 an rtx rather than an int. Take the delta as a poly_int64
4305 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4306 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4307 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4308 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4309 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4311 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4312 aarch64_add_constant.
4314 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4316 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4317 Use scalar_float_mode.
4319 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4321 * config/aarch64/aarch64-simd.md
4322 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4323 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4324 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4325 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4326 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4327 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4328 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4329 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4330 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4331 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4333 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4336 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4337 targ_options->x_arm_arch_string is non NULL.
4339 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4341 * config/aarch64/aarch64.h
4342 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4344 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4347 * expmed.c (emit_store_flag_force): Swap if const op0
4348 and change VOIDmode to mode of op0.
4350 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4352 PR rtl-optimization/83761
4353 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4354 than bytes to mode_for_size.
4356 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4359 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4360 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4363 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4366 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4367 when in layout mode.
4368 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4369 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4372 2018-01-10 Michael Collison <michael.collison@arm.com>
4374 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4375 * config/aarch64/aarch64-option-extension.def: Add
4376 AARCH64_OPT_EXTENSION of 'fp16fml'.
4377 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4378 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4379 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4380 * config/aarch64/constraints.md (Ui7): New constraint.
4381 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4382 (VFMLA_SEL_W): Ditto.
4385 (VFMLA16_LOW): New int iterator.
4386 (VFMLA16_HIGH): Ditto.
4387 (UNSPEC_FMLAL): New unspec.
4388 (UNSPEC_FMLSL): Ditto.
4389 (UNSPEC_FMLAL2): Ditto.
4390 (UNSPEC_FMLSL2): Ditto.
4391 (f16mac): New code attribute.
4392 * config/aarch64/aarch64-simd-builtins.def
4393 (aarch64_fmlal_lowv2sf): Ditto.
4394 (aarch64_fmlsl_lowv2sf): Ditto.
4395 (aarch64_fmlalq_lowv4sf): Ditto.
4396 (aarch64_fmlslq_lowv4sf): Ditto.
4397 (aarch64_fmlal_highv2sf): Ditto.
4398 (aarch64_fmlsl_highv2sf): Ditto.
4399 (aarch64_fmlalq_highv4sf): Ditto.
4400 (aarch64_fmlslq_highv4sf): Ditto.
4401 (aarch64_fmlal_lane_lowv2sf): Ditto.
4402 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4403 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4404 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4405 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4406 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4407 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4408 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4409 (aarch64_fmlal_lane_highv2sf): Ditto.
4410 (aarch64_fmlsl_lane_highv2sf): Ditto.
4411 (aarch64_fmlal_laneq_highv2sf): Ditto.
4412 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4413 (aarch64_fmlalq_lane_highv4sf): Ditto.
4414 (aarch64_fmlsl_lane_highv4sf): Ditto.
4415 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4416 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4417 * config/aarch64/aarch64-simd.md:
4418 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4419 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4420 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4421 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4422 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4423 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4424 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4425 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4426 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4427 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4428 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4429 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4430 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4431 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4432 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4433 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4434 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4435 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4436 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4437 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4438 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4439 (vfmlsl_low_u32): Ditto.
4440 (vfmlalq_low_u32): Ditto.
4441 (vfmlslq_low_u32): Ditto.
4442 (vfmlal_high_u32): Ditto.
4443 (vfmlsl_high_u32): Ditto.
4444 (vfmlalq_high_u32): Ditto.
4445 (vfmlslq_high_u32): Ditto.
4446 (vfmlal_lane_low_u32): Ditto.
4447 (vfmlsl_lane_low_u32): Ditto.
4448 (vfmlal_laneq_low_u32): Ditto.
4449 (vfmlsl_laneq_low_u32): Ditto.
4450 (vfmlalq_lane_low_u32): Ditto.
4451 (vfmlslq_lane_low_u32): Ditto.
4452 (vfmlalq_laneq_low_u32): Ditto.
4453 (vfmlslq_laneq_low_u32): Ditto.
4454 (vfmlal_lane_high_u32): Ditto.
4455 (vfmlsl_lane_high_u32): Ditto.
4456 (vfmlal_laneq_high_u32): Ditto.
4457 (vfmlsl_laneq_high_u32): Ditto.
4458 (vfmlalq_lane_high_u32): Ditto.
4459 (vfmlslq_lane_high_u32): Ditto.
4460 (vfmlalq_laneq_high_u32): Ditto.
4461 (vfmlslq_laneq_high_u32): Ditto.
4462 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4463 (AARCH64_FL_FOR_ARCH8_4): New.
4464 (AARCH64_ISA_F16FML): New ISA flag.
4465 (TARGET_F16FML): New feature flag for fp16fml.
4466 (doc/invoke.texi): Document new fp16fml option.
4468 2018-01-10 Michael Collison <michael.collison@arm.com>
4470 * config/aarch64/aarch64-builtins.c:
4471 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4472 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4473 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4474 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4475 (AARCH64_ISA_SHA3): New ISA flag.
4476 (TARGET_SHA3): New feature flag for sha3.
4477 * config/aarch64/iterators.md (sha512_op): New int attribute.
4478 (CRYPTO_SHA512): New int iterator.
4479 (UNSPEC_SHA512H): New unspec.
4480 (UNSPEC_SHA512H2): Ditto.
4481 (UNSPEC_SHA512SU0): Ditto.
4482 (UNSPEC_SHA512SU1): Ditto.
4483 * config/aarch64/aarch64-simd-builtins.def
4484 (aarch64_crypto_sha512hqv2di): New builtin.
4485 (aarch64_crypto_sha512h2qv2di): Ditto.
4486 (aarch64_crypto_sha512su0qv2di): Ditto.
4487 (aarch64_crypto_sha512su1qv2di): Ditto.
4488 (aarch64_eor3qv8hi): Ditto.
4489 (aarch64_rax1qv2di): Ditto.
4490 (aarch64_xarqv2di): Ditto.
4491 (aarch64_bcaxqv8hi): Ditto.
4492 * config/aarch64/aarch64-simd.md:
4493 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4494 (aarch64_crypto_sha512su0qv2di): Ditto.
4495 (aarch64_crypto_sha512su1qv2di): Ditto.
4496 (aarch64_eor3qv8hi): Ditto.
4497 (aarch64_rax1qv2di): Ditto.
4498 (aarch64_xarqv2di): Ditto.
4499 (aarch64_bcaxqv8hi): Ditto.
4500 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4501 (vsha512h2q_u64): Ditto.
4502 (vsha512su0q_u64): Ditto.
4503 (vsha512su1q_u64): Ditto.
4504 (veor3q_u16): Ditto.
4505 (vrax1q_u64): Ditto.
4507 (vbcaxq_u16): Ditto.
4508 * config/arm/types.md (crypto_sha512): New type attribute.
4509 (crypto_sha3): Ditto.
4510 (doc/invoke.texi): Document new sha3 option.
4512 2018-01-10 Michael Collison <michael.collison@arm.com>
4514 * config/aarch64/aarch64-builtins.c:
4515 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4516 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4517 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4518 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4519 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4520 (AARCH64_ISA_SM4): New ISA flag.
4521 (TARGET_SM4): New feature flag for sm4.
4522 * config/aarch64/aarch64-simd-builtins.def
4523 (aarch64_sm3ss1qv4si): Ditto.
4524 (aarch64_sm3tt1aq4si): Ditto.
4525 (aarch64_sm3tt1bq4si): Ditto.
4526 (aarch64_sm3tt2aq4si): Ditto.
4527 (aarch64_sm3tt2bq4si): Ditto.
4528 (aarch64_sm3partw1qv4si): Ditto.
4529 (aarch64_sm3partw2qv4si): Ditto.
4530 (aarch64_sm4eqv4si): Ditto.
4531 (aarch64_sm4ekeyqv4si): Ditto.
4532 * config/aarch64/aarch64-simd.md:
4533 (aarch64_sm3ss1qv4si): Ditto.
4534 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4535 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4536 (aarch64_sm4eqv4si): Ditto.
4537 (aarch64_sm4ekeyqv4si): Ditto.
4538 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4539 (sm3part_op): Ditto.
4540 (CRYPTO_SM3TT): Ditto.
4541 (CRYPTO_SM3PART): Ditto.
4542 (UNSPEC_SM3SS1): New unspec.
4543 (UNSPEC_SM3TT1A): Ditto.
4544 (UNSPEC_SM3TT1B): Ditto.
4545 (UNSPEC_SM3TT2A): Ditto.
4546 (UNSPEC_SM3TT2B): Ditto.
4547 (UNSPEC_SM3PARTW1): Ditto.
4548 (UNSPEC_SM3PARTW2): Ditto.
4549 (UNSPEC_SM4E): Ditto.
4550 (UNSPEC_SM4EKEY): Ditto.
4551 * config/aarch64/constraints.md (Ui2): New constraint.
4552 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4553 * config/arm/types.md (crypto_sm3): New type attribute.
4554 (crypto_sm4): Ditto.
4555 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4556 (vsm3tt1aq_u32): Ditto.
4557 (vsm3tt1bq_u32): Ditto.
4558 (vsm3tt2aq_u32): Ditto.
4559 (vsm3tt2bq_u32): Ditto.
4560 (vsm3partw1q_u32): Ditto.
4561 (vsm3partw2q_u32): Ditto.
4562 (vsm4eq_u32): Ditto.
4563 (vsm4ekeyq_u32): Ditto.
4564 (doc/invoke.texi): Document new sm4 option.
4566 2018-01-10 Michael Collison <michael.collison@arm.com>
4568 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4569 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4570 (AARCH64_FL_FOR_ARCH8_4): New.
