1 2018-02-13 Jakub Jelinek <jakub@redhat.com>
4 * stor-layout.c (place_field): For variable length fields, adjust
5 offset_align afterwards not just based on the field's alignment,
9 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
10 of exps and logs in the use_exp2 case.
12 2018-02-13 Jeff Law <law@redhat.com>
14 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
17 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
20 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
22 * common.opt (gas-loc-support, gas-locview-support): New.
23 (ginline-points, ginternal-reset-location-views): New.
24 * doc/invoke.texi: Document them. Use @itemx where intended.
25 (gvariable-location-views): Adjust.
26 * target.def (reset_location_view): New.
27 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
28 (TARGET_RESET_LOCATION_VIEW): New.
29 * doc/tm.texi: Rebuilt.
30 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
31 (dwarf2out_default_as_locview_support): New.
32 (output_asm_line_debug_info): Use option variables.
33 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
34 (output_loc_list): Likewise.
35 (add_high_low_attributes): Check option variables.
36 Don't output entry view attribute in strict mode.
37 (gen_inlined_subroutine_die): Check option variables.
38 (dwarf2out_inline_entry): Likewise.
39 (init_sections_and_labels): Likewise.
40 (dwarf2out_early_finish): Likewise.
41 (maybe_reset_location_view): New, from...
42 (dwarf2out_var_location): ... here. Call it.
43 * debug.h (dwarf2out_default_as_loc_support): Declare.
44 (dwarf2out_default_as_locview_support): Declare.
45 * hooks.c (hook_int_rtx_insn_0): New.
46 * hooks.h (hook_int_rtx_insn_0): Declare.
47 * toplev.c (process_options): Take -gas-loc-support and
48 -gas-locview-support from dwarf2out. Enable
49 -gvariable-location-views by default only with locview
50 assembler support. Enable -ginternal-reset-location-views by
51 default only if the target defines the corresponding hook.
52 Enable -ginline-points by default if location views are
53 enabled; force it disabled if statement frontiers are
55 * tree-inline.c (expand_call_inline): Check option variables.
56 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
58 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
60 PR tree-optimization/84321
61 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
62 handling. Also check whether the anti-range contains any values
63 that satisfy the mask; switch to a VR_RANGE if not.
65 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
68 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
70 2018-02-13 Martin Jambor <mjambor@suse.cz>
73 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
74 of call statements, also set location of a load to a temporary.
76 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
78 * config/rl78/rl78.c (add_vector_labels): New function.
79 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
80 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
81 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
82 which checks that no arguments are passed.
83 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
84 * doc/extend.texi: Documentation for the new attribute.
86 2018-02-13 Andreas Schwab <schwab@suse.de>
88 * config/riscv/linux.h (CPP_SPEC): Define.
90 2018-02-13 Jakub Jelinek <jakub@redhat.com>
93 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
94 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
95 OPTION_MASK_ISA_AES as first argument to def_builtin_const
96 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
97 instead of OPTION_MASK_ISA_PCLMUL as first argument to
98 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
99 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
100 temporarily for AES and PCLMUL builtins.
102 PR tree-optimization/84339
103 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
104 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
108 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
109 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
110 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
112 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
114 * omp-simd-clone.h: New file.
115 * omp-simd-clone.c: Include omp-simd-clone.h.
116 (expand_simd_clones): No longer static.
117 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
118 cgraph.h and omp-simd-clone.h.
119 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
120 (vect_recog_widen_shift_pattern): Formatting fix.
121 (vect_pattern_recog_1): Don't check optab for calls.
124 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
125 operands[2] into a REG before using gen_lowpart on it.
127 2018-02-12 Jeff Law <law@redhat.com>
130 * config/sh/sh.c (find_barrier): Consider a sibling call
133 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
134 successfully back substituting a reg.
136 2018-02-12 Richard Biener <rguenther@suse.de>
138 PR tree-optimization/84037
139 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
140 parameter, move visited init to caller.
141 (vect_slp_analyze_operations): Separate cost from validity
142 check, initialize visited once for all instances.
143 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
145 * tree-vect-stmts.c (vect_model_simple_cost): Make early
147 (vect_model_promotion_demotion_cost): Likewise.
148 (vectorizable_bswap): Guard cost modeling with !slp_node
149 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
151 (vectorizable_call): Likewise.
152 (vectorizable_conversion): Likewise.
153 (vectorizable_assignment): Likewise.
154 (vectorizable_shift): Likewise.
155 (vectorizable_operation): Likewise.
156 (vectorizable_store): Likewise.
157 (vectorizable_load): Likewise.
158 (vectorizable_condition): Likewise.
159 (vectorizable_comparison): Likewise.
161 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
164 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
165 (ASAN_MARK): Fix fnspec to account for return value, change pointer
166 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
168 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
171 * params.def (inline-min-speedup): Increase from 8 to 15.
172 (max-inline-insns-auto): Decrease from 40 to 30.
173 * ipa-split.c (consider_split): Add some buffer for function to
174 be considered inlining candidate.
175 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
178 2018-02-12 Richard Biener <rguenther@suse.de>
180 PR tree-optimization/84037
181 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
182 matched stmts if we cannot swap the non-matched ones.
184 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
186 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
187 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
188 _mm_maskz_scalef_round_ss): New intrinsics.
189 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
190 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
191 __builtin_ia32_scalefss_round): Remove.
192 (__builtin_ia32_scalefsd_mask_round,
193 __builtin_ia32_scalefss_mask_round): New intrinsics.
194 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
195 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
196 ((match_operand:VF_128 2 "<round_nimm_predicate>"
197 "<round_constraint>")): Changed to ...
198 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
199 "<round_scalar_constraint>")): ... this.
200 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
201 %0, %1, %2<round_op3>}"): Changed to ...
202 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
203 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
204 %2<round_scalar_mask_op3>}"): ... this.
205 * config/i386/subst.md (round_scalar_nimm_predicate): New.
207 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
209 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
210 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
211 (_mm_maskz_sqrt_round_ss): New intrinsics.
212 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
213 (__builtin_ia32_sqrtsd_mask_round)
214 (__builtin_ia32_sqrtss_mask_round): New builtins.
215 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
216 (__builtin_ia32_sqrtss_round): Remove.
217 (__builtin_ia32_sqrtsd_mask_round)
218 (__builtin_ia32_sqrtss_mask_round): New builtins.
219 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
220 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
221 ((match_operand:VF_128 1 "vector_operand"
222 "xBm,<round_constraint>")): Changed to ...
223 ((match_operand:VF_128 1 "vector_operand"
224 "xBm,<round_scalar_constraint>")): ... this.
225 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
226 %0, %2, %<iptr>1<round_op3>}): Changed to ...
227 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
228 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
229 %<iptr>1<round_scalar_mask_op3>}): ... this.
230 ((set_attr "prefix" "<round_prefix>")): Changed to ...
231 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
233 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
236 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
237 Cast vec_cmpeq result to correct type.
238 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
239 Cast vec_cmpgt result to correct type.
241 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
243 * final.c (final_scan_insn_1): Renamed from...
244 (final_scan_insn): ... this. New wrapper, to recover
245 seen from the outermost call in recursive ones.
246 * config/sparc/sparc.c (output_return): Drop seen from call.
247 (output_sibcall): Likewise.
248 * config/visium/visium.c (output_branch): Likewise.
250 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
252 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
255 2018-02-10 Alan Modra <amodra@gmail.com>
258 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
259 Specify LR as an input.
261 2018-02-10 Jakub Jelinek <jakub@redhat.com>
264 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
265 remove_member_access_dummy_vars): New functions.
266 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
267 lower_omp_1, execute_lower_omp): Use them.
269 PR rtl-optimization/84308
270 * shrink-wrap.c (spread_components): Release todo vector.
272 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
274 PR rtl-optimization/57193
275 * ira-color.c (struct allocno_color_data): Add member
276 conflict_allocno_hard_prefs.
277 (update_conflict_allocno_hard_prefs): New.
278 (bucket_allocno_compare_func): Add a preference based on
279 conflict_allocno_hard_prefs.
280 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
281 (color_allocnos): Remove a dead code. Initiate
282 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
284 2018-02-09 Jakub Jelinek <jakub@redhat.com>
287 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
288 constraint from =wa to wa. Avoid a subreg on the output operand,
289 instead use a pseudo and subreg it in a move.
290 (p9_xxbrd_<mode>): Changed to ...
291 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
292 (p9_xxbrd_v2df): New expander.
293 (p9_xxbrw_<mode>): Changed to ...
294 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
295 (p9_xxbrw_v4sf): New expander.
297 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
299 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
301 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
304 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
305 multiply in 32-bit mode.
306 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
307 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
310 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
312 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
313 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
314 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
315 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
317 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
320 * dwarf2out.c (is_trivial_indirect_ref): New function.
321 (dwarf2out_late_global_decl): Do not generate a location
322 attribute for variables that have a non-trivial DECL_VALUE_EXPR
323 and that are not defined in the current unit.
325 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
327 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
328 instead of a libcall for UNORDERED.
330 2018-02-09 Tamar Christina <tamar.christina@arm.com>
333 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
334 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
336 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
339 * config/s390/s390.c (s390_set_current_function): Invoke
340 s390_indirect_branch_settings also if fndecl didn't change.
342 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
344 * config/rs6000/rs6000.md (blockage): Set length to zero.
346 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
348 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
350 2018-02-09 Jakub Jelinek <jakub@redhat.com>
353 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
354 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
358 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
359 PARALLEL incoming that failed vt_get_decl_and_offset check.
362 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
363 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
365 (get_variable_section): For decls in named .bss* sections pass true as
366 second argument to bss_initializer_p.
368 2018-02-09 Marek Polacek <polacek@redhat.com>
369 Jakub Jelinek <jakub@redhat.com>
372 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
373 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
374 Sync some changes from cxx_fold_indirect_ref.
376 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
378 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
380 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
381 (BLOCK_INLINE_ENTRY_LABEL): New.
382 (dwarf2out_var_location): Disregard inline entry markers.
383 (inline_entry_data): New struct.
384 (inline_entry_data_hasher): New hashtable type.
385 (inline_entry_data_hasher::hash): New.
386 (inline_entry_data_hasher::equal): New.
387 (inline_entry_data_table): New variable.
388 (add_high_low_attributes): Add DW_AT_entry_pc and
389 DW_AT_GNU_entry_view attributes if a pending entry is found
390 in inline_entry_data_table. Add old entry_pc attribute only
391 if debug nonbinding markers are disabled.
392 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
394 (block_within_block_p, dwarf2out_inline_entry): New.
395 (dwarf2out_finish): Check that no entries remained in
396 inline_entry_data_table.
397 * final.c (reemit_insn_block_notes): Handle inline entry notes.
398 (final_scan_insn, notice_source_line): Likewise.
399 (rest_of_clean_state): Skip inline entry markers.
400 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
402 * gimple.c (gimple_build_debug_inline_entry): New.
403 * gimple.h (enum gimple_debug_subcode): Add
404 GIMPLE_DEBUG_INLINE_ENTRY.
405 (gimple_build_debug_inline_entry): Declare.
406 (gimple_debug_inline_entry_p): New.
407 (gimple_debug_nonbind_marker_p): Adjust.
408 * insn-notes.def (INLINE_ENTRY): New.
409 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
410 inline entry marker notes.
411 (print_insn): Likewise.
412 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
413 (INSN_DEBUG_MARKER_KIND): Likewise.
414 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
415 * tree-inline.c (expand_call_inline): Build and insert
416 debug_inline_entry stmt.
417 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
418 inline entry blocks early, if nonbind markers are enabled.
419 (dump_scope_block): Dump fragment info.
420 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
421 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
422 (gimple_build_debug_inline_entry): New.
423 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
424 Enable/disable inline entry points too.
425 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
426 (DEBUG_INSN): Describe inline entry markers.
428 * common.opt (gvariable-location-views): New.
429 (gvariable-location-views=incompat5): New.
430 * config.in: Rebuilt.
431 * configure: Rebuilt.
432 * configure.ac: Test assembler for view support.
433 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
434 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
435 * dwarf2out.c (var_loc_view): New typedef.
436 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
437 (dwarf2out_locviews_in_attribute): New.
438 (dwarf2out_locviews_in_loclist): New.
439 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
440 (enum dw_line_info_opcode): Add LI_adv_address.
441 (struct dw_line_info_table): Add view.
442 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
443 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
444 (zero_view_p): New variable.
445 (ZERO_VIEW_P): New macro.
446 (output_asm_line_debug_info): New.
447 (struct var_loc_node): Add view.
448 (add_AT_view_list, AT_loc_list): New.
449 (add_var_loc_to_decl): Add view param. Test it against last.
450 (new_loc_list): Add view params. Record them.
451 (AT_loc_list_ptr): Handle loc and view lists.
452 (view_list_to_loc_list_val_node): New.
453 (print_dw_val): Handle dw_val_class_view_list.
454 (size_of_die): Likewise.
455 (value_format): Likewise.
456 (loc_list_has_views): New.
457 (gen_llsym): Set vl_symbol too.
458 (maybe_gen_llsym, skip_loc_list_entry): New.
459 (dwarf2out_maybe_output_loclist_view_pair): New.
460 (output_loc_list): Output view list or entries too.
461 (output_view_list_offset): New.
462 (output_die): Handle dw_val_class_view_list.
463 (output_dwarf_version): New.
464 (output_compilation_unit_header): Use it.
465 (output_skeleton_debug_sections): Likewise.
466 (output_rnglists, output_line_info): Likewise.
467 (output_pubnames, output_aranges): Update version comments.
468 (output_one_line_info_table): Output view numbers in asm comments.
469 (dw_loc_list): Determine current endview, pass it to new_loc_list.
470 Call maybe_gen_llsym.
471 (loc_list_from_tree_1): Adjust.
472 (add_AT_location_description): Create view list attribute if
473 needed, check it's absent otherwise.
474 (convert_cfa_to_fb_loc_list): Adjust.
475 (maybe_emit_file): Call output_asm_line_debug_info for test.
476 (dwarf2out_var_location): Reset views as needed. Precompute
477 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
479 (new_line_info_table): Reset next view.
480 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
481 (dwarf2out_source_line): Likewise. Output view resets and labels to
482 the assembler, or select appropriate line info opcodes.
483 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
484 (optimize_string_length): Catch it. Adjust.
485 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
486 dw_val_class_view_list, and remove it if no longer needed.
487 (hash_loc_list): Hash view numbers.
488 (loc_list_hasher::equal): Compare them.
489 (optimize_location_lists): Check whether a view list symbol is
490 needed, and whether the locview attribute is present, and
491 whether they match. Remove the locview attribute if no longer
493 (index_location_lists): Call skip_loc_list_entry for test.
494 (dwarf2out_finish): Call output_asm_line_debug_info for test.
495 Use output_dwarf_version.
496 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
497 (struct dw_val_node): Add val_view_list.
498 * final.c (SEEN_NEXT_VIEW): New.
499 (set_next_view_needed): New.
500 (clear_next_view_needed): New.
501 (maybe_output_next_view): New.
502 (final_start_function): Rename to...
503 (final_start_function_1): ... this. Take pointer to FIRST,
504 add SEEN parameter. Emit param bindings in the initial view.
505 (final_start_function): Reintroduce SEEN-less interface.
506 (final): Rename to...
507 (final_1): ... this. Take SEEN parameter. Output final pending
508 next view at the end.
509 (final): Reintroduce seen-less interface.
510 (final_scan_insn): Output pending next view before switching
511 sections or ending a block. Mark the next view as needed when
512 outputting variable locations. Notify debug backend of section
513 changes, and of location view changes.
514 (rest_of_handle_final): Adjust.
515 * toplev.c (process_options): Autodetect value for debug variable
516 location views option. Warn on incompat5 without -gdwarf-5.
517 * doc/invoke.texi (gvariable-location-views): New.
518 (gvariable-location-views=incompat5): New.
519 (gno-variable-location-views): New.
521 2018-02-08 David Malcolm <dmalcolm@redhat.com>
523 PR tree-optimization/84136
524 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
525 that the result of find_edge is non-NULL.
527 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
530 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
531 storing integer register in SImode. Fix cost of 256 and 512
532 byte aligned SSE register store.
534 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
536 * config/i386/i386.c (ix86_multiplication_cost): Fix
537 multiplication cost for TARGET_AVX512DQ.
539 2018-02-08 Marek Polacek <polacek@redhat.com>
541 PR tree-optimization/84238
542 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
545 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
547 PR tree-optimization/84265
548 * tree-vect-stmts.c (vectorizable_store): Don't treat
549 VMAT_CONTIGUOUS accesses as grouped.
550 (vectorizable_load): Likewise.
552 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
554 PR tree-optimization/81635
555 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
556 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
557 (test_round_for_mask): New functions.
558 (wide_int_cc_tests): Call test_round_for_mask.
559 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
560 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
561 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
562 range returned by get_range_info.
564 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
567 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
568 * symtab.c: Include builtins.h
569 (symtab_node::output_to_lto_symbol_table_p): Move here
570 from lto-streamer-out.c:output_symbol_p.
571 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
572 (output_symbol_p): Move all logic to symtab.c
573 (produce_symtab): Update.
575 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
577 * config/s390/s390-opts.h (enum indirect_branch): Define.
578 * config/s390/s390-protos.h (s390_return_addr_from_memory)
579 (s390_indirect_branch_via_thunk)
580 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
581 (enum s390_indirect_branch_type): Define.
582 * config/s390/s390.c (struct s390_frame_layout, struct
583 machine_function): Remove.
584 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
585 (indirect_branch_table_label_no, indirect_branch_table_name):
587 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
588 (enum s390_indirect_branch_option): Define.
589 (s390_return_addr_from_memory): New function.
590 (s390_handle_string_attribute): New function.
591 (s390_attribute_table): Add new attribute handler.
592 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
593 (s390_indirect_branch_via_thunk): New function.
594 (s390_indirect_branch_via_inline_thunk): New function.
595 (s390_function_ok_for_sibcall): When jumping via thunk disallow
596 sibling call optimization for non z10 compiles.
597 (s390_emit_call): Force indirect branch target to be a single
598 register. Add r1 clobber for non-z10 compiles.
599 (s390_emit_epilogue): Emit return jump via return_use expander.
600 (s390_reorg): Handle JUMP_INSNs as execute targets.
601 (s390_option_override_internal): Perform validity checks for the
602 new command line options.
603 (s390_indirect_branch_attrvalue): New function.
604 (s390_indirect_branch_settings): New function.
605 (s390_set_current_function): Invoke s390_indirect_branch_settings.
606 (s390_output_indirect_thunk_function): New function.
607 (s390_code_end): Implement target hook.
608 (s390_case_values_threshold): Implement target hook.
609 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
611 * config/s390/s390.h (struct s390_frame_layout)
612 (struct machine_function): Move here from s390.c.
613 (TARGET_INDIRECT_BRANCH_NOBP_RET)
614 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
615 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
616 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
617 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
618 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
619 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
620 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
621 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
622 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
623 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
624 (mnemonic attribute): Add values which aren't recognized
626 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
627 pattern for branch conversion. Fix mnemonic attribute.
628 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
629 indirect branch via thunk if requested.
630 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
631 ("*indirect_jump"): Disable for branch conversion using out of
633 ("indirect_jump_via_thunk<mode>_z10")
634 ("indirect_jump_via_thunk<mode>")
635 ("indirect_jump_via_inlinethunk<mode>_z10")
636 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
637 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
638 ("casesi_jump_via_inlinethunk<mode>_z10")
639 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
640 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
641 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
642 ("*indirect2_jump"): Disable for branch conversion.
643 ("casesi_jump"): Turn into expander and expand patterns for branch
645 ("return_use"): New expander.
646 ("*return"): Emit return via thunk and rename it to ...
647 ("*return<mode>"): ... this one.
648 * config/s390/s390.opt: Add new options and and enum for the
651 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
653 * lra-constraints.c (match_reload): Unconditionally use
654 gen_lowpart_SUBREG, rather than selecting between that
655 and equivalent gen_rtx_SUBREG code.
657 2018-02-08 Richard Biener <rguenther@suse.de>
659 PR tree-optimization/84233
660 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
661 changed flag instead of boguously re-using phi_inserted.
663 2018-02-08 Martin Jambor <mjambor@suse.cz>
665 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
666 static local variables.
668 2018-02-08 Richard Biener <rguenther@suse.de>
670 PR tree-optimization/84278
671 * tree-vect-stmts.c (vectorizable_store): When looking for
672 smaller vector types to perform grouped strided loads/stores
673 make sure the mode is supported by the target.
674 (vectorizable_load): Likewise.
676 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
678 * config/aarch64/aarch64.c (aarch64_components_for_bb):
679 Increase LDP/STP opportunities by adding adjacent callee-saves.
681 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
683 PR rtl-optimization/84068
684 PR rtl-optimization/83459
685 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
687 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
689 PR tree-optimization/84224
690 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
691 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
694 2018-02-07 Iain Sandoe <iain@codesourcery.com>
697 * config/rs6000/altivec.md (*restore_world): Remove LR use.
698 * config/rs6000/predicates.md (restore_world_operation): Adjust op
699 count, remove one USE.
701 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
703 * doc/install.texi (Configuration): Document the
704 --with-long-double-format={ibm,ieee} PowerPC configuration
708 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
709 Convert from define_expand to be define_insn_and_split. Rework
710 float/double/_Float128 conversions to QI/HI/SImode to work with
711 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
712 conversions to QI/HImode types did a store and then a load to
713 truncate the value. For conversions to VSX registers, don't split
714 the insn, instead emit the code directly. Use the code iterator
715 any_fix to combine signed and unsigned conversions.
716 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
717 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
718 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
719 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
720 (fix_<mode>di2_hw): Likewise.
721 (fixuns_<mode>di2_hw): Likewise.
722 (fix_<mode>si2_hw): Likewise.
723 (fixuns_<mode>si2_hw): Likewise.
724 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
725 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
726 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
727 fix<uns>_trunc<SFDF:mode>si2_p8.
728 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
730 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
731 (fix<uns>_<mode>_mem): Likewise.
732 (fctiw<u>z_<mode>_mem): Likewise.
733 (fix<uns>_<mode>_mem): Likewise.
734 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
735 the register allocator from doing a direct move to the GPRs to do
736 a store, and instead use the ISA 3.0 store byte/half-word from
737 vector register instruction. For IEEE 128-bit floating point,
738 also optimize stores of 32-bit ints.
739 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
741 2018-02-07 Alan Hayward <alan.hayward@arm.com>
743 * genextract.c (push_pathstr_operand): New function to support
745 (walk_rtx): Call push_pathstr_operand.
746 (print_path): Support [a-zA-Z].
748 2018-02-07 Richard Biener <rguenther@suse.de>
750 PR tree-optimization/84037
751 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
752 (cse_and_gimplify_to_preheader): Declare.
753 (vect_get_place_in_interleaving_chain): Likewise.
754 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
756 (_loop_vec_info::~_loop_vec_info): Delete it.
757 (cse_and_gimplify_to_preheader): New function.
758 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
759 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
760 (vectorizable_load): Likewise. For grouped stores always base
761 the IV on the first element.
762 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
763 condition before gimplifying.
765 2018-02-07 Jakub Jelinek <jakub@redhat.com>
767 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
768 *DIV_EXPR and *MOD_EXPR.
770 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
773 * config/i386/i386.c (ix86_option_override_internal): Mask out
774 the CF_SET bit when checking -fcf-protection.
776 2018-02-07 Tom de Vries <tom@codesourcery.com>
779 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
782 2018-02-07 Richard Biener <rguenther@suse.de>
784 PR tree-optimization/84204
785 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
788 PR tree-optimization/84205
789 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
790 special-case isl_ast_op_zdiv_r.
792 PR tree-optimization/84223
793 * graphite-scop-detection.c (gather_bbs::before_dom_children):
794 Only add conditions from within the region.
795 (gather_bbs::after_dom_children): Adjust.
797 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
800 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
801 * config/avr/avr.md: Only post-reload split REG-REG moves if
802 either register is REGERAL_REG_P.
804 2018-02-07 Jakub Jelinek <jakub@redhat.com>
806 PR tree-optimization/84235
807 * tree-ssa-scopedtables.c
808 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
809 if the subtraction is performed in floating point type where NaNs are
810 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
811 build 1. Formatting fix.
813 2018-02-06 Jakub Jelinek <jakub@redhat.com>
816 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
817 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
818 and skip it regardless of bb boundaries. Use CALL_P macro,
819 don't test INSN_P (insn) together with CALL_P or JUMP_P check
820 unnecessarily, formatting fix.
822 2018-02-06 Michael Collison <michael.collison@arm.com>
824 * config/arm/thumb2.md:
825 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
826 (*thumb_mov_notscc): Ditto.
