1 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
3 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
6 2018-02-10 Alan Modra <amodra@gmail.com>
9 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
10 Specify LR as an input.
12 2018-02-10 Jakub Jelinek <jakub@redhat.com>
15 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
16 remove_member_access_dummy_vars): New functions.
17 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
18 lower_omp_1, execute_lower_omp): Use them.
20 PR rtl-optimization/84308
21 * shrink-wrap.c (spread_components): Release todo vector.
23 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
25 PR rtl-optimization/57193
26 * ira-color.c (struct allocno_color_data): Add member
27 conflict_allocno_hard_prefs.
28 (update_conflict_allocno_hard_prefs): New.
29 (bucket_allocno_compare_func): Add a preference based on
30 conflict_allocno_hard_prefs.
31 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
32 (color_allocnos): Remove a dead code. Initiate
33 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
35 2018-02-09 Jakub Jelinek <jakub@redhat.com>
38 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
39 constraint from =wa to wa. Avoid a subreg on the output operand,
40 instead use a pseudo and subreg it in a move.
41 (p9_xxbrd_<mode>): Changed to ...
42 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
43 (p9_xxbrd_v2df): New expander.
44 (p9_xxbrw_<mode>): Changed to ...
45 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
46 (p9_xxbrw_v4sf): New expander.
48 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
50 * config/rx.md: updated "movsicc" expand to be matched by GCC
51 * testsuite/gcc.target/rx/movsicc.c: new test case
53 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
56 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
57 multiply in 32-bit mode.
58 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
59 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
62 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
64 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
65 to allow or block "symbol_ref" depending on value of TARGET_JSR
66 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
67 call_value_internal insns
69 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
72 * dwarf2out.c (is_trivial_indirect_ref): New function.
73 (dwarf2out_late_global_decl): Do not generate a location
74 attribute for variables that have a non-trivial DECL_VALUE_EXPR
75 and that are not defined in the current unit.
77 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
79 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
80 instead of a libcall for UNORDERED.
82 2018-02-09 Tamar Christina <tamar.christina@arm.com>
85 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
86 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
88 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
91 * config/s390/s390.c (s390_set_current_function): Invoke
92 s390_indirect_branch_settings also if fndecl didn't change.
94 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
96 * config/rs6000/rs6000.md (blockage): Set length to zero.
98 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
100 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
102 2018-02-09 Jakub Jelinek <jakub@redhat.com>
105 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
106 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
110 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
111 PARALLEL incoming that failed vt_get_decl_and_offset check.
114 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
115 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
117 (get_variable_section): For decls in named .bss* sections pass true as
118 second argument to bss_initializer_p.
120 2018-02-09 Marek Polacek <polacek@redhat.com>
121 Jakub Jelinek <jakub@redhat.com>
124 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
125 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
126 Sync some changes from cxx_fold_indirect_ref.
128 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
130 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
132 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
133 (BLOCK_INLINE_ENTRY_LABEL): New.
134 (dwarf2out_var_location): Disregard inline entry markers.
135 (inline_entry_data): New struct.
136 (inline_entry_data_hasher): New hashtable type.
137 (inline_entry_data_hasher::hash): New.
138 (inline_entry_data_hasher::equal): New.
139 (inline_entry_data_table): New variable.
140 (add_high_low_attributes): Add DW_AT_entry_pc and
141 DW_AT_GNU_entry_view attributes if a pending entry is found
142 in inline_entry_data_table. Add old entry_pc attribute only
143 if debug nonbinding markers are disabled.
144 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
146 (block_within_block_p, dwarf2out_inline_entry): New.
147 (dwarf2out_finish): Check that no entries remained in
148 inline_entry_data_table.
149 * final.c (reemit_insn_block_notes): Handle inline entry notes.
150 (final_scan_insn, notice_source_line): Likewise.
151 (rest_of_clean_state): Skip inline entry markers.
152 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
154 * gimple.c (gimple_build_debug_inline_entry): New.
155 * gimple.h (enum gimple_debug_subcode): Add
156 GIMPLE_DEBUG_INLINE_ENTRY.
157 (gimple_build_debug_inline_entry): Declare.
158 (gimple_debug_inline_entry_p): New.
159 (gimple_debug_nonbind_marker_p): Adjust.
160 * insn-notes.def (INLINE_ENTRY): New.
161 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
162 inline entry marker notes.
163 (print_insn): Likewise.
164 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
165 (INSN_DEBUG_MARKER_KIND): Likewise.
166 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
167 * tree-inline.c (expand_call_inline): Build and insert
168 debug_inline_entry stmt.
169 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
170 inline entry blocks early, if nonbind markers are enabled.
171 (dump_scope_block): Dump fragment info.
172 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
173 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
174 (gimple_build_debug_inline_entry): New.
175 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
176 Enable/disable inline entry points too.
177 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
178 (DEBUG_INSN): Describe inline entry markers.
180 * common.opt (gvariable-location-views): New.
181 (gvariable-location-views=incompat5): New.
182 * config.in: Rebuilt.
183 * configure: Rebuilt.
184 * configure.ac: Test assembler for view support.
185 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
186 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
187 * dwarf2out.c (var_loc_view): New typedef.
188 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
189 (dwarf2out_locviews_in_attribute): New.
190 (dwarf2out_locviews_in_loclist): New.
191 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
192 (enum dw_line_info_opcode): Add LI_adv_address.
193 (struct dw_line_info_table): Add view.
194 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
195 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
196 (zero_view_p): New variable.
197 (ZERO_VIEW_P): New macro.
198 (output_asm_line_debug_info): New.
199 (struct var_loc_node): Add view.
200 (add_AT_view_list, AT_loc_list): New.
201 (add_var_loc_to_decl): Add view param. Test it against last.
202 (new_loc_list): Add view params. Record them.
203 (AT_loc_list_ptr): Handle loc and view lists.
204 (view_list_to_loc_list_val_node): New.
205 (print_dw_val): Handle dw_val_class_view_list.
206 (size_of_die): Likewise.
207 (value_format): Likewise.
208 (loc_list_has_views): New.
209 (gen_llsym): Set vl_symbol too.
210 (maybe_gen_llsym, skip_loc_list_entry): New.
211 (dwarf2out_maybe_output_loclist_view_pair): New.
212 (output_loc_list): Output view list or entries too.
213 (output_view_list_offset): New.
214 (output_die): Handle dw_val_class_view_list.
215 (output_dwarf_version): New.
216 (output_compilation_unit_header): Use it.
217 (output_skeleton_debug_sections): Likewise.
218 (output_rnglists, output_line_info): Likewise.
219 (output_pubnames, output_aranges): Update version comments.
220 (output_one_line_info_table): Output view numbers in asm comments.
221 (dw_loc_list): Determine current endview, pass it to new_loc_list.
222 Call maybe_gen_llsym.
223 (loc_list_from_tree_1): Adjust.
224 (add_AT_location_description): Create view list attribute if
225 needed, check it's absent otherwise.
226 (convert_cfa_to_fb_loc_list): Adjust.
227 (maybe_emit_file): Call output_asm_line_debug_info for test.
228 (dwarf2out_var_location): Reset views as needed. Precompute
229 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
231 (new_line_info_table): Reset next view.
232 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
233 (dwarf2out_source_line): Likewise. Output view resets and labels to
234 the assembler, or select appropriate line info opcodes.
235 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
236 (optimize_string_length): Catch it. Adjust.
237 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
238 dw_val_class_view_list, and remove it if no longer needed.
239 (hash_loc_list): Hash view numbers.
240 (loc_list_hasher::equal): Compare them.
241 (optimize_location_lists): Check whether a view list symbol is
242 needed, and whether the locview attribute is present, and
243 whether they match. Remove the locview attribute if no longer
245 (index_location_lists): Call skip_loc_list_entry for test.
246 (dwarf2out_finish): Call output_asm_line_debug_info for test.
247 Use output_dwarf_version.
248 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
249 (struct dw_val_node): Add val_view_list.
250 * final.c (SEEN_NEXT_VIEW): New.
251 (set_next_view_needed): New.
252 (clear_next_view_needed): New.
253 (maybe_output_next_view): New.
254 (final_start_function): Rename to...
255 (final_start_function_1): ... this. Take pointer to FIRST,
256 add SEEN parameter. Emit param bindings in the initial view.
257 (final_start_function): Reintroduce SEEN-less interface.
258 (final): Rename to...
259 (final_1): ... this. Take SEEN parameter. Output final pending
260 next view at the end.
261 (final): Reintroduce seen-less interface.
262 (final_scan_insn): Output pending next view before switching
263 sections or ending a block. Mark the next view as needed when
264 outputting variable locations. Notify debug backend of section
265 changes, and of location view changes.
266 (rest_of_handle_final): Adjust.
267 * toplev.c (process_options): Autodetect value for debug variable
268 location views option. Warn on incompat5 without -gdwarf-5.
269 * doc/invoke.texi (gvariable-location-views): New.
270 (gvariable-location-views=incompat5): New.
271 (gno-variable-location-views): New.
273 2018-02-08 David Malcolm <dmalcolm@redhat.com>
275 PR tree-optimization/84136
276 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
277 that the result of find_edge is non-NULL.
279 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
282 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
283 storing integer register in SImode. Fix cost of 256 and 512
284 byte aligned SSE register store.
286 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
288 * config/i386/i386.c (ix86_multiplication_cost): Fix
289 multiplication cost for TARGET_AVX512DQ.
291 2018-02-08 Marek Polacek <polacek@redhat.com>
293 PR tree-optimization/84238
294 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
297 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
299 PR tree-optimization/84265
300 * tree-vect-stmts.c (vectorizable_store): Don't treat
301 VMAT_CONTIGUOUS accesses as grouped.
302 (vectorizable_load): Likewise.
304 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
306 PR tree-optimization/81635
307 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
308 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
309 (test_round_for_mask): New functions.
310 (wide_int_cc_tests): Call test_round_for_mask.
311 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
312 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
313 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
314 range returned by get_range_info.
316 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
319 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
320 * symtab.c: Include builtins.h
321 (symtab_node::output_to_lto_symbol_table_p): Move here
322 from lto-streamer-out.c:output_symbol_p.
323 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
324 (output_symbol_p): Move all logic to symtab.c
325 (produce_symtab): Update.
327 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
329 * config/s390/s390-opts.h (enum indirect_branch): Define.
330 * config/s390/s390-protos.h (s390_return_addr_from_memory)
331 (s390_indirect_branch_via_thunk)
332 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
333 (enum s390_indirect_branch_type): Define.
334 * config/s390/s390.c (struct s390_frame_layout, struct
335 machine_function): Remove.
336 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
337 (indirect_branch_table_label_no, indirect_branch_table_name):
339 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
340 (enum s390_indirect_branch_option): Define.
341 (s390_return_addr_from_memory): New function.
342 (s390_handle_string_attribute): New function.
343 (s390_attribute_table): Add new attribute handler.
344 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
345 (s390_indirect_branch_via_thunk): New function.
346 (s390_indirect_branch_via_inline_thunk): New function.
347 (s390_function_ok_for_sibcall): When jumping via thunk disallow
348 sibling call optimization for non z10 compiles.
349 (s390_emit_call): Force indirect branch target to be a single
350 register. Add r1 clobber for non-z10 compiles.
351 (s390_emit_epilogue): Emit return jump via return_use expander.
352 (s390_reorg): Handle JUMP_INSNs as execute targets.
353 (s390_option_override_internal): Perform validity checks for the
354 new command line options.
355 (s390_indirect_branch_attrvalue): New function.
356 (s390_indirect_branch_settings): New function.
357 (s390_set_current_function): Invoke s390_indirect_branch_settings.
358 (s390_output_indirect_thunk_function): New function.
359 (s390_code_end): Implement target hook.
360 (s390_case_values_threshold): Implement target hook.
361 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
363 * config/s390/s390.h (struct s390_frame_layout)
364 (struct machine_function): Move here from s390.c.
365 (TARGET_INDIRECT_BRANCH_NOBP_RET)
366 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
367 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
368 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
369 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
370 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
371 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
372 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
373 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
374 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
375 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
376 (mnemonic attribute): Add values which aren't recognized
378 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
379 pattern for branch conversion. Fix mnemonic attribute.
380 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
381 indirect branch via thunk if requested.
382 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
383 ("*indirect_jump"): Disable for branch conversion using out of
385 ("indirect_jump_via_thunk<mode>_z10")
386 ("indirect_jump_via_thunk<mode>")
387 ("indirect_jump_via_inlinethunk<mode>_z10")
388 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
389 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
390 ("casesi_jump_via_inlinethunk<mode>_z10")
391 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
392 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
393 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
394 ("*indirect2_jump"): Disable for branch conversion.
395 ("casesi_jump"): Turn into expander and expand patterns for branch
397 ("return_use"): New expander.
398 ("*return"): Emit return via thunk and rename it to ...
399 ("*return<mode>"): ... this one.
400 * config/s390/s390.opt: Add new options and and enum for the
403 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
405 * lra-constraints.c (match_reload): Unconditionally use
406 gen_lowpart_SUBREG, rather than selecting between that
407 and equivalent gen_rtx_SUBREG code.
409 2018-02-08 Richard Biener <rguenther@suse.de>
411 PR tree-optimization/84233
412 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
413 changed flag instead of boguously re-using phi_inserted.
415 2018-02-08 Martin Jambor <mjambor@suse.cz>
417 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
418 static local variables.
420 2018-02-08 Richard Biener <rguenther@suse.de>
422 PR tree-optimization/84278
423 * tree-vect-stmts.c (vectorizable_store): When looking for
424 smaller vector types to perform grouped strided loads/stores
425 make sure the mode is supported by the target.
426 (vectorizable_load): Likewise.
428 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
430 * config/aarch64/aarch64.c (aarch64_components_for_bb):
431 Increase LDP/STP opportunities by adding adjacent callee-saves.
433 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
435 PR rtl-optimization/84068
436 PR rtl-optimization/83459
437 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
439 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
441 PR tree-optimization/84224
442 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
443 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
446 2018-02-07 Iain Sandoe <iain@codesourcery.com>
449 * config/rs6000/altivec.md (*restore_world): Remove LR use.