4571 (AARCH64_FL_V8_4): New flag.
4572 (doc/invoke.texi): Document new armv8.4-a option.
4574 2018-01-10 Michael Collison <michael.collison@arm.com>
4576 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4577 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4578 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4579 * config/aarch64/aarch64-option-extension.def: Add
4580 AARCH64_OPT_EXTENSION of 'sha2'.
4581 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4582 (crypto): Disable sha2 and aes if crypto disabled.
4583 (crypto): Enable aes and sha2 if enabled.
4584 (simd): Disable sha2 and aes if simd disabled.
4585 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4587 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4588 (TARGET_SHA2): New feature flag for sha2.
4589 (TARGET_AES): New feature flag for aes.
4590 * config/aarch64/aarch64-simd.md:
4591 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4592 conditional on TARGET_AES.
4593 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4594 (aarch64_crypto_sha1hsi): Make pattern conditional
4596 (aarch64_crypto_sha1hv4si): Ditto.
4597 (aarch64_be_crypto_sha1hv4si): Ditto.
4598 (aarch64_crypto_sha1su1v4si): Ditto.
4599 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4600 (aarch64_crypto_sha1su0v4si): Ditto.
4601 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4602 (aarch64_crypto_sha256su0v4si): Ditto.
4603 (aarch64_crypto_sha256su1v4si): Ditto.
4604 (doc/invoke.texi): Document new aes and sha2 options.
4606 2018-01-10 Martin Sebor <msebor@redhat.com>
4608 PR tree-optimization/83781
4609 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4612 2018-01-11 Martin Sebor <msebor@gmail.com>
4613 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4615 PR tree-optimization/83501
4616 PR tree-optimization/81703
4618 * tree-ssa-strlen.c (get_string_cst): Rename...
4619 (get_string_len): ...to this. Handle global constants.
4620 (handle_char_store): Adjust.
4622 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4623 Jim Wilson <jimw@sifive.com>
4625 * config/riscv/riscv-protos.h (riscv_output_return): New.
4626 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4627 (riscv_attribute_table, riscv_output_return),
4628 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4629 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4630 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4632 (riscv_expand_prologue): Add early return for naked function.
4633 (riscv_expand_epilogue): Likewise.
4634 (riscv_function_ok_for_sibcall): Return false for naked function.
4635 (riscv_set_current_function): New.
4636 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4637 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4638 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4639 * doc/extend.texi (RISC-V Function Attributes): New.
4641 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4643 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4644 check for 128-bit long double before checking TCmode.
4645 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4646 128-bit long doubles before checking TFmode or TCmode.
4647 (FLOAT128_IBM_P): Likewise.
4649 2018-01-10 Martin Sebor <msebor@redhat.com>
4651 PR tree-optimization/83671
4652 * builtins.c (c_strlen): Unconditionally return zero for the empty
4654 Use -Warray-bounds for warnings.
4655 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4656 for non-constant array indices with COMPONENT_REF, arrays of
4657 arrays, and pointers to arrays.
4658 (gimple_fold_builtin_strlen): Determine and set length range for
4659 non-constant character arrays.
4661 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4664 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4667 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4669 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4671 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4674 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4675 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4676 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4677 indexed_or_indirect_operand predicate.
4678 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4679 (*vsx_le_perm_load_v8hi): Likewise.
4680 (*vsx_le_perm_load_v16qi): Likewise.
4681 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4682 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4683 (*vsx_le_perm_store_v8hi): Likewise.
4684 (*vsx_le_perm_store_v16qi): Likewise.
4685 (eight unnamed splitters): Likewise.
4687 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4689 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4690 * config/rs6000/emmintrin.h: Likewise.
4691 * config/rs6000/mmintrin.h: Likewise.
4692 * config/rs6000/xmmintrin.h: Likewise.
4694 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4697 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4699 * tree.c (tree_nop_conversion): Return true for location wrapper
4701 (maybe_wrap_with_location): New function.
4702 (selftest::check_strip_nops): New function.
4703 (selftest::test_location_wrappers): New function.
4704 (selftest::tree_c_tests): Call it.
4705 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4706 (maybe_wrap_with_location): New decl.
4707 (EXPR_LOCATION_WRAPPER_P): New macro.
4708 (location_wrapper_p): New inline function.
4709 (tree_strip_any_location_wrapper): New inline function.
4711 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4714 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4715 stack_realign_offset for the largest alignment of stack slot
4717 (ix86_find_max_used_stack_alignment): New function.
4718 (ix86_finalize_stack_frame_flags): Use it. Set
4719 max_used_stack_alignment if we don't realign stack.
4720 * config/i386/i386.h (machine_function): Add
4721 max_used_stack_alignment.
4723 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4725 * config/arm/arm.opt (-mbranch-cost): New option.
4726 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4729 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4732 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4733 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4735 2018-01-10 Richard Biener <rguenther@suse.de>
4738 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4739 early out so it also covers the case where we have a non-NULL
4742 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4744 PR tree-optimization/83753
4745 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4746 for non-strided grouped accesses if the number of elements is 1.
4748 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4751 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4752 * i386.h (TARGET_USE_GATHER): Define.
4753 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4755 2018-01-10 Martin Liska <mliska@suse.cz>
4758 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4759 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4761 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4762 CLEANUP_NO_PARTITIONING is not set.
4764 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4766 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4767 for vectors, as a partial revert of r254296.
4768 * rtl.h (const_vec_p): Delete.
4769 (const_vec_duplicate_p): Don't test for vector CONSTs.
4770 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4771 * expmed.c (make_tree): Likewise.
4774 * common.md (E, F): Use CONSTANT_P instead of checking for
4776 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4777 checking for CONST_VECTOR.
4779 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4782 * predict.c (force_edge_cold): Handle in more sane way edges
4785 2018-01-09 Carl Love <cel@us.ibm.com>
4787 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4789 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4790 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4791 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4792 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4793 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4794 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4795 * config/rs6000/rs6000-protos.h: Add extern defition for
4796 rs6000_generate_float2_double_code.
4797 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4799 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4800 (float2_v2df): Add define_expand.
4802 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4805 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4806 op_mode in the force_to_mode call.
4808 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4810 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4811 instead of checking each element individually.
4812 (aarch64_evpc_uzp): Likewise.
4813 (aarch64_evpc_zip): Likewise.
4814 (aarch64_evpc_ext): Likewise.
4815 (aarch64_evpc_rev): Likewise.
4816 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4817 instead of checking each element individually. Return true without
4819 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4820 whether all selected elements come from the same input, instead of
4821 checking each element individually. Remove calls to gen_rtx_REG,
4822 start_sequence and end_sequence and instead assert that no rtl is
4825 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4827 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4828 order of HIGH and CONST checks.
4830 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4832 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4833 if the destination isn't an SSA_NAME.
4835 2018-01-09 Richard Biener <rguenther@suse.de>
4837 PR tree-optimization/83668
4838 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4840 (canonicalize_loop_form): ... here, renamed from ...
4841 (canonicalize_loop_closed_ssa_form): ... this and amended to
4842 swap successor edges for loop exit blocks to make us use
4843 the RPO order we need for initial schedule generation.
4845 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4847 PR tree-optimization/64811
4848 * match.pd: When optimizing comparisons with Inf, avoid
4849 introducing or losing exceptions from comparisons with NaN.
4851 2018-01-09 Martin Liska <mliska@suse.cz>
4854 * asan.c (shadow_mem_size): Add gcc_assert.
4856 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4858 Don't save registers in main().
4861 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4862 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4863 * config/avr/avr.c (avr_set_current_function): Don't error if
4864 naked, OS_task or OS_main are specified at the same time.
4865 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4867 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4869 * common/config/avr/avr-common.c (avr_option_optimization_table):
4870 Switch on -mmain-is-OS_task for optimizing compilations.
4872 2018-01-09 Richard Biener <rguenther@suse.de>
4874 PR tree-optimization/83572
4875 * graphite.c: Include cfganal.h.
4876 (graphite_transform_loops): Connect infinite loops to exit
4877 and remove fake edges at the end.
4879 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4881 * ipa-inline.c (edge_badness): Revert accidental checkin.
4883 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4886 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4887 symbols; not inline clones.
4889 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4892 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4893 hard registers. Formatting fixes.
4895 PR preprocessor/83722
4896 * gcc.c (try_generate_repro): Pass
4897 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4898 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4901 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4902 Kito Cheng <kito.cheng@gmail.com>
4904 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4905 (riscv_leaf_function_p): Delete.
4906 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4908 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4910 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4912 (do_ifelse): New function.
4913 (do_isel): New function.
4914 (do_sub3): New function.
4915 (do_add3): New function.
4916 (do_load_mask_compare): New function.
4917 (do_overlap_load_compare): New function.
4918 (expand_compare_loop): New function.
4919 (expand_block_compare): Call expand_compare_loop() when appropriate.
4920 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4922 (-mblock-compare-inline-loop-limit): New option.
4924 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4927 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4928 Reverse order of second and third operands in first alternative.
4929 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4930 of first and second elements in UNSPEC_VPERMR vector.
4931 (altivec_expand_vec_perm_le): Likewise.
4933 2018-01-08 Jeff Law <law@redhat.com>
4935 PR rtl-optimizatin/81308
4936 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4937 (process_switch): If group_case_labels makes a change, then set
4939 (pass_convert_switch::execute): If a switch is converted, then
4940 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4942 PR rtl-optimization/81308
4943 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4946 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4948 PR target/83663 - Revert r255946
4949 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4950 generation for cases where splatting a value is not useful.
4951 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4952 across a vec_duplicate and a paradoxical subreg forming a vector
4953 mode to a vec_concat.