828 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
831 * config/rs6000/rs6000.md (su code attribute): Use "u" for
832 unsigned_fix, not "s".
834 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
836 * configure.ac (gcc_fn_eh_frame_ro): New function.
837 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
838 correct .eh_frame permissions.
839 * configure: Regenerate.
841 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
843 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
846 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
848 * config/rs6000/rs6000.c (rs6000_option_override_internal):
849 Display warning message for -mno-speculate-indirect-jumps.
851 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
853 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
855 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
857 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
859 PR tree-optimization/84225
860 * tree-eh.c (find_trapping_overflow): Only call
861 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
863 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
866 * config/i386/i386.c: Reimplement the check of possible options
867 -mibt/-mshstk conbination. Change error messages.
868 * doc/invoke.texi: Fix a typo: remove extra '='.
870 2018-02-06 Marek Polacek <polacek@redhat.com>
872 PR tree-optimization/84228
873 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
875 2018-02-06 Tamar Christina <tamar.christina@arm.com>
878 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
879 emitted arch directives.
880 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
881 __ARM_FEATURE_COPROC before changing architectures.
883 2018-02-06 Richard Biener <rguenther@suse.de>
885 * config/i386/i386.c (print_reg): Fix typo.
886 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
888 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
890 * configure: Regenerate.
892 2018-02-05 Martin Sebor <msebor@redhat.com>
894 PR tree-optimization/83369
895 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
898 2018-02-05 Martin Liska <mliska@suse.cz>
900 * doc/invoke.texi: Cherry-pick upstream r323995.
902 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
904 * ira.c (ira_init_register_move_cost): Adjust comment.
906 2018-02-05 Martin Liska <mliska@suse.cz>
908 PR gcov-profile/84137
909 * doc/gcov.texi: Fix typo in documentation.
911 2018-02-05 Martin Liska <mliska@suse.cz>
913 PR gcov-profile/83879
914 * doc/gcov.texi: Document necessity of --dynamic-list-data when
915 using dlopen functionality.
917 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
919 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
920 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
921 _mm_maskz_range_ss, _mm_mask_range_round_ss,
922 _mm_maskz_range_round_ss): New intrinsics.
923 (__builtin_ia32_rangesd128_round)
924 (__builtin_ia32_rangess128_round): Remove.
925 (__builtin_ia32_rangesd128_mask_round,
926 __builtin_ia32_rangess128_mask_round): New builtins.
927 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
928 __builtin_ia32_rangess128_round): Remove.
929 (__builtin_ia32_rangesd128_mask_round,
930 __builtin_ia32_rangess128_mask_round): New builtins.
931 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
932 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
933 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
934 "<round_saeonly_constraint>")): Changed to ...
935 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
936 "<round_saeonly_scalar_constraint>")): ... this.
937 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
938 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
939 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
940 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
941 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
943 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
945 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
947 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
948 Remove all values except native, 8540 and 8548.
950 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
952 * config/i386/i386.c (ix86_output_function_return): Pass
953 INVALID_REGNUM, instead of -1, as invalid register number to
954 indirect_thunk_name and output_indirect_thunk.
956 2018-02-02 Julia Koval <julia.koval@intel.com>
958 * config.gcc: Add -march=icelake.
959 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
960 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
961 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
962 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
963 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
964 (processor_target_table): Add icelake.
965 (ix86_option_override_internal): Handle new PTAs.
966 (get_builtin_code_for_version): Handle icelake.
967 (M_INTEL_COREI7_ICELAKE): New.
968 (fold_builtin_cpu): Handle icelake.
969 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
970 * doc/invoke.texi: Add -march=icelake.
972 2018-02-02 Julia Koval <julia.koval@intel.com>
974 * config/i386/i386.c (ix86_option_override_internal): Change flags type
976 * wide-int-bitmask.h: New.
978 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
981 * config/i386/i386.md: Replace Pmode with word_mode in
982 builtin_setjmp_setup and builtin_longjmp to support x32.
984 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
988 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
990 (rs6000_supported_cpu_names): New static variable.
991 (linux_cpu_translation_table): Likewise.
992 (elf_platform) <cpu>: Define new static variable and use it.
993 Translate kernel AT_PLATFORM name to canonical name if needed.
994 Error if platform name is unknown.
996 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
999 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1001 2018-02-01 Jeff Law <law@redhat.com>
1004 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1005 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1006 the scratch if RELEASE_VIA_POP is false.
1007 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1008 If we have to save a temporary register, decrement SIZE appropriately.
1009 Pass new arguments to release_scratch_register_on_entry.
1010 (ix86_adjust_stack_and_probe): Likewise.
1011 (ix86_emit_probe_stack_range): Pass new arguments to
1012 release_scratch_register_on_entry.
1014 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1016 PR rtl-optimization/84157
1017 * combine.c (change_zero_ext): Use REG_P predicate in
1018 front of HARD_REGISTER_P predicate.
1020 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1022 * config/avr/avr.c (avr_option_override): Move disabling of
1023 -fdelete-null-pointer-checks to...
1024 * common/config/avr/avr-common.c (avr_option_optimization_table):
1027 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1029 PR tree-optimization/81635
1030 * tree-data-ref.c (split_constant_offset_1): For types that
1031 wrap on overflow, try to use range info to prove that wrapping
1034 2018-02-01 Renlin Li <renlin.li@arm.com>
1037 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1039 (aarch64_register_move_cost): Likewise.
1040 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1042 (REG_CLASS_NAMES): Likewise.
1043 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1044 TAILCALL_ADDR_REGS. Remove IP registers.
1045 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1047 2018-02-01 Richard Biener <rguenther@suse.de>
1049 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1050 for specifying RPO order and allow NULL for that.
1051 * domwalk.c (dom_walker::dom_walker): Likewise.
1052 (dom_walker::walk): Handle NULL RPO order.
1053 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1055 (rewrite_update_dom_walker): Likewise.
1056 (mark_def_dom_walker): Likewise.
1058 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1060 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1061 (aarch64_maybe_expand_sve_subreg_move): Declare.
1062 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1063 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1065 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1066 that are semantically a reverse operation.
1067 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1068 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1069 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1071 (aarch64_can_change_mode_class): For big-endian, forbid changes
1072 between two SVE modes if they have different element sizes.
1074 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1076 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1077 the TImode handling for big-endian targets.
1079 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1081 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1082 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1084 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1085 Remove BSWAP handing for big-endian targets and use the form of
1086 LD1RQ appropariate for the mode.
1088 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1090 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1091 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1094 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1097 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1098 check for operands that need to go through aarch64_sve_reload_be.
1100 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1102 PR tree-optimization/81661
1103 PR tree-optimization/84117
1104 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1105 * tree-eh.c: Include gimplify.h.
1106 (find_trapping_overflow, replace_trapping_overflow,
1107 rewrite_to_non_trapping_overflow): New functions.
1108 * tree-vect-loop.c: Include tree-eh.h.
1109 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1110 * tree-data-ref.c: Include tree-eh.h.
1111 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1113 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1115 PR rtl-optimization/84123
1116 * combine.c (change_zero_ext): Check if hard register satisfies
1117 can_change_dest_mode before calling gen_lowpart_SUBREG.
1119 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1122 * ira.c (ira_init_register_move_cost): Remove assert.
1124 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1126 PR rtl-optimization/84071
1127 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1128 * doc/tm.texi: Regenerate.
1130 2018-01-31 Richard Biener <rguenther@suse.de>
1132 PR tree-optimization/84132
1133 * tree-data-ref.c (analyze_miv_subscript): Properly
1134 check whether evolution_function_is_affine_multivariate_p
1135 before calling gcd_of_steps_may_divide_p.
1137 2018-01-31 Julia Koval <julia.koval@intel.com>
1140 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1141 * config/i386/i386.md (rdpid_rex64) New.
1142 (rdpid): Make 32bit only.
1144 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1147 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1148 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1150 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1153 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1155 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1157 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1159 PR rtl-optimization/84071
1160 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1161 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1163 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1165 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1166 (arc_attribute_table): Add 'aux' attribute.
1167 (arc_in_small_data_p): Consider aux like variables.
1168 (arc_is_aux_reg_p): New function.
1169 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1170 (arc_get_aux_arg): New function.
1171 (prepare_move_operands): Handle aux-register access.
1172 (arc_handle_aux_attribute): New function.
1173 * doc/extend.texi (ARC Variable attributes): Add subsection.
1175 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1177 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1178 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1179 (arc_attribute_table): Add 'uncached' attribute.
1180 (arc_print_operand): Print '.di' flag for uncached memory
1182 (arc_in_small_data_p): Do not consider for small data the uncached
1184 (arc_is_uncached_mem_p): New function.
1185 * config/arc/predicates.md (compact_store_memory_operand): Check
1186 for uncached memory accesses.
1187 (nonvol_nonimm_operand): Likewise.
1188 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1190 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1193 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1194 falign-loops=): Add Optimization flag.
1196 2018-01-30 Jeff Law <law@redhat.com>
1199 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1200 INT_REGISTERS_SAVED. Check it prior to calling
1201 get_scratch_register_on_entry.
1202 (ix86_adjust_stack_and_probe): Similarly.
1203 (ix86_emit_probe_stack_range): Similarly.
1204 (ix86_expand_prologue): Corresponding changes.
1206 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1209 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1210 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1212 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1215 * lra-constraints.c (curr_insn_transform): Process AND in the
1218 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1220 PR rtl-optimization/83986
1221 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1222 dependence against last_pending_memory_flush in addition to
1225 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1227 PR tree-optimization/81611
1228 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1231 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1233 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1236 2018-01-30 Richard Biener <rguenther@suse.de>
1237 Jakub Jelinek <jakub@redhat.com>
1239 PR tree-optimization/84111
1240 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1241 inner loops added during recursion, as they don't have up-to-date
1244 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1247 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1248 (can_inline_edge_by_limits_p): ... here.
1249 (can_early_inline_edge_p, check_callers,
1250 update_caller_keys, update_callee_keys, recursive_inlining,
1251 add_new_edges_to_heap, speculation_useful_p,
1252 inline_small_functions,
1253 inline_small_functions, flatten_function,
1254 inline_to_all_callers_1): Update.
1256 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1258 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1261 2018-01-30 Richard Biener <rguenther@suse.de>
1263 PR tree-optimization/83008
1264 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1265 invariant and constant vector uses in stmts when they need
1268 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1271 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1272 * configure: Regenerate.
1274 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1276 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1278 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1279 Use gen_rtx_REG rather than gen_lowpart.
1281 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1283 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1284 rather than 0 when creating partial subregs.
1286 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1288 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1291 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1294 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1295 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1296 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1297 flags. This restores the settings used before the 2017-07-24.
1298 Turning off pre increment/decrement/modify allows IVOPTS to
1299 optimize DF/SF loops where the index is an int.
1301 2018-01-29 Richard Biener <rguenther@suse.de>
1302 Kelvin Nilsen <kelvin@gcc.gnu.org>
1305 * tree-vect-stmts.c (vectorizable_call): Don't call
1306 targetm.vectorize_builtin_md_vectorized_function if callee is
1309 2018-01-22 Carl Love <cel@us.ibm.com>
1311 * doc/extend.tex: Fix typo in second arg in
1312 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1314 2018-01-29 Richard Biener <rguenther@suse.de>
1316 PR tree-optimization/84086
1317 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1318 (flush_ssaname_freelist): When SSA names were released reset
1319 the SCEV hash table.
1321 2018-01-29 Richard Biener <rguenther@suse.de>
1323 PR tree-optimization/84057
1324 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1325 removed paths when removing edges.
1327 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1329 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1330 -mfunction-return=@var{choice}.
1332 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1335 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1337 (layout::print_source_line): Likewise.
1338 (test_get_line_width_without_trailing_whitespace): Add test cases.
1340 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1343 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1346 2018-01-26 Jim Wilson <jimw@sifive.com>
1348 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1350 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1353 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1355 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1356 and CMP + SUB-immediate -> SUBS.
1358 2018-01-26 Martin Sebor <msebor@redhat.com>
1360 PR tree-optimization/83896
1361 * tree-ssa-strlen.c (get_string_len): Rename...
1362 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1363 Avoid assuming length is constant.
1364 (handle_char_store): Use HOST_WIDE_INT for string length.
1366 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1369 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1370 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1372 2018-01-26 Richard Biener <rguenther@suse.de>
1374 PR rtl-optimization/84003
1375 * dse.c (record_store): Only record redundant stores when
1376 the earlier store aliases at least all accesses the later one does.
1378 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1380 PR rtl-optimization/83985
1381 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1382 REG_CFA_RESTORE insns.
1383 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1384 REG_CFA_RESTORE insns here.
1387 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1388 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1390 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1392 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1393 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1394 (arc_init): Likewise.
1395 (arc_override_options): Likewise.
1396 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1398 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1400 * config/arc/arc.h (TARGET_DBNZ): Define.
1401 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1402 properly set the tune attribute.
1403 (dbnz): Use TARGET_DBNZ guard.
1404 * config/arc/arc.opt (mtune): Add core3 option.
1406 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1408 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1409 recognize new pic like addresses.
1410 (arc_delegitimize_address): Clean up.
1412 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1414 * config/arc/arc-arches.def: Option mrf16 valid for all
1416 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1417 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1418 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1419 * config/arc/arc-tables.opt: Regenerate.
1420 * config/arc/arc.c (arc_conditional_register_usage): Handle
1421 reduced register file case.
1422 (arc_file_start): Set must have build attributes.
1423 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1425 * config/arc/arc.opt (mrf16): Add new option.
1426 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1427 * config/arc/genmultilib.awk: Handle new mrf16 option.
1428 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1429 * config/arc/t-multilib: Regenerate.
1430 * doc/invoke.texi (ARC Options): Document mrf16 option.
1432 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1434 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1435 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1436 (arc_attribute_table): Add 'secure_call' attribute.
1437 (arc_print_operand): Print secure call operand.
1438 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1440 (arc_is_secure_call_p): New function. * config/arc/arc.md
1441 (call_i): Add support for sjli instruction.
1442 (call_value_i): Likewise.
1443 * config/arc/constraints.md (Csc): New constraint.
1445 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1446 John Eric Martin <John.Martin@emmicro-us.com>
1448 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1449 * config/arc/arc.c (_arc_jli_section): New struct.
1450 (arc_jli_section): New type.
1451 (rc_jli_sections): New static variable.
1452 (arc_handle_jli_attribute): New function.
1453 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1454 (arc_file_end): New function.
1455 (TARGET_ASM_FILE_END): Define.
1456 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1457 (arc_add_jli_section): New function.
1458 (jli_call_scan): Likewise.
1459 (arc_reorg): Call jli_call_scan.
1460 (arc_output_addsi): Remove 'S' from printing asm operand.
1461 (arc_is_jli_call_p): New function.
1462 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1464 (movhi_insn): Likewise.
1465 (movsi_insn): Likewise.
1466 (movsi_set_cc_insn): Likewise.
1467 (loadqi_update): Likewise.
1468 (load_zeroextendqisi_update): Likewise.
1469 (load_signextendqisi_update): Likewise.
1470 (loadhi_update): Likewise.
1471 (load_zeroextendhisi_update): Likewise.
1472 (load_signextendhisi_update): Likewise.
1473 (loadsi_update): Likewise.
1474 (loadsf_update): Likewise.
1475 (movsicc_insn): Likewise.
1476 (bset_insn): Likewise.
1477 (bxor_insn): Likewise.
1478 (bclr_insn): Likewise.
1479 (bmsk_insn): Likewise.
1480 (bicsi3_insn): Likewise.
1481 (cmpsi_cc_c_insn): Likewise.
1482 (movsi_ne): Likewise.
1483 (movsi_cond_exec): Likewise.
1484 (clrsbsi2): Likewise.
1495 (call_i): Remove 'S' asm letter, add jli instruction.
1496 (call_value_i): Likewise.
1497 * config/arc/arc.op (mjli-always): New option.
1498 * config/arc/constraints.md (Cji): New constraint.
1499 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1501 (subsf3_fpx): Likewise.
1502 (mulsf3_fpx): Likewise.
1503 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1505 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1507 * doc/invoke.texi (ARC): Document mjli-always option.
1509 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1511 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1512 avoid addition with 0 and use incw and decw where possible.
1514 2018-01-26 Richard Biener <rguenther@suse.de>
1516 PR tree-optimization/81082
1517 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1518 association if it requires casting to unsigned.
1519 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1520 from fold_plusminus_mult_expr to catch important cases late when
1521 range info is available.
1523 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1525 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1526 * configure.ac (hidden_linkonce): New test.
1527 * configure: Regenerate.
1528 * config.in: Regenerate.
1530 2018-01-26 Julia Koval <julia.koval@intel.com>
1532 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1533 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1534 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1535 _mm_mask_bitshuffle_epi64_mask): Fix type.
1536 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1537 USI_FTYPE_V4DI_V4DI_USI): Remove.
1538 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1539 __builtin_ia32_vpshufbitqmb256_mask,
1540 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1541 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1542 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1544 2018-01-26 Alan Modra <amodra@gmail.com>
1547 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1548 UNSPEC_VBPERMQ. Sort other unspecs.
1550 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1552 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1554 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1557 * predict.c (drop_profile): Do not push/pop cfun; update also
1559 (handle_missing_profiles): Fix logic looking for zero profiles.
1561 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1564 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1565 on functions with #pragma omp declare simd or functions with simd
1567 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1568 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1569 Remove trailing \n from warning_at calls.
1571 2018-01-25 Tom de Vries <tom@codesourcery.com>
1574 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1575 for neutered workers.
1577 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1580 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1581 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1583 2018-01-24 Jeff Law <law@redhat.com>
1586 * i386.c (get_probe_interval): Move to earlier point.
1587 (ix86_compute_frame_layout): If -fstack-clash-protection and
1588 the frame is larger than the probe interval, then use pushes
1589 to save registers rather than reg->mem moves.
1590 (ix86_expand_prologue): Remove conditional for int_registers_saved
1593 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1596 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1597 min/max for never referenced object.
1599 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1602 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1604 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1605 attributes from DECL_ATTRIBUTES (decl) without affecting
1606 DECL_ATTRIBUTES (current_function_decl).
1607 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1608 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1610 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1612 PR tree-optimization/83979
1613 * fold-const.c (fold_comparison): Use constant_boolean_node
1614 instead of boolean_{true,false}_node.
1616 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1618 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1621 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1623 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1624 Simplify the clause that sets the length attribute.
1625 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1626 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1627 clause that sets the length attribute.
1628 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1630 2018-01-24 Tom de Vries <tom@codesourcery.com>
1633 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1634 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1635 Add strict parameter.
1636 (prevent_branch_around_nothing): Insert dummy insn between branch to
1637 label and label with no ptx insn inbetween.
1638 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1640 2018-01-24 Tom de Vries <tom@codesourcery.com>
1643 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1644 for neutered threads in warp.
1645 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1647 2018-01-24 Richard Biener <rguenther@suse.de>
1649 PR tree-optimization/83176
1650 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1653 2018-01-24 Richard Biener <rguenther@suse.de>
1655 PR tree-optimization/82819
1656 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1657 code generating pluses that are no-ops in the target precision.
1659 2018-01-24 Richard Biener <rguenther@suse.de>
1662 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1664 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1666 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1667 to merge probabilities.
1668 * predict.c (probably_never_executed): Also mark as cold functions
1669 with global 0 profile and guessed local profile.
1670 * profile-count.c (profile_probability::combine_with_count): New
1672 * profile-count.h (profile_probability::operator*,
1673 profile_probability::operator*=, profile_probability::operator/,
1674 profile_probability::operator/=): Reduce precision to adjusted
1675 and set value to guessed on contradictory divisions.
1676 (profile_probability::combine_with_freq): Remove.
1677 (profile_probability::combine_wiht_count): Declare.
1678 (profile_count::force_nonzero):: Set to adjusted.
1679 (profile_count::probability_in):: Set quality to adjusted.
1680 * tree-ssa-tail-merge.c (replace_block_by): Use
1683 2018-01-23 Andrew Waterman <andrew@sifive.com>
1684 Jim Wilson <jimw@sifive.com>
1686 * config/riscv/riscv.c (riscv_stack_boundary): New.
1687 (riscv_option_override): Set riscv_stack_boundary. Handle
1688 riscv_preferred_stack_boundary_arg.
1689 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1690 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1691 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1692 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1693 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1694 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1696 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1699 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1700 of struct ix86_frame.
1701 (ix86_expand_epilogue): Likewise. Add a local variable for
1702 the reg_save_offset field in struct ix86_frame.
1704 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1706 PR tree-optimization/82604
1707 * tree-loop-distribution.c (enum partition_kind): New enum item
1708 PKIND_PARTIAL_MEMSET.
1709 (partition_builtin_p): Support above new enum item.
1710 (generate_code_for_partition): Ditto.
1711 (compute_access_range): Differentiate cases that equality can be
1712 proven at all loops, the innermost loops or no loops.
1713 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1714 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1715 (finalize_partitions, distribute_loop): Don't fuse partition of
1716 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1717 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1720 2018-01-23 Martin Liska <mliska@suse.cz>
1722 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1723 order to ignore the predictor.
1724 (PRED_POLYMORPHIC_CALL): Likewise.
1725 (PRED_RECURSIVE_CALL): Likewise.
1727 2018-01-23 Martin Liska <mliska@suse.cz>
1729 * tree-profile.c (tree_profiling): Print function header to
1730 aware reader which function we are working on.
1731 * value-prof.c (gimple_find_values_to_profile): Do not print
1732 not interesting value histograms.
1734 2018-01-23 Martin Liska <mliska@suse.cz>
1736 * profile-count.h (enum profile_quality): Add
1737 profile_uninitialized as the first value. Do not number values
1738 as they are zero based.
1739 (profile_count::verify): Update sanity check.
1740 (profile_probability::verify): Likewise.
1742 2018-01-23 Nathan Sidwell <nathan@acm.org>
1744 * doc/invoke.texi (ffor-scope): Deprecate.
1746 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1748 PR tree-optimization/83510
1749 * domwalk.c (set_all_edges_as_executable): New function.
1750 (dom_walker::dom_walker): Convert bool param
1751 "skip_unreachable_blocks" to enum reachability. Move setup of
1752 edge flags to set_all_edges_as_executable and only do it when
1753 reachability is REACHABLE_BLOCKS.
1754 * domwalk.h (enum dom_walker::reachability): New enum.
1755 (dom_walker::dom_walker): Convert bool param
1756 "skip_unreachable_blocks" to enum reachability.
1757 (set_all_edges_as_executable): New decl.
1758 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1759 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1761 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1762 but converting true to REACHABLE_BLOCKS.
1763 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1765 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1766 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1767 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1769 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1770 if check_all_array_refs will be called.
1772 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1774 * tree.c (selftest::test_location_wrappers): Add more test
1777 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1779 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1780 (selftest::test_bit_in_range): Likewise.
1782 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1785 * doc/sourcebuild.texi (vect_float): Say that the selector
1786 only describes the situation when -funsafe-math-optimizations is on.
1787 (vect_float_strict): Document.
1789 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1791 PR tree-optimization/83965
1792 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1793 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1794 instead of checking only for a reduction.
1795 (vect_recog_widen_sum_pattern): Likewise.
1797 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1799 * predict.c (probably_never_executed): Only use precise profile info.
1800 (compute_function_frequency): Skip after inlining hack since we now
1801 have quality checking.
1803 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1805 * profile-count.h (profile_probability::very_unlikely,
1806 profile_probability::unlikely, profile_probability::even): Set
1807 precision to guessed.
1809 2018-01-23 Richard Biener <rguenther@suse.de>
1811 PR tree-optimization/83963
1812 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1813 Properly terminate dominator walk when crossing the exit edge not
1814 when visiting its source block.
1816 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1819 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1820 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1822 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1824 PR tree-optimization/83957
1825 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1826 semicolon after for body surrounded by braces.
1828 PR tree-optimization/83081
1829 * profile-count.h (profile_probability::split): New method.
1830 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1831 Use profile_probability::split.
1832 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1833 when splitting a single conditional jump into 2.
1835 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1837 PR tree-optimization/69452
1838 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1841 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1843 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1844 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1845 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1847 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1849 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1850 * config/rl78/rl78.md (movdi): New define_expand.
1851 * config/rl78/rl78.c (rl78_split_movdi): New function.
1853 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1856 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1858 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1859 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1860 128-bit to produce an UNSPEC move to get the double word with the
1861 signbit and then a shift directly to do signbit.
1862 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1863 implementation with a new version that just does either a direct
1864 move or a regular move. Move memory interface to separate insns.