450 * config/rs6000/predicates.md (restore_world_operation): Adjust op
451 count, remove one USE.
453 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
455 * doc/install.texi (Configuration): Document the
456 --with-long-double-format={ibm,ieee} PowerPC configuration
460 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
461 Convert from define_expand to be define_insn_and_split. Rework
462 float/double/_Float128 conversions to QI/HI/SImode to work with
463 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
464 conversions to QI/HImode types did a store and then a load to
465 truncate the value. For conversions to VSX registers, don't split
466 the insn, instead emit the code directly. Use the code iterator
467 any_fix to combine signed and unsigned conversions.
468 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
469 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
470 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
471 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
472 (fix_<mode>di2_hw): Likewise.
473 (fixuns_<mode>di2_hw): Likewise.
474 (fix_<mode>si2_hw): Likewise.
475 (fixuns_<mode>si2_hw): Likewise.
476 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
477 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
478 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
479 fix<uns>_trunc<SFDF:mode>si2_p8.
480 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
482 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
483 (fix<uns>_<mode>_mem): Likewise.
484 (fctiw<u>z_<mode>_mem): Likewise.
485 (fix<uns>_<mode>_mem): Likewise.
486 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
487 the register allocator from doing a direct move to the GPRs to do
488 a store, and instead use the ISA 3.0 store byte/half-word from
489 vector register instruction. For IEEE 128-bit floating point,
490 also optimize stores of 32-bit ints.
491 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
493 2018-02-07 Alan Hayward <alan.hayward@arm.com>
495 * genextract.c (push_pathstr_operand): New function to support
497 (walk_rtx): Call push_pathstr_operand.
498 (print_path): Support [a-zA-Z].
500 2018-02-07 Richard Biener <rguenther@suse.de>
502 PR tree-optimization/84037
503 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
504 (cse_and_gimplify_to_preheader): Declare.
505 (vect_get_place_in_interleaving_chain): Likewise.
506 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
508 (_loop_vec_info::~_loop_vec_info): Delete it.
509 (cse_and_gimplify_to_preheader): New function.
510 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
511 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
512 (vectorizable_load): Likewise. For grouped stores always base
513 the IV on the first element.
514 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
515 condition before gimplifying.
517 2018-02-07 Jakub Jelinek <jakub@redhat.com>
519 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
520 *DIV_EXPR and *MOD_EXPR.
522 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
525 * config/i386/i386.c (ix86_option_override_internal): Mask out
526 the CF_SET bit when checking -fcf-protection.
528 2018-02-07 Tom de Vries <tom@codesourcery.com>
531 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
534 2018-02-07 Richard Biener <rguenther@suse.de>
536 PR tree-optimization/84204
537 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
540 PR tree-optimization/84205
541 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
542 special-case isl_ast_op_zdiv_r.
544 PR tree-optimization/84223
545 * graphite-scop-detection.c (gather_bbs::before_dom_children):
546 Only add conditions from within the region.
547 (gather_bbs::after_dom_children): Adjust.
549 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
552 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
553 * config/avr/avr.md: Only post-reload split REG-REG moves if
554 either register is REGERAL_REG_P.
556 2018-02-07 Jakub Jelinek <jakub@redhat.com>
558 PR tree-optimization/84235
559 * tree-ssa-scopedtables.c
560 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
561 if the subtraction is performed in floating point type where NaNs are
562 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
563 build 1. Formatting fix.
565 2018-02-06 Jakub Jelinek <jakub@redhat.com>
568 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
569 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
570 and skip it regardless of bb boundaries. Use CALL_P macro,
571 don't test INSN_P (insn) together with CALL_P or JUMP_P check
572 unnecessarily, formatting fix.
574 2018-02-06 Michael Collison <michael.collison@arm.com>
576 * config/arm/thumb2.md:
577 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
578 (*thumb_mov_notscc): Ditto.
580 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
583 * config/rs6000/rs6000.md (su code attribute): Use "u" for
584 unsigned_fix, not "s".
586 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
588 * configure.ac (gcc_fn_eh_frame_ro): New function.
589 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
590 correct .eh_frame permissions.
591 * configure: Regenerate.
593 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
595 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
598 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
600 * config/rs6000/rs6000.c (rs6000_option_override_internal):
601 Display warning message for -mno-speculate-indirect-jumps.
603 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
605 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
607 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
609 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
611 PR tree-optimization/84225
612 * tree-eh.c (find_trapping_overflow): Only call
613 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
615 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
618 * config/i386/i386.c: Reimplement the check of possible options
619 -mibt/-mshstk conbination. Change error messages.
620 * doc/invoke.texi: Fix a typo: remove extra '='.
622 2018-02-06 Marek Polacek <polacek@redhat.com>
624 PR tree-optimization/84228
625 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
627 2018-02-06 Tamar Christina <tamar.christina@arm.com>
630 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
631 emitted arch directives.
632 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
633 __ARM_FEATURE_COPROC before changing architectures.
635 2018-02-06 Richard Biener <rguenther@suse.de>
637 * config/i386/i386.c (print_reg): Fix typo.
638 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
640 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
642 * configure: Regenerate.
644 2018-02-05 Martin Sebor <msebor@redhat.com>
646 PR tree-optimization/83369
647 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
650 2018-02-05 Martin Liska <mliska@suse.cz>
652 * doc/invoke.texi: Cherry-pick upstream r323995.
654 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
656 * ira.c (ira_init_register_move_cost): Adjust comment.
658 2018-02-05 Martin Liska <mliska@suse.cz>
660 PR gcov-profile/84137
661 * doc/gcov.texi: Fix typo in documentation.
663 2018-02-05 Martin Liska <mliska@suse.cz>
665 PR gcov-profile/83879
666 * doc/gcov.texi: Document necessity of --dynamic-list-data when
667 using dlopen functionality.
669 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
671 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
672 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
673 _mm_maskz_range_ss, _mm_mask_range_round_ss,
674 _mm_maskz_range_round_ss): New intrinsics.
675 (__builtin_ia32_rangesd128_round)
676 (__builtin_ia32_rangess128_round): Remove.
677 (__builtin_ia32_rangesd128_mask_round,
678 __builtin_ia32_rangess128_mask_round): New builtins.
679 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
680 __builtin_ia32_rangess128_round): Remove.
681 (__builtin_ia32_rangesd128_mask_round,
682 __builtin_ia32_rangess128_mask_round): New builtins.
683 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
684 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
685 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
686 "<round_saeonly_constraint>")): Changed to ...
687 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
688 "<round_saeonly_scalar_constraint>")): ... this.
689 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
690 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
691 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
692 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
693 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
695 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
697 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
699 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
700 Remove all values except native, 8540 and 8548.
702 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
704 * config/i386/i386.c (ix86_output_function_return): Pass
705 INVALID_REGNUM, instead of -1, as invalid register number to
706 indirect_thunk_name and output_indirect_thunk.
708 2018-02-02 Julia Koval <julia.koval@intel.com>
710 * config.gcc: Add -march=icelake.
711 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
712 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
713 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
714 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
715 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
716 (processor_target_table): Add icelake.
717 (ix86_option_override_internal): Handle new PTAs.
718 (get_builtin_code_for_version): Handle icelake.
719 (M_INTEL_COREI7_ICELAKE): New.
720 (fold_builtin_cpu): Handle icelake.
721 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
722 * doc/invoke.texi: Add -march=icelake.
724 2018-02-02 Julia Koval <julia.koval@intel.com>
726 * config/i386/i386.c (ix86_option_override_internal): Change flags type
728 * wide-int-bitmask.h: New.
730 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
733 * config/i386/i386.md: Replace Pmode with word_mode in
734 builtin_setjmp_setup and builtin_longjmp to support x32.
736 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
740 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
742 (rs6000_supported_cpu_names): New static variable.
743 (linux_cpu_translation_table): Likewise.
744 (elf_platform) <cpu>: Define new static variable and use it.
745 Translate kernel AT_PLATFORM name to canonical name if needed.
746 Error if platform name is unknown.
748 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
751 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
753 2018-02-01 Jeff Law <law@redhat.com>
756 * config/i386/i386.c (release_scratch_register_on_entry): Add new
757 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
758 the scratch if RELEASE_VIA_POP is false.
759 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
760 If we have to save a temporary register, decrement SIZE appropriately.
761 Pass new arguments to release_scratch_register_on_entry.
762 (ix86_adjust_stack_and_probe): Likewise.
763 (ix86_emit_probe_stack_range): Pass new arguments to
764 release_scratch_register_on_entry.
766 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
768 PR rtl-optimization/84157
769 * combine.c (change_zero_ext): Use REG_P predicate in
770 front of HARD_REGISTER_P predicate.
772 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
774 * config/avr/avr.c (avr_option_override): Move disabling of
775 -fdelete-null-pointer-checks to...
776 * common/config/avr/avr-common.c (avr_option_optimization_table):
779 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
781 PR tree-optimization/81635
782 * tree-data-ref.c (split_constant_offset_1): For types that
783 wrap on overflow, try to use range info to prove that wrapping
786 2018-02-01 Renlin Li <renlin.li@arm.com>
789 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
791 (aarch64_register_move_cost): Likewise.
792 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
794 (REG_CLASS_NAMES): Likewise.
795 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
796 TAILCALL_ADDR_REGS. Remove IP registers.
797 * config/aarch64/aarch64.md (Ucs): Update register constraint.
799 2018-02-01 Richard Biener <rguenther@suse.de>
801 * domwalk.h (dom_walker::dom_walker): Add additional constructor
802 for specifying RPO order and allow NULL for that.
803 * domwalk.c (dom_walker::dom_walker): Likewise.
804 (dom_walker::walk): Handle NULL RPO order.
805 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
807 (rewrite_update_dom_walker): Likewise.
808 (mark_def_dom_walker): Likewise.
810 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
812 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
813 (aarch64_maybe_expand_sve_subreg_move): Declare.
814 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
815 * config/aarch64/predicates.md (aarch64_any_register_operand): New
817 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
818 that are semantically a reverse operation.
819 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
820 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
821 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
823 (aarch64_can_change_mode_class): For big-endian, forbid changes
824 between two SVE modes if they have different element sizes.
826 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
828 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
829 the TImode handling for big-endian targets.
831 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
833 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
834 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
836 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
837 Remove BSWAP handing for big-endian targets and use the form of
838 LD1RQ appropariate for the mode.
840 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
842 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
843 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
846 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
849 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
850 check for operands that need to go through aarch64_sve_reload_be.
852 2018-02-01 Jakub Jelinek <jakub@redhat.com>
854 PR tree-optimization/81661
855 PR tree-optimization/84117
856 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
857 * tree-eh.c: Include gimplify.h.
858 (find_trapping_overflow, replace_trapping_overflow,
859 rewrite_to_non_trapping_overflow): New functions.
860 * tree-vect-loop.c: Include tree-eh.h.
861 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
862 * tree-data-ref.c: Include tree-eh.h.
863 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
865 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
867 PR rtl-optimization/84123
868 * combine.c (change_zero_ext): Check if hard register satisfies
869 can_change_dest_mode before calling gen_lowpart_SUBREG.
871 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
874 * ira.c (ira_init_register_move_cost): Remove assert.
876 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
878 PR rtl-optimization/84071
879 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
880 * doc/tm.texi: Regenerate.
882 2018-01-31 Richard Biener <rguenther@suse.de>
884 PR tree-optimization/84132
885 * tree-data-ref.c (analyze_miv_subscript): Properly
886 check whether evolution_function_is_affine_multivariate_p
887 before calling gcd_of_steps_may_divide_p.
889 2018-01-31 Julia Koval <julia.koval@intel.com>
892 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
893 * config/i386/i386.md (rdpid_rex64) New.
894 (rdpid): Make 32bit only.
896 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
899 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
900 an IDENTIFIER_NODE for FUNCTION_TYPE's.
902 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
905 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
907 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
909 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
911 PR rtl-optimization/84071
912 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
913 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
915 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
917 * config/arc/arc.c (arc_handle_aux_attribute): New function.
918 (arc_attribute_table): Add 'aux' attribute.
919 (arc_in_small_data_p): Consider aux like variables.
920 (arc_is_aux_reg_p): New function.
921 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
922 (arc_get_aux_arg): New function.
923 (prepare_move_operands): Handle aux-register access.
924 (arc_handle_aux_attribute): New function.
925 * doc/extend.texi (ARC Variable attributes): Add subsection.
927 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
929 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
930 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
931 (arc_attribute_table): Add 'uncached' attribute.
932 (arc_print_operand): Print '.di' flag for uncached memory
934 (arc_in_small_data_p): Do not consider for small data the uncached
936 (arc_is_uncached_mem_p): New function.
937 * config/arc/predicates.md (compact_store_memory_operand): Check
938 for uncached memory accesses.
939 (nonvol_nonimm_operand): Likewise.
940 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
942 2018-01-31 Jakub Jelinek <jakub@redhat.com>
945 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
946 falign-loops=): Add Optimization flag.
948 2018-01-30 Jeff Law <law@redhat.com>
951 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
952 INT_REGISTERS_SAVED. Check it prior to calling
953 get_scratch_register_on_entry.
954 (ix86_adjust_stack_and_probe): Similarly.
955 (ix86_emit_probe_stack_range): Similarly.
956 (ix86_expand_prologue): Corresponding changes.
958 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
961 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
962 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
964 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
967 * lra-constraints.c (curr_insn_transform): Process AND in the
970 2018-01-30 Jakub Jelinek <jakub@redhat.com>
972 PR rtl-optimization/83986
973 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
974 dependence against last_pending_memory_flush in addition to
977 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
979 PR tree-optimization/81611
980 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
983 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
985 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
988 2018-01-30 Richard Biener <rguenther@suse.de>
989 Jakub Jelinek <jakub@redhat.com>
991 PR tree-optimization/84111
992 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
993 inner loops added during recursion, as they don't have up-to-date
996 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
999 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1000 (can_inline_edge_by_limits_p): ... here.