4955 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4957 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4958 -march=armv8.3-a variants.
4959 * config/arm/t-multilib: Likewise.
4960 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4962 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4964 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4966 (cceq_ior_compare_complement): Give it a name so I can use it, and
4967 change boolean_or_operator predicate to boolean_operator so it can
4968 be used to generate a crand.
4969 (eqne): New code iterator.
4970 (bd/bd_neg): New code_attrs.
4971 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4972 a single define_insn.
4973 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4974 decrement (bdnzt/bdnzf/bdzt/bdzf).
4975 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4976 with the new names of the branch decrement patterns, and added the
4977 names of the branch decrement conditional patterns.
4979 2018-01-08 Richard Biener <rguenther@suse.de>
4981 PR tree-optimization/83563
4982 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4985 2018-01-08 Richard Biener <rguenther@suse.de>
4988 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4990 2018-01-08 Richard Biener <rguenther@suse.de>
4992 PR tree-optimization/83685
4993 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4994 references to abnormals.
4996 2018-01-08 Richard Biener <rguenther@suse.de>
4999 * dwarf2out.c (output_indirect_strings): Handle empty
5000 skeleton_debug_str_hash.
5001 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5003 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5005 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5006 (emit_store_direct): Likewise.
5007 (arc_trampoline_adjust_address): Likewise.
5008 (arc_asm_trampoline_template): New function.
5009 (arc_initialize_trampoline): Use asm_trampoline_template.
5010 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5011 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5012 * config/arc/arc.md (flush_icache): Delete pattern.
5014 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5016 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5017 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5020 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5023 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5024 by not USED_FOR_TARGET.
5025 (make_pass_resolve_sw_modes): Likewise.
5027 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5029 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5032 2018-01-08 Richard Biener <rguenther@suse.de>
5035 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5037 2018-01-08 Richard Biener <rguenther@suse.de>
5040 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5042 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5045 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5046 basic blocks with a small number of successors.
5047 (convert_control_dep_chain_into_preds): Improve handling of
5049 (dump_predicates): Split apart into...
5050 (dump_pred_chain): ...here...
5051 (dump_pred_info): ...and here.
5052 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5053 (can_chain_union_be_invalidated_p): Improve check for invalidation
5055 (uninit_uses_cannot_happen): Avoid unnecessary if
5056 convert_control_dep_chain_into_preds yielded nothing.
5058 2018-01-06 Martin Sebor <msebor@redhat.com>
5060 PR tree-optimization/83640
5061 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5062 subtracting negative offset from size.
5063 (builtin_access::overlap): Adjust offset bounds of the access to fall
5064 within the size of the object if possible.
5066 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5068 PR rtl-optimization/83699
5069 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5070 extract_bit_field_as_subreg to cases in which the extracted
5071 value is also a vector.
5073 * lra-constraints.c (process_alt_operands): Test for the equivalence
5074 substitutions when detecting a possible reload cycle.
5076 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5079 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5080 by default if flag_selective_schedling{,2}. Formatting fixes.
5082 PR rtl-optimization/83682
5083 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5084 if it has non-VECTOR_MODE element mode.
5085 (vec_duplicate_p): Likewise.
5088 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5089 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5091 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5094 * config/i386/i386-builtin.def
5095 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5096 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5097 Require also OPTION_MASK_ISA_AVX512F in addition to
5098 OPTION_MASK_ISA_GFNI.
5099 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5100 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5101 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5102 to OPTION_MASK_ISA_GFNI.
5103 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5104 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5105 OPTION_MASK_ISA_AVX512BW.
5106 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5107 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5108 addition to OPTION_MASK_ISA_GFNI.
5109 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5110 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5111 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5112 to OPTION_MASK_ISA_GFNI.
5113 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5114 a requirement for all ISAs rather than any of them with a few
5116 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5118 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5119 bitmasks to be enabled with 3 exceptions, instead of requiring any
5120 enabled ISA with lots of exceptions.
5121 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5122 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5123 Change avx512bw in isa attribute to avx512f.
5124 * config/i386/sgxintrin.h: Add license boilerplate.
5125 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5126 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5127 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5128 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5130 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5131 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5132 temporarily sse2 rather than sse if not enabled already.
5135 * config/i386/sse.md (VI248_VLBW): Rename to ...
5136 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5137 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5138 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5139 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5140 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5141 mode iterator instead of VI248_VLBW.
5143 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5145 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5146 (record_modified): Skip clobbers; add debug output.
5147 (param_change_prob): Use sreal frequencies.
5149 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5151 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5152 punt for user-aligned variables.
5154 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5156 * tree-chrec.c (chrec_contains_symbols): Return true for
5159 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5162 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5163 of (x|y) == x for BICS pattern.
5165 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5167 PR tree-optimization/83605
5168 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5169 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5172 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5174 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5175 * config/epiphany/rtems.h: New file.
5177 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5178 Uros Bizjak <ubizjak@gmail.com>
5181 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5182 QIreg_operand instead of register_operand predicate.
5183 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5184 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5185 comments instead of -fmitigate[-_]rop.
5187 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5190 * cgraphunit.c (symbol_table::compile): Switch to text_section
5191 before calling assembly_start debug hook.
5192 * run-rtl-passes.c (run_rtl_passes): Likewise.
5195 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5197 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5198 range_int_cst_p rather than !symbolic_range_p before calling
5199 extract_range_from_multiplicative_op_1.
5201 2018-01-04 Jeff Law <law@redhat.com>
5203 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5204 redundant test in assertion.
5206 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5208 * doc/rtl.texi: Document machine_mode wrapper classes.
5210 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5212 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5215 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5217 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5218 the VEC_PERM_EXPR fold to fail.
5220 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5223 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5224 to switched_sections.
5226 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5229 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5232 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5235 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5236 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5238 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5241 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5242 is BLKmode and bitpos not zero or mode change is needed.
5244 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5247 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5250 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5253 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5254 instead of MULT rtx. Update all corresponding splitters.
5256 (*ssub<modesuffix>): Ditto.
5258 (*cmp_sadd_di): Update split patterns.
5259 (*cmp_sadd_si): Ditto.
5260 (*cmp_sadd_sidi): Ditto.
5261 (*cmp_ssub_di): Ditto.
5262 (*cmp_ssub_si): Ditto.
5263 (*cmp_ssub_sidi): Ditto.
5264 * config/alpha/predicates.md (const23_operand): New predicate.
5265 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5266 Look for ASHIFT, not MULT inner operand.
5267 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5269 2018-01-04 Martin Liska <mliska@suse.cz>
5271 PR gcov-profile/83669
5272 * gcov.c (output_intermediate_file): Add version to intermediate
5274 * doc/gcov.texi: Document new field 'version' in intermediate
5275 file format. Fix location of '-k' option of gcov command.
5277 2018-01-04 Martin Liska <mliska@suse.cz>
5280 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5282 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5284 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5286 2018-01-03 Martin Sebor <msebor@redhat.com>
5288 PR tree-optimization/83655
5289 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5290 checking calls with invalid arguments.
5292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5294 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5295 (vectorizable_mask_load_store): Delete.
5296 (vectorizable_call): Return false for masked loads and stores.
5297 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5298 instead of gimple_assign_rhs1.
5299 (vectorizable_load): Handle IFN_MASK_LOAD.
5300 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5302 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5304 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5306 (vectorizable_mask_load_store): ...here.
5307 (vectorizable_load): ...and here.
5309 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5311 * tree-vect-stmts.c (vect_build_all_ones_mask)
5312 (vect_build_zero_merge_argument): New functions, split out from...
5313 (vectorizable_load): ...here.
5315 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5317 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5319 (vectorizable_mask_load_store): ...here.
5320 (vectorizable_store): ...and here.
5322 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5324 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5326 (vectorizable_mask_load_store): ...here.
5328 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5330 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5331 (vect_model_store_cost): Take a vec_load_store_type instead of a
5333 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5334 (vect_model_store_cost): Take a vec_load_store_type instead of a
5336 (vectorizable_mask_load_store): Update accordingly.
5337 (vectorizable_store): Likewise.
5338 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5340 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5342 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5343 IFN_MASK_LOAD calls here rather than...
5344 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5347 Alan Hayward <alan.hayward@arm.com>
5348 David Sherwood <david.sherwood@arm.com>
5350 * expmed.c (extract_bit_field_1): For vector extracts,
5351 fall back to extract_bit_field_as_subreg if vec_extract
5354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5355 Alan Hayward <alan.hayward@arm.com>
5356 David Sherwood <david.sherwood@arm.com>
5358 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5359 they are variable or constant sized.
5360 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5361 slots for constant-sized data.
5363 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5364 Alan Hayward <alan.hayward@arm.com>
5365 David Sherwood <david.sherwood@arm.com>
5367 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5368 handling COND_EXPRs with boolean comparisons, try to find a better
5369 basis for the mask type than the boolean itself.
5371 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5373 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5374 is calculated and how it can be overridden.
5375 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5376 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5378 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5382 Alan Hayward <alan.hayward@arm.com>
5383 David Sherwood <david.sherwood@arm.com>
5385 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5386 Remove the mode argument.
5387 (aarch64_simd_valid_immediate): Remove the mode and inverse
5389 * config/aarch64/iterators.md (bitsize): New iterator.
5390 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5391 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5392 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5393 aarch64_simd_valid_immediate.
5394 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5395 (aarch64_reg_or_bic_imm): Likewise.
5396 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5397 with an insn_type enum and msl with a modifier_type enum.
5398 Replace element_width with a scalar_mode. Change the shift
5399 to unsigned int. Add constructors for scalar_float_mode and
5400 scalar_int_mode elements.