1865 Move insns so they are next to the expander.
1866 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1867 with signbit move. Split big and little endian case.
1868 (signbit<mode>2_dm_mem_le): Likewise.
1869 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1870 (signbit<mode>2_dm2): Likewise.
1872 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1874 * config/rl78/rl78.md (anddi3): New define_expand.
1876 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1878 * config/rl78/rl78.md (umindi3): New define_expand.
1880 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1882 * config/rl78/rl78.md (smindi3): New define_expand.
1884 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1886 * config/rl78/rl78.md (smaxdi3): New define_expand.
1888 2018-01-22 Carl Love <cel@us.ibm.com>
1890 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1891 LVX_V1TI): Add macro expansion.
1892 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1893 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1894 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1895 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1896 Change check to determine if the instruction is a byte reversing
1897 entry. Fix typo in comment.
1898 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1899 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1900 Add def_builtin calls for new builtins.
1901 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1902 Add define_insn expansion.
1904 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1906 * config/rl78/rl78.md (umaxdi3): New define_expand.
1908 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1910 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
1911 for non-QImode registers.
1913 2018-01-22 Richard Biener <rguenther@suse.de>
1915 PR tree-optimization/83963
1916 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1917 including the loop exit block.
1918 (scop_detection::merge_sese): Likewise.
1919 (scop_detection::add_scop): Do it here instead.
1921 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1923 * doc/sourcebuild.texi (arm_softfloat): Document.
1925 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1928 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1929 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1930 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1932 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1933 David Edelsohn <dje.gcc@gmail.com>
1936 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1937 Change "crset eq" to "crset 2".
1938 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1939 (*call_indirect_aix<mode>_nospec): Likewise.
1940 (*call_value_indirect_aix<mode>_nospec): Likewise.
1941 (*call_indirect_elfv2<mode>_nospec): Likewise.
1942 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1943 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1944 change assembly output from . to $.
1945 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1946 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1947 (*tablejump<mode>_internal1_nospec): Likewise.
1949 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1952 * config/sh/sh_optimize_sett_clrt.cc:
1953 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1955 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1957 PR tree-optimization/83940
1958 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1959 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1960 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1961 use it to pass back the definition type.
1962 (vect_check_store_rhs): Likewise rhs_dt_out.
1963 (vect_build_gather_load_calls): Add a mask_dt argument and use
1964 it instead of a call to vect_is_simple_use.
1965 (vectorizable_store): Update calls to vect_check_load_store_mask
1966 and vect_check_store_rhs. Use the dt returned by the latter instead
1967 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1968 instead of calls to vect_is_simple_use. Pass the scalar rather
1969 than the vector operand to vect_is_simple_use when handling
1970 second and subsequent copies of an rhs value.
1971 (vectorizable_load): Update calls to vect_check_load_store_mask
1972 and vect_build_gather_load_calls. Use the cached mask_dt and
1973 gs_info.offset_dt instead of calls to vect_is_simple_use.
1975 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1978 * tree-emutls.c: Include gimplify.h.
1979 (lower_emutls_2): New function.
1980 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1981 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1982 it before further processing.
1985 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1986 UINTVAL (trueop1) instead of INTVAL (op1).
1988 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1992 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1993 INCOMING_FRAME_SP_OFFSET if not defined.
1994 (scan_trace): Add ENTRY argument. If true and
1995 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1996 emit a note to adjust the CFA offset.
1997 (create_cfi_notes): Adjust scan_trace callers.
1998 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1999 INCOMING_FRAME_SP_OFFSET in the CIE.
2000 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2001 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2003 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2004 * doc/tm.texi: Regenerated.
2006 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2008 PR rtl-optimization/83147
2009 * lra-constraints.c (remove_inheritance_pseudos): Use
2010 lra_substitute_pseudo_within_insn.
2012 2018-01-19 Tom de Vries <tom@codesourcery.com>
2013 Cesar Philippidis <cesar@codesourcery.com>
2016 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2018 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2021 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2022 spaces for function labels.
2024 2018-01-19 Martin Liska <mliska@suse.cz>
2026 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2027 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2028 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2029 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2030 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2031 (PRED_CONST_RETURN): Change from 69 to 65.
2032 (PRED_NULL_RETURN): Change from 91 to 71.
2033 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2034 (PRED_LOOP_GUARD): Change from 66 to 73.
2036 2018-01-19 Martin Liska <mliska@suse.cz>
2038 * predict.c (predict_insn_def): Add new assert.
2039 (struct branch_predictor): Change type to signed integer.
2040 (test_prediction_value_range): Amend test to cover
2042 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2043 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2044 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2045 (PRED_LOOP_IV_COMPARE): Likewise.
2046 * predict.h (PROB_UNINITIALIZED): Define new constant.
2048 2018-01-19 Martin Liska <mliska@suse.cz>
2050 * predict.c (dump_prediction): Add new format for
2051 analyze_brprob.py script which is enabled with -details
2053 * profile-count.h (precise_p): New function.
2055 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2057 PR tree-optimization/83922
2058 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2059 there are no statements that need masking.
2060 (vect_active_double_reduction_p): New function.
2061 (vect_analyze_loop_operations): Use it when handling phis that
2062 are not in the loop header.
2064 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2066 PR tree-optimization/83914
2067 * tree-vect-loop.c (vectorizable_induction): Don't convert
2068 init_expr or apply the peeling adjustment for inductions
2069 that are nested within the vectorized loop.
2071 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2073 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2076 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2080 * function.h (gimplify_parameters): Add gimple_seq * argument.
2081 * function.c: Include gimple.h and options.h.
2082 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2083 for the added local temporaries if needed.
2084 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2085 if there are any parameter cleanups, wrap whole body into a
2086 try/finally with the cleanups.
2088 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2091 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2092 Use GET_MODE_CLASS for scalar floating point.
2094 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2098 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2099 Fix call of call_cgraph_insertion_hooks.
2101 2018-01-18 Martin Sebor <msebor@redhat.com>
2103 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2105 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2108 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2111 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2114 * common.opt: (-ffile-prefix-map): New option.
2115 * opts.c (common_handle_option): Defer it.
2116 * opts-global.c (handle_common_deferred_options): Handle it.
2117 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2118 * file-prefix-map.h: New file.
2119 (remap_debug_filename, add_debug_prefix_map): ...here.
2120 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2121 * final.c (debug_prefix_map, add_debug_prefix_map
2122 remap_debug_filename): Move to...
2123 * file-prefix-map.c: New file.
2124 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2125 generalize, get rid of alloca(), use strrchr() instead of strchr().
2126 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2127 Implement in terms of add_prefix_map().
2128 (remap_macro_filename, remap_debug_filename): Implement in term of
2130 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2131 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2132 * dbxout.c: Include file-prefix-map.h.
2133 * varasm.c: Likewise.
2134 * vmsdbgout.c: Likewise.
2135 * xcoffout.c: Likewise.
2136 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2137 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2138 * doc/invoke.texi (-ffile-prefix-map): Document.
2139 (-fdebug-prefix-map): Update description.
2141 2018-01-18 Martin Liska <mliska@suse.cz>
2143 * config/i386/i386.c (indirect_thunk_name): Document that also
2145 (output_indirect_thunk): Document why both instructions
2146 (pause and lfence) are generated.
2148 2018-01-18 Richard Biener <rguenther@suse.de>
2150 PR tree-optimization/83887
2151 * graphite-scop-detection.c
2152 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2153 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2154 (scop_detection::merge_sese): Re-implement with a flood-fill
2155 algorithm that properly finds a SESE region if it exists.
2157 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2160 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2161 pointer_diff optimizations use view_convert instead of convert.
2163 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2165 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2166 Generate different code for -mno-speculate-indirect-jumps.
2167 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2168 (*call_indirect_aix<mode>): Disable for
2169 -mno-speculate-indirect-jumps.
2170 (*call_indirect_aix<mode>_nospec): New define_insn.
2171 (*call_value_indirect_aix<mode>): Disable for
2172 -mno-speculate-indirect-jumps.
2173 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2174 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2175 -mno-speculate-indirect-jumps.
2176 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2178 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2180 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2181 long double type, set the flags for noting the default long double
2182 type, even if we don't pass or return a long double type.
2184 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2187 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2190 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2192 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2193 support for merge[hl].
2194 (fold_mergehl_helper): New helper function.
2195 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2196 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2197 (altivec_vmrglw_direct): Add xxmrglw insn.
2199 2018-01-17 Andrew Waterman <andrew@sifive.com>
2201 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2202 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2204 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2207 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2208 call the lto_location_cache before reading the
2209 DECL_SOURCE_LOCATION of the types.
2211 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2212 Richard Sandiford <richard.sandiford@linaro.org>
2214 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2215 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2216 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2217 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2218 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2220 * config/aarch64/constraints.md (aarch64_movti_operand):
2222 * config/aarch64/predicates.md (Uti): Add new constraint.
2224 2018-01-17 Carl Love <cel@us.ibm.com>
2225 * config/rs6000/vsx.md (define_expand xl_len_r,
2226 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2227 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2229 (define_expand, define_insn): Move the shift left from the
2230 define_insn to the define_expand for lxvl and stxvl instructions.
2231 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2232 and XL_LEN_R definitions to PURE.
2234 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2236 * config/i386/i386.c (indirect_thunk_name): Declare regno
2237 as unsigned int. Compare regno with INVALID_REGNUM.
2238 (output_indirect_thunk): Ditto.
2239 (output_indirect_thunk_function): Ditto.
2240 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2241 in the call to output_indirect_thunk_function.
2243 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2246 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2247 rather than the size of inner_type to determine the stack slot size
2248 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2250 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2253 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2256 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2258 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2259 endian Linux systems to optionally enable multilibs for selecting
2260 the long double type if the user configured an explicit type.
2261 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2262 have no long double multilibs if not defined.
2263 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2264 warn if the user used -mabi={ieee,ibm}longdouble and we built
2265 multilibs for long double.
2266 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2267 appropriate multilib option.
2268 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2270 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2271 for building long double multilibs.
2272 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2274 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2276 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2279 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2281 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2284 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2287 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2290 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2292 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2293 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2295 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2297 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2298 different rtl trees depending on TARGET_64BIT.
2299 (rs6000_gen_lvx): Likewise.
2301 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2303 * config/visium/visium.md (nop): Tweak comment.
2304 (hazard_nop): Likewise.
2306 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2308 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2309 -mspeculate-indirect-jumps.
2310 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2311 for -mno-speculate-indirect-jumps.
2312 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2313 (*call_value_indirect_elfv2<mode>): Disable for
2314 -mno-speculate-indirect-jumps.
2315 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2316 (indirect_jump): Emit different RTL for
2317 -mno-speculate-indirect-jumps.
2318 (*indirect_jump<mode>): Disable for
2319 -mno-speculate-indirect-jumps.
2320 (*indirect_jump<mode>_nospec): New define_insn.
2321 (tablejump): Emit different RTL for
2322 -mno-speculate-indirect-jumps.
2323 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2324 (tablejumpsi_nospec): New define_expand.
2325 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2326 (tablejumpdi_nospec): New define_expand.
2327 (*tablejump<mode>_internal1): Disable for
2328 -mno-speculate-indirect-jumps.
2329 (*tablejump<mode>_internal1_nospec): New define_insn.
2330 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2333 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2335 * caller-save.c (insert_save): Drop unnecessary parameter. All
2338 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2339 Richard Biener <rguenth@suse.de>
2342 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2343 return early, inline manually is_gimple_sizepos. Make sure if we
2344 call gimplify_expr we don't end up with a gimple constant.
2345 * tree.c (variably_modified_type_p): Don't return true for
2346 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2347 * gimplify.h (is_gimple_sizepos): Remove.
2349 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2351 PR tree-optimization/83857
2352 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2353 vectorizable_live_operation for pure SLP statements.
2354 (vectorizable_live_operation): Handle PHIs.
2356 2018-01-16 Richard Biener <rguenther@suse.de>
2358 PR tree-optimization/83867
2359 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2360 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2362 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2365 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2366 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2367 If off is not INTEGER_CST, issue a may not be aligned warning
2368 rather than isn't aligned. Use isn%'t rather than isn't.
2369 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2371 <case MULT_EXPR>: Improve the case when bottom and one of the
2372 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2373 operand, in that case check if the other operand is multiple of
2374 bottom divided by the INTEGER_CST operand.
2376 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2379 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2380 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2381 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2382 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2383 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2384 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2385 (pa_function_arg_size): New function.
2387 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2389 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2390 in a separate statement.
2392 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2394 PR tree-optimization/83847
2395 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2396 group gathers and scatters.
2398 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2400 PR rtl-optimization/86620
2401 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2403 PR rtl-optimization/83213
2404 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2405 to last if both are JUMP_INSNs.
2407 PR tree-optimization/83843
2408 * gimple-ssa-store-merging.c
2409 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2410 store_immediate_info for bswap/nop orig_stores.
2412 2018-01-15 Andrew Waterman <andrew@sifive.com>
2414 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2416 <UDIV>: Increase cost if !TARGET_DIV.
2418 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2420 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2421 (define_attr "cr_logical_3op"): New.
2422 (cceq_ior_compare): Adjust.
2423 (cceq_ior_compare_complement): Adjust.
2424 (*cceq_rev_compare): Adjust.
2425 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2426 (is_cracked_insn): Adjust.
2427 (insn_must_be_first_in_group): Adjust.
2428 * config/rs6000/40x.md: Adjust.
2429 * config/rs6000/440.md: Adjust.
2430 * config/rs6000/476.md: Adjust.
2431 * config/rs6000/601.md: Adjust.
2432 * config/rs6000/603.md: Adjust.
2433 * config/rs6000/6xx.md: Adjust.
2434 * config/rs6000/7450.md: Adjust.
2435 * config/rs6000/7xx.md: Adjust.
2436 * config/rs6000/8540.md: Adjust.
2437 * config/rs6000/cell.md: Adjust.
2438 * config/rs6000/e300c2c3.md: Adjust.
2439 * config/rs6000/e500mc.md: Adjust.
2440 * config/rs6000/e500mc64.md: Adjust.
2441 * config/rs6000/e5500.md: Adjust.
2442 * config/rs6000/e6500.md: Adjust.
2443 * config/rs6000/mpc.md: Adjust.
2444 * config/rs6000/power4.md: Adjust.
2445 * config/rs6000/power5.md: Adjust.
2446 * config/rs6000/power6.md: Adjust.
2447 * config/rs6000/power7.md: Adjust.
2448 * config/rs6000/power8.md: Adjust.
2449 * config/rs6000/power9.md: Adjust.
2450 * config/rs6000/rs64.md: Adjust.
2451 * config/rs6000/titan.md: Adjust.
2453 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2455 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2456 ix86_indirect_branch_register logic.
2458 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2460 * config/i386/constraints.md (Bs): Update
2461 ix86_indirect_branch_register check. Don't check
2462 ix86_indirect_branch_register with GOT_memory_operand.
2464 * config/i386/predicates.md (GOT_memory_operand): Don't check
2465 ix86_indirect_branch_register here.
2466 (GOT32_symbol_operand): Likewise.
2468 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2470 * config/i386/predicates.md (constant_call_address_operand):
2471 Rewrite ix86_indirect_branch_register logic.
2472 (sibcall_insn_operand): Likewise.
2474 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2476 * config/i386/constraints.md (Bs): Replace
2477 ix86_indirect_branch_thunk_register with
2478 ix86_indirect_branch_register.
2480 * config/i386/i386.md (indirect_jump): Likewise.
2481 (tablejump): Likewise.
2482 (*sibcall_memory): Likewise.
2483 (*sibcall_value_memory): Likewise.
2484 Peepholes of indirect call and jump via memory: Likewise.
2485 * config/i386/i386.opt: Likewise.
2486 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2487 (GOT_memory_operand): Likewise.
2488 (call_insn_operand): Likewise.
2489 (sibcall_insn_operand): Likewise.
2490 (GOT32_symbol_operand): Likewise.
2492 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2495 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2496 type rather than type addr's type points to.
2497 (expand_omp_atomic_mutex): Likewise.
2498 (expand_omp_atomic): Likewise.
2500 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2503 * config/i386/i386.c (output_indirect_thunk_function): Use
2504 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2505 for __x86_return_thunk.
2507 2018-01-15 Richard Biener <rguenther@suse.de>
2510 * expmed.c (extract_bit_field_1): Fix typo.
2512 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2515 * config/arm/iterators.md (VF): New mode iterator.
2516 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2517 Remove integer-related logic from pattern.
2518 (neon_vabd<mode>_3): Likewise.
2520 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2523 * common.opt (fstrict-overflow): No longer an alias.
2524 (fwrapv-pointer): New option.
2525 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2526 also for pointer types based on flag_wrapv_pointer.
2527 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2528 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2529 opts->x_flag_wrapv got set.
2530 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2531 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2532 POINTER_TYPE_OVERFLOW_UNDEFINED.
2533 * match.pd: Likewise in address comparison pattern.
2534 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2536 2018-01-15 Richard Biener <rguenther@suse.de>
2539 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2540 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2541 Reset type names to their identifier if their TYPE_DECL doesn't
2542 have linkage (and thus is used for ODR and devirt).
2543 (save_debug_info_for_decl): Remove.
2544 (save_debug_info_for_type): Likewise.
2545 (add_tree_to_fld_list): Adjust.
2546 * tree-pretty-print.c (dump_generic_node): Make dumping of
2547 type names more robust.
2549 2018-01-15 Richard Biener <rguenther@suse.de>
2551 * BASE-VER: Bump to 8.0.1.
2553 2018-01-14 Martin Sebor <msebor@redhat.com>
2556 * builtins.c (check_access): Avoid warning when the no-warning bit
2559 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2561 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2562 * ira-color (allocno_hard_regs_compare): Likewise.
2564 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2567 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2568 Use .pushsection/.popsection.
2570 2018-01-14 Martin Sebor <msebor@redhat.com>
2573 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2575 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2577 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2578 entry from extra_headers.
2579 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2580 extra_headers, make the list bitwise identical to the i?86-*-* one.
2582 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2584 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2585 -mcmodel=large with -mindirect-branch=thunk,
2586 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2587 -mfunction-return=thunk-extern.
2588 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2589 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2590 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2592 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2594 * config/i386/i386.c (print_reg): Print the name of the full
2595 integer register without '%'.
2596 (ix86_print_operand): Handle 'V'.
2597 * doc/extend.texi: Document 'V' modifier.
2599 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2601 * config/i386/constraints.md (Bs): Disallow memory operand for
2602 -mindirect-branch-register.
2604 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2605 (GOT_memory_operand): Likewise.
2606 (call_insn_operand): Likewise.
2607 (sibcall_insn_operand): Likewise.
2608 (GOT32_symbol_operand): Likewise.
2609 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2610 for -mindirect-branch-register.
2611 (tablejump): Likewise.
2612 (*sibcall_memory): Likewise.
2613 (*sibcall_value_memory): Likewise.
2614 Disallow peepholes of indirect call and jump via memory for
2615 -mindirect-branch-register.
2616 (*call_pop): Replace m with Bw.
2617 (*call_value_pop): Likewise.
2618 (*sibcall_pop_memory): Replace m with Bs.
2619 * config/i386/i386.opt (mindirect-branch-register): New option.
2620 * doc/invoke.texi: Document -mindirect-branch-register option.
2622 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2624 * config/i386/i386-protos.h (ix86_output_function_return): New.
2625 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2626 set function_return_type.
2627 (indirect_thunk_name): Add ret_p to indicate thunk for function
2629 (output_indirect_thunk_function): Pass false to
2630 indirect_thunk_name.
2631 (ix86_output_indirect_branch_via_reg): Likewise.
2632 (ix86_output_indirect_branch_via_push): Likewise.
2633 (output_indirect_thunk_function): Create alias for function
2634 return thunk if regno < 0.
2635 (ix86_output_function_return): New function.
2636 (ix86_handle_fndecl_attribute): Handle function_return.
2637 (ix86_attribute_table): Add function_return.
2638 * config/i386/i386.h (machine_function): Add
2639 function_return_type.
2640 * config/i386/i386.md (simple_return_internal): Use
2641 ix86_output_function_return.
2642 (simple_return_internal_long): Likewise.
2643 * config/i386/i386.opt (mfunction-return=): New option.
2644 (indirect_branch): Mention -mfunction-return=.
2645 * doc/extend.texi: Document function_return function attribute.
2646 * doc/invoke.texi: Document -mfunction-return= option.
2648 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2650 * config/i386/i386-opts.h (indirect_branch): New.
2651 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2652 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2653 with local indirect jump when converting indirect call and jump.
2654 (ix86_set_indirect_branch_type): New.
2655 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2656 (indirectlabelno): New.
2657 (indirect_thunk_needed): Likewise.
2658 (indirect_thunk_bnd_needed): Likewise.
2659 (indirect_thunks_used): Likewise.
2660 (indirect_thunks_bnd_used): Likewise.
2661 (INDIRECT_LABEL): Likewise.
2662 (indirect_thunk_name): Likewise.
2663 (output_indirect_thunk): Likewise.
2664 (output_indirect_thunk_function): Likewise.
2665 (ix86_output_indirect_branch_via_reg): Likewise.
2666 (ix86_output_indirect_branch_via_push): Likewise.
2667 (ix86_output_indirect_branch): Likewise.
2668 (ix86_output_indirect_jmp): Likewise.
2669 (ix86_code_end): Call output_indirect_thunk_function if needed.
2670 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2672 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2673 (ix86_attribute_table): Add indirect_branch.
2674 * config/i386/i386.h (machine_function): Add indirect_branch_type
2675 and has_local_indirect_jump.
2676 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2678 (tablejump): Likewise.
2679 (*indirect_jump): Use ix86_output_indirect_jmp.
2680 (*tablejump_1): Likewise.
2681 (simple_return_indirect_internal): Likewise.
2682 * config/i386/i386.opt (mindirect-branch=): New option.
2683 (indirect_branch): New.
2686 (thunk-inline): Likewise.
2687 (thunk-extern): Likewise.
2688 * doc/extend.texi: Document indirect_branch function attribute.
2689 * doc/invoke.texi: Document -mindirect-branch= option.
2691 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2694 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2696 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2698 * ipa-inline.c (want_inline_small_function_p): Return false if
2699 inlining has already failed with CIF_FINAL_ERROR.
2700 (update_caller_keys): Call want_inline_small_function_p before
2702 (update_callee_keys): Likewise.
2704 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2706 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2708 (rs6000_quadword_masked_address_p): Likewise.
2709 (quad_aligned_load_p): Likewise.
2710 (quad_aligned_store_p): Likewise.
2711 (const_load_sequence_p): Add comment to describe the outer-most loop.
2712 (mimic_memory_attributes_and_flags): New function.
2713 (rs6000_gen_stvx): Likewise.
2714 (replace_swapped_aligned_store): Likewise.
2715 (rs6000_gen_lvx): Likewise.
2716 (replace_swapped_aligned_load): Likewise.
2717 (replace_swapped_load_constant): Capitalize argument name in
2718 comment describing this function.
2719 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2720 and stores that access quad-word aligned addresses and replace
2721 with stvx or lvx instructions when appropriate.
2722 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2723 New function prototype.
2724 (rs6000_quadword_masked_address_p): Likewise.
2725 (rs6000_gen_lvx): Likewise.
2726 (rs6000_gen_stvx): Likewise.
2727 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2728 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2729 when memory address is aligned.
2730 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2731 this split to select lvx instruction when memory address is aligned.
2732 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2733 instruction when memory address is aligned.
2734 (*vsx_le_perm_load_v16qi): Likewise.
2735 (four unnamed splitters): Modify to select the stvx instruction
2736 when memory is aligned.
2738 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2740 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2741 which appears in the queue multiple times.
2743 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2744 Alan Hayward <alan.hayward@arm.com>
2745 David Sherwood <david.sherwood@arm.com>
2747 * tree-vectorizer.h (vec_lower_bound): New structure.
2748 (_loop_vec_info): Add check_nonzero and lower_bounds.
2749 (LOOP_VINFO_CHECK_NONZERO): New macro.
2750 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2751 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2752 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2753 fields. Make seg_len the distance travelled, not including the
2755 (dr_direction_indicator): Declare.
2756 (dr_zero_step_indicator): Likewise.
2757 (dr_known_forward_stride_p): Likewise.
2758 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2760 (runtime_alias_check_p): Allow runtime alias checks with
2762 (operator ==): Compare access_size and align.
2763 (prune_runtime_alias_test_list): Rework for new distinction between
2764 the access_size and seg_len.
2765 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2767 (get_segment_min_max): New function.
2768 (create_intersect_range_checks): Use it.
2769 (dr_step_indicator): New function.
2770 (dr_direction_indicator): Likewise.