1001 (can_early_inline_edge_p, check_callers,
1002 update_caller_keys, update_callee_keys, recursive_inlining,
1003 add_new_edges_to_heap, speculation_useful_p,
1004 inline_small_functions,
1005 inline_small_functions, flatten_function,
1006 inline_to_all_callers_1): Update.
1008 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1010 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1013 2018-01-30 Richard Biener <rguenther@suse.de>
1015 PR tree-optimization/83008
1016 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1017 invariant and constant vector uses in stmts when they need
1020 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1023 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1024 * configure: Regenerate.
1026 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1028 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1030 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1031 Use gen_rtx_REG rather than gen_lowpart.
1033 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1035 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1036 rather than 0 when creating partial subregs.
1038 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1040 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1043 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1046 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1047 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1048 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1049 flags. This restores the settings used before the 2017-07-24.
1050 Turning off pre increment/decrement/modify allows IVOPTS to
1051 optimize DF/SF loops where the index is an int.
1053 2018-01-29 Richard Biener <rguenther@suse.de>
1054 Kelvin Nilsen <kelvin@gcc.gnu.org>
1057 * tree-vect-stmts.c (vectorizable_call): Don't call
1058 targetm.vectorize_builtin_md_vectorized_function if callee is
1061 2018-01-22 Carl Love <cel@us.ibm.com>
1063 * doc/extend.tex: Fix typo in second arg in
1064 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1066 2018-01-29 Richard Biener <rguenther@suse.de>
1068 PR tree-optimization/84086
1069 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1070 (flush_ssaname_freelist): When SSA names were released reset
1071 the SCEV hash table.
1073 2018-01-29 Richard Biener <rguenther@suse.de>
1075 PR tree-optimization/84057
1076 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1077 removed paths when removing edges.
1079 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1081 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1082 -mfunction-return=@var{choice}.
1084 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1087 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1089 (layout::print_source_line): Likewise.
1090 (test_get_line_width_without_trailing_whitespace): Add test cases.
1092 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1095 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1098 2018-01-26 Jim Wilson <jimw@sifive.com>
1100 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1102 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1105 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1107 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1108 and CMP + SUB-immediate -> SUBS.
1110 2018-01-26 Martin Sebor <msebor@redhat.com>
1112 PR tree-optimization/83896
1113 * tree-ssa-strlen.c (get_string_len): Rename...
1114 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1115 Avoid assuming length is constant.
1116 (handle_char_store): Use HOST_WIDE_INT for string length.
1118 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1121 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1122 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1124 2018-01-26 Richard Biener <rguenther@suse.de>
1126 PR rtl-optimization/84003
1127 * dse.c (record_store): Only record redundant stores when
1128 the earlier store aliases at least all accesses the later one does.
1130 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1132 PR rtl-optimization/83985
1133 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1134 REG_CFA_RESTORE insns.
1135 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1136 REG_CFA_RESTORE insns here.
1139 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1140 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1142 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1144 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1145 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1146 (arc_init): Likewise.
1147 (arc_override_options): Likewise.
1148 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1150 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1152 * config/arc/arc.h (TARGET_DBNZ): Define.
1153 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1154 properly set the tune attribute.
1155 (dbnz): Use TARGET_DBNZ guard.
1156 * config/arc/arc.opt (mtune): Add core3 option.
1158 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1160 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1161 recognize new pic like addresses.
1162 (arc_delegitimize_address): Clean up.
1164 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1166 * config/arc/arc-arches.def: Option mrf16 valid for all
1168 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1169 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1170 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1171 * config/arc/arc-tables.opt: Regenerate.
1172 * config/arc/arc.c (arc_conditional_register_usage): Handle
1173 reduced register file case.
1174 (arc_file_start): Set must have build attributes.
1175 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1177 * config/arc/arc.opt (mrf16): Add new option.
1178 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1179 * config/arc/genmultilib.awk: Handle new mrf16 option.
1180 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1181 * config/arc/t-multilib: Regenerate.
1182 * doc/invoke.texi (ARC Options): Document mrf16 option.
1184 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1186 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1187 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1188 (arc_attribute_table): Add 'secure_call' attribute.
1189 (arc_print_operand): Print secure call operand.
1190 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1192 (arc_is_secure_call_p): New function. * config/arc/arc.md
1193 (call_i): Add support for sjli instruction.
1194 (call_value_i): Likewise.
1195 * config/arc/constraints.md (Csc): New constraint.
1197 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1198 John Eric Martin <John.Martin@emmicro-us.com>
1200 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1201 * config/arc/arc.c (_arc_jli_section): New struct.
1202 (arc_jli_section): New type.
1203 (rc_jli_sections): New static variable.
1204 (arc_handle_jli_attribute): New function.
1205 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1206 (arc_file_end): New function.
1207 (TARGET_ASM_FILE_END): Define.
1208 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1209 (arc_add_jli_section): New function.
1210 (jli_call_scan): Likewise.
1211 (arc_reorg): Call jli_call_scan.
1212 (arc_output_addsi): Remove 'S' from printing asm operand.
1213 (arc_is_jli_call_p): New function.
1214 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1216 (movhi_insn): Likewise.
1217 (movsi_insn): Likewise.
1218 (movsi_set_cc_insn): Likewise.
1219 (loadqi_update): Likewise.
1220 (load_zeroextendqisi_update): Likewise.
1221 (load_signextendqisi_update): Likewise.
1222 (loadhi_update): Likewise.
1223 (load_zeroextendhisi_update): Likewise.
1224 (load_signextendhisi_update): Likewise.
1225 (loadsi_update): Likewise.
1226 (loadsf_update): Likewise.
1227 (movsicc_insn): Likewise.
1228 (bset_insn): Likewise.
1229 (bxor_insn): Likewise.
1230 (bclr_insn): Likewise.
1231 (bmsk_insn): Likewise.
1232 (bicsi3_insn): Likewise.
1233 (cmpsi_cc_c_insn): Likewise.
1234 (movsi_ne): Likewise.
1235 (movsi_cond_exec): Likewise.
1236 (clrsbsi2): Likewise.
1247 (call_i): Remove 'S' asm letter, add jli instruction.
1248 (call_value_i): Likewise.
1249 * config/arc/arc.op (mjli-always): New option.
1250 * config/arc/constraints.md (Cji): New constraint.
1251 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1253 (subsf3_fpx): Likewise.
1254 (mulsf3_fpx): Likewise.
1255 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1257 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1259 * doc/invoke.texi (ARC): Document mjli-always option.
1261 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1263 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1264 and use incw and decw where possible
1265 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1267 2018-01-26 Richard Biener <rguenther@suse.de>
1269 PR tree-optimization/81082
1270 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1271 association if it requires casting to unsigned.
1272 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1273 from fold_plusminus_mult_expr to catch important cases late when
1274 range info is available.
1276 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1278 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1279 * configure.ac (hidden_linkonce): New test.
1280 * configure: Regenerate.
1281 * config.in: Regenerate.
1283 2018-01-26 Julia Koval <julia.koval@intel.com>
1285 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1286 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1287 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1288 _mm_mask_bitshuffle_epi64_mask): Fix type.
1289 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1290 USI_FTYPE_V4DI_V4DI_USI): Remove.
1291 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1292 __builtin_ia32_vpshufbitqmb256_mask,
1293 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1294 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1295 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1297 2018-01-26 Alan Modra <amodra@gmail.com>
1300 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1301 UNSPEC_VBPERMQ. Sort other unspecs.
1303 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1305 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1307 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1310 * predict.c (drop_profile): Do not push/pop cfun; update also
1312 (handle_missing_profiles): Fix logic looking for zero profiles.
1314 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1317 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1318 on functions with #pragma omp declare simd or functions with simd
1320 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1321 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1322 Remove trailing \n from warning_at calls.
1324 2018-01-25 Tom de Vries <tom@codesourcery.com>
1327 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1328 for neutered workers.
1330 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1333 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1334 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1336 2018-01-24 Jeff Law <law@redhat.com>
1339 * i386.c (get_probe_interval): Move to earlier point.
1340 (ix86_compute_frame_layout): If -fstack-clash-protection and
1341 the frame is larger than the probe interval, then use pushes
1342 to save registers rather than reg->mem moves.
1343 (ix86_expand_prologue): Remove conditional for int_registers_saved
1346 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1349 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1350 min/max for never referenced object.
1352 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1355 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1357 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1358 attributes from DECL_ATTRIBUTES (decl) without affecting
1359 DECL_ATTRIBUTES (current_function_decl).
1360 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1361 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1363 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1365 PR tree-optimization/83979
1366 * fold-const.c (fold_comparison): Use constant_boolean_node
1367 instead of boolean_{true,false}_node.
1369 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1371 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1374 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1376 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1377 Simplify the clause that sets the length attribute.
1378 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1379 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1380 clause that sets the length attribute.
1381 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1383 2018-01-24 Tom de Vries <tom@codesourcery.com>
1386 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1387 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1388 Add strict parameter.
1389 (prevent_branch_around_nothing): Insert dummy insn between branch to
1390 label and label with no ptx insn inbetween.
1391 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1393 2018-01-24 Tom de Vries <tom@codesourcery.com>
1396 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1397 for neutered threads in warp.
1398 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1400 2018-01-24 Richard Biener <rguenther@suse.de>
1402 PR tree-optimization/83176
1403 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1406 2018-01-24 Richard Biener <rguenther@suse.de>
1408 PR tree-optimization/82819
1409 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1410 code generating pluses that are no-ops in the target precision.
1412 2018-01-24 Richard Biener <rguenther@suse.de>
1415 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1417 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1419 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1420 to merge probabilities.
1421 * predict.c (probably_never_executed): Also mark as cold functions
1422 with global 0 profile and guessed local profile.
1423 * profile-count.c (profile_probability::combine_with_count): New
1425 * profile-count.h (profile_probability::operator*,
1426 profile_probability::operator*=, profile_probability::operator/,
1427 profile_probability::operator/=): Reduce precision to adjusted
1428 and set value to guessed on contradictory divisions.
1429 (profile_probability::combine_with_freq): Remove.
1430 (profile_probability::combine_wiht_count): Declare.
1431 (profile_count::force_nonzero):: Set to adjusted.
1432 (profile_count::probability_in):: Set quality to adjusted.
1433 * tree-ssa-tail-merge.c (replace_block_by): Use
1436 2018-01-23 Andrew Waterman <andrew@sifive.com>
1437 Jim Wilson <jimw@sifive.com>
1439 * config/riscv/riscv.c (riscv_stack_boundary): New.
1440 (riscv_option_override): Set riscv_stack_boundary. Handle
1441 riscv_preferred_stack_boundary_arg.
1442 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1443 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1444 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1445 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1446 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1447 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1449 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1452 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1453 of struct ix86_frame.
1454 (ix86_expand_epilogue): Likewise. Add a local variable for
1455 the reg_save_offset field in struct ix86_frame.
1457 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1459 PR tree-optimization/82604
1460 * tree-loop-distribution.c (enum partition_kind): New enum item
1461 PKIND_PARTIAL_MEMSET.
1462 (partition_builtin_p): Support above new enum item.
1463 (generate_code_for_partition): Ditto.
1464 (compute_access_range): Differentiate cases that equality can be
1465 proven at all loops, the innermost loops or no loops.
1466 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1467 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1468 (finalize_partitions, distribute_loop): Don't fuse partition of
1469 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1470 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1473 2018-01-23 Martin Liska <mliska@suse.cz>
1475 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1476 order to ignore the predictor.
1477 (PRED_POLYMORPHIC_CALL): Likewise.
1478 (PRED_RECURSIVE_CALL): Likewise.
1480 2018-01-23 Martin Liska <mliska@suse.cz>
1482 * tree-profile.c (tree_profiling): Print function header to
1483 aware reader which function we are working on.
1484 * value-prof.c (gimple_find_values_to_profile): Do not print
1485 not interesting value histograms.
1487 2018-01-23 Martin Liska <mliska@suse.cz>
1489 * profile-count.h (enum profile_quality): Add
1490 profile_uninitialized as the first value. Do not number values
1491 as they are zero based.
1492 (profile_count::verify): Update sanity check.
1493 (profile_probability::verify): Likewise.
1495 2018-01-23 Nathan Sidwell <nathan@acm.org>
1497 * doc/invoke.texi (ffor-scope): Deprecate.
1499 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1501 PR tree-optimization/83510
1502 * domwalk.c (set_all_edges_as_executable): New function.
1503 (dom_walker::dom_walker): Convert bool param
1504 "skip_unreachable_blocks" to enum reachability. Move setup of
1505 edge flags to set_all_edges_as_executable and only do it when
1506 reachability is REACHABLE_BLOCKS.
1507 * domwalk.h (enum dom_walker::reachability): New enum.
1508 (dom_walker::dom_walker): Convert bool param
1509 "skip_unreachable_blocks" to enum reachability.
1510 (set_all_edges_as_executable): New decl.
1511 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1512 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1514 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1515 but converting true to REACHABLE_BLOCKS.
1516 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1518 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1519 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1520 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1522 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1523 if check_all_array_refs will be called.
1525 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1527 * tree.c (selftest::test_location_wrappers): Add more test
1530 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1532 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1533 (selftest::test_bit_in_range): Likewise.
1535 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1538 * doc/sourcebuild.texi (vect_float): Say that the selector
1539 only describes the situation when -funsafe-math-optimizations is on.
1540 (vect_float_strict): Document.
1542 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1544 PR tree-optimization/83965
1545 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1546 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1547 instead of checking only for a reduction.
1548 (vect_recog_widen_sum_pattern): Likewise.
1550 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1552 * predict.c (probably_never_executed): Only use precise profile info.
1553 (compute_function_frequency): Skip after inlining hack since we now
1554 have quality checking.
1556 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1558 * profile-count.h (profile_probability::very_unlikely,
1559 profile_probability::unlikely, profile_probability::even): Set
1560 precision to guessed.
1562 2018-01-23 Richard Biener <rguenther@suse.de>
1564 PR tree-optimization/83963
1565 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1566 Properly terminate dominator walk when crossing the exit edge not
1567 when visiting its source block.
1569 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1572 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1573 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1575 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1577 PR tree-optimization/83957
1578 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1579 semicolon after for body surrounded by braces.