5401 (aarch64_vect_float_const_representable_p): Delete.
5402 (aarch64_can_const_movi_rtx_p)
5403 (aarch64_simd_scalar_immediate_valid_for_move)
5404 (aarch64_simd_make_constant): Update call to
5405 aarch64_simd_valid_immediate.
5406 (aarch64_advsimd_valid_immediate_hs): New function.
5407 (aarch64_advsimd_valid_immediate): Likewise.
5408 (aarch64_simd_valid_immediate): Remove mode and inverse
5409 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5410 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5411 and aarch64_float_const_representable_p on the result.
5412 (aarch64_output_simd_mov_immediate): Remove mode argument.
5413 Update call to aarch64_simd_valid_immediate and use of
5414 simd_immediate_info.
5415 (aarch64_output_scalar_simd_mov_immediate): Update call
5418 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5419 Alan Hayward <alan.hayward@arm.com>
5420 David Sherwood <david.sherwood@arm.com>
5422 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5423 (mode_nunits): Likewise CONST_MODE_NUNITS.
5424 * machmode.def (ADJUST_NUNITS): Document.
5425 * genmodes.c (mode_data::need_nunits_adj): New field.
5426 (blank_mode): Update accordingly.
5427 (adj_nunits): New variable.
5428 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5430 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5431 listed in adj_nunits.
5432 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5433 listed in adj_nunits. Don't emit case statements for such modes.
5434 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5435 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5436 nothing if adj_nunits is nonnull.
5437 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5438 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5439 (emit_mode_fbit): Update use of print_maybe_const_decl.
5440 (emit_move_size): Likewise. Treat the array as non-const
5442 (emit_mode_adjustments): Handle adj_nunits.
5444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5446 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5447 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5448 (VECTOR_MODES): Use it.
5449 (make_vector_modes): Take the prefix as an argument.
5451 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5452 Alan Hayward <alan.hayward@arm.com>
5453 David Sherwood <david.sherwood@arm.com>
5455 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5456 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5457 for MODE_VECTOR_BOOL.
5458 * machmode.def (VECTOR_BOOL_MODE): Document.
5459 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5460 (make_vector_bool_mode): New function.
5461 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5463 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5464 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5466 * stor-layout.c (int_mode_for_mode): Likewise.
5467 * tree.c (build_vector_type_for_mode): Likewise.
5468 * varasm.c (output_constant_pool_2): Likewise.
5469 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5470 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5471 for MODE_VECTOR_BOOL.
5472 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5473 of mode class checks.
5474 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5475 instead of a list of mode class checks.
5476 (expand_vector_scalar_condition): Likewise.
5477 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5479 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5480 Alan Hayward <alan.hayward@arm.com>
5481 David Sherwood <david.sherwood@arm.com>
5483 * machmode.h (mode_size): Change from unsigned short to
5485 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5486 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5487 or if measurement_type is not polynomial.
5488 (fixed_size_mode::includes_p): Check for constant-sized modes.
5489 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5490 return a poly_uint16 rather than an unsigned short.
5491 (emit_mode_size): Change the type of mode_size from unsigned short
5492 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5493 (emit_mode_adjustments): Cope with polynomial vector sizes.
5494 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5496 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5498 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5499 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5500 * caller-save.c (setup_save_areas): Likewise.
5501 (replace_reg_with_saved_mem): Likewise.
5502 * calls.c (emit_library_call_value_1): Likewise.
5503 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5504 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5505 (gen_lowpart_for_combine): Likewise.
5506 * convert.c (convert_to_integer_1): Likewise.
5507 * cse.c (equiv_constant, cse_insn): Likewise.
5508 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5509 (cselib_subst_to_values): Likewise.
5510 * dce.c (word_dce_process_block): Likewise.
5511 * df-problems.c (df_word_lr_mark_ref): Likewise.
5512 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5513 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5514 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5515 (rtl_for_decl_location): Likewise.
5516 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5517 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5518 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5519 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5520 (expand_expr_real_1): Likewise.
5521 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5522 (pad_below): Likewise.
5523 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5524 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5525 * ira.c (get_subreg_tracking_sizes): Likewise.
5526 * ira-build.c (ira_create_allocno_objects): Likewise.
5527 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5528 (ira_sort_regnos_for_alter_reg): Likewise.
5529 * ira-costs.c (record_operand_costs): Likewise.
5530 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5531 (resolve_simple_move): Likewise.
5532 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5533 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5534 (lra_constraints): Likewise.
5535 (CONST_POOL_OK_P): Reject variable-sized modes.
5536 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5537 (add_pseudo_to_slot, lra_spill): Likewise.
5538 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5539 * optabs-query.c (get_best_extraction_insn): Likewise.
5540 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5541 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5542 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5543 * recog.c (offsettable_address_addr_space_p): Likewise.
5544 * regcprop.c (maybe_mode_change): Likewise.
5545 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5546 * regrename.c (build_def_use): Likewise.
5547 * regstat.c (dump_reg_info): Likewise.
5548 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5549 (find_reloads, find_reloads_subreg_address): Likewise.
5550 * reload1.c (eliminate_regs_1): Likewise.
5551 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5552 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5553 (simplify_binary_operation_1, simplify_subreg): Likewise.
5554 * targhooks.c (default_function_arg_padding): Likewise.
5555 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5556 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5557 (verify_gimple_assign_ternary): Likewise.
5558 * tree-inline.c (estimate_move_cost): Likewise.
5559 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5560 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5561 (get_address_cost_ainc): Likewise.
5562 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5563 (vect_supportable_dr_alignment): Likewise.
5564 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5565 (vectorizable_reduction): Likewise.
5566 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5567 (vectorizable_operation, vectorizable_load): Likewise.
5568 * tree.c (build_same_sized_truth_vector_type): Likewise.
5569 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5570 * var-tracking.c (emit_note_insn_var_location): Likewise.
5571 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5572 (ADDR_VEC_ALIGN): Likewise.
5574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5575 Alan Hayward <alan.hayward@arm.com>
5576 David Sherwood <david.sherwood@arm.com>
5578 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5580 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5581 or if measurement_type is polynomial.
5582 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5583 * combine.c (make_extraction): Likewise.
5584 * dse.c (find_shift_sequence): Likewise.
5585 * dwarf2out.c (mem_loc_descriptor): Likewise.
5586 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5587 (extract_bit_field, extract_low_bits): Likewise.
5588 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5589 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5590 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5591 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5592 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5593 * reload.c (find_reloads): Likewise.
5594 * reload1.c (alter_reg): Likewise.
5595 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5596 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5597 * tree-if-conv.c (predicate_mem_writes): Likewise.
5598 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5599 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5600 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5601 * valtrack.c (dead_debug_insert_temp): Likewise.
5602 * varasm.c (mergeable_constant_section): Likewise.
5603 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5605 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5606 Alan Hayward <alan.hayward@arm.com>
5607 David Sherwood <david.sherwood@arm.com>
5609 * expr.c (expand_assignment): Cope with polynomial mode sizes
5610 when assigning to a CONCAT.
5612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5613 Alan Hayward <alan.hayward@arm.com>
5614 David Sherwood <david.sherwood@arm.com>
5616 * machmode.h (mode_precision): Change from unsigned short to
5618 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5620 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5621 or if measurement_type is not polynomial.
5622 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5623 in which the mode is already known to be a scalar_int_mode.
5624 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5625 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5627 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5628 for GET_MODE_PRECISION.
5629 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5630 for GET_MODE_PRECISION.
5631 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5633 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5634 (expand_field_assignment, make_extraction): Likewise.
5635 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5636 (get_last_value): Likewise.
5637 * convert.c (convert_to_integer_1): Likewise.
5638 * cse.c (cse_insn): Likewise.
5639 * expr.c (expand_expr_real_1): Likewise.
5640 * lra-constraints.c (simplify_operand_subreg): Likewise.
5641 * optabs-query.c (can_atomic_load_p): Likewise.
5642 * optabs.c (expand_atomic_load): Likewise.
5643 (expand_atomic_store): Likewise.
5644 * ree.c (combine_reaching_defs): Likewise.
5645 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5646 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5647 * tree.h (type_has_mode_precision_p): Likewise.
5648 * ubsan.c (instrument_si_overflow): Likewise.
5650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5651 Alan Hayward <alan.hayward@arm.com>
5652 David Sherwood <david.sherwood@arm.com>
5654 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5655 polynomial numbers of units.
5656 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5657 (valid_vector_subparts_p): New function.
5658 (build_vector_type): Remove temporary shim and take the number
5659 of units as a poly_uint64 rather than an int.
5660 (build_opaque_vector_type): Take the number of units as a
5661 poly_uint64 rather than an int.
5662 * tree.c (build_vector_from_ctor): Handle polynomial
5663 TYPE_VECTOR_SUBPARTS.
5664 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5665 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5666 (build_vector_from_val): If the number of units is variable,
5667 use build_vec_duplicate_cst for constant operands and
5668 VEC_DUPLICATE_EXPR otherwise.
5669 (make_vector_type): Remove temporary is_constant ().
5670 (build_vector_type, build_opaque_vector_type): Take the number of
5671 units as a poly_uint64 rather than an int.
5672 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5674 * cfgexpand.c (expand_debug_expr): Likewise.
5675 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5676 (store_constructor, expand_expr_real_1): Likewise.
5677 (const_scalar_mask_from_tree): Likewise.
5678 * fold-const-call.c (fold_const_reduction): Likewise.
5679 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5680 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5681 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5682 (fold_relational_const): Likewise.
5683 (native_interpret_vector): Likewise. Change the size from an
5684 int to an unsigned int.