2771 (dr_zero_step_indicator): Likewise.
2772 (dr_known_forward_stride_p): Likewise.
2773 * tree-loop-distribution.c (data_ref_segment_size): Return
2774 DR_STEP * (niters - 1).
2775 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2777 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2778 (vect_preserves_scalar_order_p): New function, split out from...
2779 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2780 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2781 (vect_vfa_access_size): New function.
2782 (vect_vfa_align): Likewise.
2783 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2784 (dump_lower_bound): New function.
2785 (vect_check_lower_bound): Likewise.
2786 (vect_small_gap_p): Likewise.
2787 (vectorizable_with_step_bound_p): Likewise.
2788 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2789 depencies if the vectorization factor is 1. Convert the checks
2790 for nonzero steps into checks on the bounds of DR_STEP. Try using
2791 a bunds check for variable steps if the minimum required step is
2792 relatively small. Update calls to the dr_with_seg_len
2793 constructor and to vect_compile_time_alias.
2794 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2796 (vect_loop_versioning): Call it.
2797 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2799 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2801 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2802 Alan Hayward <alan.hayward@arm.com>
2803 David Sherwood <david.sherwood@arm.com>
2805 * doc/sourcebuild.texi (vect_scatter_store): Document.
2806 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2808 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2810 * genopinit.c (main): Add supports_vec_scatter_store and
2811 supports_vec_scatter_store_cached to target_optabs.
2812 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2813 IFN_MASK_SCATTER_STORE.
2814 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2816 * internal-fn.h (internal_store_fn_p): Declare.
2817 (internal_fn_stored_value_index): Likewise.
2818 * internal-fn.c (scatter_store_direct): New macro.
2819 (expand_scatter_store_optab_fn): New function.
2820 (direct_scatter_store_optab_supported_p): New macro.
2821 (internal_store_fn_p): New function.
2822 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2823 IFN_MASK_SCATTER_STORE.
2824 (internal_fn_mask_index): Likewise.
2825 (internal_fn_stored_value_index): New function.
2826 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2828 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2829 * optabs-query.c (supports_vec_scatter_store_p): New function.
2830 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2831 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2832 true for scatter stores.
2833 (vect_gather_scatter_fn_p): Handle scatter stores too.
2834 (vect_check_gather_scatter): Consider using scatter stores if
2835 supports_vec_scatter_store_p.
2836 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2838 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2839 internal_fn_stored_value_index.
2840 (check_load_store_masking): Handle scatter stores too.
2841 (vect_get_store_rhs): Make public.
2842 (vectorizable_call): Use internal_store_fn_p.
2843 (vectorizable_store): Handle scatter store internal functions.
2844 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2845 when deciding whether the end of the group has been reached.
2846 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2847 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2848 (mask_scatter_store<mode>): New insns.
2850 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2851 Alan Hayward <alan.hayward@arm.com>
2852 David Sherwood <david.sherwood@arm.com>
2854 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2855 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2856 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2858 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2859 Use vect_truncate_gather_scatter_offset if we can't treat the
2860 operation as a normal gather load or scatter store.
2861 (get_group_load_store_type): Take the gather_scatter_info
2862 as argument. Try using a gather load or scatter store for
2863 single-element groups.
2864 (get_load_store_type): Update calls to get_group_load_store_type
2865 and vect_use_strided_gather_scatters_p.
2867 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2868 Alan Hayward <alan.hayward@arm.com>
2869 David Sherwood <david.sherwood@arm.com>
2871 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2872 optional tree argument.
2873 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2875 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2876 but continue to use the current value as a fallback.
2877 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2878 to compare the updates.
2879 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2880 (get_load_store_type): Use it when handling a strided access.
2881 (vect_get_strided_load_store_ops): New function.
2882 (vect_get_data_ptr_increment): Likewise.
2883 (vectorizable_load): Handle strided gather loads. Always pass
2884 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2886 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2887 Alan Hayward <alan.hayward@arm.com>
2888 David Sherwood <david.sherwood@arm.com>
2890 * doc/md.texi (gather_load@var{m}): Document.
2891 (mask_gather_load@var{m}): Likewise.
2892 * genopinit.c (main): Add supports_vec_gather_load and
2893 supports_vec_gather_load_cached to target_optabs.
2894 * optabs-tree.c (init_tree_optimization_optabs): Use
2895 ggc_cleared_alloc to allocate target_optabs.
2896 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2897 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2899 * internal-fn.h (internal_load_fn_p): Declare.
2900 (internal_gather_scatter_fn_p): Likewise.
2901 (internal_fn_mask_index): Likewise.
2902 (internal_gather_scatter_fn_supported_p): Likewise.
2903 * internal-fn.c (gather_load_direct): New macro.
2904 (expand_gather_load_optab_fn): New function.
2905 (direct_gather_load_optab_supported_p): New macro.
2906 (direct_internal_fn_optab): New function.
2907 (internal_load_fn_p): Likewise.
2908 (internal_gather_scatter_fn_p): Likewise.
2909 (internal_fn_mask_index): Likewise.
2910 (internal_gather_scatter_fn_supported_p): Likewise.
2911 * optabs-query.c (supports_at_least_one_mode_p): New function.
2912 (supports_vec_gather_load_p): Likewise.
2913 * optabs-query.h (supports_vec_gather_load_p): Declare.
2914 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2915 and memory_type field.
2916 (NUM_PATTERNS): Bump to 15.
2917 * tree-vect-data-refs.c: Include internal-fn.h.
2918 (vect_gather_scatter_fn_p): New function.
2919 (vect_describe_gather_scatter_call): Likewise.
2920 (vect_check_gather_scatter): Try using internal functions for
2921 gather loads. Recognize existing calls to a gather load function.
2922 (vect_analyze_data_refs): Consider using gather loads if
2923 supports_vec_gather_load_p.
2924 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2925 (vect_get_gather_scatter_offset_type): Likewise.
2926 (vect_convert_mask_for_vectype): Likewise.
2927 (vect_add_conversion_to_patterm): Likewise.
2928 (vect_try_gather_scatter_pattern): Likewise.
2929 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2930 (vect_vect_recog_func_ptrs): Add it.
2931 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2932 internal_fn_mask_index and internal_gather_scatter_fn_p.
2933 (check_load_store_masking): Take the gather_scatter_info as an
2934 argument and handle gather loads.
2935 (vect_get_gather_scatter_ops): New function.
2936 (vectorizable_call): Check internal_load_fn_p.
2937 (vectorizable_load): Likewise. Handle gather load internal
2939 (vectorizable_store): Update call to check_load_store_masking.
2940 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2941 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2942 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2943 (aarch64_gather_scale_operand_d): New predicates.
2944 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2945 (mask_gather_load<mode>): New insns.
2947 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2948 Alan Hayward <alan.hayward@arm.com>
2949 David Sherwood <david.sherwood@arm.com>
2951 * optabs.def (fold_left_plus_optab): New optab.
2952 * doc/md.texi (fold_left_plus_@var{m}): Document.
2953 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2954 * internal-fn.c (fold_left_direct): Define.
2955 (expand_fold_left_optab_fn): Likewise.
2956 (direct_fold_left_optab_supported_p): Likewise.
2957 * fold-const-call.c (fold_const_fold_left): New function.
2958 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2959 * tree-parloops.c (valid_reduction_p): New function.
2960 (gather_scalar_reductions): Use it.
2961 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2962 (vect_finish_replace_stmt): Declare.
2963 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2964 (needs_fold_left_reduction_p): New function, split out from...
2965 (vect_is_simple_reduction): ...here. Accept reductions that
2966 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2967 (vect_force_simple_reduction): Also store the reduction type in
2968 the assignment's STMT_VINFO_REDUC_TYPE.
2969 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2970 (merge_with_identity): New function.
2971 (vect_expand_fold_left): Likewise.
2972 (vectorize_fold_left_reduction): Likewise.
2973 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2974 scalar phi in place for it. Check for target support and reject
2975 cases that would reassociate the operation. Defer the transform
2976 phase to vectorize_fold_left_reduction.
2977 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2978 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2979 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2981 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2983 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2984 call to ifc_temp_var.
2986 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2987 Alan Hayward <alan.hayward@arm.com>
2988 David Sherwood <david.sherwood@arm.com>
2990 * target.def (legitimize_address_displacement): Take the original
2991 offset as a poly_int.
2992 * targhooks.h (default_legitimize_address_displacement): Update
2994 * targhooks.c (default_legitimize_address_displacement): Likewise.
2995 * doc/tm.texi: Regenerate.
2996 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2997 as an argument, moving assert of ad->disp == ad->disp_term to...
2998 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2999 Try calling targetm.legitimize_address_displacement before expanding
3000 the address rather than afterwards, and adjust for the new interface.
3001 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3002 Match the new hook interface. Handle SVE addresses.
3003 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3006 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3008 * Makefile.in (OBJS): Add early-remat.o.
3009 * target.def (select_early_remat_modes): New hook.
3010 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3011 * doc/tm.texi: Regenerate.
3012 * targhooks.h (default_select_early_remat_modes): Declare.
3013 * targhooks.c (default_select_early_remat_modes): New function.
3014 * timevar.def (TV_EARLY_REMAT): New timevar.
3015 * passes.def (pass_early_remat): New pass.
3016 * tree-pass.h (make_pass_early_remat): Declare.
3017 * early-remat.c: New file.
3018 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3020 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3022 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3023 Alan Hayward <alan.hayward@arm.com>
3024 David Sherwood <david.sherwood@arm.com>
3026 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3027 vfm1 with a bound_epilog parameter.
3028 (vect_do_peeling): Update calls accordingly, and move the prologue
3029 call earlier in the function. Treat the base bound_epilog as 0 for
3030 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3031 this base when peeling for gaps.
3032 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3033 with fully-masked loops.
3034 (vect_estimate_min_profitable_iters): Handle the single peeled
3035 iteration in that case.
3037 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3038 Alan Hayward <alan.hayward@arm.com>
3039 David Sherwood <david.sherwood@arm.com>
3041 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3042 single-element interleaving even if the size is not a power of 2.
3043 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3044 accesses for single-element interleaving if the group size is
3047 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3048 Alan Hayward <alan.hayward@arm.com>
3049 David Sherwood <david.sherwood@arm.com>
3051 * doc/md.texi (fold_extract_last_@var{m}): Document.
3052 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3053 * optabs.def (fold_extract_last_optab): New optab.
3054 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3055 * internal-fn.c (fold_extract_direct): New macro.
3056 (expand_fold_extract_optab_fn): Likewise.
3057 (direct_fold_extract_optab_supported_p): Likewise.
3058 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3059 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3060 EXTRACT_LAST_REDUCTION.
3061 (get_initial_def_for_reduction): Do not create an initial vector
3062 for EXTRACT_LAST_REDUCTION reductions.
3063 (vectorizable_reduction): Leave the scalar phi in place for
3064 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3065 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3066 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3067 transform phase to vectorizable_condition.
3068 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3070 (vect_finish_stmt_generation): ...here.
3071 (vect_finish_replace_stmt): New function.
3072 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3073 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3075 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3077 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3078 Alan Hayward <alan.hayward@arm.com>
3079 David Sherwood <david.sherwood@arm.com>
3081 * doc/md.texi (extract_last_@var{m}): Document.
3082 * optabs.def (extract_last_optab): New optab.
3083 * internal-fn.def (EXTRACT_LAST): New internal function.
3084 * internal-fn.c (cond_unary_direct): New macro.
3085 (expand_cond_unary_optab_fn): Likewise.
3086 (direct_cond_unary_optab_supported_p): Likewise.
3087 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3088 loops using EXTRACT_LAST.
3089 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3090 (extract_last_<mode>): ...this optab.
3091 (vec_extract<mode><Vel>): Update accordingly.
3093 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3094 Alan Hayward <alan.hayward@arm.com>
3095 David Sherwood <david.sherwood@arm.com>
3097 * target.def (empty_mask_is_expensive): New hook.
3098 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3099 * doc/tm.texi: Regenerate.
3100 * targhooks.h (default_empty_mask_is_expensive): Declare.
3101 * targhooks.c (default_empty_mask_is_expensive): New function.
3102 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3103 if the target says that empty masks are expensive.
3104 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3106 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3108 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3109 Alan Hayward <alan.hayward@arm.com>
3110 David Sherwood <david.sherwood@arm.com>
3112 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3113 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3114 (vect_use_loop_mask_for_alignment_p): New function.
3115 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3116 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3117 niters_skip argument. Make sure that the first niters_skip elements
3118 of the first iteration are inactive.
3119 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3120 Update call to vect_set_loop_masks_directly.
3121 (get_misalign_in_elems): New function, split out from...
3122 (vect_gen_prolog_loop_niters): ...here.
3123 (vect_update_init_of_dr): Take a code argument that specifies whether
3124 the adjustment should be added or subtracted.
3125 (vect_update_init_of_drs): Likewise.
3126 (vect_prepare_for_masked_peels): New function.
3127 (vect_do_peeling): Skip prologue peeling if we're using a mask
3128 instead. Update call to vect_update_inits_of_drs.
3129 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3131 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3132 alignment. Do not include the number of peeled iterations in
3133 the minimum threshold in that case.
3134 (vectorizable_induction): Adjust the start value down by
3135 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3136 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3137 Take the number of skipped iterations into account when calculating
3139 * tree-vect-stmts.c (vect_gen_while_not): New function.
3141 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3142 Alan Hayward <alan.hayward@arm.com>
3143 David Sherwood <david.sherwood@arm.com>
3145 * doc/sourcebuild.texi (vect_fully_masked): Document.
3146 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3148 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3150 (vect_analyze_loop_2): ...here. Don't check the vectorization
3151 factor against the number of loop iterations if the loop is
3154 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3155 Alan Hayward <alan.hayward@arm.com>
3156 David Sherwood <david.sherwood@arm.com>
3158 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3159 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3160 (dump_groups): Update accordingly.
3161 (iv_use::mem_type): New member variable.
3162 (address_p): New function.
3163 (record_use): Add a mem_type argument and initialize the new
3165 (record_group_use): Add a mem_type argument. Use address_p.
3166 Remove obsolete null checks of base_object. Update call to record_use.
3167 (find_interesting_uses_op): Update call to record_group_use.
3168 (find_interesting_uses_cond): Likewise.
3169 (find_interesting_uses_address): Likewise.
3170 (get_mem_type_for_internal_fn): New function.
3171 (find_address_like_use): Likewise.
3172 (find_interesting_uses_stmt): Try find_address_like_use before
3173 calling find_interesting_uses_op.
3174 (addr_offset_valid_p): Use the iv mem_type field as the type
3175 of the addressed memory.
3176 (add_autoinc_candidates): Likewise.
3177 (get_address_cost): Likewise.
3178 (split_small_address_groups_p): Use address_p.
3179 (split_address_groups): Likewise.
3180 (add_iv_candidate_for_use): Likewise.
3181 (autoinc_possible_for_pair): Likewise.
3182 (rewrite_groups): Likewise.
3183 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3184 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3185 (get_alias_ptr_type_for_ptr_address): New function.
3186 (rewrite_use_address): Rewrite address uses in calls that were
3187 identified by find_address_like_use.
3189 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3190 Alan Hayward <alan.hayward@arm.com>
3191 David Sherwood <david.sherwood@arm.com>
3193 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3195 * gimple-expr.h (is_gimple_addressable: Likewise.
3196 * gimple-expr.c (is_gimple_address): Likewise.
3197 * internal-fn.c (expand_call_mem_ref): New function.
3198 (expand_mask_load_optab_fn): Use it.
3199 (expand_mask_store_optab_fn): Likewise.
3201 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3202 Alan Hayward <alan.hayward@arm.com>
3203 David Sherwood <david.sherwood@arm.com>
3205 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3206 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3207 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3208 (cond_umax@var{mode}): Document.
3209 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3210 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3211 (cond_umin_optab, cond_umax_optab): New optabs.
3212 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3213 (COND_IOR, COND_XOR): New internal functions.
3214 * internal-fn.h (get_conditional_internal_fn): Declare.
3215 * internal-fn.c (cond_binary_direct): New macro.
3216 (expand_cond_binary_optab_fn): Likewise.
3217 (direct_cond_binary_optab_supported_p): Likewise.
3218 (get_conditional_internal_fn): New function.
3219 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3220 Cope with reduction statements that are vectorized as calls rather
3222 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3223 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3224 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3225 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3226 (UNSPEC_COND_EOR): New unspecs.
3227 (optab): Add mappings for them.
3228 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3229 (sve_int_op, sve_fp_op): New int attributes.
3231 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3232 Alan Hayward <alan.hayward@arm.com>
3233 David Sherwood <david.sherwood@arm.com>
3235 * optabs.def (while_ult_optab): New optab.
3236 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3237 * internal-fn.def (WHILE_ULT): New internal function.
3238 * internal-fn.h (direct_internal_fn_supported_p): New override
3239 that takes two types as argument.
3240 * internal-fn.c (while_direct): New macro.
3241 (expand_while_optab_fn): New function.
3242 (convert_optab_supported_p): Likewise.
3243 (direct_while_optab_supported_p): New macro.
3244 * wide-int.h (wi::udiv_ceil): New function.
3245 * tree-vectorizer.h (rgroup_masks): New structure.
3246 (vec_loop_masks): New typedef.
3247 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3249 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3250 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3251 (vect_max_vf): New function.
3252 (slpeel_make_loop_iterate_ntimes): Delete.
3253 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3254 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3255 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3256 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3257 internal-fn.h, stor-layout.h and optabs-query.h.
3258 (vect_set_loop_mask): New function.
3259 (add_preheader_seq): Likewise.
3260 (add_header_seq): Likewise.
3261 (interleave_supported_p): Likewise.
3262 (vect_maybe_permute_loop_masks): Likewise.
3263 (vect_set_loop_masks_directly): Likewise.
3264 (vect_set_loop_condition_masked): Likewise.
3265 (vect_set_loop_condition_unmasked): New function, split out from
3266 slpeel_make_loop_iterate_ntimes.
3267 (slpeel_make_loop_iterate_ntimes): Rename to..
3268 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3269 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3270 (vect_do_peeling): Update call accordingly.
3271 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3273 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3274 mask_compare_type, can_fully_mask_p and fully_masked_p.
3275 (release_vec_loop_masks): New function.
3276 (_loop_vec_info): Use it to free the loop masks.
3277 (can_produce_all_loop_masks_p): New function.
3278 (vect_get_max_nscalars_per_iter): Likewise.
3279 (vect_verify_full_masking): Likewise.
3280 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3281 retries, and free the mask rgroups before retrying. Check loop-wide
3282 reasons for disallowing fully-masked loops. Make the final decision
3283 about whether use a fully-masked loop or not.
3284 (vect_estimate_min_profitable_iters): Do not assume that peeling
3285 for the number of iterations will be needed for fully-masked loops.
3286 (vectorizable_reduction): Disable fully-masked loops.
3287 (vectorizable_live_operation): Likewise.
3288 (vect_halve_mask_nunits): New function.
3289 (vect_double_mask_nunits): Likewise.
3290 (vect_record_loop_mask): Likewise.
3291 (vect_get_loop_mask): Likewise.
3292 (vect_transform_loop): Handle the case in which the final loop
3293 iteration might handle a partial vector. Call vect_set_loop_condition
3294 instead of slpeel_make_loop_iterate_ntimes.
3295 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3296 (check_load_store_masking): New function.
3297 (prepare_load_store_mask): Likewise.
3298 (vectorizable_store): Handle fully-masked loops.
3299 (vectorizable_load): Likewise.
3300 (supportable_widening_operation): Use vect_halve_mask_nunits for
3302 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3303 (vect_gen_while): New function.
3304 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3305 (aarch64_uqdec<mode>): New insn.
3307 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3308 Alan Hayward <alan.hayward@arm.com>
3309 David Sherwood <david.sherwood@arm.com>
3311 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3312 (reduc_xor_scal_optab): New optabs.
3313 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3314 (reduc_xor_scal_@var{m}): Document.
3315 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3316 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3318 * fold-const-call.c (fold_const_call): Handle them.
3319 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3320 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3321 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3322 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3323 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3324 (UNSPEC_XORV): New unspecs.
3325 (optab): Add entries for them.
3326 (BITWISEV): New int iterator.
3327 (bit_reduc_op): New int attributes.
3329 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3330 Alan Hayward <alan.hayward@arm.com>
3331 David Sherwood <david.sherwood@arm.com>
3333 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3334 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3335 * optabs.def (vec_shl_insert_optab): New optab.
3336 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3337 (duplicate_and_interleave): Likewise.
3338 * tree-vect-loop.c: Include internal-fn.h.
3339 (neutral_op_for_slp_reduction): New function, split out from
3340 get_initial_defs_for_reduction.
3341 (get_initial_def_for_reduction): Handle option 2 for variable-length
3342 vectors by loading the neutral value into a vector and then shifting
3343 the initial value into element 0.
3344 (get_initial_defs_for_reduction): Replace the code argument with
3345 the neutral value calculated by neutral_op_for_slp_reduction.
3346 Use gimple_build_vector for constant-length vectors.
3347 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3348 but the first group_size elements have a neutral value.
3349 Use duplicate_and_interleave otherwise.
3350 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3351 Update call to get_initial_defs_for_reduction. Handle SLP
3352 reductions for variable-length vectors by creating one vector
3353 result for each scalar result, with the elements associated
3354 with other scalar results stubbed out with the neutral value.
3355 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3356 Require IFN_VEC_SHL_INSERT for double reductions on
3357 variable-length vectors, or SLP reductions that have
3358 a neutral value. Require can_duplicate_and_interleave_p
3359 support for variable-length unchained SLP reductions if there
3360 is no neutral value, such as for MIN/MAX reductions. Also require
3361 the number of vector elements to be a multiple of the number of
3362 SLP statements when doing variable-length unchained SLP reductions.
3363 Update call to vect_create_epilog_for_reduction.
3364 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3365 and remove initial values.
3366 (duplicate_and_interleave): Make public.
3367 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3368 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3370 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3371 Alan Hayward <alan.hayward@arm.com>
3372 David Sherwood <david.sherwood@arm.com>
3374 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3375 (can_duplicate_and_interleave_p): New function.
3376 (vect_get_and_check_slp_defs): Take the vector of statements
3377 rather than just the current one. Remove excess parentheses.
3378 Restriction rejectinon of vect_constant_def and vect_external_def
3379 for variable-length vectors to boolean types, or types for which
3380 can_duplicate_and_interleave_p is false.
3381 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3382 (duplicate_and_interleave): New function.
3383 (vect_get_constant_vectors): Use gimple_build_vector for
3384 constant-length vectors and suitable variable-length constant
3385 vectors. Use duplicate_and_interleave for other variable-length
3386 vectors. Don't defer the update when inserting new statements.
3388 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3389 Alan Hayward <alan.hayward@arm.com>
3390 David Sherwood <david.sherwood@arm.com>
3392 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3393 min_profitable_iters doesn't go negative.
3395 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3396 Alan Hayward <alan.hayward@arm.com>
3397 David Sherwood <david.sherwood@arm.com>
3399 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3400 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3401 * optabs.def (vec_mask_load_lanes_optab): New optab.
3402 (vec_mask_store_lanes_optab): Likewise.
3403 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3404 (MASK_STORE_LANES): Likewise.
3405 * internal-fn.c (mask_load_lanes_direct): New macro.
3406 (mask_store_lanes_direct): Likewise.
3407 (expand_mask_load_optab_fn): Handle masked operations.
3408 (expand_mask_load_lanes_optab_fn): New macro.
3409 (expand_mask_store_optab_fn): Handle masked operations.
3410 (expand_mask_store_lanes_optab_fn): New macro.
3411 (direct_mask_load_lanes_optab_supported_p): Likewise.
3412 (direct_mask_store_lanes_optab_supported_p): Likewise.
3413 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3415 (vect_load_lanes_supported): Likewise.
3416 * tree-vect-data-refs.c (strip_conversion): New function.
3417 (can_group_stmts_p): Likewise.
3418 (vect_analyze_data_ref_accesses): Use it instead of checking
3419 for a pair of assignments.
3420 (vect_store_lanes_supported): Take a masked_p parameter.
3421 (vect_load_lanes_supported): Likewise.
3422 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3423 vect_store_lanes_supported and vect_load_lanes_supported.
3424 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3425 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3426 parameter. Don't allow gaps for masked accesses.
3427 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3428 and vect_load_lanes_supported.
3429 (get_load_store_type): Take a masked_p parameter and update
3430 call to get_group_load_store_type.
3431 (vectorizable_store): Update call to get_load_store_type.
3432 Handle IFN_MASK_STORE_LANES.
3433 (vectorizable_load): Update call to get_load_store_type.