1581 PR tree-optimization/83081
1582 * profile-count.h (profile_probability::split): New method.
1583 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1584 Use profile_probability::split.
1585 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1586 when splitting a single conditional jump into 2.
1588 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1590 PR tree-optimization/69452
1591 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1594 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1596 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1597 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1598 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1600 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1602 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1603 * config/rl78/rl78.md: New define_expand "movdi"
1604 * config/rl78/rl78.c: New function definition rl78_split_movdi
1606 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1609 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1611 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1612 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1613 128-bit to produce an UNSPEC move to get the double word with the
1614 signbit and then a shift directly to do signbit.
1615 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1616 implementation with a new version that just does either a direct
1617 move or a regular move. Move memory interface to separate insns.
1618 Move insns so they are next to the expander.
1619 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1620 with signbit move. Split big and little endian case.
1621 (signbit<mode>2_dm_mem_le): Likewise.
1622 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1623 (signbit<mode>2_dm2): Likewise.
1625 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1627 * config/rl78/rl78.md: New define_expand "anddi3".
1629 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1631 * config/rl78/rl78.md: New define_expand "umindi3".
1633 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1635 * config/rl78/rl78.md: New define_expand "smindi3".
1637 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1639 * config/rl78/rl78.md: New define_expand "smaxdi3".
1641 2018-01-22 Carl Love <cel@us.ibm.com>
1643 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1644 LVX_V1TI): Add macro expansion.
1645 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1646 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1647 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1648 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1649 Change check to determine if the instruction is a byte reversing
1650 entry. Fix typo in comment.
1651 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1652 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1653 Add def_builtin calls for new builtins.
1654 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1655 Add define_insn expansion.
1657 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1659 * config/rl78/rl78.md: New define_expand "umaxdi3".
1661 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1663 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1664 for non-QImode registers
1666 2018-01-22 Richard Biener <rguenther@suse.de>
1668 PR tree-optimization/83963
1669 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1670 including the loop exit block.
1671 (scop_detection::merge_sese): Likewise.
1672 (scop_detection::add_scop): Do it here instead.
1674 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1676 * doc/sourcebuild.texi (arm_softfloat): Document.
1678 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1681 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1682 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1683 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1685 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1686 David Edelsohn <dje.gcc@gmail.com>
1689 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1690 Change "crset eq" to "crset 2".
1691 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1692 (*call_indirect_aix<mode>_nospec): Likewise.
1693 (*call_value_indirect_aix<mode>_nospec): Likewise.
1694 (*call_indirect_elfv2<mode>_nospec): Likewise.
1695 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1696 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1697 change assembly output from . to $.
1698 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1699 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1700 (*tablejump<mode>_internal1_nospec): Likewise.
1702 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1705 * config/sh/sh_optimize_sett_clrt.cc:
1706 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1708 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1710 PR tree-optimization/83940
1711 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1712 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1713 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1714 use it to pass back the definition type.
1715 (vect_check_store_rhs): Likewise rhs_dt_out.
1716 (vect_build_gather_load_calls): Add a mask_dt argument and use
1717 it instead of a call to vect_is_simple_use.
1718 (vectorizable_store): Update calls to vect_check_load_store_mask
1719 and vect_check_store_rhs. Use the dt returned by the latter instead
1720 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1721 instead of calls to vect_is_simple_use. Pass the scalar rather
1722 than the vector operand to vect_is_simple_use when handling
1723 second and subsequent copies of an rhs value.
1724 (vectorizable_load): Update calls to vect_check_load_store_mask
1725 and vect_build_gather_load_calls. Use the cached mask_dt and
1726 gs_info.offset_dt instead of calls to vect_is_simple_use.
1728 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1731 * tree-emutls.c: Include gimplify.h.
1732 (lower_emutls_2): New function.
1733 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1734 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1735 it before further processing.
1738 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1739 UINTVAL (trueop1) instead of INTVAL (op1).
1741 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1745 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1746 INCOMING_FRAME_SP_OFFSET if not defined.
1747 (scan_trace): Add ENTRY argument. If true and
1748 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1749 emit a note to adjust the CFA offset.
1750 (create_cfi_notes): Adjust scan_trace callers.
1751 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1752 INCOMING_FRAME_SP_OFFSET in the CIE.
1753 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1754 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1756 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1757 * doc/tm.texi: Regenerated.
1759 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1761 PR rtl-optimization/83147
1762 * lra-constraints.c (remove_inheritance_pseudos): Use
1763 lra_substitute_pseudo_within_insn.
1765 2018-01-19 Tom de Vries <tom@codesourcery.com>
1766 Cesar Philippidis <cesar@codesourcery.com>
1769 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1771 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1774 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1775 spaces for function labels.
1777 2018-01-19 Martin Liska <mliska@suse.cz>
1779 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1780 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1781 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1782 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1783 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1784 (PRED_CONST_RETURN): Change from 69 to 65.
1785 (PRED_NULL_RETURN): Change from 91 to 71.
1786 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1787 (PRED_LOOP_GUARD): Change from 66 to 73.
1789 2018-01-19 Martin Liska <mliska@suse.cz>
1791 * predict.c (predict_insn_def): Add new assert.
1792 (struct branch_predictor): Change type to signed integer.
1793 (test_prediction_value_range): Amend test to cover
1795 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1796 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1797 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1798 (PRED_LOOP_IV_COMPARE): Likewise.
1799 * predict.h (PROB_UNINITIALIZED): Define new constant.
1801 2018-01-19 Martin Liska <mliska@suse.cz>
1803 * predict.c (dump_prediction): Add new format for
1804 analyze_brprob.py script which is enabled with -details
1806 * profile-count.h (precise_p): New function.
1808 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1810 PR tree-optimization/83922
1811 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1812 there are no statements that need masking.
1813 (vect_active_double_reduction_p): New function.
1814 (vect_analyze_loop_operations): Use it when handling phis that
1815 are not in the loop header.
1817 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1819 PR tree-optimization/83914
1820 * tree-vect-loop.c (vectorizable_induction): Don't convert
1821 init_expr or apply the peeling adjustment for inductions
1822 that are nested within the vectorized loop.
1824 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1826 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1829 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1833 * function.h (gimplify_parameters): Add gimple_seq * argument.
1834 * function.c: Include gimple.h and options.h.
1835 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1836 for the added local temporaries if needed.
1837 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1838 if there are any parameter cleanups, wrap whole body into a
1839 try/finally with the cleanups.
1841 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1844 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1845 Use GET_MODE_CLASS for scalar floating point.
1847 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1851 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1852 Fix call of call_cgraph_insertion_hooks.
1854 2018-01-18 Martin Sebor <msebor@redhat.com>
1856 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1858 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1861 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1864 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1867 * common.opt: (-ffile-prefix-map): New option.
1868 * opts.c (common_handle_option): Defer it.
1869 * opts-global.c (handle_common_deferred_options): Handle it.
1870 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1871 * file-prefix-map.h: New file.
1872 (remap_debug_filename, add_debug_prefix_map): ...here.
1873 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1874 * final.c (debug_prefix_map, add_debug_prefix_map
1875 remap_debug_filename): Move to...
1876 * file-prefix-map.c: New file.
1877 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1878 generalize, get rid of alloca(), use strrchr() instead of strchr().
1879 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1880 Implement in terms of add_prefix_map().
1881 (remap_macro_filename, remap_debug_filename): Implement in term of
1883 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1884 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1885 * dbxout.c: Include file-prefix-map.h.
1886 * varasm.c: Likewise.
1887 * vmsdbgout.c: Likewise.
1888 * xcoffout.c: Likewise.
1889 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1890 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1891 * doc/invoke.texi (-ffile-prefix-map): Document.
1892 (-fdebug-prefix-map): Update description.
1894 2018-01-18 Martin Liska <mliska@suse.cz>
1896 * config/i386/i386.c (indirect_thunk_name): Document that also
1898 (output_indirect_thunk): Document why both instructions
1899 (pause and lfence) are generated.
1901 2018-01-18 Richard Biener <rguenther@suse.de>
1903 PR tree-optimization/83887
1904 * graphite-scop-detection.c
1905 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1906 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1907 (scop_detection::merge_sese): Re-implement with a flood-fill
1908 algorithm that properly finds a SESE region if it exists.
1910 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1913 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1914 pointer_diff optimizations use view_convert instead of convert.
1916 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1918 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1919 Generate different code for -mno-speculate-indirect-jumps.
1920 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1921 (*call_indirect_aix<mode>): Disable for
1922 -mno-speculate-indirect-jumps.
1923 (*call_indirect_aix<mode>_nospec): New define_insn.
1924 (*call_value_indirect_aix<mode>): Disable for
1925 -mno-speculate-indirect-jumps.
1926 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1927 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1928 -mno-speculate-indirect-jumps.
1929 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1931 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1933 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1934 long double type, set the flags for noting the default long double
1935 type, even if we don't pass or return a long double type.
1937 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1940 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1943 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1945 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1946 support for merge[hl].
1947 (fold_mergehl_helper): New helper function.
1948 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1949 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1950 (altivec_vmrglw_direct): Add xxmrglw insn.
1952 2018-01-17 Andrew Waterman <andrew@sifive.com>
1954 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1955 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1957 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1960 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1961 call the lto_location_cache before reading the
1962 DECL_SOURCE_LOCATION of the types.
1964 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1965 Richard Sandiford <richard.sandiford@linaro.org>
1967 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1968 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1969 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1970 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1971 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1973 * config/aarch64/constraints.md (aarch64_movti_operand):
1975 * config/aarch64/predicates.md (Uti): Add new constraint.
1977 2018-01-17 Carl Love <cel@us.ibm.com>
1978 * config/rs6000/vsx.md (define_expand xl_len_r,
1979 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1980 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1982 (define_expand, define_insn): Move the shift left from the
1983 define_insn to the define_expand for lxvl and stxvl instructions.
1984 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1985 and XL_LEN_R definitions to PURE.
1987 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1989 * config/i386/i386.c (indirect_thunk_name): Declare regno
1990 as unsigned int. Compare regno with INVALID_REGNUM.
1991 (output_indirect_thunk): Ditto.
1992 (output_indirect_thunk_function): Ditto.
1993 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1994 in the call to output_indirect_thunk_function.
1996 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1999 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2000 rather than the size of inner_type to determine the stack slot size
2001 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2003 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2006 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2009 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2011 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2012 endian Linux systems to optionally enable multilibs for selecting
2013 the long double type if the user configured an explicit type.
2014 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2015 have no long double multilibs if not defined.
2016 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2017 warn if the user used -mabi={ieee,ibm}longdouble and we built
2018 multilibs for long double.
2019 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2020 appropriate multilib option.
2021 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2023 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2024 for building long double multilibs.
2025 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2027 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2029 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2032 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2034 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2037 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2040 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2043 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2045 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2046 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2048 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2050 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2051 different rtl trees depending on TARGET_64BIT.
2052 (rs6000_gen_lvx): Likewise.
2054 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2056 * config/visium/visium.md (nop): Tweak comment.
2057 (hazard_nop): Likewise.
2059 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2061 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2062 -mspeculate-indirect-jumps.
2063 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2064 for -mno-speculate-indirect-jumps.
2065 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2066 (*call_value_indirect_elfv2<mode>): Disable for
2067 -mno-speculate-indirect-jumps.
2068 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2069 (indirect_jump): Emit different RTL for
2070 -mno-speculate-indirect-jumps.
2071 (*indirect_jump<mode>): Disable for
2072 -mno-speculate-indirect-jumps.
2073 (*indirect_jump<mode>_nospec): New define_insn.
2074 (tablejump): Emit different RTL for
2075 -mno-speculate-indirect-jumps.
2076 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2077 (tablejumpsi_nospec): New define_expand.
2078 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2079 (tablejumpdi_nospec): New define_expand.
2080 (*tablejump<mode>_internal1): Disable for
2081 -mno-speculate-indirect-jumps.
2082 (*tablejump<mode>_internal1_nospec): New define_insn.
2083 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2086 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2088 * caller-save.c (insert_save): Drop unnecessary parameter. All
2091 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2092 Richard Biener <rguenth@suse.de>
2095 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2096 return early, inline manually is_gimple_sizepos. Make sure if we
2097 call gimplify_expr we don't end up with a gimple constant.
2098 * tree.c (variably_modified_type_p): Don't return true for
2099 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2100 * gimplify.h (is_gimple_sizepos): Remove.
2102 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2104 PR tree-optimization/83857
2105 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2106 vectorizable_live_operation for pure SLP statements.
2107 (vectorizable_live_operation): Handle PHIs.
2109 2018-01-16 Richard Biener <rguenther@suse.de>
2111 PR tree-optimization/83867
2112 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2113 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2115 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2118 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2119 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2120 If off is not INTEGER_CST, issue a may not be aligned warning
2121 rather than isn't aligned. Use isn%'t rather than isn't.
2122 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2124 <case MULT_EXPR>: Improve the case when bottom and one of the
2125 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2126 operand, in that case check if the other operand is multiple of
2127 bottom divided by the INTEGER_CST operand.
2129 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2132 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2133 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2134 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2135 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2136 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2137 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2138 (pa_function_arg_size): New function.
2140 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2142 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2143 in a separate statement.
2145 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2147 PR tree-optimization/83847
2148 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2149 group gathers and scatters.
2151 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2153 PR rtl-optimization/86620
2154 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2156 PR rtl-optimization/83213
2157 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2158 to last if both are JUMP_INSNs.
2160 PR tree-optimization/83843
2161 * gimple-ssa-store-merging.c
2162 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2163 store_immediate_info for bswap/nop orig_stores.
2165 2018-01-15 Andrew Waterman <andrew@sifive.com>
2167 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2169 <UDIV>: Increase cost if !TARGET_DIV.
2171 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2173 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2174 (define_attr "cr_logical_3op"): New.
2175 (cceq_ior_compare): Adjust.
2176 (cceq_ior_compare_complement): Adjust.
2177 (*cceq_rev_compare): Adjust.
2178 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2179 (is_cracked_insn): Adjust.
2180 (insn_must_be_first_in_group): Adjust.
2181 * config/rs6000/40x.md: Adjust.