5685 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5686 TYPE_VECTOR_SUBPARTS.
5687 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5688 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5689 duplicating a non-constant operand into a variable-length vector.
5690 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5691 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5692 * ipa-icf.c (sem_variable::equals): Likewise.
5693 * match.pd: Likewise.
5694 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5695 * print-tree.c (print_node): Likewise.
5696 * stor-layout.c (layout_type): Likewise.
5697 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5698 * tree-cfg.c (verify_gimple_comparison): Likewise.
5699 (verify_gimple_assign_binary): Likewise.
5700 (verify_gimple_assign_ternary): Likewise.
5701 (verify_gimple_assign_single): Likewise.
5702 * tree-pretty-print.c (dump_generic_node): Likewise.
5703 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5704 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5705 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5706 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5707 (vect_shift_permute_load_chain): Likewise.
5708 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5709 (expand_vector_condition, optimize_vector_constructor): Likewise.
5710 (lower_vec_perm, get_compute_type): Likewise.
5711 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5712 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5713 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5714 (vect_recog_mask_conversion_pattern): Likewise.
5715 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5716 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5717 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5718 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5719 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5720 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5721 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5722 (supportable_widening_operation): Likewise.
5723 (supportable_narrowing_operation): Likewise.
5724 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5726 * varasm.c (output_constant): Likewise.
5728 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5729 Alan Hayward <alan.hayward@arm.com>
5730 David Sherwood <david.sherwood@arm.com>
5732 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5733 so that both the length == 3 and length != 3 cases set up their
5734 own permute vectors. Add comments explaining why we know the
5735 number of elements is constant.
5736 (vect_permute_load_chain): Likewise.
5738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5739 Alan Hayward <alan.hayward@arm.com>
5740 David Sherwood <david.sherwood@arm.com>
5742 * machmode.h (mode_nunits): Change from unsigned char to
5744 (ONLY_FIXED_SIZE_MODES): New macro.
5745 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5746 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5747 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5749 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5750 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5751 or if measurement_type is not polynomial.
5752 * genmodes.c (ZERO_COEFFS): New macro.
5753 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5755 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5756 Use ZERO_COEFFS when emitting initializers.
5757 * data-streamer.h (bp_pack_poly_value): New function.
5758 (bp_unpack_poly_value): Likewise.
5759 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5760 for GET_MODE_NUNITS.
5761 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5762 for GET_MODE_NUNITS.
5763 * tree.c (make_vector_type): Remove temporary shim and make
5764 the real function take the number of units as a poly_uint64
5766 (build_vector_type_for_mode): Handle polynomial nunits.
5767 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5768 * emit-rtl.c (const_vec_series_p_1): Likewise.
5769 (gen_rtx_CONST_VECTOR): Likewise.
5770 * fold-const.c (test_vec_duplicate_folding): Likewise.
5771 * genrecog.c (validate_pattern): Likewise.
5772 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5773 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5774 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5775 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5776 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5777 * rtlanal.c (subreg_get_info): Likewise.
5778 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5779 (vect_grouped_load_supported): Likewise.
5780 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5781 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5782 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5783 (simplify_const_unary_operation, simplify_binary_operation_1)
5784 (simplify_const_binary_operation, simplify_ternary_operation)
5785 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5786 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5787 instead of CONST_VECTOR_NUNITS.
5788 * varasm.c (output_constant_pool_2): Likewise.
5789 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5790 explicit-encoded elements in the XVEC for variable-length vectors.
5792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5794 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5797 Alan Hayward <alan.hayward@arm.com>
5798 David Sherwood <david.sherwood@arm.com>
5800 * coretypes.h (fixed_size_mode): Declare.
5801 (fixed_size_mode_pod): New typedef.
5802 * builtins.h (target_builtins::x_apply_args_mode)
5803 (target_builtins::x_apply_result_mode): Change type to
5804 fixed_size_mode_pod.
5805 * builtins.c (apply_args_size, apply_result_size, result_vector)
5806 (expand_builtin_apply_args_1, expand_builtin_apply)
5807 (expand_builtin_return): Update accordingly.
5809 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5811 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5812 * cselib.c (cselib_hash_rtx): Likewise.
5813 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5814 CONST_VECTOR encoding.
5816 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5817 Jeff Law <law@redhat.com>
5820 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5821 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5822 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5823 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5826 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5827 explicitly probe *sp in a noreturn function if there were any callee
5828 register saves or frame pointer is needed.
5830 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5833 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5834 BLKmode for ternary, binary or unary expressions.
5837 * var-tracking.c (delete_vta_debug_insn): New inline function.
5838 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5839 insns from get_insns () to NULL instead of each bb separately.
5840 Use delete_vta_debug_insn. No longer static.
5841 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5842 delete_vta_debug_insns callers.
5843 * rtl.h (delete_vta_debug_insns): Declare.
5844 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5845 instead of variable_tracking_main.
5847 2018-01-03 Martin Sebor <msebor@redhat.com>
5849 PR tree-optimization/83603
5850 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5851 arguments past the endof the argument list in functions declared
5852 without a prototype.
5853 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5854 Avoid checking when arguments are null.
5856 2018-01-03 Martin Sebor <msebor@redhat.com>
5859 * doc/extend.texi (attribute const): Fix a typo.
5860 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5861 issuing -Wsuggest-attribute for void functions.
5863 2018-01-03 Martin Sebor <msebor@redhat.com>
5865 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5866 offset_int::from instead of wide_int::to_shwi.
5867 (maybe_diag_overlap): Remove assertion.
5868 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5869 * gimple-ssa-sprintf.c (format_directive): Same.
5870 (parse_directive): Same.
5871 (sprintf_dom_walker::compute_format_length): Same.
5872 (try_substitute_return_value): Same.
5874 2018-01-03 Jeff Law <law@redhat.com>
5877 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5878 non-constant residual for zero at runtime and avoid probing in
5879 that case. Reorganize code for trailing problem to mirror handling
5882 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5884 PR tree-optimization/83501
5885 * tree-ssa-strlen.c (get_string_cst): New.
5886 (handle_char_store): Call get_string_cst.
5888 2018-01-03 Martin Liska <mliska@suse.cz>
5890 PR tree-optimization/83593
5891 * tree-ssa-strlen.c: Include tree-cfg.h.
5892 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5893 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5894 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5896 (strlen_dom_walker::before_dom_children): Call
5897 gimple_purge_dead_eh_edges. Dump tranformation with details
5899 (strlen_dom_walker::before_dom_children): Update call by adding
5900 new argument cleanup_eh.
5901 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5903 2018-01-03 Martin Liska <mliska@suse.cz>
5906 * cif-code.def (VARIADIC_THUNK): New enum value.
5907 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5910 2018-01-03 Jan Beulich <jbeulich@suse.com>
5912 * sse.md (mov<mode>_internal): Tighten condition for when to use
5913 vmovdqu<ssescalarsize> for TI and OI modes.
5915 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5917 Update copyright years.
5919 2018-01-03 Martin Liska <mliska@suse.cz>
5922 * ipa-visibility.c (function_and_variable_visibility): Skip
5923 functions with noipa attribure.
5925 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5927 * gcc.c (process_command): Update copyright notice dates.
5928 * gcov-dump.c (print_version): Ditto.
5929 * gcov.c (print_version): Ditto.
5930 * gcov-tool.c (print_version): Ditto.
5931 * gengtype.c (create_file): Ditto.
5932 * doc/cpp.texi: Bump @copying's copyright year.
5933 * doc/cppinternals.texi: Ditto.
5934 * doc/gcc.texi: Ditto.
5935 * doc/gccint.texi: Ditto.
5936 * doc/gcov.texi: Ditto.
5937 * doc/install.texi: Ditto.
5938 * doc/invoke.texi: Ditto.
5940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5942 * vector-builder.h (vector_builder::m_full_nelts): Change from
5943 unsigned int to poly_uint64.
5944 (vector_builder::full_nelts): Update prototype accordingly.
5945 (vector_builder::new_vector): Likewise.
5946 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5947 (vector_builder::operator ==): Likewise.
5948 (vector_builder::finalize): Likewise.
5949 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5950 Take the number of elements as a poly_uint64 rather than an
5952 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5953 from unsigned int to poly_uint64.
5954 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5955 (vec_perm_indices::new_vector): Likewise.
5956 (vec_perm_indices::length): Likewise.
5957 (vec_perm_indices::nelts_per_input): Likewise.
5958 (vec_perm_indices::input_nelts): Likewise.
5959 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5960 number of elements per input as a poly_uint64 rather than an
5961 unsigned int. Use the original encoding for variable-length
5962 vectors, rather than clamping each individual element.
5963 For the second and subsequent elements in each pattern,
5964 clamp the step and base before clamping their sum.
5965 (vec_perm_indices::series_p): Handle polynomial element counts.
5966 (vec_perm_indices::all_in_range_p): Likewise.
5967 (vec_perm_indices_to_tree): Likewise.
5968 (vec_perm_indices_to_rtx): Likewise.
5969 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5970 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5971 (tree_vector_builder::new_binary_operation): Handle polynomial
5972 element counts. Return false if we need to know the number
5973 of elements at compile time.
5974 * fold-const.c (fold_vec_perm): Punt if the number of elements
5975 isn't known at compile time.
5977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5979 * vec-perm-indices.h (vec_perm_builder): Change element type
5980 from HOST_WIDE_INT to poly_int64.
5981 (vec_perm_indices::element_type): Update accordingly.
5982 (vec_perm_indices::clamp): Handle polynomial element_types.
5983 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5984 (vec_perm_indices::all_in_range_p): Likewise.