3434 Handle IFN_MASK_LOAD_LANES.
3436 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3437 Alan Hayward <alan.hayward@arm.com>
3438 David Sherwood <david.sherwood@arm.com>
3440 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3442 * config/aarch64/aarch64-protos.h
3443 (aarch64_sve_struct_memory_operand_p): Declare.
3444 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3445 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3446 (VPRED, vpred): Handle SVE structure modes.
3447 * config/aarch64/constraints.md (Utx): New constraint.
3448 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3449 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3450 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3451 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3452 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3453 structure modes. Split into pieces after RA.
3454 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3455 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3457 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3458 SVE structure modes.
3459 (aarch64_classify_address): Likewise.
3460 (sizetochar): Move earlier in file.
3461 (aarch64_print_operand): Handle SVE register lists.
3462 (aarch64_array_mode): New function.
3463 (aarch64_sve_struct_memory_operand_p): Likewise.
3464 (TARGET_ARRAY_MODE): Redefine.
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 * target.def (array_mode): New target hook.
3471 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3472 * doc/tm.texi: Regenerate.
3473 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3474 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3475 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3477 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3480 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3481 Alan Hayward <alan.hayward@arm.com>
3482 David Sherwood <david.sherwood@arm.com>
3484 * fold-const.c (fold_binary_loc): Check the argument types
3485 rather than the result type when testing for a vector operation.
3487 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3489 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3490 * doc/tm.texi: Regenerate.
3492 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3493 Alan Hayward <alan.hayward@arm.com>
3494 David Sherwood <david.sherwood@arm.com>
3496 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3497 (sve): Document new AArch64 extension.
3498 * doc/md.texi (w): Extend the description of the AArch64
3499 constraint to include SVE vectors.
3500 (Upl, Upa): Document new AArch64 predicate constraints.
3501 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3503 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3504 (msve-vector-bits=): New option.
3505 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3506 SVE when these are disabled.
3507 (sve): New extension.
3508 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3509 modes. Adjust their number of units based on aarch64_sve_vg.
3510 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3511 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3512 aarch64_addr_query_type.
3513 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3514 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3515 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3516 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3517 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3518 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3519 (aarch64_simd_imm_zero_p): Delete.
3520 (aarch64_check_zero_based_sve_index_immediate): Declare.
3521 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3522 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3523 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3524 (aarch64_sve_float_mul_immediate_p): Likewise.
3525 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3527 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3528 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3529 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3530 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3531 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3532 (aarch64_regmode_natural_size): Likewise.
3533 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3534 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3536 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3537 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3538 for VG and the SVE predicate registers.
3539 (V_ALIASES): Add a "z"-prefixed alias.
3540 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3541 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3542 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3543 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3544 (REG_CLASS_NAMES): Add entries for them.
3545 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3546 and the predicate registers.
3547 (aarch64_sve_vg): Declare.
3548 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3549 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3550 (REGMODE_NATURAL_SIZE): Define.
3551 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3553 * config/aarch64/aarch64.c: Include cfgrtl.h.
3554 (simd_immediate_info): Add a constructor for series vectors,
3555 and an associated step field.
3556 (aarch64_sve_vg): New variable.
3557 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3558 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3559 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3560 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3561 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3562 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3563 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3564 (aarch64_get_mask_mode): New functions.
3565 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3566 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3567 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3568 predicate modes and predicate registers. Explicitly restrict
3569 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3570 to store a vector mode if it is recognized by
3571 aarch64_classify_vector_mode.
3572 (aarch64_regmode_natural_size): New function.
3573 (aarch64_hard_regno_caller_save_mode): Return the original mode
3575 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3576 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3577 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3578 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3580 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3581 does not overlap dest if the function is frame-related. Handle
3583 (aarch64_split_add_offset): New function.
3584 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3585 them aarch64_add_offset.
3586 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3587 and update call to aarch64_sub_sp.
3588 (aarch64_add_cfa_expression): New function.
3589 (aarch64_expand_prologue): Pass extra temporary registers to the
3590 functions above. Handle the case in which we need to emit new
3591 DW_CFA_expressions for registers that were originally saved
3592 relative to the stack pointer, but now have to be expressed
3593 relative to the frame pointer.
3594 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3596 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3597 IP0 and IP1 values for SVE frames.
3598 (aarch64_expand_vec_series): New function.
3599 (aarch64_expand_sve_widened_duplicate): Likewise.
3600 (aarch64_expand_sve_const_vector): Likewise.
3601 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3602 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3603 into the register, rather than emitting a SET directly.
3604 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3605 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3606 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3607 (offset_9bit_signed_scaled_p): New functions.
3608 (aarch64_replicate_bitmask_imm): New function.
3609 (aarch64_bitmask_imm): Use it.
3610 (aarch64_cannot_force_const_mem): Reject expressions involving
3611 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3612 (aarch64_classify_index): Handle SVE indices, by requiring
3613 a plain register index with a scale that matches the element size.
3614 (aarch64_classify_address): Handle SVE addresses. Assert that
3615 the mode of the address is VOIDmode or an integer mode.
3616 Update call to aarch64_classify_symbol.
3617 (aarch64_classify_symbolic_expression): Update call to
3618 aarch64_classify_symbol.
3619 (aarch64_const_vec_all_in_range_p): New function.
3620 (aarch64_print_vector_float_operand): Likewise.
3621 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3622 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3623 and the FP immediates 1.0 and 0.5.
3624 (aarch64_print_address_internal): Handle SVE addresses.
3625 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3626 (aarch64_regno_regclass): Handle predicate registers.
3627 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3629 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3630 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3631 (aarch64_convert_sve_vector_bits): New function.
3632 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3633 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3635 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3636 Handle SVE vector and predicate modes. Accept VL-based constants
3637 that need only one temporary register, and VL offsets that require
3638 no temporary registers.
3639 (aarch64_conditional_register_usage): Mark the predicate registers
3640 as fixed if SVE isn't available.
3641 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3642 Return true for SVE vector and predicate modes.
3643 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3644 rather than an unsigned int. Handle SVE modes.
3645 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3647 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3649 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3650 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3651 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3652 (aarch64_sve_float_mul_immediate_p): New functions.
3653 (aarch64_sve_valid_immediate): New function.
3654 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3655 Explicitly reject structure modes. Check for INDEX constants.
3656 Handle PTRUE and PFALSE constants.
3657 (aarch64_check_zero_based_sve_index_immediate): New function.
3658 (aarch64_simd_imm_zero_p): Delete.
3659 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3660 vector modes. Accept constants in the range of CNT[BHWD].
3661 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3662 ask for an Advanced SIMD mode.
3663 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3664 (aarch64_simd_vector_alignment): Handle SVE predicates.
3665 (aarch64_vectorize_preferred_vector_alignment): New function.
3666 (aarch64_simd_vector_alignment_reachable): Use it instead of
3668 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3669 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3671 (MAX_VECT_LEN): Delete.
3672 (expand_vec_perm_d): Add a vec_flags field.
3673 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3674 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3675 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3677 (aarch64_evpc_rev): Rename to...
3678 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3679 (aarch64_evpc_rev_global): New function.
3680 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3681 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3683 (aarch64_evpc_sve_tbl): New function.
3684 (aarch64_expand_vec_perm_const_1): Update after rename of
3685 aarch64_evpc_rev. Handle SVE permutes too, trying
3686 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3687 than aarch64_evpc_tbl.
3688 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3689 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3690 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3691 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3692 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3693 (aarch64_expand_sve_vcond): New functions.
3694 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3695 of aarch64_vector_mode_p.
3696 (aarch64_dwarf_poly_indeterminate_value): New function.
3697 (aarch64_compute_pressure_classes): Likewise.
3698 (aarch64_can_change_mode_class): Likewise.
3699 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3700 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3701 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3702 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3703 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3704 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3705 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3706 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3708 (Dn, Dl, Dr): Accept const as well as const_vector.
3709 (Dz): Likewise. Compare against CONST0_RTX.
3710 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3711 of "vector" where appropriate.
3712 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3713 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3714 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3715 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3716 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3717 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3718 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3719 (v_int_equiv): Extend to SVE modes.
3720 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3722 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3723 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3724 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3725 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3726 (SVE_COND_FP_CMP): New int iterators.
3727 (perm_hilo): Handle the new unpack unspecs.
3728 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3730 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3731 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3732 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3733 (aarch64_equality_operator, aarch64_constant_vector_operand)
3734 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3735 (aarch64_sve_nonimmediate_operand): Likewise.
3736 (aarch64_sve_general_operand): Likewise.
3737 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3738 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3739 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3740 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3741 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3742 (aarch64_sve_float_arith_immediate): Likewise.
3743 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3744 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3745 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3746 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3747 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3748 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3749 (aarch64_sve_float_arith_operand): Likewise.
3750 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3751 (aarch64_sve_float_mul_operand): Likewise.
3752 (aarch64_sve_vec_perm_operand): Likewise.
3753 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3754 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3755 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3756 as well as const_vector.
3757 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3758 in file. Use CONST0_RTX and CONSTM1_RTX.
3759 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3760 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3761 Use aarch64_simd_imm_zero.
3762 * config/aarch64/aarch64-sve.md: New file.
3763 * config/aarch64/aarch64.md: Include it.
3764 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3765 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3766 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3767 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3768 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3769 (sve): New attribute.
3770 (enabled): Disable instructions with the sve attribute unless
3772 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3773 aarch64_expand_mov_immediate.
3774 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3775 CNT[BHSD] immediates.
3776 (movti): Split CONST_POLY_INT moves into two halves.
3777 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3778 Split additions that need a temporary here if the destination
3779 is the stack pointer.
3780 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3781 (*add<mode>3_poly_1): New instruction.
3782 (set_clobber_cc): New expander.
3784 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3786 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3787 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3788 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3789 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3790 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3791 Change innermode from fixed_mode_size to machine_mode.
3792 (simplify_subreg): Update call accordingly. Handle a constant-sized
3793 subreg of a variable-length CONST_VECTOR.
3795 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3796 Alan Hayward <alan.hayward@arm.com>
3797 David Sherwood <david.sherwood@arm.com>
3799 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3800 (add_offset_to_base): New function, split out from...
3801 (create_mem_ref): ...here. When handling a scale other than 1,
3802 check first whether the address is valid without the offset.
3803 Add it into the base if so, leaving the index and scale as-is.
3805 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3808 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3809 fold_for_warn before checking if arg2 is INTEGER_CST.
3811 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3813 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3814 (store_multiple_operation): Delete.
3815 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3816 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3817 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3818 guarded by TARGET_STRING.
3819 (rs6000_output_load_multiple): Delete.
3820 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3821 OPTION_MASK_STRING / TARGET_STRING handling.
3822 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3823 (const rs6000_opt_masks) <"string">: Change mask to 0.
3824 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3825 (MASK_STRING): Delete.
3826 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3828 (load_multiple): Delete.
3835 (store_multiple): Delete.
3842 (movmemsi_8reg): Delete.
3843 (corresponding unnamed define_insn): Delete.
3844 (movmemsi_6reg): Delete.
3845 (corresponding unnamed define_insn): Delete.
3846 (movmemsi_4reg): Delete.
3847 (corresponding unnamed define_insn): Delete.
3848 (movmemsi_2reg): Delete.
3849 (corresponding unnamed define_insn): Delete.
3850 (movmemsi_1reg): Delete.
3851 (corresponding unnamed define_insn): Delete.
3852 * config/rs6000/rs6000.opt (mno-string): New.
3853 (mstring): Replace by deprecation warning stub.
3854 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3856 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3858 * regrename.c (regrename_do_replace): If replacing the same
3859 reg multiple times, try to reuse last created gen_raw_REG.
3862 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3863 main to workaround a bug in GDB.
3865 2018-01-12 Tom de Vries <tom@codesourcery.com>
3868 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3870 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3872 PR rtl-optimization/80481
3873 * ira-color.c (get_cap_member): New function.
3874 (allocnos_conflict_by_live_ranges_p): Use it.
3875 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3876 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3878 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3881 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3882 (*saddl_se_1): Ditto.
3884 (*ssubl_se_1): Ditto.
3886 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3888 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3889 rather than wi::to_widest for DR_INITs.
3890 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3891 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3892 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3894 (vect_analyze_group_access_1): Note that here.
3896 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3898 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3899 polynomial type sizes.
3901 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3903 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3904 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3905 (gimple_add_tmp_var): Likewise.
3907 2018-01-12 Martin Liska <mliska@suse.cz>
3909 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3910 (gimple_alloc_sizes): Likewise.
3911 (dump_gimple_statistics): Use PRIu64 in printf format.
3912 * gimple.h: Change uint64_t to int.
3914 2018-01-12 Martin Liska <mliska@suse.cz>
3916 * tree-core.h: Use uint64_t instead of int.
3917 * tree.c (tree_node_counts): Likewise.
3918 (tree_node_sizes): Likewise.
3919 (dump_tree_statistics): Use PRIu64 in printf format.
3921 2018-01-12 Martin Liska <mliska@suse.cz>
3923 * Makefile.in: As qsort_chk is implemented in vec.c, add
3924 vec.o to linkage of gencfn-macros.
3925 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3926 passing the info to record_node_allocation_statistics.
3927 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3929 * ggc-common.c (struct ggc_usage): Add operator== and use
3930 it in operator< and compare function.
3931 * mem-stats.h (struct mem_usage): Likewise.
3932 * vec.c (struct vec_usage): Remove operator< and compare
3933 function. Can be simply inherited.
3935 2018-01-12 Martin Jambor <mjambor@suse.cz>
3938 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3939 * tree-ssa-math-opts.c: Include domwalk.h.
3940 (convert_mult_to_fma_1): New function.
3941 (fma_transformation_info): New type.
3942 (fma_deferring_state): Likewise.
3943 (cancel_fma_deferring): New function.
3944 (result_of_phi): Likewise.
3945 (last_fma_candidate_feeds_initial_phi): Likewise.
3946 (convert_mult_to_fma): Added deferring logic, split actual
3947 transformation to convert_mult_to_fma_1.
3948 (math_opts_dom_walker): New type.
3949 (math_opts_dom_walker::after_dom_children): New method, body moved
3950 here from pass_optimize_widening_mul::execute, added deferring logic
3952 (pass_optimize_widening_mul::execute): Moved most of code to
3953 math_opts_dom_walker::after_dom_children.
3954 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3955 * config/i386/i386.c (ix86_option_override_internal): Added
3956 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3958 2018-01-12 Richard Biener <rguenther@suse.de>
3961 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3962 inline instance vars.
3964 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3967 * config/rx/rx.c (rx_is_restricted_memory_address):
3970 2018-01-12 Richard Biener <rguenther@suse.de>
3972 PR tree-optimization/80846
3973 * target.def (split_reduction): New target hook.
3974 * targhooks.c (default_split_reduction): New function.
3975 * targhooks.h (default_split_reduction): Declare.
3976 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3977 target requests first reduce vectors by combining low and high
3979 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3980 (get_vectype_for_scalar_type_and_size): Export.
3981 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3982 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3983 * doc/tm.texi: Regenerate.
3984 * config/i386/i386.c (ix86_split_reduction): Implement
3985 TARGET_VECTORIZE_SPLIT_REDUCTION.
3987 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3990 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3991 in PIC mode except for TARGET_VXWORKS_RTP.
3992 * config/sparc/sparc.c: Include cfgrtl.h.
3993 (TARGET_INIT_PIC_REG): Define.
3994 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3995 (sparc_pic_register_p): New predicate.
3996 (sparc_legitimate_address_p): Use it.
3997 (sparc_legitimize_pic_address): Likewise.
3998 (sparc_delegitimize_address): Likewise.
3999 (sparc_mode_dependent_address_p): Likewise.
4000 (gen_load_pcrel_sym): Remove 4th parameter.
4001 (load_got_register): Adjust call to above. Remove obsolete stuff.
4002 (sparc_expand_prologue): Do not call load_got_register here.
4003 (sparc_flat_expand_prologue): Likewise.
4004 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4005 (sparc_use_pseudo_pic_reg): New function.
4006 (sparc_init_pic_reg): Likewise.
4007 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4008 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4010 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4012 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4013 Add item for branch_cost.
4015 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4017 PR rtl-optimization/83565
4018 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4019 not extend the result to a larger mode for rotate operations.
4020 (num_sign_bit_copies1): Likewise.
4022 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4025 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4027 Use values-Xc.o for -pedantic.
4028 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4030 2018-01-12 Martin Liska <mliska@suse.cz>
4033 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4035 (possible_polymorphic_call_targets): Use it.
4036 (ipa_devirt): Likewise.
4038 2018-01-12 Martin Liska <mliska@suse.cz>
4040 * profile-count.h (enum profile_quality): Use 0 as invalid
4041 enum value of profile_quality.
4043 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4045 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4046 -mext-string options.
4048 2018-01-12 Richard Biener <rguenther@suse.de>
4050 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4051 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4052 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4054 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4056 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4058 * configure.ac (--with-long-double-format): Add support for the
4059 configuration option to change the default long double format on
4061 * config.gcc (powerpc*-linux*-*): Likewise.
4062 * configure: Regenerate.
4063 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4064 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4065 used without modification.
4067 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4069 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4070 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4071 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4072 MISC_BUILTIN_SPEC_BARRIER.
4073 (rs6000_init_builtins): Likewise.
4074 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4076 (speculation_barrier): New define_insn.
4077 * doc/extend.texi: Document __builtin_speculation_barrier.
4079 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4082 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4083 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4084 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4086 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4087 integral modes instead of "ss" and "sd".
4088 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4089 vectors with 32-bit and 64-bit elements.
4090 (vecdupssescalarmodesuffix): New mode attribute.
4091 (vec_dup<mode>): Use it.
4093 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4096 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4097 frame if argument is passed on stack.
4099 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4102 * ree.c (combine_reaching_defs): Optimize also
4103 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4104 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4106 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4109 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4111 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4114 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4115 after they are computed.
4117 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4119 PR tree-optimization/83695
4120 * gimple-loop-linterchange.cc
4121 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4122 reset cached scev information after interchange.
4123 (pass_linterchange::execute): Remove call to scev_reset_htab.
4125 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4127 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4128 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4129 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4130 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4131 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4132 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4133 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4134 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4135 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4136 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4137 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4138 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4139 (V_lane_reg): Likewise.
4140 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4142 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4143 (vfmal_lane_low<mode>_intrinsic,
4144 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4145 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4146 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4147 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4148 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4149 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4151 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4153 * config/arm/arm-cpus.in (fp16fml): New feature.
4154 (ALL_SIMD): Add fp16fml.
4155 (armv8.2-a): Add fp16fml as an option.
4156 (armv8.3-a): Likewise.
4157 (armv8.4-a): Add fp16fml as part of fp16.
4158 * config/arm/arm.h (TARGET_FP16FML): Define.
4159 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4161 * config/arm/arm-modes.def (V2HF): Define.
4162 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4163 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4164 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4165 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4166 vfmsl_low, vfmsl_high): New set of builtins.
4167 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4168 (vfml_op): New code attribute.
4169 (VFMLHALVES): New int iterator.
4170 (VFML, VFMLSEL): New mode attributes.
4171 (V_reg): Define mapping for V2HF.
4172 (V_hi, V_lo): New mode attributes.
4173 (VF_constraint): Likewise.
4174 (vfml_half, vfml_half_selector): New int attributes.
4175 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4177 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4178 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4180 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4181 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4182 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4183 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4185 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4186 Document new effective target and option set.
4188 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4190 * config/arm/arm-cpus.in (armv8_4): New feature.
4191 (ARMv8_4a): New fgroup.
4192 (armv8.4-a): New arch.
4193 * config/arm/arm-tables.opt: Regenerate.
4194 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4195 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4196 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4197 Add matching rules for -march=armv8.4-a and extensions.
4198 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4200 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4203 * config/rx/rx.md (BW): New mode attribute.
4204 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4206 2018-01-11 Richard Biener <rguenther@suse.de>
4208 PR tree-optimization/83435
4209 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4210 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4211 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4213 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4214 Alan Hayward <alan.hayward@arm.com>
4215 David Sherwood <david.sherwood@arm.com>
4217 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4219 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4220 (aarch64_print_address_internal): Use it to check for a zero offset.
4222 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4223 Alan Hayward <alan.hayward@arm.com>
4224 David Sherwood <david.sherwood@arm.com>
4226 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4227 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4228 Return a poly_int64 rather than a HOST_WIDE_INT.
4229 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4230 rather than a HOST_WIDE_INT.
4231 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4232 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4233 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4234 final_offset from HOST_WIDE_INT to poly_int64.
4235 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4236 to_constant when getting the number of units in an Advanced SIMD
4238 (aarch64_builtin_vectorized_function): Check for a constant number
4240 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4242 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4243 attribute instead of GET_MODE_NUNITS.
4244 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4245 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4246 GET_MODE_SIZE for fixed-size registers.
4247 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4248 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4249 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4250 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4251 (aarch64_print_operand, aarch64_print_address_internal)
4252 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4253 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4254 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4255 Handle polynomial GET_MODE_SIZE.
4256 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4257 wider than SImode without modification.
4258 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4259 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4260 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4261 passing and returning SVE modes.
4262 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4263 rather than GEN_INT.
4264 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4265 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4266 (aarch64_allocate_and_probe_stack_space): Likewise.
4267 (aarch64_layout_frame): Cope with polynomial offsets.
4268 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4269 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4271 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4272 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4273 poly_int64 rather than a HOST_WIDE_INT.
4274 (aarch64_get_separate_components, aarch64_process_components)
4275 (aarch64_expand_prologue, aarch64_expand_epilogue)
4276 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4277 (aarch64_anchor_offset): New function, split out from...
4278 (aarch64_legitimize_address): ...here.
4279 (aarch64_builtin_vectorization_cost): Handle polynomial
4280 TYPE_VECTOR_SUBPARTS.
4281 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4283 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4284 number of elements from the PARALLEL rather than the mode.
4285 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4286 rather than GET_MODE_BITSIZE.
4287 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4288 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4289 (aarch64_expand_vec_perm_const_1): Handle polynomial
4290 d->perm.length () and d->perm elements.
4291 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4292 Apply to_constant to d->perm elements.
4293 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4294 polynomial CONST_VECTOR_NUNITS.
4295 (aarch64_move_pointer): Take amount as a poly_int64 rather
4297 (aarch64_progress_pointer): Avoid temporary variable.
4298 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4299 the mode attribute instead of GET_MODE.
4301 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4302 Alan Hayward <alan.hayward@arm.com>
4303 David Sherwood <david.sherwood@arm.com>
4305 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4306 x exists before using it.
4307 (aarch64_add_constant_internal): Rename to...
4308 (aarch64_add_offset_1): ...this. Replace regnum with separate
4309 src and dest rtxes. Handle the case in which they're different,
4310 including when the offset is zero. Replace scratchreg with an rtx.
4311 Use 2 additions if there is no spare register into which we can
4312 move a 16-bit constant.
4313 (aarch64_add_constant): Delete.
4314 (aarch64_add_offset): Replace reg with separate src and dest
4315 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4316 Use aarch64_add_offset_1.
4317 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4318 an rtx rather than an int. Take the delta as a poly_int64
4319 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4320 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4321 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4322 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4323 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4325 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4326 aarch64_add_constant.
4328 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4330 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4331 Use scalar_float_mode.
4333 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4335 * config/aarch64/aarch64-simd.md
4336 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4337 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4338 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4339 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4340 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4341 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4342 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4343 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4344 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4345 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4347 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4350 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4351 targ_options->x_arm_arch_string is non NULL.
4353 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4355 * config/aarch64/aarch64.h
4356 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4358 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4361 * expmed.c (emit_store_flag_force): Swap if const op0
4362 and change VOIDmode to mode of op0.
4364 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4366 PR rtl-optimization/83761
4367 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4368 than bytes to mode_for_size.
4370 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4373 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4374 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4377 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4380 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4381 when in layout mode.
4382 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4383 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4386 2018-01-10 Michael Collison <michael.collison@arm.com>
4388 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4389 * config/aarch64/aarch64-option-extension.def: Add
4390 AARCH64_OPT_EXTENSION of 'fp16fml'.
4391 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4392 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4393 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4394 * config/aarch64/constraints.md (Ui7): New constraint.
4395 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4396 (VFMLA_SEL_W): Ditto.
4399 (VFMLA16_LOW): New int iterator.
4400 (VFMLA16_HIGH): Ditto.
4401 (UNSPEC_FMLAL): New unspec.
4402 (UNSPEC_FMLSL): Ditto.
4403 (UNSPEC_FMLAL2): Ditto.
4404 (UNSPEC_FMLSL2): Ditto.
4405 (f16mac): New code attribute.