2182 * config/rs6000/440.md: Adjust.
2183 * config/rs6000/476.md: Adjust.
2184 * config/rs6000/601.md: Adjust.
2185 * config/rs6000/603.md: Adjust.
2186 * config/rs6000/6xx.md: Adjust.
2187 * config/rs6000/7450.md: Adjust.
2188 * config/rs6000/7xx.md: Adjust.
2189 * config/rs6000/8540.md: Adjust.
2190 * config/rs6000/cell.md: Adjust.
2191 * config/rs6000/e300c2c3.md: Adjust.
2192 * config/rs6000/e500mc.md: Adjust.
2193 * config/rs6000/e500mc64.md: Adjust.
2194 * config/rs6000/e5500.md: Adjust.
2195 * config/rs6000/e6500.md: Adjust.
2196 * config/rs6000/mpc.md: Adjust.
2197 * config/rs6000/power4.md: Adjust.
2198 * config/rs6000/power5.md: Adjust.
2199 * config/rs6000/power6.md: Adjust.
2200 * config/rs6000/power7.md: Adjust.
2201 * config/rs6000/power8.md: Adjust.
2202 * config/rs6000/power9.md: Adjust.
2203 * config/rs6000/rs64.md: Adjust.
2204 * config/rs6000/titan.md: Adjust.
2206 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2208 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2209 ix86_indirect_branch_register logic.
2211 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2213 * config/i386/constraints.md (Bs): Update
2214 ix86_indirect_branch_register check. Don't check
2215 ix86_indirect_branch_register with GOT_memory_operand.
2217 * config/i386/predicates.md (GOT_memory_operand): Don't check
2218 ix86_indirect_branch_register here.
2219 (GOT32_symbol_operand): Likewise.
2221 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2223 * config/i386/predicates.md (constant_call_address_operand):
2224 Rewrite ix86_indirect_branch_register logic.
2225 (sibcall_insn_operand): Likewise.
2227 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2229 * config/i386/constraints.md (Bs): Replace
2230 ix86_indirect_branch_thunk_register with
2231 ix86_indirect_branch_register.
2233 * config/i386/i386.md (indirect_jump): Likewise.
2234 (tablejump): Likewise.
2235 (*sibcall_memory): Likewise.
2236 (*sibcall_value_memory): Likewise.
2237 Peepholes of indirect call and jump via memory: Likewise.
2238 * config/i386/i386.opt: Likewise.
2239 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2240 (GOT_memory_operand): Likewise.
2241 (call_insn_operand): Likewise.
2242 (sibcall_insn_operand): Likewise.
2243 (GOT32_symbol_operand): Likewise.
2245 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2248 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2249 type rather than type addr's type points to.
2250 (expand_omp_atomic_mutex): Likewise.
2251 (expand_omp_atomic): Likewise.
2253 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2256 * config/i386/i386.c (output_indirect_thunk_function): Use
2257 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2258 for __x86_return_thunk.
2260 2018-01-15 Richard Biener <rguenther@suse.de>
2263 * expmed.c (extract_bit_field_1): Fix typo.
2265 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2268 * config/arm/iterators.md (VF): New mode iterator.
2269 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2270 Remove integer-related logic from pattern.
2271 (neon_vabd<mode>_3): Likewise.
2273 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2276 * common.opt (fstrict-overflow): No longer an alias.
2277 (fwrapv-pointer): New option.
2278 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2279 also for pointer types based on flag_wrapv_pointer.
2280 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2281 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2282 opts->x_flag_wrapv got set.
2283 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2284 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2285 POINTER_TYPE_OVERFLOW_UNDEFINED.
2286 * match.pd: Likewise in address comparison pattern.
2287 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2289 2018-01-15 Richard Biener <rguenther@suse.de>
2292 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2293 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2294 Reset type names to their identifier if their TYPE_DECL doesn't
2295 have linkage (and thus is used for ODR and devirt).
2296 (save_debug_info_for_decl): Remove.
2297 (save_debug_info_for_type): Likewise.
2298 (add_tree_to_fld_list): Adjust.
2299 * tree-pretty-print.c (dump_generic_node): Make dumping of
2300 type names more robust.
2302 2018-01-15 Richard Biener <rguenther@suse.de>
2304 * BASE-VER: Bump to 8.0.1.
2306 2018-01-14 Martin Sebor <msebor@redhat.com>
2309 * builtins.c (check_access): Avoid warning when the no-warning bit
2312 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2314 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2315 * ira-color (allocno_hard_regs_compare): Likewise.
2317 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2320 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2321 Use .pushsection/.popsection.
2323 2018-01-14 Martin Sebor <msebor@redhat.com>
2326 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2328 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2330 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2331 entry from extra_headers.
2332 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2333 extra_headers, make the list bitwise identical to the i?86-*-* one.
2335 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2337 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2338 -mcmodel=large with -mindirect-branch=thunk,
2339 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2340 -mfunction-return=thunk-extern.
2341 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2342 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2343 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2345 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2347 * config/i386/i386.c (print_reg): Print the name of the full
2348 integer register without '%'.
2349 (ix86_print_operand): Handle 'V'.
2350 * doc/extend.texi: Document 'V' modifier.
2352 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2354 * config/i386/constraints.md (Bs): Disallow memory operand for
2355 -mindirect-branch-register.
2357 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2358 (GOT_memory_operand): Likewise.
2359 (call_insn_operand): Likewise.
2360 (sibcall_insn_operand): Likewise.
2361 (GOT32_symbol_operand): Likewise.
2362 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2363 for -mindirect-branch-register.
2364 (tablejump): Likewise.
2365 (*sibcall_memory): Likewise.
2366 (*sibcall_value_memory): Likewise.
2367 Disallow peepholes of indirect call and jump via memory for
2368 -mindirect-branch-register.
2369 (*call_pop): Replace m with Bw.
2370 (*call_value_pop): Likewise.
2371 (*sibcall_pop_memory): Replace m with Bs.
2372 * config/i386/i386.opt (mindirect-branch-register): New option.
2373 * doc/invoke.texi: Document -mindirect-branch-register option.
2375 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2377 * config/i386/i386-protos.h (ix86_output_function_return): New.
2378 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2379 set function_return_type.
2380 (indirect_thunk_name): Add ret_p to indicate thunk for function
2382 (output_indirect_thunk_function): Pass false to
2383 indirect_thunk_name.
2384 (ix86_output_indirect_branch_via_reg): Likewise.
2385 (ix86_output_indirect_branch_via_push): Likewise.
2386 (output_indirect_thunk_function): Create alias for function
2387 return thunk if regno < 0.
2388 (ix86_output_function_return): New function.
2389 (ix86_handle_fndecl_attribute): Handle function_return.
2390 (ix86_attribute_table): Add function_return.
2391 * config/i386/i386.h (machine_function): Add
2392 function_return_type.
2393 * config/i386/i386.md (simple_return_internal): Use
2394 ix86_output_function_return.
2395 (simple_return_internal_long): Likewise.
2396 * config/i386/i386.opt (mfunction-return=): New option.
2397 (indirect_branch): Mention -mfunction-return=.
2398 * doc/extend.texi: Document function_return function attribute.
2399 * doc/invoke.texi: Document -mfunction-return= option.
2401 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2403 * config/i386/i386-opts.h (indirect_branch): New.
2404 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2405 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2406 with local indirect jump when converting indirect call and jump.
2407 (ix86_set_indirect_branch_type): New.
2408 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2409 (indirectlabelno): New.
2410 (indirect_thunk_needed): Likewise.
2411 (indirect_thunk_bnd_needed): Likewise.
2412 (indirect_thunks_used): Likewise.
2413 (indirect_thunks_bnd_used): Likewise.
2414 (INDIRECT_LABEL): Likewise.
2415 (indirect_thunk_name): Likewise.
2416 (output_indirect_thunk): Likewise.
2417 (output_indirect_thunk_function): Likewise.
2418 (ix86_output_indirect_branch_via_reg): Likewise.
2419 (ix86_output_indirect_branch_via_push): Likewise.
2420 (ix86_output_indirect_branch): Likewise.
2421 (ix86_output_indirect_jmp): Likewise.
2422 (ix86_code_end): Call output_indirect_thunk_function if needed.
2423 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2425 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2426 (ix86_attribute_table): Add indirect_branch.
2427 * config/i386/i386.h (machine_function): Add indirect_branch_type
2428 and has_local_indirect_jump.
2429 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2431 (tablejump): Likewise.
2432 (*indirect_jump): Use ix86_output_indirect_jmp.
2433 (*tablejump_1): Likewise.
2434 (simple_return_indirect_internal): Likewise.
2435 * config/i386/i386.opt (mindirect-branch=): New option.
2436 (indirect_branch): New.
2439 (thunk-inline): Likewise.
2440 (thunk-extern): Likewise.
2441 * doc/extend.texi: Document indirect_branch function attribute.
2442 * doc/invoke.texi: Document -mindirect-branch= option.
2444 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2447 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2449 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2451 * ipa-inline.c (want_inline_small_function_p): Return false if
2452 inlining has already failed with CIF_FINAL_ERROR.
2453 (update_caller_keys): Call want_inline_small_function_p before
2455 (update_callee_keys): Likewise.
2457 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2459 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2461 (rs6000_quadword_masked_address_p): Likewise.
2462 (quad_aligned_load_p): Likewise.
2463 (quad_aligned_store_p): Likewise.
2464 (const_load_sequence_p): Add comment to describe the outer-most loop.
2465 (mimic_memory_attributes_and_flags): New function.
2466 (rs6000_gen_stvx): Likewise.
2467 (replace_swapped_aligned_store): Likewise.
2468 (rs6000_gen_lvx): Likewise.
2469 (replace_swapped_aligned_load): Likewise.
2470 (replace_swapped_load_constant): Capitalize argument name in
2471 comment describing this function.
2472 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2473 and stores that access quad-word aligned addresses and replace
2474 with stvx or lvx instructions when appropriate.
2475 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2476 New function prototype.
2477 (rs6000_quadword_masked_address_p): Likewise.
2478 (rs6000_gen_lvx): Likewise.
2479 (rs6000_gen_stvx): Likewise.
2480 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2481 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2482 when memory address is aligned.
2483 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2484 this split to select lvx instruction when memory address is aligned.
2485 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2486 instruction when memory address is aligned.
2487 (*vsx_le_perm_load_v16qi): Likewise.
2488 (four unnamed splitters): Modify to select the stvx instruction
2489 when memory is aligned.
2491 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2493 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2494 which appears in the queue multiple times.
2496 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2497 Alan Hayward <alan.hayward@arm.com>
2498 David Sherwood <david.sherwood@arm.com>
2500 * tree-vectorizer.h (vec_lower_bound): New structure.
2501 (_loop_vec_info): Add check_nonzero and lower_bounds.
2502 (LOOP_VINFO_CHECK_NONZERO): New macro.
2503 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2504 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2505 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2506 fields. Make seg_len the distance travelled, not including the
2508 (dr_direction_indicator): Declare.
2509 (dr_zero_step_indicator): Likewise.
2510 (dr_known_forward_stride_p): Likewise.
2511 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2513 (runtime_alias_check_p): Allow runtime alias checks with
2515 (operator ==): Compare access_size and align.
2516 (prune_runtime_alias_test_list): Rework for new distinction between
2517 the access_size and seg_len.
2518 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2520 (get_segment_min_max): New function.
2521 (create_intersect_range_checks): Use it.
2522 (dr_step_indicator): New function.
2523 (dr_direction_indicator): Likewise.
2524 (dr_zero_step_indicator): Likewise.
2525 (dr_known_forward_stride_p): Likewise.
2526 * tree-loop-distribution.c (data_ref_segment_size): Return
2527 DR_STEP * (niters - 1).
2528 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2530 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2531 (vect_preserves_scalar_order_p): New function, split out from...
2532 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2533 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2534 (vect_vfa_access_size): New function.
2535 (vect_vfa_align): Likewise.
2536 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2537 (dump_lower_bound): New function.
2538 (vect_check_lower_bound): Likewise.
2539 (vect_small_gap_p): Likewise.
2540 (vectorizable_with_step_bound_p): Likewise.
2541 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2542 depencies if the vectorization factor is 1. Convert the checks
2543 for nonzero steps into checks on the bounds of DR_STEP. Try using
2544 a bunds check for variable steps if the minimum required step is
2545 relatively small. Update calls to the dr_with_seg_len
2546 constructor and to vect_compile_time_alias.
2547 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2549 (vect_loop_versioning): Call it.
2550 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2552 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2554 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2555 Alan Hayward <alan.hayward@arm.com>
2556 David Sherwood <david.sherwood@arm.com>
2558 * doc/sourcebuild.texi (vect_scatter_store): Document.
2559 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2561 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2563 * genopinit.c (main): Add supports_vec_scatter_store and
2564 supports_vec_scatter_store_cached to target_optabs.
2565 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2566 IFN_MASK_SCATTER_STORE.
2567 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2569 * internal-fn.h (internal_store_fn_p): Declare.
2570 (internal_fn_stored_value_index): Likewise.
2571 * internal-fn.c (scatter_store_direct): New macro.
2572 (expand_scatter_store_optab_fn): New function.
2573 (direct_scatter_store_optab_supported_p): New macro.
2574 (internal_store_fn_p): New function.
2575 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2576 IFN_MASK_SCATTER_STORE.
2577 (internal_fn_mask_index): Likewise.
2578 (internal_fn_stored_value_index): New function.
2579 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2581 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2582 * optabs-query.c (supports_vec_scatter_store_p): New function.
2583 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2584 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2585 true for scatter stores.
2586 (vect_gather_scatter_fn_p): Handle scatter stores too.
2587 (vect_check_gather_scatter): Consider using scatter stores if
2588 supports_vec_scatter_store_p.
2589 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2591 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2592 internal_fn_stored_value_index.
2593 (check_load_store_masking): Handle scatter stores too.
2594 (vect_get_store_rhs): Make public.
2595 (vectorizable_call): Use internal_store_fn_p.
2596 (vectorizable_store): Handle scatter store internal functions.
2597 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2598 when deciding whether the end of the group has been reached.