5985 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5987 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5988 polynomial vec_perm_indices element types.
5989 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5990 * fold-const.c (fold_vec_perm): Likewise.
5991 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5992 * tree-vect-generic.c (lower_vec_perm): Likewise.
5993 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5994 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5995 element type to HOST_WIDE_INT.
5997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5998 Alan Hayward <alan.hayward@arm.com>
5999 David Sherwood <david.sherwood@arm.com>
6001 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6002 rather than an int. Use plus_constant.
6003 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6004 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6006 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6007 Alan Hayward <alan.hayward@arm.com>
6008 David Sherwood <david.sherwood@arm.com>
6010 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6011 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
6018 mode sizes. Require a constant size for BLKmode parameters
6019 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6020 forces a parameter to be padded at the lsb end in order to
6021 fill a complete number of words, require the parameter size
6022 to be ordered wrt UNITS_PER_WORD.
6024 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6025 Alan Hayward <alan.hayward@arm.com>
6026 David Sherwood <david.sherwood@arm.com>
6028 * reload1.c (spill_stack_slot_width): Change element type
6029 from unsigned int to poly_uint64_pod.
6030 (alter_reg): Treat mode sizes as polynomial.
6032 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6033 Alan Hayward <alan.hayward@arm.com>
6034 David Sherwood <david.sherwood@arm.com>
6036 * reload.c (complex_word_subreg_p): New function.
6037 (reload_inner_reg_of_subreg, push_reload): Use it.
6039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6040 Alan Hayward <alan.hayward@arm.com>
6041 David Sherwood <david.sherwood@arm.com>
6043 * lra-constraints.c (process_alt_operands): Reject matched
6044 operands whose sizes aren't ordered.
6045 (match_reload): Refer to this check here.
6047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6048 Alan Hayward <alan.hayward@arm.com>
6049 David Sherwood <david.sherwood@arm.com>
6051 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6052 that the mode size is in the set {1, 2, 4, 8, 16}.
6054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6055 Alan Hayward <alan.hayward@arm.com>
6056 David Sherwood <david.sherwood@arm.com>
6058 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6059 Use plus_constant instead of gen_rtx_PLUS.
6061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6062 Alan Hayward <alan.hayward@arm.com>
6063 David Sherwood <david.sherwood@arm.com>
6065 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6066 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6067 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6068 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6069 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6070 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6071 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6072 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6073 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6074 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6076 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6077 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6078 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6079 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6080 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6081 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6082 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6083 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6084 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6085 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6087 * expr.c (emit_move_resolve_push): Treat the input and result
6088 of PUSH_ROUNDING as a poly_int64.
6089 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6090 (emit_push_insn): Likewise.
6091 * lra-eliminations.c (mark_not_eliminable): Likewise.
6092 * recog.c (push_operand): Likewise.
6093 * reload1.c (elimination_effects): Likewise.
6094 * rtlanal.c (nonzero_bits1): Likewise.
6095 * calls.c (store_one_arg): Likewise. Require the padding to be
6096 known at compile time.
6098 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6099 Alan Hayward <alan.hayward@arm.com>
6100 David Sherwood <david.sherwood@arm.com>
6102 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6103 Use plus_constant instead of gen_rtx_PLUS.
6105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6106 Alan Hayward <alan.hayward@arm.com>
6107 David Sherwood <david.sherwood@arm.com>
6109 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6113 Alan Hayward <alan.hayward@arm.com>
6114 David Sherwood <david.sherwood@arm.com>
6116 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6117 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6118 via stack temporaries. Treat the mode size as polynomial too.
6120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6121 Alan Hayward <alan.hayward@arm.com>
6122 David Sherwood <david.sherwood@arm.com>
6124 * expr.c (expand_expr_real_2): When handling conversions involving
6125 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6126 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6127 as a poly_uint64 too.
6129 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6130 Alan Hayward <alan.hayward@arm.com>
6131 David Sherwood <david.sherwood@arm.com>
6133 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6136 Alan Hayward <alan.hayward@arm.com>
6137 David Sherwood <david.sherwood@arm.com>
6139 * combine.c (can_change_dest_mode): Handle polynomial
6140 REGMODE_NATURAL_SIZE.
6141 * expmed.c (store_bit_field_1): Likewise.
6142 * expr.c (store_constructor): Likewise.
6143 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6144 and polynomial REGMODE_NATURAL_SIZE.
6145 (gen_lowpart_common): Likewise.
6146 * reginfo.c (record_subregs_of_mode): Likewise.
6147 * rtlanal.c (read_modify_subreg_p): Likewise.
6149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6150 Alan Hayward <alan.hayward@arm.com>
6151 David Sherwood <david.sherwood@arm.com>
6153 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6154 numbers of elements.
6156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6157 Alan Hayward <alan.hayward@arm.com>
6158 David Sherwood <david.sherwood@arm.com>
6160 * match.pd: Cope with polynomial numbers of vector elements.
6162 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6163 Alan Hayward <alan.hayward@arm.com>
6164 David Sherwood <david.sherwood@arm.com>
6166 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6167 in a POINTER_PLUS_EXPR.
6169 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6170 Alan Hayward <alan.hayward@arm.com>
6171 David Sherwood <david.sherwood@arm.com>
6173 * omp-simd-clone.c (simd_clone_subparts): New function.
6174 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6175 (ipa_simd_modify_function_body): Likewise.
6177 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6178 Alan Hayward <alan.hayward@arm.com>
6179 David Sherwood <david.sherwood@arm.com>
6181 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6182 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6183 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6184 (expand_vector_condition, vector_element): Likewise.
6185 (subparts_gt): New function.
6186 (get_compute_type): Use subparts_gt.
6187 (count_type_subparts): Delete.
6188 (expand_vector_operations_1): Use subparts_gt instead of
6189 count_type_subparts.
6191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6192 Alan Hayward <alan.hayward@arm.com>
6193 David Sherwood <david.sherwood@arm.com>
6195 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6196 (vect_compile_time_alias): ...this new function. Do the calculation
6197 on poly_ints rather than trees.
6198 (vect_prune_runtime_alias_test_list): Update call accordingly.
6200 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6201 Alan Hayward <alan.hayward@arm.com>
6202 David Sherwood <david.sherwood@arm.com>
6204 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6206 (vect_schedule_slp_instance): Likewise.
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 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6213 constant and extern definitions for variable-length vectors.
6214 (vect_get_constant_vectors): Note that the number of units
6215 is known to be constant.
6217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6218 Alan Hayward <alan.hayward@arm.com>
6219 David Sherwood <david.sherwood@arm.com>
6221 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6222 of units as polynomial. Choose between WIDE and NARROW based
6225 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6226 Alan Hayward <alan.hayward@arm.com>
6227 David Sherwood <david.sherwood@arm.com>
6229 * tree-vect-stmts.c (simd_clone_subparts): New function.
6230 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6232 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6233 Alan Hayward <alan.hayward@arm.com>
6234 David Sherwood <david.sherwood@arm.com>
6236 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6237 vectors as polynomial. Use build_index_vector for
6240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6241 Alan Hayward <alan.hayward@arm.com>
6242 David Sherwood <david.sherwood@arm.com>
6244 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6245 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6246 for variable-length vectors.
6247 (vectorizable_mask_load_store): Treat the number of units as
6248 polynomial, asserting that it is constant if the condition has
6249 already been enforced.
6250 (vectorizable_store, vectorizable_load): Likewise.
6252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6253 Alan Hayward <alan.hayward@arm.com>
6254 David Sherwood <david.sherwood@arm.com>
6256 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6257 of units as polynomial. Punt if we can't tell at compile time
6258 which vector contains the final result.
6260 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6261 Alan Hayward <alan.hayward@arm.com>
6262 David Sherwood <david.sherwood@arm.com>
6264 * tree-vect-loop.c (vectorizable_induction): Treat the number
6265 of units as polynomial. Punt on SLP inductions. Use an integer
6266 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6267 cast of such a series for variable-length floating-point
6270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6271 Alan Hayward <alan.hayward@arm.com>
6272 David Sherwood <david.sherwood@arm.com>
6274 * tree.h (build_index_vector): Declare.
6275 * tree.c (build_index_vector): New function.
6276 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6277 of units as polynomial, forcibly converting it to a constant if
6278 vectorizable_reduction has already enforced the condition.
6279 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6280 to create a {1,2,3,...} vector.
6281 (vectorizable_reduction): Treat the number of units as polynomial.
6282 Choose vectype_in based on the largest scalar element size rather
6283 than the smallest number of units. Enforce the restrictions
6286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6287 Alan Hayward <alan.hayward@arm.com>
6288 David Sherwood <david.sherwood@arm.com>
6290 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6291 number of units as polynomial.
6293 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6294 Alan Hayward <alan.hayward@arm.com>
6295 David Sherwood <david.sherwood@arm.com>
6297 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6298 * target.def (autovectorize_vector_sizes): Return the vector sizes
6299 by pointer, using vector_sizes rather than a bitmask.
6300 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6301 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6302 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6304 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6305 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6306 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6307 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6308 * omp-general.c (omp_max_vf): Likewise.
6309 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6310 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6311 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6312 * tree-vect-slp.c (vect_slp_bb): Likewise.
6313 * doc/tm.texi: Regenerate.
6314 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6316 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6317 the vector size as a poly_uint64 rather than an unsigned int.
6318 (current_vector_size): Change from an unsigned int to a poly_uint64.
6319 (get_vectype_for_scalar_type): Update accordingly.
6320 * tree.h (build_truth_vector_type): Take the size and number of
6321 units as a poly_uint64 rather than an unsigned int.