4406 * config/aarch64/aarch64-simd-builtins.def
4407 (aarch64_fmlal_lowv2sf): Ditto.
4408 (aarch64_fmlsl_lowv2sf): Ditto.
4409 (aarch64_fmlalq_lowv4sf): Ditto.
4410 (aarch64_fmlslq_lowv4sf): Ditto.
4411 (aarch64_fmlal_highv2sf): Ditto.
4412 (aarch64_fmlsl_highv2sf): Ditto.
4413 (aarch64_fmlalq_highv4sf): Ditto.
4414 (aarch64_fmlslq_highv4sf): Ditto.
4415 (aarch64_fmlal_lane_lowv2sf): Ditto.
4416 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4417 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4418 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4419 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4420 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4421 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4422 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4423 (aarch64_fmlal_lane_highv2sf): Ditto.
4424 (aarch64_fmlsl_lane_highv2sf): Ditto.
4425 (aarch64_fmlal_laneq_highv2sf): Ditto.
4426 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4427 (aarch64_fmlalq_lane_highv4sf): Ditto.
4428 (aarch64_fmlsl_lane_highv4sf): Ditto.
4429 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4430 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4431 * config/aarch64/aarch64-simd.md:
4432 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4433 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4434 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4435 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4436 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4437 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4438 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4439 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4440 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4441 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4442 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4443 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4444 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4445 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4446 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4447 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4448 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4449 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4450 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4451 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4452 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4453 (vfmlsl_low_u32): Ditto.
4454 (vfmlalq_low_u32): Ditto.
4455 (vfmlslq_low_u32): Ditto.
4456 (vfmlal_high_u32): Ditto.
4457 (vfmlsl_high_u32): Ditto.
4458 (vfmlalq_high_u32): Ditto.
4459 (vfmlslq_high_u32): Ditto.
4460 (vfmlal_lane_low_u32): Ditto.
4461 (vfmlsl_lane_low_u32): Ditto.
4462 (vfmlal_laneq_low_u32): Ditto.
4463 (vfmlsl_laneq_low_u32): Ditto.
4464 (vfmlalq_lane_low_u32): Ditto.
4465 (vfmlslq_lane_low_u32): Ditto.
4466 (vfmlalq_laneq_low_u32): Ditto.
4467 (vfmlslq_laneq_low_u32): Ditto.
4468 (vfmlal_lane_high_u32): Ditto.
4469 (vfmlsl_lane_high_u32): Ditto.
4470 (vfmlal_laneq_high_u32): Ditto.
4471 (vfmlsl_laneq_high_u32): Ditto.
4472 (vfmlalq_lane_high_u32): Ditto.
4473 (vfmlslq_lane_high_u32): Ditto.
4474 (vfmlalq_laneq_high_u32): Ditto.
4475 (vfmlslq_laneq_high_u32): Ditto.
4476 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4477 (AARCH64_FL_FOR_ARCH8_4): New.
4478 (AARCH64_ISA_F16FML): New ISA flag.
4479 (TARGET_F16FML): New feature flag for fp16fml.
4480 (doc/invoke.texi): Document new fp16fml option.
4482 2018-01-10 Michael Collison <michael.collison@arm.com>
4484 * config/aarch64/aarch64-builtins.c:
4485 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4486 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4487 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4488 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4489 (AARCH64_ISA_SHA3): New ISA flag.
4490 (TARGET_SHA3): New feature flag for sha3.
4491 * config/aarch64/iterators.md (sha512_op): New int attribute.
4492 (CRYPTO_SHA512): New int iterator.
4493 (UNSPEC_SHA512H): New unspec.
4494 (UNSPEC_SHA512H2): Ditto.
4495 (UNSPEC_SHA512SU0): Ditto.
4496 (UNSPEC_SHA512SU1): Ditto.
4497 * config/aarch64/aarch64-simd-builtins.def
4498 (aarch64_crypto_sha512hqv2di): New builtin.
4499 (aarch64_crypto_sha512h2qv2di): Ditto.
4500 (aarch64_crypto_sha512su0qv2di): Ditto.
4501 (aarch64_crypto_sha512su1qv2di): Ditto.
4502 (aarch64_eor3qv8hi): Ditto.
4503 (aarch64_rax1qv2di): Ditto.
4504 (aarch64_xarqv2di): Ditto.
4505 (aarch64_bcaxqv8hi): Ditto.
4506 * config/aarch64/aarch64-simd.md:
4507 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4508 (aarch64_crypto_sha512su0qv2di): Ditto.
4509 (aarch64_crypto_sha512su1qv2di): Ditto.
4510 (aarch64_eor3qv8hi): Ditto.
4511 (aarch64_rax1qv2di): Ditto.
4512 (aarch64_xarqv2di): Ditto.
4513 (aarch64_bcaxqv8hi): Ditto.
4514 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4515 (vsha512h2q_u64): Ditto.
4516 (vsha512su0q_u64): Ditto.
4517 (vsha512su1q_u64): Ditto.
4518 (veor3q_u16): Ditto.
4519 (vrax1q_u64): Ditto.
4521 (vbcaxq_u16): Ditto.
4522 * config/arm/types.md (crypto_sha512): New type attribute.
4523 (crypto_sha3): Ditto.
4524 (doc/invoke.texi): Document new sha3 option.
4526 2018-01-10 Michael Collison <michael.collison@arm.com>
4528 * config/aarch64/aarch64-builtins.c:
4529 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4530 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4531 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4532 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4533 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4534 (AARCH64_ISA_SM4): New ISA flag.
4535 (TARGET_SM4): New feature flag for sm4.
4536 * config/aarch64/aarch64-simd-builtins.def
4537 (aarch64_sm3ss1qv4si): Ditto.
4538 (aarch64_sm3tt1aq4si): Ditto.
4539 (aarch64_sm3tt1bq4si): Ditto.
4540 (aarch64_sm3tt2aq4si): Ditto.
4541 (aarch64_sm3tt2bq4si): Ditto.
4542 (aarch64_sm3partw1qv4si): Ditto.
4543 (aarch64_sm3partw2qv4si): Ditto.
4544 (aarch64_sm4eqv4si): Ditto.
4545 (aarch64_sm4ekeyqv4si): Ditto.
4546 * config/aarch64/aarch64-simd.md:
4547 (aarch64_sm3ss1qv4si): Ditto.
4548 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4549 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4550 (aarch64_sm4eqv4si): Ditto.
4551 (aarch64_sm4ekeyqv4si): Ditto.
4552 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4553 (sm3part_op): Ditto.
4554 (CRYPTO_SM3TT): Ditto.
4555 (CRYPTO_SM3PART): Ditto.
4556 (UNSPEC_SM3SS1): New unspec.
4557 (UNSPEC_SM3TT1A): Ditto.
4558 (UNSPEC_SM3TT1B): Ditto.
4559 (UNSPEC_SM3TT2A): Ditto.
4560 (UNSPEC_SM3TT2B): Ditto.
4561 (UNSPEC_SM3PARTW1): Ditto.
4562 (UNSPEC_SM3PARTW2): Ditto.
4563 (UNSPEC_SM4E): Ditto.
4564 (UNSPEC_SM4EKEY): Ditto.
4565 * config/aarch64/constraints.md (Ui2): New constraint.
4566 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4567 * config/arm/types.md (crypto_sm3): New type attribute.
4568 (crypto_sm4): Ditto.
4569 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4570 (vsm3tt1aq_u32): Ditto.
4571 (vsm3tt1bq_u32): Ditto.
4572 (vsm3tt2aq_u32): Ditto.
4573 (vsm3tt2bq_u32): Ditto.
4574 (vsm3partw1q_u32): Ditto.
4575 (vsm3partw2q_u32): Ditto.
4576 (vsm4eq_u32): Ditto.
4577 (vsm4ekeyq_u32): Ditto.
4578 (doc/invoke.texi): Document new sm4 option.
4580 2018-01-10 Michael Collison <michael.collison@arm.com>
4582 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4583 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4584 (AARCH64_FL_FOR_ARCH8_4): New.
4585 (AARCH64_FL_V8_4): New flag.
4586 (doc/invoke.texi): Document new armv8.4-a option.
4588 2018-01-10 Michael Collison <michael.collison@arm.com>
4590 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4591 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4592 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4593 * config/aarch64/aarch64-option-extension.def: Add
4594 AARCH64_OPT_EXTENSION of 'sha2'.
4595 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4596 (crypto): Disable sha2 and aes if crypto disabled.
4597 (crypto): Enable aes and sha2 if enabled.
4598 (simd): Disable sha2 and aes if simd disabled.
4599 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4601 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4602 (TARGET_SHA2): New feature flag for sha2.
4603 (TARGET_AES): New feature flag for aes.
4604 * config/aarch64/aarch64-simd.md:
4605 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4606 conditional on TARGET_AES.
4607 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4608 (aarch64_crypto_sha1hsi): Make pattern conditional
4610 (aarch64_crypto_sha1hv4si): Ditto.
4611 (aarch64_be_crypto_sha1hv4si): Ditto.
4612 (aarch64_crypto_sha1su1v4si): Ditto.
4613 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4614 (aarch64_crypto_sha1su0v4si): Ditto.
4615 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4616 (aarch64_crypto_sha256su0v4si): Ditto.
4617 (aarch64_crypto_sha256su1v4si): Ditto.
4618 (doc/invoke.texi): Document new aes and sha2 options.
4620 2018-01-10 Martin Sebor <msebor@redhat.com>
4622 PR tree-optimization/83781
4623 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4626 2018-01-11 Martin Sebor <msebor@gmail.com>
4627 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4629 PR tree-optimization/83501
4630 PR tree-optimization/81703
4632 * tree-ssa-strlen.c (get_string_cst): Rename...
4633 (get_string_len): ...to this. Handle global constants.
4634 (handle_char_store): Adjust.
4636 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4637 Jim Wilson <jimw@sifive.com>
4639 * config/riscv/riscv-protos.h (riscv_output_return): New.
4640 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4641 (riscv_attribute_table, riscv_output_return),
4642 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4643 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4644 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4646 (riscv_expand_prologue): Add early return for naked function.
4647 (riscv_expand_epilogue): Likewise.
4648 (riscv_function_ok_for_sibcall): Return false for naked function.
4649 (riscv_set_current_function): New.
4650 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4651 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4652 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4653 * doc/extend.texi (RISC-V Function Attributes): New.
4655 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4657 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4658 check for 128-bit long double before checking TCmode.
4659 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4660 128-bit long doubles before checking TFmode or TCmode.
4661 (FLOAT128_IBM_P): Likewise.
4663 2018-01-10 Martin Sebor <msebor@redhat.com>
4665 PR tree-optimization/83671
4666 * builtins.c (c_strlen): Unconditionally return zero for the empty
4668 Use -Warray-bounds for warnings.
4669 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4670 for non-constant array indices with COMPONENT_REF, arrays of
4671 arrays, and pointers to arrays.
4672 (gimple_fold_builtin_strlen): Determine and set length range for
4673 non-constant character arrays.
4675 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4678 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4681 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4683 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4685 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4688 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4689 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4690 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4691 indexed_or_indirect_operand predicate.
4692 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4693 (*vsx_le_perm_load_v8hi): Likewise.
4694 (*vsx_le_perm_load_v16qi): Likewise.
4695 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4696 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4697 (*vsx_le_perm_store_v8hi): Likewise.
4698 (*vsx_le_perm_store_v16qi): Likewise.
4699 (eight unnamed splitters): Likewise.
4701 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4703 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4704 * config/rs6000/emmintrin.h: Likewise.
4705 * config/rs6000/mmintrin.h: Likewise.
4706 * config/rs6000/xmmintrin.h: Likewise.
4708 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4711 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4713 * tree.c (tree_nop_conversion): Return true for location wrapper
4715 (maybe_wrap_with_location): New function.
4716 (selftest::check_strip_nops): New function.
4717 (selftest::test_location_wrappers): New function.
4718 (selftest::tree_c_tests): Call it.
4719 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4720 (maybe_wrap_with_location): New decl.
4721 (EXPR_LOCATION_WRAPPER_P): New macro.
4722 (location_wrapper_p): New inline function.
4723 (tree_strip_any_location_wrapper): New inline function.
4725 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4728 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4729 stack_realign_offset for the largest alignment of stack slot
4731 (ix86_find_max_used_stack_alignment): New function.
4732 (ix86_finalize_stack_frame_flags): Use it. Set
4733 max_used_stack_alignment if we don't realign stack.
4734 * config/i386/i386.h (machine_function): Add
4735 max_used_stack_alignment.
4737 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4739 * config/arm/arm.opt (-mbranch-cost): New option.
4740 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4743 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4746 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4747 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4749 2018-01-10 Richard Biener <rguenther@suse.de>
4752 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4753 early out so it also covers the case where we have a non-NULL
4756 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4758 PR tree-optimization/83753
4759 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4760 for non-strided grouped accesses if the number of elements is 1.
4762 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4765 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4766 * i386.h (TARGET_USE_GATHER): Define.
4767 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4769 2018-01-10 Martin Liska <mliska@suse.cz>
4772 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4773 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4775 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4776 CLEANUP_NO_PARTITIONING is not set.
4778 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4780 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4781 for vectors, as a partial revert of r254296.
4782 * rtl.h (const_vec_p): Delete.
4783 (const_vec_duplicate_p): Don't test for vector CONSTs.
4784 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4785 * expmed.c (make_tree): Likewise.
4788 * common.md (E, F): Use CONSTANT_P instead of checking for
4790 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4791 checking for CONST_VECTOR.
4793 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4796 * predict.c (force_edge_cold): Handle in more sane way edges
4799 2018-01-09 Carl Love <cel@us.ibm.com>
4801 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4803 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4804 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4805 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4806 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4807 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4808 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4809 * config/rs6000/rs6000-protos.h: Add extern defition for
4810 rs6000_generate_float2_double_code.
4811 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4813 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4814 (float2_v2df): Add define_expand.
4816 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4819 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4820 op_mode in the force_to_mode call.
4822 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4824 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4825 instead of checking each element individually.
4826 (aarch64_evpc_uzp): Likewise.
4827 (aarch64_evpc_zip): Likewise.
4828 (aarch64_evpc_ext): Likewise.
4829 (aarch64_evpc_rev): Likewise.
4830 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4831 instead of checking each element individually. Return true without
4833 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4834 whether all selected elements come from the same input, instead of
4835 checking each element individually. Remove calls to gen_rtx_REG,
4836 start_sequence and end_sequence and instead assert that no rtl is
4839 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4841 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4842 order of HIGH and CONST checks.
4844 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4846 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4847 if the destination isn't an SSA_NAME.
4849 2018-01-09 Richard Biener <rguenther@suse.de>
4851 PR tree-optimization/83668
4852 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4854 (canonicalize_loop_form): ... here, renamed from ...
4855 (canonicalize_loop_closed_ssa_form): ... this and amended to
4856 swap successor edges for loop exit blocks to make us use
4857 the RPO order we need for initial schedule generation.
4859 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4861 PR tree-optimization/64811
4862 * match.pd: When optimizing comparisons with Inf, avoid
4863 introducing or losing exceptions from comparisons with NaN.
4865 2018-01-09 Martin Liska <mliska@suse.cz>
4868 * asan.c (shadow_mem_size): Add gcc_assert.
4870 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4872 Don't save registers in main().
4875 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4876 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4877 * config/avr/avr.c (avr_set_current_function): Don't error if
4878 naked, OS_task or OS_main are specified at the same time.
4879 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4881 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4883 * common/config/avr/avr-common.c (avr_option_optimization_table):
4884 Switch on -mmain-is-OS_task for optimizing compilations.
4886 2018-01-09 Richard Biener <rguenther@suse.de>
4888 PR tree-optimization/83572
4889 * graphite.c: Include cfganal.h.
4890 (graphite_transform_loops): Connect infinite loops to exit
4891 and remove fake edges at the end.
4893 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4895 * ipa-inline.c (edge_badness): Revert accidental checkin.
4897 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4900 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4901 symbols; not inline clones.
4903 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4906 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4907 hard registers. Formatting fixes.
4909 PR preprocessor/83722
4910 * gcc.c (try_generate_repro): Pass
4911 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4912 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4915 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4916 Kito Cheng <kito.cheng@gmail.com>
4918 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4919 (riscv_leaf_function_p): Delete.
4920 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4922 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4924 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4926 (do_ifelse): New function.
4927 (do_isel): New function.
4928 (do_sub3): New function.
4929 (do_add3): New function.
4930 (do_load_mask_compare): New function.
4931 (do_overlap_load_compare): New function.
4932 (expand_compare_loop): New function.
4933 (expand_block_compare): Call expand_compare_loop() when appropriate.
4934 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4936 (-mblock-compare-inline-loop-limit): New option.
4938 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4941 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4942 Reverse order of second and third operands in first alternative.
4943 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4944 of first and second elements in UNSPEC_VPERMR vector.
4945 (altivec_expand_vec_perm_le): Likewise.
4947 2018-01-08 Jeff Law <law@redhat.com>
4949 PR rtl-optimizatin/81308
4950 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4951 (process_switch): If group_case_labels makes a change, then set
4953 (pass_convert_switch::execute): If a switch is converted, then
4954 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4956 PR rtl-optimization/81308
4957 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4960 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4962 PR target/83663 - Revert r255946
4963 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4964 generation for cases where splatting a value is not useful.
4965 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4966 across a vec_duplicate and a paradoxical subreg forming a vector
4967 mode to a vec_concat.
4969 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4971 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4972 -march=armv8.3-a variants.
4973 * config/arm/t-multilib: Likewise.
4974 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4976 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4978 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4980 (cceq_ior_compare_complement): Give it a name so I can use it, and
4981 change boolean_or_operator predicate to boolean_operator so it can
4982 be used to generate a crand.
4983 (eqne): New code iterator.
4984 (bd/bd_neg): New code_attrs.
4985 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4986 a single define_insn.
4987 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4988 decrement (bdnzt/bdnzf/bdzt/bdzf).
4989 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4990 with the new names of the branch decrement patterns, and added the
4991 names of the branch decrement conditional patterns.
4993 2018-01-08 Richard Biener <rguenther@suse.de>
4995 PR tree-optimization/83563
4996 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4999 2018-01-08 Richard Biener <rguenther@suse.de>
5002 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5004 2018-01-08 Richard Biener <rguenther@suse.de>
5006 PR tree-optimization/83685
5007 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5008 references to abnormals.
5010 2018-01-08 Richard Biener <rguenther@suse.de>
5013 * dwarf2out.c (output_indirect_strings): Handle empty
5014 skeleton_debug_str_hash.
5015 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5017 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5019 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5020 (emit_store_direct): Likewise.
5021 (arc_trampoline_adjust_address): Likewise.
5022 (arc_asm_trampoline_template): New function.
5023 (arc_initialize_trampoline): Use asm_trampoline_template.
5024 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5025 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5026 * config/arc/arc.md (flush_icache): Delete pattern.
5028 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5030 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5031 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5034 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5037 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5038 by not USED_FOR_TARGET.
5039 (make_pass_resolve_sw_modes): Likewise.
5041 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5043 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5046 2018-01-08 Richard Biener <rguenther@suse.de>
5049 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5051 2018-01-08 Richard Biener <rguenther@suse.de>
5054 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5056 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5059 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5060 basic blocks with a small number of successors.
5061 (convert_control_dep_chain_into_preds): Improve handling of
5063 (dump_predicates): Split apart into...
5064 (dump_pred_chain): ...here...
5065 (dump_pred_info): ...and here.
5066 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5067 (can_chain_union_be_invalidated_p): Improve check for invalidation
5069 (uninit_uses_cannot_happen): Avoid unnecessary if
5070 convert_control_dep_chain_into_preds yielded nothing.
5072 2018-01-06 Martin Sebor <msebor@redhat.com>
5074 PR tree-optimization/83640
5075 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5076 subtracting negative offset from size.
5077 (builtin_access::overlap): Adjust offset bounds of the access to fall
5078 within the size of the object if possible.
5080 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5082 PR rtl-optimization/83699
5083 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5084 extract_bit_field_as_subreg to cases in which the extracted
5085 value is also a vector.
5087 * lra-constraints.c (process_alt_operands): Test for the equivalence
5088 substitutions when detecting a possible reload cycle.
5090 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5093 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5094 by default if flag_selective_schedling{,2}. Formatting fixes.
5096 PR rtl-optimization/83682
5097 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5098 if it has non-VECTOR_MODE element mode.
5099 (vec_duplicate_p): Likewise.
5102 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5103 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5105 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5108 * config/i386/i386-builtin.def
5109 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5110 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5111 Require also OPTION_MASK_ISA_AVX512F in addition to
5112 OPTION_MASK_ISA_GFNI.
5113 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5114 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5115 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5116 to OPTION_MASK_ISA_GFNI.
5117 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5118 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5119 OPTION_MASK_ISA_AVX512BW.
5120 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5121 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5122 addition to OPTION_MASK_ISA_GFNI.
5123 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5124 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5125 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5126 to OPTION_MASK_ISA_GFNI.
5127 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5128 a requirement for all ISAs rather than any of them with a few
5130 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5132 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5133 bitmasks to be enabled with 3 exceptions, instead of requiring any
5134 enabled ISA with lots of exceptions.
5135 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5136 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5137 Change avx512bw in isa attribute to avx512f.
5138 * config/i386/sgxintrin.h: Add license boilerplate.
5139 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5140 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5141 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5142 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5144 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5145 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5146 temporarily sse2 rather than sse if not enabled already.
5149 * config/i386/sse.md (VI248_VLBW): Rename to ...
5150 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5151 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5152 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5153 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5154 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5155 mode iterator instead of VI248_VLBW.
5157 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5159 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5160 (record_modified): Skip clobbers; add debug output.
5161 (param_change_prob): Use sreal frequencies.
5163 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5165 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5166 punt for user-aligned variables.
5168 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5170 * tree-chrec.c (chrec_contains_symbols): Return true for
5173 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5176 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5177 of (x|y) == x for BICS pattern.
5179 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5181 PR tree-optimization/83605
5182 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5183 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5186 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5188 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5189 * config/epiphany/rtems.h: New file.
5191 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5192 Uros Bizjak <ubizjak@gmail.com>
5195 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5196 QIreg_operand instead of register_operand predicate.
5197 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5198 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5199 comments instead of -fmitigate[-_]rop.
5201 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5204 * cgraphunit.c (symbol_table::compile): Switch to text_section
5205 before calling assembly_start debug hook.
5206 * run-rtl-passes.c (run_rtl_passes): Likewise.
5209 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5211 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5212 range_int_cst_p rather than !symbolic_range_p before calling
5213 extract_range_from_multiplicative_op_1.
5215 2018-01-04 Jeff Law <law@redhat.com>
5217 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5218 redundant test in assertion.
5220 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5222 * doc/rtl.texi: Document machine_mode wrapper classes.
5224 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5226 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5229 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5231 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5232 the VEC_PERM_EXPR fold to fail.
5234 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5237 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5238 to switched_sections.
5240 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5243 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5246 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5249 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5250 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5252 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5255 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5256 is BLKmode and bitpos not zero or mode change is needed.
5258 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5261 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5264 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5267 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5268 instead of MULT rtx. Update all corresponding splitters.
5270 (*ssub<modesuffix>): Ditto.
5272 (*cmp_sadd_di): Update split patterns.
5273 (*cmp_sadd_si): Ditto.
5274 (*cmp_sadd_sidi): Ditto.
5275 (*cmp_ssub_di): Ditto.
5276 (*cmp_ssub_si): Ditto.
5277 (*cmp_ssub_sidi): Ditto.
5278 * config/alpha/predicates.md (const23_operand): New predicate.
5279 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5280 Look for ASHIFT, not MULT inner operand.
5281 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5283 2018-01-04 Martin Liska <mliska@suse.cz>
5285 PR gcov-profile/83669
5286 * gcov.c (output_intermediate_file): Add version to intermediate
5288 * doc/gcov.texi: Document new field 'version' in intermediate
5289 file format. Fix location of '-k' option of gcov command.
5291 2018-01-04 Martin Liska <mliska@suse.cz>
5294 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5296 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5298 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5300 2018-01-03 Martin Sebor <msebor@redhat.com>
5302 PR tree-optimization/83655
5303 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5304 checking calls with invalid arguments.
5306 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5308 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5309 (vectorizable_mask_load_store): Delete.
5310 (vectorizable_call): Return false for masked loads and stores.
5311 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5312 instead of gimple_assign_rhs1.
5313 (vectorizable_load): Handle IFN_MASK_LOAD.
5314 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5316 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5318 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5320 (vectorizable_mask_load_store): ...here.
5321 (vectorizable_load): ...and here.