2599 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2600 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2601 (mask_scatter_store<mode>): New insns.
2603 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2604 Alan Hayward <alan.hayward@arm.com>
2605 David Sherwood <david.sherwood@arm.com>
2607 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2608 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2609 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2611 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2612 Use vect_truncate_gather_scatter_offset if we can't treat the
2613 operation as a normal gather load or scatter store.
2614 (get_group_load_store_type): Take the gather_scatter_info
2615 as argument. Try using a gather load or scatter store for
2616 single-element groups.
2617 (get_load_store_type): Update calls to get_group_load_store_type
2618 and vect_use_strided_gather_scatters_p.
2620 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2621 Alan Hayward <alan.hayward@arm.com>
2622 David Sherwood <david.sherwood@arm.com>
2624 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2625 optional tree argument.
2626 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2628 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2629 but continue to use the current value as a fallback.
2630 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2631 to compare the updates.
2632 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2633 (get_load_store_type): Use it when handling a strided access.
2634 (vect_get_strided_load_store_ops): New function.
2635 (vect_get_data_ptr_increment): Likewise.
2636 (vectorizable_load): Handle strided gather loads. Always pass
2637 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2639 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2640 Alan Hayward <alan.hayward@arm.com>
2641 David Sherwood <david.sherwood@arm.com>
2643 * doc/md.texi (gather_load@var{m}): Document.
2644 (mask_gather_load@var{m}): Likewise.
2645 * genopinit.c (main): Add supports_vec_gather_load and
2646 supports_vec_gather_load_cached to target_optabs.
2647 * optabs-tree.c (init_tree_optimization_optabs): Use
2648 ggc_cleared_alloc to allocate target_optabs.
2649 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2650 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2652 * internal-fn.h (internal_load_fn_p): Declare.
2653 (internal_gather_scatter_fn_p): Likewise.
2654 (internal_fn_mask_index): Likewise.
2655 (internal_gather_scatter_fn_supported_p): Likewise.
2656 * internal-fn.c (gather_load_direct): New macro.
2657 (expand_gather_load_optab_fn): New function.
2658 (direct_gather_load_optab_supported_p): New macro.
2659 (direct_internal_fn_optab): New function.
2660 (internal_load_fn_p): Likewise.
2661 (internal_gather_scatter_fn_p): Likewise.
2662 (internal_fn_mask_index): Likewise.
2663 (internal_gather_scatter_fn_supported_p): Likewise.
2664 * optabs-query.c (supports_at_least_one_mode_p): New function.
2665 (supports_vec_gather_load_p): Likewise.
2666 * optabs-query.h (supports_vec_gather_load_p): Declare.
2667 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2668 and memory_type field.
2669 (NUM_PATTERNS): Bump to 15.
2670 * tree-vect-data-refs.c: Include internal-fn.h.
2671 (vect_gather_scatter_fn_p): New function.
2672 (vect_describe_gather_scatter_call): Likewise.
2673 (vect_check_gather_scatter): Try using internal functions for
2674 gather loads. Recognize existing calls to a gather load function.
2675 (vect_analyze_data_refs): Consider using gather loads if
2676 supports_vec_gather_load_p.
2677 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2678 (vect_get_gather_scatter_offset_type): Likewise.
2679 (vect_convert_mask_for_vectype): Likewise.
2680 (vect_add_conversion_to_patterm): Likewise.
2681 (vect_try_gather_scatter_pattern): Likewise.
2682 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2683 (vect_vect_recog_func_ptrs): Add it.
2684 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2685 internal_fn_mask_index and internal_gather_scatter_fn_p.
2686 (check_load_store_masking): Take the gather_scatter_info as an
2687 argument and handle gather loads.
2688 (vect_get_gather_scatter_ops): New function.
2689 (vectorizable_call): Check internal_load_fn_p.
2690 (vectorizable_load): Likewise. Handle gather load internal
2692 (vectorizable_store): Update call to check_load_store_masking.
2693 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2694 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2695 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2696 (aarch64_gather_scale_operand_d): New predicates.
2697 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2698 (mask_gather_load<mode>): New insns.
2700 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2701 Alan Hayward <alan.hayward@arm.com>
2702 David Sherwood <david.sherwood@arm.com>
2704 * optabs.def (fold_left_plus_optab): New optab.
2705 * doc/md.texi (fold_left_plus_@var{m}): Document.
2706 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2707 * internal-fn.c (fold_left_direct): Define.
2708 (expand_fold_left_optab_fn): Likewise.
2709 (direct_fold_left_optab_supported_p): Likewise.
2710 * fold-const-call.c (fold_const_fold_left): New function.
2711 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2712 * tree-parloops.c (valid_reduction_p): New function.
2713 (gather_scalar_reductions): Use it.
2714 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2715 (vect_finish_replace_stmt): Declare.
2716 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2717 (needs_fold_left_reduction_p): New function, split out from...
2718 (vect_is_simple_reduction): ...here. Accept reductions that
2719 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2720 (vect_force_simple_reduction): Also store the reduction type in
2721 the assignment's STMT_VINFO_REDUC_TYPE.
2722 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2723 (merge_with_identity): New function.
2724 (vect_expand_fold_left): Likewise.
2725 (vectorize_fold_left_reduction): Likewise.
2726 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2727 scalar phi in place for it. Check for target support and reject
2728 cases that would reassociate the operation. Defer the transform
2729 phase to vectorize_fold_left_reduction.
2730 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2731 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2732 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2734 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2736 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2737 call to ifc_temp_var.
2739 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2740 Alan Hayward <alan.hayward@arm.com>
2741 David Sherwood <david.sherwood@arm.com>
2743 * target.def (legitimize_address_displacement): Take the original
2744 offset as a poly_int.
2745 * targhooks.h (default_legitimize_address_displacement): Update
2747 * targhooks.c (default_legitimize_address_displacement): Likewise.
2748 * doc/tm.texi: Regenerate.
2749 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2750 as an argument, moving assert of ad->disp == ad->disp_term to...
2751 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2752 Try calling targetm.legitimize_address_displacement before expanding
2753 the address rather than afterwards, and adjust for the new interface.
2754 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2755 Match the new hook interface. Handle SVE addresses.
2756 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2759 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2761 * Makefile.in (OBJS): Add early-remat.o.
2762 * target.def (select_early_remat_modes): New hook.
2763 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2764 * doc/tm.texi: Regenerate.
2765 * targhooks.h (default_select_early_remat_modes): Declare.
2766 * targhooks.c (default_select_early_remat_modes): New function.
2767 * timevar.def (TV_EARLY_REMAT): New timevar.
2768 * passes.def (pass_early_remat): New pass.
2769 * tree-pass.h (make_pass_early_remat): Declare.
2770 * early-remat.c: New file.
2771 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2773 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2775 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2776 Alan Hayward <alan.hayward@arm.com>
2777 David Sherwood <david.sherwood@arm.com>
2779 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2780 vfm1 with a bound_epilog parameter.
2781 (vect_do_peeling): Update calls accordingly, and move the prologue
2782 call earlier in the function. Treat the base bound_epilog as 0 for
2783 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2784 this base when peeling for gaps.
2785 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2786 with fully-masked loops.
2787 (vect_estimate_min_profitable_iters): Handle the single peeled
2788 iteration in that case.
2790 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2791 Alan Hayward <alan.hayward@arm.com>
2792 David Sherwood <david.sherwood@arm.com>
2794 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2795 single-element interleaving even if the size is not a power of 2.
2796 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2797 accesses for single-element interleaving if the group size is
2800 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2801 Alan Hayward <alan.hayward@arm.com>
2802 David Sherwood <david.sherwood@arm.com>
2804 * doc/md.texi (fold_extract_last_@var{m}): Document.
2805 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2806 * optabs.def (fold_extract_last_optab): New optab.
2807 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2808 * internal-fn.c (fold_extract_direct): New macro.
2809 (expand_fold_extract_optab_fn): Likewise.
2810 (direct_fold_extract_optab_supported_p): Likewise.
2811 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2812 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2813 EXTRACT_LAST_REDUCTION.
2814 (get_initial_def_for_reduction): Do not create an initial vector
2815 for EXTRACT_LAST_REDUCTION reductions.
2816 (vectorizable_reduction): Leave the scalar phi in place for
2817 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2818 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2819 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2820 transform phase to vectorizable_condition.
2821 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2823 (vect_finish_stmt_generation): ...here.
2824 (vect_finish_replace_stmt): New function.
2825 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2826 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2828 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2830 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2831 Alan Hayward <alan.hayward@arm.com>
2832 David Sherwood <david.sherwood@arm.com>
2834 * doc/md.texi (extract_last_@var{m}): Document.
2835 * optabs.def (extract_last_optab): New optab.
2836 * internal-fn.def (EXTRACT_LAST): New internal function.
2837 * internal-fn.c (cond_unary_direct): New macro.
2838 (expand_cond_unary_optab_fn): Likewise.
2839 (direct_cond_unary_optab_supported_p): Likewise.
2840 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2841 loops using EXTRACT_LAST.
2842 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2843 (extract_last_<mode>): ...this optab.
2844 (vec_extract<mode><Vel>): Update accordingly.
2846 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2847 Alan Hayward <alan.hayward@arm.com>
2848 David Sherwood <david.sherwood@arm.com>
2850 * target.def (empty_mask_is_expensive): New hook.
2851 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2852 * doc/tm.texi: Regenerate.
2853 * targhooks.h (default_empty_mask_is_expensive): Declare.
2854 * targhooks.c (default_empty_mask_is_expensive): New function.
2855 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2856 if the target says that empty masks are expensive.
2857 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2859 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2861 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2862 Alan Hayward <alan.hayward@arm.com>
2863 David Sherwood <david.sherwood@arm.com>
2865 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2866 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2867 (vect_use_loop_mask_for_alignment_p): New function.
2868 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2869 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2870 niters_skip argument. Make sure that the first niters_skip elements
2871 of the first iteration are inactive.
2872 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2873 Update call to vect_set_loop_masks_directly.
2874 (get_misalign_in_elems): New function, split out from...
2875 (vect_gen_prolog_loop_niters): ...here.
2876 (vect_update_init_of_dr): Take a code argument that specifies whether
2877 the adjustment should be added or subtracted.
2878 (vect_update_init_of_drs): Likewise.
2879 (vect_prepare_for_masked_peels): New function.
2880 (vect_do_peeling): Skip prologue peeling if we're using a mask
2881 instead. Update call to vect_update_inits_of_drs.
2882 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2884 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2885 alignment. Do not include the number of peeled iterations in
2886 the minimum threshold in that case.
2887 (vectorizable_induction): Adjust the start value down by
2888 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2889 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2890 Take the number of skipped iterations into account when calculating
2892 * tree-vect-stmts.c (vect_gen_while_not): New function.
2894 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2895 Alan Hayward <alan.hayward@arm.com>
2896 David Sherwood <david.sherwood@arm.com>
2898 * doc/sourcebuild.texi (vect_fully_masked): Document.
2899 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2901 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2903 (vect_analyze_loop_2): ...here. Don't check the vectorization
2904 factor against the number of loop iterations if the loop is
2907 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2908 Alan Hayward <alan.hayward@arm.com>
2909 David Sherwood <david.sherwood@arm.com>
2911 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2912 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2913 (dump_groups): Update accordingly.
2914 (iv_use::mem_type): New member variable.
2915 (address_p): New function.
2916 (record_use): Add a mem_type argument and initialize the new
2918 (record_group_use): Add a mem_type argument. Use address_p.
2919 Remove obsolete null checks of base_object. Update call to record_use.
2920 (find_interesting_uses_op): Update call to record_group_use.
2921 (find_interesting_uses_cond): Likewise.
2922 (find_interesting_uses_address): Likewise.
2923 (get_mem_type_for_internal_fn): New function.
2924 (find_address_like_use): Likewise.
2925 (find_interesting_uses_stmt): Try find_address_like_use before
2926 calling find_interesting_uses_op.
2927 (addr_offset_valid_p): Use the iv mem_type field as the type
2928 of the addressed memory.
2929 (add_autoinc_candidates): Likewise.
2930 (get_address_cost): Likewise.
2931 (split_small_address_groups_p): Use address_p.
2932 (split_address_groups): Likewise.
2933 (add_iv_candidate_for_use): Likewise.
2934 (autoinc_possible_for_pair): Likewise.
2935 (rewrite_groups): Likewise.
2936 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2937 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2938 (get_alias_ptr_type_for_ptr_address): New function.
2939 (rewrite_use_address): Rewrite address uses in calls that were
2940 identified by find_address_like_use.
2942 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2943 Alan Hayward <alan.hayward@arm.com>
2944 David Sherwood <david.sherwood@arm.com>
2946 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2948 * gimple-expr.h (is_gimple_addressable: Likewise.
2949 * gimple-expr.c (is_gimple_address): Likewise.
2950 * internal-fn.c (expand_call_mem_ref): New function.
2951 (expand_mask_load_optab_fn): Use it.
2952 (expand_mask_store_optab_fn): Likewise.
2954 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2955 Alan Hayward <alan.hayward@arm.com>
2956 David Sherwood <david.sherwood@arm.com>
2958 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2959 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2960 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2961 (cond_umax@var{mode}): Document.
2962 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2963 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2964 (cond_umin_optab, cond_umax_optab): New optabs.
2965 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2966 (COND_IOR, COND_XOR): New internal functions.
2967 * internal-fn.h (get_conditional_internal_fn): Declare.
2968 * internal-fn.c (cond_binary_direct): New macro.
2969 (expand_cond_binary_optab_fn): Likewise.
2970 (direct_cond_binary_optab_supported_p): Likewise.
2971 (get_conditional_internal_fn): New function.
2972 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2973 Cope with reduction statements that are vectorized as calls rather
2975 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2976 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2977 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2978 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2979 (UNSPEC_COND_EOR): New unspecs.
2980 (optab): Add mappings for them.
2981 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2982 (sve_int_op, sve_fp_op): New int attributes.
2984 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2985 Alan Hayward <alan.hayward@arm.com>
2986 David Sherwood <david.sherwood@arm.com>
2988 * optabs.def (while_ult_optab): New optab.