6322 (build_vector_type): Add a temporary overload that takes
6323 the number of units as a poly_uint64 rather than an unsigned int.
6324 * tree.c (make_vector_type): Likewise.
6325 (build_truth_vector_type): Take the number of units as a poly_uint64
6326 rather than an unsigned int.
6328 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6329 Alan Hayward <alan.hayward@arm.com>
6330 David Sherwood <david.sherwood@arm.com>
6332 * target.def (get_mask_mode): Take the number of units and length
6333 as poly_uint64s rather than unsigned ints.
6334 * targhooks.h (default_get_mask_mode): Update accordingly.
6335 * targhooks.c (default_get_mask_mode): Likewise.
6336 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6337 * doc/tm.texi: Regenerate.
6339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6340 Alan Hayward <alan.hayward@arm.com>
6341 David Sherwood <david.sherwood@arm.com>
6343 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6344 * omp-general.c (omp_max_vf): Likewise.
6345 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6346 (expand_omp_simd): Handle polynomial safelen.
6347 * omp-low.c (omplow_simd_context): Add a default constructor.
6348 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6349 (lower_rec_simd_input_clauses): Update accordingly.
6350 (lower_rec_input_clauses): Likewise.
6352 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6353 Alan Hayward <alan.hayward@arm.com>
6354 David Sherwood <david.sherwood@arm.com>
6356 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6357 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6358 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6359 (vect_analyze_slp_cost): Likewise.
6360 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6361 (vect_model_load_cost): Likewise.
6363 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6364 Alan Hayward <alan.hayward@arm.com>
6365 David Sherwood <david.sherwood@arm.com>
6367 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6368 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6369 from an unsigned int * to a poly_uint64_pod *.
6370 (calculate_unrolling_factor): New function.
6371 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6374 Alan Hayward <alan.hayward@arm.com>
6375 David Sherwood <david.sherwood@arm.com>
6377 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6378 from an unsigned int to a poly_uint64.
6379 (_loop_vec_info::slp_unrolling_factor): Likewise.
6380 (_loop_vec_info::vectorization_factor): Change from an int
6382 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6383 (vect_get_num_vectors): New function.
6384 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6385 (vect_get_num_copies): Use vect_get_num_vectors.
6386 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6387 to an unsigned int *.
6388 (vect_analyze_data_refs): Change min_vf from an int * to a
6390 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6391 than an unsigned HOST_WIDE_INT.
6392 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6393 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6394 to an unsigned int *.
6395 (vect_analyze_data_ref_dependences): Likewise.
6396 (vect_compute_data_ref_alignment): Handle polynomial vf.
6397 (vect_enhance_data_refs_alignment): Likewise.
6398 (vect_prune_runtime_alias_test_list): Likewise.
6399 (vect_shift_permute_load_chain): Likewise.
6400 (vect_supportable_dr_alignment): Likewise.
6401 (dependence_distance_ge_vf): Take the vectorization factor as a
6402 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6403 (vect_analyze_data_refs): Change min_vf from an int * to a
6405 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6406 vfm1 as a poly_uint64 rather than an int. Make the same change
6407 for the returned bound_scalar.
6408 (vect_gen_vector_loop_niters): Handle polynomial vf.
6409 (vect_do_peeling): Likewise. Update call to
6410 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6411 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6413 * tree-vect-loop.c (vect_determine_vectorization_factor)
6414 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6415 (vect_get_known_peeling_cost): Likewise.
6416 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6417 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6418 (vect_transform_loop): Likewise. Use the lowest possible VF when
6419 updating the upper bounds of the loop.
6420 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6422 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6423 polynomial unroll factors.
6424 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6425 (vect_make_slp_decision): Likewise.
6426 (vect_supported_load_permutation_p): Likewise, and polynomial
6428 (vect_analyze_slp_cost): Handle polynomial vf.
6429 (vect_slp_analyze_node_operations): Likewise.
6430 (vect_slp_analyze_bb_1): Likewise.
6431 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6432 than an unsigned HOST_WIDE_INT.
6433 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6434 (vectorizable_load): Handle polynomial vf.
6435 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6437 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6440 Alan Hayward <alan.hayward@arm.com>
6441 David Sherwood <david.sherwood@arm.com>
6443 * match.pd: Handle bit operations involving three constants
6444 and try to fold one pair.
6446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6448 * tree-vect-loop-manip.c: Include gimple-fold.h.
6449 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6450 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6451 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6452 Add a path that uses a step of VF instead of 1, but disable it
6454 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6455 and niters_no_overflow parameters. Update calls to
6456 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6457 Create a new SSA name if the latter choses to use a ste other
6458 than zero, and return it via niters_vector_mult_vf_var.
6459 * tree-vect-loop.c (vect_transform_loop): Update calls to
6460 vect_do_peeling, vect_gen_vector_loop_niters and
6461 slpeel_make_loop_iterate_ntimes.
6462 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6463 (vect_gen_vector_loop_niters): Update declarations after above changes.
6465 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6467 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6468 128-bit round to integer instructions.
6469 (ceil<mode>2): Likewise.
6470 (btrunc<mode>2): Likewise.
6471 (round<mode>2): Likewise.
6473 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6475 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6476 unaligned VSX load/store on P8/P9.
6477 (expand_block_clear): Allow the use of unaligned VSX
6478 load/store on P8/P9.
6480 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6482 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6484 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6485 swap associated with both a load and a store.
6487 2018-01-02 Andrew Waterman <andrew@sifive.com>
6489 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6490 * config/riscv/riscv.md (clear_cache): Use it.
6492 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6494 * web.c: Remove out-of-date comment.
6496 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6498 * expr.c (fixup_args_size_notes): Check that any existing
6499 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6500 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6501 (emit_single_push_insn): ...here.
6503 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6505 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6506 (const_vector_encoded_nelts): New function.
6507 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6508 (const_vector_int_elt, const_vector_elt): Declare.
6509 * emit-rtl.c (const_vector_int_elt_1): New function.
6510 (const_vector_elt): Likewise.
6511 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6512 of CONST_VECTOR_ELT.
6514 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6516 * expr.c: Include rtx-vector-builder.h.
6517 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6518 directly on the tree encoding.
6519 (const_vector_from_tree): Likewise.
6520 * optabs.c: Include rtx-vector-builder.h.
6521 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6522 sequence of "u" values.
6523 * vec-perm-indices.c: Include rtx-vector-builder.h.
6524 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6525 directly on the vec_perm_indices encoding.
6527 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6529 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6530 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6531 * rtx-vector-builder.h: New file.
6532 * rtx-vector-builder.c: Likewise.
6533 * rtl.h (rtx_def::u2): Add a const_vector field.
6534 (CONST_VECTOR_NPATTERNS): New macro.
6535 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6536 (CONST_VECTOR_DUPLICATE_P): Likewise.
6537 (CONST_VECTOR_STEPPED_P): Likewise.
6538 (CONST_VECTOR_ENCODED_ELT): Likewise.
6539 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6540 (unwrap_const_vec_duplicate): Likewise.
6541 (const_vec_series_p): Check for a non-duplicated vector encoding.
6542 Say that the function only returns true for integer vectors.
6543 * emit-rtl.c: Include rtx-vector-builder.h.
6544 (gen_const_vec_duplicate_1): Delete.
6545 (gen_const_vector): Call gen_const_vec_duplicate instead of
6546 gen_const_vec_duplicate_1.
6547 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6548 (gen_const_vec_duplicate): Use rtx_vector_builder.
6549 (gen_const_vec_series): Likewise.
6550 (gen_rtx_CONST_VECTOR): Likewise.
6551 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6552 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6553 Build a new vector rather than modifying a CONST_VECTOR in-place.
6554 (handle_special_swappables): Update call accordingly.
6555 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6556 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6557 Build a new vector rather than modifying a CONST_VECTOR in-place.
6558 (handle_special_swappables): Update call accordingly.
6560 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6562 * simplify-rtx.c (simplify_const_binary_operation): Use
6563 CONST_VECTOR_ELT instead of XVECEXP.
6565 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6567 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6568 the selector elements to be different from the data elements
6569 if the selector is a VECTOR_CST.
6570 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6571 ssizetype for the selector.
6573 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6575 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6576 before testing each element individually.
6577 * tree-vect-generic.c (lower_vec_perm): Likewise.
6579 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6581 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6582 * selftest-run-tests.c (selftest::run_tests): Call it.
6583 * vector-builder.h (vector_builder::operator ==): New function.
6584 (vector_builder::operator !=): Likewise.
6585 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6586 (vec_perm_indices::all_from_input_p): New function.
6587 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6588 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6589 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6590 instead of reading the VECTOR_CST directly. Detect whether both
6591 vector inputs are the same before constructing the vec_perm_indices,
6592 and update the number of inputs argument accordingly. Use the
6593 utility functions added above. Only construct sel2 if we need to.
6595 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6597 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6598 the broadcast of the low byte.
6599 (expand_mult_highpart): Use an explicit encoding for the permutes.
6600 * optabs-query.c (can_mult_highpart_p): Likewise.
6601 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6602 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6603 (vectorizable_bswap): Likewise.
6604 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6605 explicit encoding for the power-of-2 permutes.
6606 (vect_permute_store_chain): Likewise.
6607 (vect_grouped_load_supported): Likewise.
6608 (vect_permute_load_chain): Likewise.
6610 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6612 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6613 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6614 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6615 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6616 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6617 (vect_gen_perm_mask_any): Likewise.
6619 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6621 * int-vector-builder.h: New file.
6622 * vec-perm-indices.h: Include int-vector-builder.h.