5323 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5325 * tree-vect-stmts.c (vect_build_all_ones_mask)
5326 (vect_build_zero_merge_argument): New functions, split out from...
5327 (vectorizable_load): ...here.
5329 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5331 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5333 (vectorizable_mask_load_store): ...here.
5334 (vectorizable_store): ...and here.
5336 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5338 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5340 (vectorizable_mask_load_store): ...here.
5342 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5344 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5345 (vect_model_store_cost): Take a vec_load_store_type instead of a
5347 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5348 (vect_model_store_cost): Take a vec_load_store_type instead of a
5350 (vectorizable_mask_load_store): Update accordingly.
5351 (vectorizable_store): Likewise.
5352 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5356 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5357 IFN_MASK_LOAD calls here rather than...
5358 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5361 Alan Hayward <alan.hayward@arm.com>
5362 David Sherwood <david.sherwood@arm.com>
5364 * expmed.c (extract_bit_field_1): For vector extracts,
5365 fall back to extract_bit_field_as_subreg if vec_extract
5368 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5369 Alan Hayward <alan.hayward@arm.com>
5370 David Sherwood <david.sherwood@arm.com>
5372 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5373 they are variable or constant sized.
5374 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5375 slots for constant-sized data.
5377 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5378 Alan Hayward <alan.hayward@arm.com>
5379 David Sherwood <david.sherwood@arm.com>
5381 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5382 handling COND_EXPRs with boolean comparisons, try to find a better
5383 basis for the mask type than the boolean itself.
5385 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5387 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5388 is calculated and how it can be overridden.
5389 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5390 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5392 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5396 Alan Hayward <alan.hayward@arm.com>
5397 David Sherwood <david.sherwood@arm.com>
5399 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5400 Remove the mode argument.
5401 (aarch64_simd_valid_immediate): Remove the mode and inverse
5403 * config/aarch64/iterators.md (bitsize): New iterator.
5404 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5405 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5406 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5407 aarch64_simd_valid_immediate.
5408 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5409 (aarch64_reg_or_bic_imm): Likewise.
5410 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5411 with an insn_type enum and msl with a modifier_type enum.
5412 Replace element_width with a scalar_mode. Change the shift
5413 to unsigned int. Add constructors for scalar_float_mode and
5414 scalar_int_mode elements.
5415 (aarch64_vect_float_const_representable_p): Delete.
5416 (aarch64_can_const_movi_rtx_p)
5417 (aarch64_simd_scalar_immediate_valid_for_move)
5418 (aarch64_simd_make_constant): Update call to
5419 aarch64_simd_valid_immediate.
5420 (aarch64_advsimd_valid_immediate_hs): New function.
5421 (aarch64_advsimd_valid_immediate): Likewise.
5422 (aarch64_simd_valid_immediate): Remove mode and inverse
5423 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5424 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5425 and aarch64_float_const_representable_p on the result.
5426 (aarch64_output_simd_mov_immediate): Remove mode argument.
5427 Update call to aarch64_simd_valid_immediate and use of
5428 simd_immediate_info.
5429 (aarch64_output_scalar_simd_mov_immediate): Update call
5432 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5433 Alan Hayward <alan.hayward@arm.com>
5434 David Sherwood <david.sherwood@arm.com>
5436 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5437 (mode_nunits): Likewise CONST_MODE_NUNITS.
5438 * machmode.def (ADJUST_NUNITS): Document.
5439 * genmodes.c (mode_data::need_nunits_adj): New field.
5440 (blank_mode): Update accordingly.
5441 (adj_nunits): New variable.
5442 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5444 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5445 listed in adj_nunits.
5446 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5447 listed in adj_nunits. Don't emit case statements for such modes.
5448 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5449 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5450 nothing if adj_nunits is nonnull.
5451 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5452 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5453 (emit_mode_fbit): Update use of print_maybe_const_decl.
5454 (emit_move_size): Likewise. Treat the array as non-const
5456 (emit_mode_adjustments): Handle adj_nunits.
5458 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5460 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5461 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5462 (VECTOR_MODES): Use it.
5463 (make_vector_modes): Take the prefix as an argument.
5465 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5466 Alan Hayward <alan.hayward@arm.com>
5467 David Sherwood <david.sherwood@arm.com>
5469 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5470 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5471 for MODE_VECTOR_BOOL.
5472 * machmode.def (VECTOR_BOOL_MODE): Document.
5473 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5474 (make_vector_bool_mode): New function.
5475 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5477 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5478 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5480 * stor-layout.c (int_mode_for_mode): Likewise.
5481 * tree.c (build_vector_type_for_mode): Likewise.
5482 * varasm.c (output_constant_pool_2): Likewise.
5483 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5484 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5485 for MODE_VECTOR_BOOL.
5486 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5487 of mode class checks.
5488 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5489 instead of a list of mode class checks.
5490 (expand_vector_scalar_condition): Likewise.
5491 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5493 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5494 Alan Hayward <alan.hayward@arm.com>
5495 David Sherwood <david.sherwood@arm.com>
5497 * machmode.h (mode_size): Change from unsigned short to
5499 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5500 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5501 or if measurement_type is not polynomial.
5502 (fixed_size_mode::includes_p): Check for constant-sized modes.
5503 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5504 return a poly_uint16 rather than an unsigned short.
5505 (emit_mode_size): Change the type of mode_size from unsigned short
5506 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5507 (emit_mode_adjustments): Cope with polynomial vector sizes.
5508 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5510 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5512 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5513 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5514 * caller-save.c (setup_save_areas): Likewise.
5515 (replace_reg_with_saved_mem): Likewise.
5516 * calls.c (emit_library_call_value_1): Likewise.
5517 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5518 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5519 (gen_lowpart_for_combine): Likewise.
5520 * convert.c (convert_to_integer_1): Likewise.
5521 * cse.c (equiv_constant, cse_insn): Likewise.
5522 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5523 (cselib_subst_to_values): Likewise.
5524 * dce.c (word_dce_process_block): Likewise.
5525 * df-problems.c (df_word_lr_mark_ref): Likewise.
5526 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5527 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5528 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5529 (rtl_for_decl_location): Likewise.
5530 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5531 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5532 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5533 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5534 (expand_expr_real_1): Likewise.
5535 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5536 (pad_below): Likewise.
5537 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5538 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5539 * ira.c (get_subreg_tracking_sizes): Likewise.
5540 * ira-build.c (ira_create_allocno_objects): Likewise.
5541 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5542 (ira_sort_regnos_for_alter_reg): Likewise.
5543 * ira-costs.c (record_operand_costs): Likewise.
5544 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5545 (resolve_simple_move): Likewise.
5546 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5547 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5548 (lra_constraints): Likewise.
5549 (CONST_POOL_OK_P): Reject variable-sized modes.
5550 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5551 (add_pseudo_to_slot, lra_spill): Likewise.
5552 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5553 * optabs-query.c (get_best_extraction_insn): Likewise.
5554 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5555 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5556 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5557 * recog.c (offsettable_address_addr_space_p): Likewise.
5558 * regcprop.c (maybe_mode_change): Likewise.
5559 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5560 * regrename.c (build_def_use): Likewise.
5561 * regstat.c (dump_reg_info): Likewise.
5562 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5563 (find_reloads, find_reloads_subreg_address): Likewise.
5564 * reload1.c (eliminate_regs_1): Likewise.
5565 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5566 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5567 (simplify_binary_operation_1, simplify_subreg): Likewise.
5568 * targhooks.c (default_function_arg_padding): Likewise.
5569 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5570 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5571 (verify_gimple_assign_ternary): Likewise.
5572 * tree-inline.c (estimate_move_cost): Likewise.
5573 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5574 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5575 (get_address_cost_ainc): Likewise.
5576 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5577 (vect_supportable_dr_alignment): Likewise.
5578 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5579 (vectorizable_reduction): Likewise.
5580 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5581 (vectorizable_operation, vectorizable_load): Likewise.
5582 * tree.c (build_same_sized_truth_vector_type): Likewise.
5583 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5584 * var-tracking.c (emit_note_insn_var_location): Likewise.
5585 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5586 (ADDR_VEC_ALIGN): Likewise.
5588 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5589 Alan Hayward <alan.hayward@arm.com>
5590 David Sherwood <david.sherwood@arm.com>
5592 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5594 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5595 or if measurement_type is polynomial.
5596 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5597 * combine.c (make_extraction): Likewise.
5598 * dse.c (find_shift_sequence): Likewise.
5599 * dwarf2out.c (mem_loc_descriptor): Likewise.
5600 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5601 (extract_bit_field, extract_low_bits): Likewise.
5602 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5603 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5604 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5605 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5606 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5607 * reload.c (find_reloads): Likewise.
5608 * reload1.c (alter_reg): Likewise.
5609 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5610 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5611 * tree-if-conv.c (predicate_mem_writes): Likewise.
5612 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5613 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5614 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5615 * valtrack.c (dead_debug_insert_temp): Likewise.
5616 * varasm.c (mergeable_constant_section): Likewise.
5617 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5620 Alan Hayward <alan.hayward@arm.com>
5621 David Sherwood <david.sherwood@arm.com>
5623 * expr.c (expand_assignment): Cope with polynomial mode sizes
5624 when assigning to a CONCAT.
5626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5627 Alan Hayward <alan.hayward@arm.com>
5628 David Sherwood <david.sherwood@arm.com>
5630 * machmode.h (mode_precision): Change from unsigned short to
5632 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5634 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5635 or if measurement_type is not polynomial.
5636 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5637 in which the mode is already known to be a scalar_int_mode.
5638 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5639 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5641 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5642 for GET_MODE_PRECISION.
5643 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5644 for GET_MODE_PRECISION.
5645 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5647 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5648 (expand_field_assignment, make_extraction): Likewise.
5649 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5650 (get_last_value): Likewise.
5651 * convert.c (convert_to_integer_1): Likewise.
5652 * cse.c (cse_insn): Likewise.
5653 * expr.c (expand_expr_real_1): Likewise.
5654 * lra-constraints.c (simplify_operand_subreg): Likewise.
5655 * optabs-query.c (can_atomic_load_p): Likewise.
5656 * optabs.c (expand_atomic_load): Likewise.
5657 (expand_atomic_store): Likewise.
5658 * ree.c (combine_reaching_defs): Likewise.
5659 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5660 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5661 * tree.h (type_has_mode_precision_p): Likewise.
5662 * ubsan.c (instrument_si_overflow): Likewise.
5664 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5665 Alan Hayward <alan.hayward@arm.com>
5666 David Sherwood <david.sherwood@arm.com>
5668 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5669 polynomial numbers of units.
5670 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5671 (valid_vector_subparts_p): New function.
5672 (build_vector_type): Remove temporary shim and take the number
5673 of units as a poly_uint64 rather than an int.
5674 (build_opaque_vector_type): Take the number of units as a
5675 poly_uint64 rather than an int.
5676 * tree.c (build_vector_from_ctor): Handle polynomial
5677 TYPE_VECTOR_SUBPARTS.
5678 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5679 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5680 (build_vector_from_val): If the number of units is variable,
5681 use build_vec_duplicate_cst for constant operands and
5682 VEC_DUPLICATE_EXPR otherwise.
5683 (make_vector_type): Remove temporary is_constant ().
5684 (build_vector_type, build_opaque_vector_type): Take the number of
5685 units as a poly_uint64 rather than an int.
5686 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5688 * cfgexpand.c (expand_debug_expr): Likewise.
5689 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5690 (store_constructor, expand_expr_real_1): Likewise.
5691 (const_scalar_mask_from_tree): Likewise.
5692 * fold-const-call.c (fold_const_reduction): Likewise.
5693 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5694 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5695 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5696 (fold_relational_const): Likewise.
5697 (native_interpret_vector): Likewise. Change the size from an
5698 int to an unsigned int.
5699 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5700 TYPE_VECTOR_SUBPARTS.
5701 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5702 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5703 duplicating a non-constant operand into a variable-length vector.
5704 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5705 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5706 * ipa-icf.c (sem_variable::equals): Likewise.
5707 * match.pd: Likewise.
5708 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5709 * print-tree.c (print_node): Likewise.
5710 * stor-layout.c (layout_type): Likewise.
5711 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5712 * tree-cfg.c (verify_gimple_comparison): Likewise.
5713 (verify_gimple_assign_binary): Likewise.
5714 (verify_gimple_assign_ternary): Likewise.
5715 (verify_gimple_assign_single): Likewise.
5716 * tree-pretty-print.c (dump_generic_node): Likewise.
5717 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5718 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5719 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5720 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5721 (vect_shift_permute_load_chain): Likewise.
5722 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5723 (expand_vector_condition, optimize_vector_constructor): Likewise.
5724 (lower_vec_perm, get_compute_type): Likewise.
5725 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5726 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5727 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5728 (vect_recog_mask_conversion_pattern): Likewise.
5729 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5730 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5731 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5732 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5733 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5734 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5735 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5736 (supportable_widening_operation): Likewise.
5737 (supportable_narrowing_operation): Likewise.
5738 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5740 * varasm.c (output_constant): Likewise.
5742 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5743 Alan Hayward <alan.hayward@arm.com>
5744 David Sherwood <david.sherwood@arm.com>
5746 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5747 so that both the length == 3 and length != 3 cases set up their
5748 own permute vectors. Add comments explaining why we know the
5749 number of elements is constant.
5750 (vect_permute_load_chain): Likewise.
5752 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5753 Alan Hayward <alan.hayward@arm.com>
5754 David Sherwood <david.sherwood@arm.com>
5756 * machmode.h (mode_nunits): Change from unsigned char to
5758 (ONLY_FIXED_SIZE_MODES): New macro.
5759 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5760 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5761 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5763 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5764 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5765 or if measurement_type is not polynomial.
5766 * genmodes.c (ZERO_COEFFS): New macro.
5767 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5769 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5770 Use ZERO_COEFFS when emitting initializers.
5771 * data-streamer.h (bp_pack_poly_value): New function.
5772 (bp_unpack_poly_value): Likewise.
5773 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5774 for GET_MODE_NUNITS.
5775 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5776 for GET_MODE_NUNITS.
5777 * tree.c (make_vector_type): Remove temporary shim and make
5778 the real function take the number of units as a poly_uint64
5780 (build_vector_type_for_mode): Handle polynomial nunits.
5781 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5782 * emit-rtl.c (const_vec_series_p_1): Likewise.
5783 (gen_rtx_CONST_VECTOR): Likewise.
5784 * fold-const.c (test_vec_duplicate_folding): Likewise.
5785 * genrecog.c (validate_pattern): Likewise.
5786 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5787 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5788 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5789 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5790 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5791 * rtlanal.c (subreg_get_info): Likewise.
5792 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5793 (vect_grouped_load_supported): Likewise.
5794 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5795 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5796 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5797 (simplify_const_unary_operation, simplify_binary_operation_1)
5798 (simplify_const_binary_operation, simplify_ternary_operation)
5799 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5800 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5801 instead of CONST_VECTOR_NUNITS.
5802 * varasm.c (output_constant_pool_2): Likewise.
5803 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5804 explicit-encoded elements in the XVEC for variable-length vectors.
5806 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5808 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5810 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5811 Alan Hayward <alan.hayward@arm.com>
5812 David Sherwood <david.sherwood@arm.com>
5814 * coretypes.h (fixed_size_mode): Declare.
5815 (fixed_size_mode_pod): New typedef.
5816 * builtins.h (target_builtins::x_apply_args_mode)
5817 (target_builtins::x_apply_result_mode): Change type to
5818 fixed_size_mode_pod.
5819 * builtins.c (apply_args_size, apply_result_size, result_vector)
5820 (expand_builtin_apply_args_1, expand_builtin_apply)
5821 (expand_builtin_return): Update accordingly.
5823 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5825 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5826 * cselib.c (cselib_hash_rtx): Likewise.
5827 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5828 CONST_VECTOR encoding.
5830 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5831 Jeff Law <law@redhat.com>
5834 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5835 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5836 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5837 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5840 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5841 explicitly probe *sp in a noreturn function if there were any callee
5842 register saves or frame pointer is needed.
5844 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5847 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5848 BLKmode for ternary, binary or unary expressions.
5851 * var-tracking.c (delete_vta_debug_insn): New inline function.
5852 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5853 insns from get_insns () to NULL instead of each bb separately.
5854 Use delete_vta_debug_insn. No longer static.
5855 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5856 delete_vta_debug_insns callers.
5857 * rtl.h (delete_vta_debug_insns): Declare.
5858 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5859 instead of variable_tracking_main.
5861 2018-01-03 Martin Sebor <msebor@redhat.com>
5863 PR tree-optimization/83603
5864 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5865 arguments past the endof the argument list in functions declared
5866 without a prototype.
5867 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5868 Avoid checking when arguments are null.
5870 2018-01-03 Martin Sebor <msebor@redhat.com>
5873 * doc/extend.texi (attribute const): Fix a typo.
5874 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5875 issuing -Wsuggest-attribute for void functions.
5877 2018-01-03 Martin Sebor <msebor@redhat.com>
5879 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5880 offset_int::from instead of wide_int::to_shwi.
5881 (maybe_diag_overlap): Remove assertion.
5882 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5883 * gimple-ssa-sprintf.c (format_directive): Same.
5884 (parse_directive): Same.
5885 (sprintf_dom_walker::compute_format_length): Same.
5886 (try_substitute_return_value): Same.
5888 2018-01-03 Jeff Law <law@redhat.com>
5891 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5892 non-constant residual for zero at runtime and avoid probing in
5893 that case. Reorganize code for trailing problem to mirror handling
5896 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5898 PR tree-optimization/83501
5899 * tree-ssa-strlen.c (get_string_cst): New.
5900 (handle_char_store): Call get_string_cst.
5902 2018-01-03 Martin Liska <mliska@suse.cz>
5904 PR tree-optimization/83593
5905 * tree-ssa-strlen.c: Include tree-cfg.h.
5906 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5907 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5908 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5910 (strlen_dom_walker::before_dom_children): Call
5911 gimple_purge_dead_eh_edges. Dump tranformation with details
5913 (strlen_dom_walker::before_dom_children): Update call by adding
5914 new argument cleanup_eh.
5915 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5917 2018-01-03 Martin Liska <mliska@suse.cz>
5920 * cif-code.def (VARIADIC_THUNK): New enum value.
5921 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5924 2018-01-03 Jan Beulich <jbeulich@suse.com>
5926 * sse.md (mov<mode>_internal): Tighten condition for when to use
5927 vmovdqu<ssescalarsize> for TI and OI modes.
5929 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5931 Update copyright years.
5933 2018-01-03 Martin Liska <mliska@suse.cz>
5936 * ipa-visibility.c (function_and_variable_visibility): Skip
5937 functions with noipa attribure.
5939 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5941 * gcc.c (process_command): Update copyright notice dates.
5942 * gcov-dump.c (print_version): Ditto.
5943 * gcov.c (print_version): Ditto.
5944 * gcov-tool.c (print_version): Ditto.
5945 * gengtype.c (create_file): Ditto.
5946 * doc/cpp.texi: Bump @copying's copyright year.
5947 * doc/cppinternals.texi: Ditto.
5948 * doc/gcc.texi: Ditto.
5949 * doc/gccint.texi: Ditto.
5950 * doc/gcov.texi: Ditto.
5951 * doc/install.texi: Ditto.
5952 * doc/invoke.texi: Ditto.
5954 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5956 * vector-builder.h (vector_builder::m_full_nelts): Change from
5957 unsigned int to poly_uint64.
5958 (vector_builder::full_nelts): Update prototype accordingly.
5959 (vector_builder::new_vector): Likewise.
5960 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5961 (vector_builder::operator ==): Likewise.
5962 (vector_builder::finalize): Likewise.
5963 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5964 Take the number of elements as a poly_uint64 rather than an
5966 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5967 from unsigned int to poly_uint64.
5968 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5969 (vec_perm_indices::new_vector): Likewise.
5970 (vec_perm_indices::length): Likewise.
5971 (vec_perm_indices::nelts_per_input): Likewise.
5972 (vec_perm_indices::input_nelts): Likewise.
5973 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5974 number of elements per input as a poly_uint64 rather than an
5975 unsigned int. Use the original encoding for variable-length
5976 vectors, rather than clamping each individual element.
5977 For the second and subsequent elements in each pattern,
5978 clamp the step and base before clamping their sum.
5979 (vec_perm_indices::series_p): Handle polynomial element counts.
5980 (vec_perm_indices::all_in_range_p): Likewise.
5981 (vec_perm_indices_to_tree): Likewise.
5982 (vec_perm_indices_to_rtx): Likewise.
5983 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5984 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5985 (tree_vector_builder::new_binary_operation): Handle polynomial
5986 element counts. Return false if we need to know the number
5987 of elements at compile time.
5988 * fold-const.c (fold_vec_perm): Punt if the number of elements
5989 isn't known at compile time.
5991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5993 * vec-perm-indices.h (vec_perm_builder): Change element type
5994 from HOST_WIDE_INT to poly_int64.
5995 (vec_perm_indices::element_type): Update accordingly.
5996 (vec_perm_indices::clamp): Handle polynomial element_types.
5997 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5998 (vec_perm_indices::all_in_range_p): Likewise.
5999 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6001 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6002 polynomial vec_perm_indices element types.
6003 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6004 * fold-const.c (fold_vec_perm): Likewise.
6005 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6006 * tree-vect-generic.c (lower_vec_perm): Likewise.
6007 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6008 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6009 element type to HOST_WIDE_INT.
6011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6012 Alan Hayward <alan.hayward@arm.com>
6013 David Sherwood <david.sherwood@arm.com>
6015 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6016 rather than an int. Use plus_constant.
6017 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6018 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6021 Alan Hayward <alan.hayward@arm.com>
6022 David Sherwood <david.sherwood@arm.com>
6024 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6025 a HOST_WIDE_INT to a poly_int64.
6027 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6028 Alan Hayward <alan.hayward@arm.com>
6029 David Sherwood <david.sherwood@arm.com>
6031 * calls.c (load_register_parameters): Cope with polynomial
6032 mode sizes. Require a constant size for BLKmode parameters
6033 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6034 forces a parameter to be padded at the lsb end in order to
6035 fill a complete number of words, require the parameter size
6036 to be ordered wrt UNITS_PER_WORD.
6038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6039 Alan Hayward <alan.hayward@arm.com>
6040 David Sherwood <david.sherwood@arm.com>
6042 * reload1.c (spill_stack_slot_width): Change element type
6043 from unsigned int to poly_uint64_pod.
6044 (alter_reg): Treat mode sizes as polynomial.
6046 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6047 Alan Hayward <alan.hayward@arm.com>
6048 David Sherwood <david.sherwood@arm.com>
6050 * reload.c (complex_word_subreg_p): New function.
6051 (reload_inner_reg_of_subreg, push_reload): Use it.
6053 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6054 Alan Hayward <alan.hayward@arm.com>
6055 David Sherwood <david.sherwood@arm.com>
6057 * lra-constraints.c (process_alt_operands): Reject matched
6058 operands whose sizes aren't ordered.
6059 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6066 that the mode size is in the set {1, 2, 4, 8, 16}.
6068 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6069 Alan Hayward <alan.hayward@arm.com>
6070 David Sherwood <david.sherwood@arm.com>
6072 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6073 Use plus_constant instead of gen_rtx_PLUS.
6075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6076 Alan Hayward <alan.hayward@arm.com>
6077 David Sherwood <david.sherwood@arm.com>
6079 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6080 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6081 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6082 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6083 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6084 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6085 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6086 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6087 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6088 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6090 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6091 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6092 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6093 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6094 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6095 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6096 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6097 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6098 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6099 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6101 * expr.c (emit_move_resolve_push): Treat the input and result
6102 of PUSH_ROUNDING as a poly_int64.
6103 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6104 (emit_push_insn): Likewise.
6105 * lra-eliminations.c (mark_not_eliminable): Likewise.
6106 * recog.c (push_operand): Likewise.
6107 * reload1.c (elimination_effects): Likewise.
6108 * rtlanal.c (nonzero_bits1): Likewise.
6109 * calls.c (store_one_arg): Likewise. Require the padding to be
6110 known at compile time.
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 (emit_single_push_insn_1): Treat mode sizes as polynomial.
6117 Use plus_constant instead of gen_rtx_PLUS.
6119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6120 Alan Hayward <alan.hayward@arm.com>
6121 David Sherwood <david.sherwood@arm.com>
6123 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6127 Alan Hayward <alan.hayward@arm.com>
6128 David Sherwood <david.sherwood@arm.com>
6130 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6131 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6132 via stack temporaries. Treat the mode size as polynomial too.
6134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6135 Alan Hayward <alan.hayward@arm.com>
6136 David Sherwood <david.sherwood@arm.com>
6138 * expr.c (expand_expr_real_2): When handling conversions involving
6139 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6140 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6141 as a poly_uint64 too.