2989 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2990 * internal-fn.def (WHILE_ULT): New internal function.
2991 * internal-fn.h (direct_internal_fn_supported_p): New override
2992 that takes two types as argument.
2993 * internal-fn.c (while_direct): New macro.
2994 (expand_while_optab_fn): New function.
2995 (convert_optab_supported_p): Likewise.
2996 (direct_while_optab_supported_p): New macro.
2997 * wide-int.h (wi::udiv_ceil): New function.
2998 * tree-vectorizer.h (rgroup_masks): New structure.
2999 (vec_loop_masks): New typedef.
3000 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3002 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3003 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3004 (vect_max_vf): New function.
3005 (slpeel_make_loop_iterate_ntimes): Delete.
3006 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3007 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3008 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3009 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3010 internal-fn.h, stor-layout.h and optabs-query.h.
3011 (vect_set_loop_mask): New function.
3012 (add_preheader_seq): Likewise.
3013 (add_header_seq): Likewise.
3014 (interleave_supported_p): Likewise.
3015 (vect_maybe_permute_loop_masks): Likewise.
3016 (vect_set_loop_masks_directly): Likewise.
3017 (vect_set_loop_condition_masked): Likewise.
3018 (vect_set_loop_condition_unmasked): New function, split out from
3019 slpeel_make_loop_iterate_ntimes.
3020 (slpeel_make_loop_iterate_ntimes): Rename to..
3021 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3022 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3023 (vect_do_peeling): Update call accordingly.
3024 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3026 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3027 mask_compare_type, can_fully_mask_p and fully_masked_p.
3028 (release_vec_loop_masks): New function.
3029 (_loop_vec_info): Use it to free the loop masks.
3030 (can_produce_all_loop_masks_p): New function.
3031 (vect_get_max_nscalars_per_iter): Likewise.
3032 (vect_verify_full_masking): Likewise.
3033 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3034 retries, and free the mask rgroups before retrying. Check loop-wide
3035 reasons for disallowing fully-masked loops. Make the final decision
3036 about whether use a fully-masked loop or not.
3037 (vect_estimate_min_profitable_iters): Do not assume that peeling
3038 for the number of iterations will be needed for fully-masked loops.
3039 (vectorizable_reduction): Disable fully-masked loops.
3040 (vectorizable_live_operation): Likewise.
3041 (vect_halve_mask_nunits): New function.
3042 (vect_double_mask_nunits): Likewise.
3043 (vect_record_loop_mask): Likewise.
3044 (vect_get_loop_mask): Likewise.
3045 (vect_transform_loop): Handle the case in which the final loop
3046 iteration might handle a partial vector. Call vect_set_loop_condition
3047 instead of slpeel_make_loop_iterate_ntimes.
3048 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3049 (check_load_store_masking): New function.
3050 (prepare_load_store_mask): Likewise.
3051 (vectorizable_store): Handle fully-masked loops.
3052 (vectorizable_load): Likewise.
3053 (supportable_widening_operation): Use vect_halve_mask_nunits for
3055 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3056 (vect_gen_while): New function.
3057 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3058 (aarch64_uqdec<mode>): New insn.
3060 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3061 Alan Hayward <alan.hayward@arm.com>
3062 David Sherwood <david.sherwood@arm.com>
3064 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3065 (reduc_xor_scal_optab): New optabs.
3066 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3067 (reduc_xor_scal_@var{m}): Document.
3068 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3069 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3071 * fold-const-call.c (fold_const_call): Handle them.
3072 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3073 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3074 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3075 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3076 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3077 (UNSPEC_XORV): New unspecs.
3078 (optab): Add entries for them.
3079 (BITWISEV): New int iterator.
3080 (bit_reduc_op): New int attributes.
3082 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3083 Alan Hayward <alan.hayward@arm.com>
3084 David Sherwood <david.sherwood@arm.com>
3086 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3087 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3088 * optabs.def (vec_shl_insert_optab): New optab.
3089 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3090 (duplicate_and_interleave): Likewise.
3091 * tree-vect-loop.c: Include internal-fn.h.
3092 (neutral_op_for_slp_reduction): New function, split out from
3093 get_initial_defs_for_reduction.
3094 (get_initial_def_for_reduction): Handle option 2 for variable-length
3095 vectors by loading the neutral value into a vector and then shifting
3096 the initial value into element 0.
3097 (get_initial_defs_for_reduction): Replace the code argument with
3098 the neutral value calculated by neutral_op_for_slp_reduction.
3099 Use gimple_build_vector for constant-length vectors.
3100 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3101 but the first group_size elements have a neutral value.
3102 Use duplicate_and_interleave otherwise.
3103 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3104 Update call to get_initial_defs_for_reduction. Handle SLP
3105 reductions for variable-length vectors by creating one vector
3106 result for each scalar result, with the elements associated
3107 with other scalar results stubbed out with the neutral value.
3108 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3109 Require IFN_VEC_SHL_INSERT for double reductions on
3110 variable-length vectors, or SLP reductions that have
3111 a neutral value. Require can_duplicate_and_interleave_p
3112 support for variable-length unchained SLP reductions if there
3113 is no neutral value, such as for MIN/MAX reductions. Also require
3114 the number of vector elements to be a multiple of the number of
3115 SLP statements when doing variable-length unchained SLP reductions.
3116 Update call to vect_create_epilog_for_reduction.
3117 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3118 and remove initial values.
3119 (duplicate_and_interleave): Make public.
3120 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3121 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3123 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3124 Alan Hayward <alan.hayward@arm.com>
3125 David Sherwood <david.sherwood@arm.com>
3127 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3128 (can_duplicate_and_interleave_p): New function.
3129 (vect_get_and_check_slp_defs): Take the vector of statements
3130 rather than just the current one. Remove excess parentheses.
3131 Restriction rejectinon of vect_constant_def and vect_external_def
3132 for variable-length vectors to boolean types, or types for which
3133 can_duplicate_and_interleave_p is false.
3134 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3135 (duplicate_and_interleave): New function.
3136 (vect_get_constant_vectors): Use gimple_build_vector for
3137 constant-length vectors and suitable variable-length constant
3138 vectors. Use duplicate_and_interleave for other variable-length
3139 vectors. Don't defer the update when inserting new statements.
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 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3146 min_profitable_iters doesn't go negative.
3148 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3149 Alan Hayward <alan.hayward@arm.com>
3150 David Sherwood <david.sherwood@arm.com>
3152 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3153 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3154 * optabs.def (vec_mask_load_lanes_optab): New optab.
3155 (vec_mask_store_lanes_optab): Likewise.
3156 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3157 (MASK_STORE_LANES): Likewise.
3158 * internal-fn.c (mask_load_lanes_direct): New macro.
3159 (mask_store_lanes_direct): Likewise.
3160 (expand_mask_load_optab_fn): Handle masked operations.
3161 (expand_mask_load_lanes_optab_fn): New macro.
3162 (expand_mask_store_optab_fn): Handle masked operations.
3163 (expand_mask_store_lanes_optab_fn): New macro.
3164 (direct_mask_load_lanes_optab_supported_p): Likewise.
3165 (direct_mask_store_lanes_optab_supported_p): Likewise.
3166 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3168 (vect_load_lanes_supported): Likewise.
3169 * tree-vect-data-refs.c (strip_conversion): New function.
3170 (can_group_stmts_p): Likewise.
3171 (vect_analyze_data_ref_accesses): Use it instead of checking
3172 for a pair of assignments.
3173 (vect_store_lanes_supported): Take a masked_p parameter.
3174 (vect_load_lanes_supported): Likewise.
3175 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3176 vect_store_lanes_supported and vect_load_lanes_supported.
3177 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3178 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3179 parameter. Don't allow gaps for masked accesses.
3180 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3181 and vect_load_lanes_supported.
3182 (get_load_store_type): Take a masked_p parameter and update
3183 call to get_group_load_store_type.
3184 (vectorizable_store): Update call to get_load_store_type.
3185 Handle IFN_MASK_STORE_LANES.
3186 (vectorizable_load): Update call to get_load_store_type.
3187 Handle IFN_MASK_LOAD_LANES.
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 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3195 * config/aarch64/aarch64-protos.h
3196 (aarch64_sve_struct_memory_operand_p): Declare.
3197 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3198 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3199 (VPRED, vpred): Handle SVE structure modes.
3200 * config/aarch64/constraints.md (Utx): New constraint.
3201 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3202 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3203 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3204 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3205 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3206 structure modes. Split into pieces after RA.
3207 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3208 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3210 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3211 SVE structure modes.
3212 (aarch64_classify_address): Likewise.
3213 (sizetochar): Move earlier in file.
3214 (aarch64_print_operand): Handle SVE register lists.
3215 (aarch64_array_mode): New function.
3216 (aarch64_sve_struct_memory_operand_p): Likewise.
3217 (TARGET_ARRAY_MODE): Redefine.
3219 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3220 Alan Hayward <alan.hayward@arm.com>
3221 David Sherwood <david.sherwood@arm.com>
3223 * target.def (array_mode): New target hook.
3224 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3225 * doc/tm.texi: Regenerate.
3226 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3227 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3228 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3230 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3233 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3234 Alan Hayward <alan.hayward@arm.com>
3235 David Sherwood <david.sherwood@arm.com>
3237 * fold-const.c (fold_binary_loc): Check the argument types
3238 rather than the result type when testing for a vector operation.
3240 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3242 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3243 * doc/tm.texi: Regenerate.
3245 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3246 Alan Hayward <alan.hayward@arm.com>
3247 David Sherwood <david.sherwood@arm.com>
3249 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3250 (sve): Document new AArch64 extension.
3251 * doc/md.texi (w): Extend the description of the AArch64
3252 constraint to include SVE vectors.
3253 (Upl, Upa): Document new AArch64 predicate constraints.
3254 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3256 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3257 (msve-vector-bits=): New option.
3258 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3259 SVE when these are disabled.
3260 (sve): New extension.
3261 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3262 modes. Adjust their number of units based on aarch64_sve_vg.
3263 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3264 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3265 aarch64_addr_query_type.
3266 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3267 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3268 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3269 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3270 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3271 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3272 (aarch64_simd_imm_zero_p): Delete.
3273 (aarch64_check_zero_based_sve_index_immediate): Declare.
3274 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3275 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3276 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3277 (aarch64_sve_float_mul_immediate_p): Likewise.
3278 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3280 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3281 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3282 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3283 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3284 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3285 (aarch64_regmode_natural_size): Likewise.
3286 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3287 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3289 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3290 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3291 for VG and the SVE predicate registers.
3292 (V_ALIASES): Add a "z"-prefixed alias.
3293 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3294 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3295 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3296 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3297 (REG_CLASS_NAMES): Add entries for them.
3298 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3299 and the predicate registers.
3300 (aarch64_sve_vg): Declare.
3301 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3302 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3303 (REGMODE_NATURAL_SIZE): Define.
3304 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3306 * config/aarch64/aarch64.c: Include cfgrtl.h.
3307 (simd_immediate_info): Add a constructor for series vectors,
3308 and an associated step field.
3309 (aarch64_sve_vg): New variable.
3310 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3311 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3312 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3313 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3314 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3315 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3316 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3317 (aarch64_get_mask_mode): New functions.
3318 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3319 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3320 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3321 predicate modes and predicate registers. Explicitly restrict
3322 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3323 to store a vector mode if it is recognized by
3324 aarch64_classify_vector_mode.
3325 (aarch64_regmode_natural_size): New function.
3326 (aarch64_hard_regno_caller_save_mode): Return the original mode
3328 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3329 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3330 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3331 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3333 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3334 does not overlap dest if the function is frame-related. Handle
3336 (aarch64_split_add_offset): New function.
3337 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3338 them aarch64_add_offset.
3339 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3340 and update call to aarch64_sub_sp.
3341 (aarch64_add_cfa_expression): New function.
3342 (aarch64_expand_prologue): Pass extra temporary registers to the
3343 functions above. Handle the case in which we need to emit new
3344 DW_CFA_expressions for registers that were originally saved
3345 relative to the stack pointer, but now have to be expressed
3346 relative to the frame pointer.
3347 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3349 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3350 IP0 and IP1 values for SVE frames.
3351 (aarch64_expand_vec_series): New function.
3352 (aarch64_expand_sve_widened_duplicate): Likewise.
3353 (aarch64_expand_sve_const_vector): Likewise.
3354 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3355 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3356 into the register, rather than emitting a SET directly.
3357 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3358 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3359 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3360 (offset_9bit_signed_scaled_p): New functions.
3361 (aarch64_replicate_bitmask_imm): New function.
3362 (aarch64_bitmask_imm): Use it.
3363 (aarch64_cannot_force_const_mem): Reject expressions involving
3364 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3365 (aarch64_classify_index): Handle SVE indices, by requiring
3366 a plain register index with a scale that matches the element size.
3367 (aarch64_classify_address): Handle SVE addresses. Assert that
3368 the mode of the address is VOIDmode or an integer mode.
3369 Update call to aarch64_classify_symbol.
3370 (aarch64_classify_symbolic_expression): Update call to
3371 aarch64_classify_symbol.
3372 (aarch64_const_vec_all_in_range_p): New function.
3373 (aarch64_print_vector_float_operand): Likewise.
3374 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3375 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3376 and the FP immediates 1.0 and 0.5.
3377 (aarch64_print_address_internal): Handle SVE addresses.
3378 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3379 (aarch64_regno_regclass): Handle predicate registers.
3380 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3382 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3383 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3384 (aarch64_convert_sve_vector_bits): New function.
3385 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3386 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3388 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3389 Handle SVE vector and predicate modes. Accept VL-based constants
3390 that need only one temporary register, and VL offsets that require
3391 no temporary registers.
3392 (aarch64_conditional_register_usage): Mark the predicate registers
3393 as fixed if SVE isn't available.
3394 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3395 Return true for SVE vector and predicate modes.
3396 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3397 rather than an unsigned int. Handle SVE modes.
3398 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3400 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3402 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3403 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3404 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3405 (aarch64_sve_float_mul_immediate_p): New functions.
3406 (aarch64_sve_valid_immediate): New function.