6623 (vec_perm_indices): Redefine as an int_vector_builder.
6624 (auto_vec_perm_indices): Delete.
6625 (vec_perm_builder): Redefine as a stand-alone class.
6626 (vec_perm_indices::vec_perm_indices): New function.
6627 (vec_perm_indices::clamp): Likewise.
6628 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6629 (vec_perm_indices::new_vector): New function.
6630 (vec_perm_indices::new_expanded_vector): Update for new
6631 vec_perm_indices class.
6632 (vec_perm_indices::rotate_inputs): New function.
6633 (vec_perm_indices::all_in_range_p): Operate directly on the
6634 encoded form, without computing elided elements.
6635 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6636 encoding. Update for new vec_perm_indices class.
6637 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6638 the given vec_perm_builder.
6639 (expand_vec_perm_var): Update vec_perm_builder constructor.
6640 (expand_mult_highpart): Use vec_perm_builder instead of
6641 auto_vec_perm_indices.
6642 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6643 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6644 or double series encoding as appropriate.
6645 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6646 vec_perm_indices instead of auto_vec_perm_indices.
6647 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6648 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6649 (vect_permute_store_chain): Likewise.
6650 (vect_grouped_load_supported): Likewise.
6651 (vect_permute_load_chain): Likewise.
6652 (vect_shift_permute_load_chain): Likewise.
6653 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6654 (vect_transform_slp_perm_load): Likewise.
6655 (vect_schedule_slp_instance): Likewise.
6656 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6657 (vectorizable_mask_load_store): Likewise.
6658 (vectorizable_bswap): Likewise.
6659 (vectorizable_store): Likewise.
6660 (vectorizable_load): Likewise.
6661 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6662 vec_perm_indices instead of auto_vec_perm_indices. Use
6663 tree_to_vec_perm_builder to read the vector from a tree.
6664 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6665 vec_perm_builder instead of a vec_perm_indices.
6666 (have_whole_vector_shift): Use vec_perm_builder and
6667 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6668 truncation to calc_vec_perm_mask_for_shift.
6669 (vect_create_epilog_for_reduction): Likewise.
6670 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6671 from auto_vec_perm_indices to vec_perm_indices.
6672 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6673 instead of changing individual elements.
6674 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6675 the vector in d.perm.
6676 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6677 from auto_vec_perm_indices to vec_perm_indices.
6678 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6679 instead of changing individual elements.
6680 (arm_vectorize_vec_perm_const): Use new_vector to install
6681 the vector in d.perm.
6682 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6683 Update vec_perm_builder constructor.
6684 (rs6000_expand_interleave): Likewise.
6685 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6686 (rs6000_expand_interleave): Likewise.
6688 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6690 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6691 to qimode could truncate the indices.
6692 * optabs.c (expand_vec_perm_var): Likewise.
6694 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6696 * Makefile.in (OBJS): Add vec-perm-indices.o.
6697 * vec-perm-indices.h: New file.
6698 * vec-perm-indices.c: Likewise.
6699 * target.h (vec_perm_indices): Replace with a forward class
6701 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6702 * optabs.h: Include vec-perm-indices.h.
6703 (expand_vec_perm): Delete.
6704 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6705 (expand_vec_perm_const): Declare.
6706 * target.def (vec_perm_const_ok): Replace with...
6707 (vec_perm_const): ...this new hook.
6708 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6709 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6710 * doc/tm.texi: Regenerate.
6711 * optabs.def (vec_perm_const): Delete.
6712 * doc/md.texi (vec_perm_const): Likewise.
6713 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6714 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6715 expand_vec_perm for constant permutation vectors. Assert that
6716 the mode of variable permutation vectors is the integer equivalent
6717 of the mode that is being permuted.
6718 * optabs-query.h (selector_fits_mode_p): Declare.
6719 * optabs-query.c: Include vec-perm-indices.h.
6720 (selector_fits_mode_p): New function.
6721 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6722 is defined, instead of checking whether the vec_perm_const_optab
6723 exists. Use targetm.vectorize.vec_perm_const instead of
6724 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6725 fit in the vector mode before using a variable permute.
6726 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6727 vec_perm_indices instead of an rtx.
6728 (expand_vec_perm): Replace with...
6729 (expand_vec_perm_const): ...this new function. Take the selector
6730 as a vec_perm_indices rather than an rtx. Also take the mode of
6731 the selector. Update call to shift_amt_for_vec_perm_mask.
6732 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6733 Use vec_perm_indices::new_expanded_vector to expand the original
6734 selector into bytes. Check whether the indices fit in the vector
6735 mode before using a variable permute.
6736 (expand_vec_perm_var): Make global.
6737 (expand_mult_highpart): Use expand_vec_perm_const.
6738 * fold-const.c: Includes vec-perm-indices.h.
6739 * tree-ssa-forwprop.c: Likewise.
6740 * tree-vect-data-refs.c: Likewise.
6741 * tree-vect-generic.c: Likewise.
6742 * tree-vect-loop.c: Likewise.
6743 * tree-vect-slp.c: Likewise.
6744 * tree-vect-stmts.c: Likewise.
6745 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6747 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6748 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6749 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6750 (aarch64_vectorize_vec_perm_const): ...this new function.
6751 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6752 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6753 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6754 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6755 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6756 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6757 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6759 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6760 check for NEON modes.
6761 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6762 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6763 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6764 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6766 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6767 the old VEC_PERM_CONST conditions.
6768 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6769 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6770 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6771 (ia64_vectorize_vec_perm_const_ok): Merge into...
6772 (ia64_vectorize_vec_perm_const): ...this new function.
6773 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6774 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6775 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6776 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6777 * config/mips/mips.c (mips_expand_vec_perm_const)
6778 (mips_vectorize_vec_perm_const_ok): Merge into...
6779 (mips_vectorize_vec_perm_const): ...this new function.
6780 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6781 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6782 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6783 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6784 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6785 (rs6000_expand_vec_perm_const): Delete.
6786 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6788 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6789 (altivec_expand_vec_perm_const_le): Take each operand individually.
6790 Operate on constant selectors rather than rtxes.
6791 (altivec_expand_vec_perm_const): Likewise. Update call to
6792 altivec_expand_vec_perm_const_le.
6793 (rs6000_expand_vec_perm_const): Delete.
6794 (rs6000_vectorize_vec_perm_const_ok): Delete.
6795 (rs6000_vectorize_vec_perm_const): New function.
6796 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6797 an element count and rtx array.
6798 (rs6000_expand_extract_even): Update call accordingly.
6799 (rs6000_expand_interleave): Likewise.
6800 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6801 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6802 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6803 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6804 (rs6000_expand_vec_perm_const): Delete.
6805 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6806 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6807 (altivec_expand_vec_perm_const_le): Take each operand individually.
6808 Operate on constant selectors rather than rtxes.
6809 (altivec_expand_vec_perm_const): Likewise. Update call to
6810 altivec_expand_vec_perm_const_le.
6811 (rs6000_expand_vec_perm_const): Delete.
6812 (rs6000_vectorize_vec_perm_const_ok): Delete.
6813 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6814 reference to the SPE evmerge intructions.
6815 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6816 an element count and rtx array.
6817 (rs6000_expand_extract_even): Update call accordingly.
6818 (rs6000_expand_interleave): Likewise.
6819 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6820 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6822 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6824 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6826 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6827 vector mode and that that mode matches the mode of the data
6829 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6830 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6831 directly using expand_vec_perm_1 when forcing selectors into
6833 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6835 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6837 * optabs-query.h (can_vec_perm_p): Delete.
6838 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6839 * optabs-query.c (can_vec_perm_p): Split into...
6840 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6841 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6842 particular selector is valid.
6843 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6844 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6845 (vect_grouped_load_supported): Likewise.
6846 (vect_shift_permute_load_chain): Likewise.
6847 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6848 (vect_transform_slp_perm_load): Likewise.
6849 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6850 (vectorizable_bswap): Likewise.
6851 (vect_gen_perm_mask_checked): Likewise.
6852 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6853 implementations of variable permutation vectors into account
6854 when deciding which selector to use.
6855 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6856 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6857 with a false third argument.
6858 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6859 to test whether the constant selector is valid and can_vec_perm_var_p
6860 to test whether a variable selector is valid.
6862 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6864 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6865 * optabs-query.c (can_vec_perm_p): Likewise.
6866 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6867 instead of vec_perm_indices.
6868 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6869 (vect_gen_perm_mask_checked): Likewise,
6870 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6871 (vect_gen_perm_mask_checked): Likewise,
6873 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6875 * optabs-query.h (qimode_for_vec_perm): Declare.
6876 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6877 (qimode_for_vec_perm): ...this new function.
6878 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6880 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6882 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6883 does not have a conditional at the top.
6885 2018-01-02 Richard Biener <rguenther@suse.de>
6887 * ipa-inline.c (big_speedup_p): Fix expression.
6889 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6892 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6895 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6899 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6900 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6901 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6902 cond_taken_branch_cost 3->4.
6904 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6906 PR tree-optimization/83581
6907 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6908 TODO_cleanup_cfg if any changes have been made.
6911 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6912 convert_modes if target mode has the right side, but different mode
6916 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6917 last argument when extracting from CONCAT. If either from_real or
6918 from_imag is NULL, use expansion through memory. If result is not
6919 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6920 the parts directly to inner mode, if even that fails, use expansion
6924 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6925 check for bswap in mode rather than HImode and use that in expand_unop
6928 Copyright (C) 2018 Free Software Foundation, Inc.
6930 Copying and distribution of this file, with or without modification,
6931 are permitted in any medium without royalty provided the copyright
6932 notice and this notice are preserved.