6143 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6144 Alan Hayward <alan.hayward@arm.com>
6145 David Sherwood <david.sherwood@arm.com>
6147 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
6154 REGMODE_NATURAL_SIZE.
6155 * expmed.c (store_bit_field_1): Likewise.
6156 * expr.c (store_constructor): Likewise.
6157 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6158 and polynomial REGMODE_NATURAL_SIZE.
6159 (gen_lowpart_common): Likewise.
6160 * reginfo.c (record_subregs_of_mode): Likewise.
6161 * rtlanal.c (read_modify_subreg_p): Likewise.
6163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6164 Alan Hayward <alan.hayward@arm.com>
6165 David Sherwood <david.sherwood@arm.com>
6167 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6168 numbers of elements.
6170 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6171 Alan Hayward <alan.hayward@arm.com>
6172 David Sherwood <david.sherwood@arm.com>
6174 * match.pd: Cope with polynomial numbers of vector elements.
6176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6177 Alan Hayward <alan.hayward@arm.com>
6178 David Sherwood <david.sherwood@arm.com>
6180 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6181 in a POINTER_PLUS_EXPR.
6183 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6184 Alan Hayward <alan.hayward@arm.com>
6185 David Sherwood <david.sherwood@arm.com>
6187 * omp-simd-clone.c (simd_clone_subparts): New function.
6188 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6189 (ipa_simd_modify_function_body): Likewise.
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-generic.c (nunits_for_known_piecewise_op): New function.
6196 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6197 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6198 (expand_vector_condition, vector_element): Likewise.
6199 (subparts_gt): New function.
6200 (get_compute_type): Use subparts_gt.
6201 (count_type_subparts): Delete.
6202 (expand_vector_operations_1): Use subparts_gt instead of
6203 count_type_subparts.
6205 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6206 Alan Hayward <alan.hayward@arm.com>
6207 David Sherwood <david.sherwood@arm.com>
6209 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6210 (vect_compile_time_alias): ...this new function. Do the calculation
6211 on poly_ints rather than trees.
6212 (vect_prune_runtime_alias_test_list): Update call accordingly.
6214 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6215 Alan Hayward <alan.hayward@arm.com>
6216 David Sherwood <david.sherwood@arm.com>
6218 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6220 (vect_schedule_slp_instance): Likewise.
6222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6223 Alan Hayward <alan.hayward@arm.com>
6224 David Sherwood <david.sherwood@arm.com>
6226 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6227 constant and extern definitions for variable-length vectors.
6228 (vect_get_constant_vectors): Note that the number of units
6229 is known to be constant.
6231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6232 Alan Hayward <alan.hayward@arm.com>
6233 David Sherwood <david.sherwood@arm.com>
6235 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6236 of units as polynomial. Choose between WIDE and NARROW based
6239 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6240 Alan Hayward <alan.hayward@arm.com>
6241 David Sherwood <david.sherwood@arm.com>
6243 * tree-vect-stmts.c (simd_clone_subparts): New function.
6244 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6246 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6247 Alan Hayward <alan.hayward@arm.com>
6248 David Sherwood <david.sherwood@arm.com>
6250 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6251 vectors as polynomial. Use build_index_vector for
6254 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6255 Alan Hayward <alan.hayward@arm.com>
6256 David Sherwood <david.sherwood@arm.com>
6258 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6259 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6260 for variable-length vectors.
6261 (vectorizable_mask_load_store): Treat the number of units as
6262 polynomial, asserting that it is constant if the condition has
6263 already been enforced.
6264 (vectorizable_store, vectorizable_load): Likewise.
6266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6267 Alan Hayward <alan.hayward@arm.com>
6268 David Sherwood <david.sherwood@arm.com>
6270 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6271 of units as polynomial. Punt if we can't tell at compile time
6272 which vector contains the final result.
6274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6275 Alan Hayward <alan.hayward@arm.com>
6276 David Sherwood <david.sherwood@arm.com>
6278 * tree-vect-loop.c (vectorizable_induction): Treat the number
6279 of units as polynomial. Punt on SLP inductions. Use an integer
6280 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6281 cast of such a series for variable-length floating-point
6284 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6285 Alan Hayward <alan.hayward@arm.com>
6286 David Sherwood <david.sherwood@arm.com>
6288 * tree.h (build_index_vector): Declare.
6289 * tree.c (build_index_vector): New function.
6290 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6291 of units as polynomial, forcibly converting it to a constant if
6292 vectorizable_reduction has already enforced the condition.
6293 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6294 to create a {1,2,3,...} vector.
6295 (vectorizable_reduction): Treat the number of units as polynomial.
6296 Choose vectype_in based on the largest scalar element size rather
6297 than the smallest number of units. Enforce the restrictions
6300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6301 Alan Hayward <alan.hayward@arm.com>
6302 David Sherwood <david.sherwood@arm.com>
6304 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6305 number of units as polynomial.
6307 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6308 Alan Hayward <alan.hayward@arm.com>
6309 David Sherwood <david.sherwood@arm.com>
6311 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6312 * target.def (autovectorize_vector_sizes): Return the vector sizes
6313 by pointer, using vector_sizes rather than a bitmask.
6314 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6315 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6316 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6318 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6319 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6320 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6321 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6322 * omp-general.c (omp_max_vf): Likewise.
6323 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6324 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6325 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6326 * tree-vect-slp.c (vect_slp_bb): Likewise.
6327 * doc/tm.texi: Regenerate.
6328 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6330 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6331 the vector size as a poly_uint64 rather than an unsigned int.
6332 (current_vector_size): Change from an unsigned int to a poly_uint64.
6333 (get_vectype_for_scalar_type): Update accordingly.
6334 * tree.h (build_truth_vector_type): Take the size and number of
6335 units as a poly_uint64 rather than an unsigned int.
6336 (build_vector_type): Add a temporary overload that takes
6337 the number of units as a poly_uint64 rather than an unsigned int.
6338 * tree.c (make_vector_type): Likewise.
6339 (build_truth_vector_type): Take the number of units as a poly_uint64
6340 rather than an unsigned int.
6342 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6343 Alan Hayward <alan.hayward@arm.com>
6344 David Sherwood <david.sherwood@arm.com>
6346 * target.def (get_mask_mode): Take the number of units and length
6347 as poly_uint64s rather than unsigned ints.
6348 * targhooks.h (default_get_mask_mode): Update accordingly.
6349 * targhooks.c (default_get_mask_mode): Likewise.
6350 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6351 * doc/tm.texi: Regenerate.
6353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6354 Alan Hayward <alan.hayward@arm.com>
6355 David Sherwood <david.sherwood@arm.com>
6357 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6358 * omp-general.c (omp_max_vf): Likewise.
6359 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6360 (expand_omp_simd): Handle polynomial safelen.
6361 * omp-low.c (omplow_simd_context): Add a default constructor.
6362 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6363 (lower_rec_simd_input_clauses): Update accordingly.
6364 (lower_rec_input_clauses): Likewise.
6366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6367 Alan Hayward <alan.hayward@arm.com>
6368 David Sherwood <david.sherwood@arm.com>
6370 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6371 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6372 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6373 (vect_analyze_slp_cost): Likewise.
6374 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6375 (vect_model_load_cost): Likewise.
6377 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6378 Alan Hayward <alan.hayward@arm.com>
6379 David Sherwood <david.sherwood@arm.com>
6381 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6382 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6383 from an unsigned int * to a poly_uint64_pod *.
6384 (calculate_unrolling_factor): New function.
6385 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6387 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6388 Alan Hayward <alan.hayward@arm.com>
6389 David Sherwood <david.sherwood@arm.com>
6391 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6392 from an unsigned int to a poly_uint64.
6393 (_loop_vec_info::slp_unrolling_factor): Likewise.
6394 (_loop_vec_info::vectorization_factor): Change from an int
6396 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6397 (vect_get_num_vectors): New function.
6398 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6399 (vect_get_num_copies): Use vect_get_num_vectors.
6400 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6401 to an unsigned int *.
6402 (vect_analyze_data_refs): Change min_vf from an int * to a
6404 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6405 than an unsigned HOST_WIDE_INT.
6406 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6407 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6408 to an unsigned int *.
6409 (vect_analyze_data_ref_dependences): Likewise.
6410 (vect_compute_data_ref_alignment): Handle polynomial vf.
6411 (vect_enhance_data_refs_alignment): Likewise.
6412 (vect_prune_runtime_alias_test_list): Likewise.
6413 (vect_shift_permute_load_chain): Likewise.
6414 (vect_supportable_dr_alignment): Likewise.
6415 (dependence_distance_ge_vf): Take the vectorization factor as a
6416 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6417 (vect_analyze_data_refs): Change min_vf from an int * to a
6419 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6420 vfm1 as a poly_uint64 rather than an int. Make the same change
6421 for the returned bound_scalar.
6422 (vect_gen_vector_loop_niters): Handle polynomial vf.
6423 (vect_do_peeling): Likewise. Update call to
6424 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6425 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6427 * tree-vect-loop.c (vect_determine_vectorization_factor)
6428 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6429 (vect_get_known_peeling_cost): Likewise.
6430 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6431 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6432 (vect_transform_loop): Likewise. Use the lowest possible VF when
6433 updating the upper bounds of the loop.
6434 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6436 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6437 polynomial unroll factors.
6438 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6439 (vect_make_slp_decision): Likewise.
6440 (vect_supported_load_permutation_p): Likewise, and polynomial
6442 (vect_analyze_slp_cost): Handle polynomial vf.
6443 (vect_slp_analyze_node_operations): Likewise.
6444 (vect_slp_analyze_bb_1): Likewise.
6445 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6446 than an unsigned HOST_WIDE_INT.
6447 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6448 (vectorizable_load): Handle polynomial vf.
6449 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6451 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6453 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6454 Alan Hayward <alan.hayward@arm.com>
6455 David Sherwood <david.sherwood@arm.com>
6457 * match.pd: Handle bit operations involving three constants
6458 and try to fold one pair.
6460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6462 * tree-vect-loop-manip.c: Include gimple-fold.h.
6463 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6464 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6465 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6466 Add a path that uses a step of VF instead of 1, but disable it
6468 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6469 and niters_no_overflow parameters. Update calls to
6470 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6471 Create a new SSA name if the latter choses to use a ste other
6472 than zero, and return it via niters_vector_mult_vf_var.
6473 * tree-vect-loop.c (vect_transform_loop): Update calls to
6474 vect_do_peeling, vect_gen_vector_loop_niters and
6475 slpeel_make_loop_iterate_ntimes.
6476 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6477 (vect_gen_vector_loop_niters): Update declarations after above changes.
6479 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6481 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6482 128-bit round to integer instructions.
6483 (ceil<mode>2): Likewise.
6484 (btrunc<mode>2): Likewise.
6485 (round<mode>2): Likewise.
6487 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6489 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6490 unaligned VSX load/store on P8/P9.
6491 (expand_block_clear): Allow the use of unaligned VSX
6492 load/store on P8/P9.
6494 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6496 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6498 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6499 swap associated with both a load and a store.
6501 2018-01-02 Andrew Waterman <andrew@sifive.com>
6503 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6504 * config/riscv/riscv.md (clear_cache): Use it.
6506 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6508 * web.c: Remove out-of-date comment.
6510 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6512 * expr.c (fixup_args_size_notes): Check that any existing
6513 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6514 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6515 (emit_single_push_insn): ...here.
6517 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6519 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6520 (const_vector_encoded_nelts): New function.
6521 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6522 (const_vector_int_elt, const_vector_elt): Declare.
6523 * emit-rtl.c (const_vector_int_elt_1): New function.
6524 (const_vector_elt): Likewise.
6525 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6526 of CONST_VECTOR_ELT.
6528 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6530 * expr.c: Include rtx-vector-builder.h.
6531 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6532 directly on the tree encoding.
6533 (const_vector_from_tree): Likewise.
6534 * optabs.c: Include rtx-vector-builder.h.
6535 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6536 sequence of "u" values.
6537 * vec-perm-indices.c: Include rtx-vector-builder.h.
6538 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6539 directly on the vec_perm_indices encoding.
6541 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6543 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6544 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6545 * rtx-vector-builder.h: New file.
6546 * rtx-vector-builder.c: Likewise.
6547 * rtl.h (rtx_def::u2): Add a const_vector field.
6548 (CONST_VECTOR_NPATTERNS): New macro.
6549 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6550 (CONST_VECTOR_DUPLICATE_P): Likewise.
6551 (CONST_VECTOR_STEPPED_P): Likewise.
6552 (CONST_VECTOR_ENCODED_ELT): Likewise.
6553 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6554 (unwrap_const_vec_duplicate): Likewise.
6555 (const_vec_series_p): Check for a non-duplicated vector encoding.
6556 Say that the function only returns true for integer vectors.
6557 * emit-rtl.c: Include rtx-vector-builder.h.
6558 (gen_const_vec_duplicate_1): Delete.
6559 (gen_const_vector): Call gen_const_vec_duplicate instead of
6560 gen_const_vec_duplicate_1.
6561 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6562 (gen_const_vec_duplicate): Use rtx_vector_builder.
6563 (gen_const_vec_series): Likewise.
6564 (gen_rtx_CONST_VECTOR): Likewise.
6565 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6566 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6567 Build a new vector rather than modifying a CONST_VECTOR in-place.
6568 (handle_special_swappables): Update call accordingly.
6569 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6570 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6571 Build a new vector rather than modifying a CONST_VECTOR in-place.
6572 (handle_special_swappables): Update call accordingly.
6574 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6576 * simplify-rtx.c (simplify_const_binary_operation): Use
6577 CONST_VECTOR_ELT instead of XVECEXP.
6579 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6581 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6582 the selector elements to be different from the data elements
6583 if the selector is a VECTOR_CST.
6584 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6585 ssizetype for the selector.
6587 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6589 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6590 before testing each element individually.
6591 * tree-vect-generic.c (lower_vec_perm): Likewise.
6593 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6595 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6596 * selftest-run-tests.c (selftest::run_tests): Call it.
6597 * vector-builder.h (vector_builder::operator ==): New function.
6598 (vector_builder::operator !=): Likewise.
6599 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6600 (vec_perm_indices::all_from_input_p): New function.
6601 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6602 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6603 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6604 instead of reading the VECTOR_CST directly. Detect whether both
6605 vector inputs are the same before constructing the vec_perm_indices,
6606 and update the number of inputs argument accordingly. Use the
6607 utility functions added above. Only construct sel2 if we need to.
6609 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6611 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6612 the broadcast of the low byte.
6613 (expand_mult_highpart): Use an explicit encoding for the permutes.
6614 * optabs-query.c (can_mult_highpart_p): Likewise.
6615 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6616 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6617 (vectorizable_bswap): Likewise.
6618 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6619 explicit encoding for the power-of-2 permutes.
6620 (vect_permute_store_chain): Likewise.
6621 (vect_grouped_load_supported): Likewise.
6622 (vect_permute_load_chain): Likewise.
6624 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6626 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6627 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6628 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6629 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6630 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6631 (vect_gen_perm_mask_any): Likewise.
6633 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6635 * int-vector-builder.h: New file.
6636 * vec-perm-indices.h: Include int-vector-builder.h.
6637 (vec_perm_indices): Redefine as an int_vector_builder.
6638 (auto_vec_perm_indices): Delete.
6639 (vec_perm_builder): Redefine as a stand-alone class.
6640 (vec_perm_indices::vec_perm_indices): New function.
6641 (vec_perm_indices::clamp): Likewise.
6642 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6643 (vec_perm_indices::new_vector): New function.
6644 (vec_perm_indices::new_expanded_vector): Update for new
6645 vec_perm_indices class.
6646 (vec_perm_indices::rotate_inputs): New function.
6647 (vec_perm_indices::all_in_range_p): Operate directly on the
6648 encoded form, without computing elided elements.
6649 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6650 encoding. Update for new vec_perm_indices class.
6651 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6652 the given vec_perm_builder.
6653 (expand_vec_perm_var): Update vec_perm_builder constructor.
6654 (expand_mult_highpart): Use vec_perm_builder instead of
6655 auto_vec_perm_indices.
6656 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6657 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6658 or double series encoding as appropriate.
6659 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6660 vec_perm_indices instead of auto_vec_perm_indices.
6661 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6662 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6663 (vect_permute_store_chain): Likewise.
6664 (vect_grouped_load_supported): Likewise.
6665 (vect_permute_load_chain): Likewise.
6666 (vect_shift_permute_load_chain): Likewise.
6667 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6668 (vect_transform_slp_perm_load): Likewise.
6669 (vect_schedule_slp_instance): Likewise.
6670 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6671 (vectorizable_mask_load_store): Likewise.
6672 (vectorizable_bswap): Likewise.
6673 (vectorizable_store): Likewise.
6674 (vectorizable_load): Likewise.
6675 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6676 vec_perm_indices instead of auto_vec_perm_indices. Use
6677 tree_to_vec_perm_builder to read the vector from a tree.
6678 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6679 vec_perm_builder instead of a vec_perm_indices.
6680 (have_whole_vector_shift): Use vec_perm_builder and
6681 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6682 truncation to calc_vec_perm_mask_for_shift.
6683 (vect_create_epilog_for_reduction): Likewise.
6684 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6685 from auto_vec_perm_indices to vec_perm_indices.
6686 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6687 instead of changing individual elements.
6688 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6689 the vector in d.perm.
6690 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6691 from auto_vec_perm_indices to vec_perm_indices.
6692 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6693 instead of changing individual elements.
6694 (arm_vectorize_vec_perm_const): Use new_vector to install
6695 the vector in d.perm.
6696 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6697 Update vec_perm_builder constructor.
6698 (rs6000_expand_interleave): Likewise.
6699 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6700 (rs6000_expand_interleave): Likewise.
6702 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6704 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6705 to qimode could truncate the indices.
6706 * optabs.c (expand_vec_perm_var): Likewise.
6708 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6710 * Makefile.in (OBJS): Add vec-perm-indices.o.
6711 * vec-perm-indices.h: New file.
6712 * vec-perm-indices.c: Likewise.
6713 * target.h (vec_perm_indices): Replace with a forward class
6715 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6716 * optabs.h: Include vec-perm-indices.h.
6717 (expand_vec_perm): Delete.
6718 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6719 (expand_vec_perm_const): Declare.
6720 * target.def (vec_perm_const_ok): Replace with...
6721 (vec_perm_const): ...this new hook.
6722 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6723 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6724 * doc/tm.texi: Regenerate.
6725 * optabs.def (vec_perm_const): Delete.
6726 * doc/md.texi (vec_perm_const): Likewise.
6727 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6728 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6729 expand_vec_perm for constant permutation vectors. Assert that
6730 the mode of variable permutation vectors is the integer equivalent
6731 of the mode that is being permuted.
6732 * optabs-query.h (selector_fits_mode_p): Declare.
6733 * optabs-query.c: Include vec-perm-indices.h.
6734 (selector_fits_mode_p): New function.
6735 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6736 is defined, instead of checking whether the vec_perm_const_optab
6737 exists. Use targetm.vectorize.vec_perm_const instead of
6738 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6739 fit in the vector mode before using a variable permute.
6740 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6741 vec_perm_indices instead of an rtx.
6742 (expand_vec_perm): Replace with...
6743 (expand_vec_perm_const): ...this new function. Take the selector
6744 as a vec_perm_indices rather than an rtx. Also take the mode of
6745 the selector. Update call to shift_amt_for_vec_perm_mask.
6746 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6747 Use vec_perm_indices::new_expanded_vector to expand the original
6748 selector into bytes. Check whether the indices fit in the vector
6749 mode before using a variable permute.
6750 (expand_vec_perm_var): Make global.
6751 (expand_mult_highpart): Use expand_vec_perm_const.
6752 * fold-const.c: Includes vec-perm-indices.h.
6753 * tree-ssa-forwprop.c: Likewise.
6754 * tree-vect-data-refs.c: Likewise.
6755 * tree-vect-generic.c: Likewise.
6756 * tree-vect-loop.c: Likewise.
6757 * tree-vect-slp.c: Likewise.
6758 * tree-vect-stmts.c: Likewise.
6759 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6761 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6762 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6763 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6764 (aarch64_vectorize_vec_perm_const): ...this new function.
6765 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6766 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6767 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6768 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6769 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6770 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6771 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6773 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6774 check for NEON modes.
6775 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6776 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6777 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6778 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6780 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6781 the old VEC_PERM_CONST conditions.
6782 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6783 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6784 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6785 (ia64_vectorize_vec_perm_const_ok): Merge into...
6786 (ia64_vectorize_vec_perm_const): ...this new function.
6787 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6788 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6789 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6790 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6791 * config/mips/mips.c (mips_expand_vec_perm_const)
6792 (mips_vectorize_vec_perm_const_ok): Merge into...
6793 (mips_vectorize_vec_perm_const): ...this new function.
6794 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6795 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6796 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6797 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6798 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6799 (rs6000_expand_vec_perm_const): Delete.
6800 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6802 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6803 (altivec_expand_vec_perm_const_le): Take each operand individually.
6804 Operate on constant selectors rather than rtxes.
6805 (altivec_expand_vec_perm_const): Likewise. Update call to
6806 altivec_expand_vec_perm_const_le.
6807 (rs6000_expand_vec_perm_const): Delete.
6808 (rs6000_vectorize_vec_perm_const_ok): Delete.
6809 (rs6000_vectorize_vec_perm_const): New function.
6810 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6811 an element count and rtx array.
6812 (rs6000_expand_extract_even): Update call accordingly.
6813 (rs6000_expand_interleave): Likewise.
6814 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6815 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6816 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6817 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6818 (rs6000_expand_vec_perm_const): Delete.
6819 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6820 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6821 (altivec_expand_vec_perm_const_le): Take each operand individually.
6822 Operate on constant selectors rather than rtxes.
6823 (altivec_expand_vec_perm_const): Likewise. Update call to
6824 altivec_expand_vec_perm_const_le.
6825 (rs6000_expand_vec_perm_const): Delete.
6826 (rs6000_vectorize_vec_perm_const_ok): Delete.
6827 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6828 reference to the SPE evmerge intructions.
6829 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6830 an element count and rtx array.
6831 (rs6000_expand_extract_even): Update call accordingly.
6832 (rs6000_expand_interleave): Likewise.
6833 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6834 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6836 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6838 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6840 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6841 vector mode and that that mode matches the mode of the data
6843 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6844 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6845 directly using expand_vec_perm_1 when forcing selectors into
6847 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6849 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6851 * optabs-query.h (can_vec_perm_p): Delete.
6852 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6853 * optabs-query.c (can_vec_perm_p): Split into...
6854 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6855 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6856 particular selector is valid.
6857 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6858 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6859 (vect_grouped_load_supported): Likewise.
6860 (vect_shift_permute_load_chain): Likewise.
6861 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6862 (vect_transform_slp_perm_load): Likewise.
6863 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6864 (vectorizable_bswap): Likewise.
6865 (vect_gen_perm_mask_checked): Likewise.
6866 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6867 implementations of variable permutation vectors into account
6868 when deciding which selector to use.
6869 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6870 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6871 with a false third argument.
6872 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6873 to test whether the constant selector is valid and can_vec_perm_var_p
6874 to test whether a variable selector is valid.
6876 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6878 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6879 * optabs-query.c (can_vec_perm_p): Likewise.
6880 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6881 instead of vec_perm_indices.
6882 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6883 (vect_gen_perm_mask_checked): Likewise,
6884 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6885 (vect_gen_perm_mask_checked): Likewise,
6887 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6889 * optabs-query.h (qimode_for_vec_perm): Declare.
6890 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6891 (qimode_for_vec_perm): ...this new function.
6892 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6894 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6896 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6897 does not have a conditional at the top.
6899 2018-01-02 Richard Biener <rguenther@suse.de>
6901 * ipa-inline.c (big_speedup_p): Fix expression.
6903 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6906 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6909 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6913 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6914 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6915 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6916 cond_taken_branch_cost 3->4.
6918 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6920 PR tree-optimization/83581
6921 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6922 TODO_cleanup_cfg if any changes have been made.
6925 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6926 convert_modes if target mode has the right side, but different mode
6930 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6931 last argument when extracting from CONCAT. If either from_real or
6932 from_imag is NULL, use expansion through memory. If result is not
6933 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6934 the parts directly to inner mode, if even that fails, use expansion
6938 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6939 check for bswap in mode rather than HImode and use that in expand_unop
6942 Copyright (C) 2018 Free Software Foundation, Inc.
6944 Copying and distribution of this file, with or without modification,
6945 are permitted in any medium without royalty provided the copyright
6946 notice and this notice are preserved.