3407 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3408 Explicitly reject structure modes. Check for INDEX constants.
3409 Handle PTRUE and PFALSE constants.
3410 (aarch64_check_zero_based_sve_index_immediate): New function.
3411 (aarch64_simd_imm_zero_p): Delete.
3412 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3413 vector modes. Accept constants in the range of CNT[BHWD].
3414 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3415 ask for an Advanced SIMD mode.
3416 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3417 (aarch64_simd_vector_alignment): Handle SVE predicates.
3418 (aarch64_vectorize_preferred_vector_alignment): New function.
3419 (aarch64_simd_vector_alignment_reachable): Use it instead of
3421 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3422 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3424 (MAX_VECT_LEN): Delete.
3425 (expand_vec_perm_d): Add a vec_flags field.
3426 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3427 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3428 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3430 (aarch64_evpc_rev): Rename to...
3431 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3432 (aarch64_evpc_rev_global): New function.
3433 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3434 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3436 (aarch64_evpc_sve_tbl): New function.
3437 (aarch64_expand_vec_perm_const_1): Update after rename of
3438 aarch64_evpc_rev. Handle SVE permutes too, trying
3439 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3440 than aarch64_evpc_tbl.
3441 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3442 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3443 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3444 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3445 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3446 (aarch64_expand_sve_vcond): New functions.
3447 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3448 of aarch64_vector_mode_p.
3449 (aarch64_dwarf_poly_indeterminate_value): New function.
3450 (aarch64_compute_pressure_classes): Likewise.
3451 (aarch64_can_change_mode_class): Likewise.
3452 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3453 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3454 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3455 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3456 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3457 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3458 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3459 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3461 (Dn, Dl, Dr): Accept const as well as const_vector.
3462 (Dz): Likewise. Compare against CONST0_RTX.
3463 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3464 of "vector" where appropriate.
3465 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3466 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3467 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3468 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3469 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3470 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3471 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3472 (v_int_equiv): Extend to SVE modes.
3473 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3475 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3476 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3477 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3478 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3479 (SVE_COND_FP_CMP): New int iterators.
3480 (perm_hilo): Handle the new unpack unspecs.
3481 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3483 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3484 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3485 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3486 (aarch64_equality_operator, aarch64_constant_vector_operand)
3487 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3488 (aarch64_sve_nonimmediate_operand): Likewise.
3489 (aarch64_sve_general_operand): Likewise.
3490 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3491 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3492 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3493 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3494 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3495 (aarch64_sve_float_arith_immediate): Likewise.
3496 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3497 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3498 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3499 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3500 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3501 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3502 (aarch64_sve_float_arith_operand): Likewise.
3503 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3504 (aarch64_sve_float_mul_operand): Likewise.
3505 (aarch64_sve_vec_perm_operand): Likewise.
3506 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3507 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3508 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3509 as well as const_vector.
3510 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3511 in file. Use CONST0_RTX and CONSTM1_RTX.
3512 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3513 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3514 Use aarch64_simd_imm_zero.
3515 * config/aarch64/aarch64-sve.md: New file.
3516 * config/aarch64/aarch64.md: Include it.
3517 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3518 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3519 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3520 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3521 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3522 (sve): New attribute.
3523 (enabled): Disable instructions with the sve attribute unless
3525 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3526 aarch64_expand_mov_immediate.
3527 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3528 CNT[BHSD] immediates.
3529 (movti): Split CONST_POLY_INT moves into two halves.
3530 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3531 Split additions that need a temporary here if the destination
3532 is the stack pointer.
3533 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3534 (*add<mode>3_poly_1): New instruction.
3535 (set_clobber_cc): New expander.
3537 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3539 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3540 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3541 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3542 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3543 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3544 Change innermode from fixed_mode_size to machine_mode.
3545 (simplify_subreg): Update call accordingly. Handle a constant-sized
3546 subreg of a variable-length CONST_VECTOR.
3548 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3549 Alan Hayward <alan.hayward@arm.com>
3550 David Sherwood <david.sherwood@arm.com>
3552 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3553 (add_offset_to_base): New function, split out from...
3554 (create_mem_ref): ...here. When handling a scale other than 1,
3555 check first whether the address is valid without the offset.
3556 Add it into the base if so, leaving the index and scale as-is.
3558 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3561 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3562 fold_for_warn before checking if arg2 is INTEGER_CST.
3564 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3566 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3567 (store_multiple_operation): Delete.
3568 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3569 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3570 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3571 guarded by TARGET_STRING.
3572 (rs6000_output_load_multiple): Delete.
3573 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3574 OPTION_MASK_STRING / TARGET_STRING handling.
3575 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3576 (const rs6000_opt_masks) <"string">: Change mask to 0.
3577 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3578 (MASK_STRING): Delete.
3579 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3581 (load_multiple): Delete.
3588 (store_multiple): Delete.
3595 (movmemsi_8reg): Delete.
3596 (corresponding unnamed define_insn): Delete.
3597 (movmemsi_6reg): Delete.
3598 (corresponding unnamed define_insn): Delete.
3599 (movmemsi_4reg): Delete.
3600 (corresponding unnamed define_insn): Delete.
3601 (movmemsi_2reg): Delete.
3602 (corresponding unnamed define_insn): Delete.
3603 (movmemsi_1reg): Delete.
3604 (corresponding unnamed define_insn): Delete.
3605 * config/rs6000/rs6000.opt (mno-string): New.
3606 (mstring): Replace by deprecation warning stub.
3607 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3609 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3611 * regrename.c (regrename_do_replace): If replacing the same
3612 reg multiple times, try to reuse last created gen_raw_REG.
3615 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3616 main to workaround a bug in GDB.
3618 2018-01-12 Tom de Vries <tom@codesourcery.com>
3621 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3623 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3625 PR rtl-optimization/80481
3626 * ira-color.c (get_cap_member): New function.
3627 (allocnos_conflict_by_live_ranges_p): Use it.
3628 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3629 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3631 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3634 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3635 (*saddl_se_1): Ditto.
3637 (*ssubl_se_1): Ditto.
3639 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3641 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3642 rather than wi::to_widest for DR_INITs.
3643 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3644 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3645 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3647 (vect_analyze_group_access_1): Note that here.
3649 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3651 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3652 polynomial type sizes.
3654 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3656 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3657 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3658 (gimple_add_tmp_var): Likewise.
3660 2018-01-12 Martin Liska <mliska@suse.cz>
3662 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3663 (gimple_alloc_sizes): Likewise.
3664 (dump_gimple_statistics): Use PRIu64 in printf format.
3665 * gimple.h: Change uint64_t to int.
3667 2018-01-12 Martin Liska <mliska@suse.cz>
3669 * tree-core.h: Use uint64_t instead of int.
3670 * tree.c (tree_node_counts): Likewise.
3671 (tree_node_sizes): Likewise.
3672 (dump_tree_statistics): Use PRIu64 in printf format.
3674 2018-01-12 Martin Liska <mliska@suse.cz>
3676 * Makefile.in: As qsort_chk is implemented in vec.c, add
3677 vec.o to linkage of gencfn-macros.
3678 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3679 passing the info to record_node_allocation_statistics.
3680 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3682 * ggc-common.c (struct ggc_usage): Add operator== and use
3683 it in operator< and compare function.
3684 * mem-stats.h (struct mem_usage): Likewise.
3685 * vec.c (struct vec_usage): Remove operator< and compare
3686 function. Can be simply inherited.
3688 2018-01-12 Martin Jambor <mjambor@suse.cz>
3691 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3692 * tree-ssa-math-opts.c: Include domwalk.h.
3693 (convert_mult_to_fma_1): New function.
3694 (fma_transformation_info): New type.
3695 (fma_deferring_state): Likewise.
3696 (cancel_fma_deferring): New function.
3697 (result_of_phi): Likewise.
3698 (last_fma_candidate_feeds_initial_phi): Likewise.
3699 (convert_mult_to_fma): Added deferring logic, split actual
3700 transformation to convert_mult_to_fma_1.
3701 (math_opts_dom_walker): New type.
3702 (math_opts_dom_walker::after_dom_children): New method, body moved
3703 here from pass_optimize_widening_mul::execute, added deferring logic
3705 (pass_optimize_widening_mul::execute): Moved most of code to
3706 math_opts_dom_walker::after_dom_children.
3707 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3708 * config/i386/i386.c (ix86_option_override_internal): Added
3709 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3711 2018-01-12 Richard Biener <rguenther@suse.de>
3714 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3715 inline instance vars.
3717 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3720 * config/rx/rx.c (rx_is_restricted_memory_address):
3723 2018-01-12 Richard Biener <rguenther@suse.de>
3725 PR tree-optimization/80846
3726 * target.def (split_reduction): New target hook.
3727 * targhooks.c (default_split_reduction): New function.
3728 * targhooks.h (default_split_reduction): Declare.
3729 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3730 target requests first reduce vectors by combining low and high
3732 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3733 (get_vectype_for_scalar_type_and_size): Export.
3734 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3735 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3736 * doc/tm.texi: Regenerate.
3737 * config/i386/i386.c (ix86_split_reduction): Implement
3738 TARGET_VECTORIZE_SPLIT_REDUCTION.
3740 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3743 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3744 in PIC mode except for TARGET_VXWORKS_RTP.
3745 * config/sparc/sparc.c: Include cfgrtl.h.
3746 (TARGET_INIT_PIC_REG): Define.
3747 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3748 (sparc_pic_register_p): New predicate.
3749 (sparc_legitimate_address_p): Use it.
3750 (sparc_legitimize_pic_address): Likewise.
3751 (sparc_delegitimize_address): Likewise.
3752 (sparc_mode_dependent_address_p): Likewise.
3753 (gen_load_pcrel_sym): Remove 4th parameter.
3754 (load_got_register): Adjust call to above. Remove obsolete stuff.
3755 (sparc_expand_prologue): Do not call load_got_register here.
3756 (sparc_flat_expand_prologue): Likewise.
3757 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3758 (sparc_use_pseudo_pic_reg): New function.
3759 (sparc_init_pic_reg): Likewise.
3760 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3761 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3763 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3765 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3766 Add item for branch_cost.
3768 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3770 PR rtl-optimization/83565
3771 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3772 not extend the result to a larger mode for rotate operations.
3773 (num_sign_bit_copies1): Likewise.
3775 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3778 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3780 Use values-Xc.o for -pedantic.
3781 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3783 2018-01-12 Martin Liska <mliska@suse.cz>
3786 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3788 (possible_polymorphic_call_targets): Use it.
3789 (ipa_devirt): Likewise.
3791 2018-01-12 Martin Liska <mliska@suse.cz>
3793 * profile-count.h (enum profile_quality): Use 0 as invalid
3794 enum value of profile_quality.
3796 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3798 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3799 -mext-string options.
3801 2018-01-12 Richard Biener <rguenther@suse.de>
3803 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3804 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3805 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3807 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3809 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3811 * configure.ac (--with-long-double-format): Add support for the
3812 configuration option to change the default long double format on
3814 * config.gcc (powerpc*-linux*-*): Likewise.
3815 * configure: Regenerate.
3816 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3817 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3818 used without modification.
3820 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3822 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3823 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3824 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3825 MISC_BUILTIN_SPEC_BARRIER.
3826 (rs6000_init_builtins): Likewise.
3827 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3829 (speculation_barrier): New define_insn.
3830 * doc/extend.texi: Document __builtin_speculation_barrier.
3832 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3835 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3836 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3837 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3839 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3840 integral modes instead of "ss" and "sd".
3841 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3842 vectors with 32-bit and 64-bit elements.
3843 (vecdupssescalarmodesuffix): New mode attribute.
3844 (vec_dup<mode>): Use it.
3846 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3849 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3850 frame if argument is passed on stack.
3852 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3855 * ree.c (combine_reaching_defs): Optimize also
3856 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3857 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3859 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3862 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3864 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3867 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3868 after they are computed.
3870 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3872 PR tree-optimization/83695
3873 * gimple-loop-linterchange.cc
3874 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3875 reset cached scev information after interchange.
3876 (pass_linterchange::execute): Remove call to scev_reset_htab.
3878 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3880 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3881 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3882 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3883 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3884 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3885 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3886 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3887 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3888 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3889 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3890 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3891 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3892 (V_lane_reg): Likewise.
3893 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3895 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3896 (vfmal_lane_low<mode>_intrinsic,
3897 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3898 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3899 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3900 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3901 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3902 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3904 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3906 * config/arm/arm-cpus.in (fp16fml): New feature.
3907 (ALL_SIMD): Add fp16fml.
3908 (armv8.2-a): Add fp16fml as an option.
3909 (armv8.3-a): Likewise.
3910 (armv8.4-a): Add fp16fml as part of fp16.
3911 * config/arm/arm.h (TARGET_FP16FML): Define.
3912 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3914 * config/arm/arm-modes.def (V2HF): Define.
3915 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3916 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3917 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3918 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3919 vfmsl_low, vfmsl_high): New set of builtins.
3920 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3921 (vfml_op): New code attribute.
3922 (VFMLHALVES): New int iterator.
3923 (VFML, VFMLSEL): New mode attributes.
3924 (V_reg): Define mapping for V2HF.
3925 (V_hi, V_lo): New mode attributes.
3926 (VF_constraint): Likewise.
3927 (vfml_half, vfml_half_selector): New int attributes.
3928 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3930 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3931 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3933 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3934 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3935 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3936 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3938 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3939 Document new effective target and option set.
3941 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3943 * config/arm/arm-cpus.in (armv8_4): New feature.
3944 (ARMv8_4a): New fgroup.
3945 (armv8.4-a): New arch.
3946 * config/arm/arm-tables.opt: Regenerate.
3947 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3948 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3949 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3950 Add matching rules for -march=armv8.4-a and extensions.
3951 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3953 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3956 * config/rx/rx.md (BW): New mode attribute.
3957 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3959 2018-01-11 Richard Biener <rguenther@suse.de>
3961 PR tree-optimization/83435
3962 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3963 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3964 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3966 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3967 Alan Hayward <alan.hayward@arm.com>
3968 David Sherwood <david.sherwood@arm.com>
3970 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3972 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3973 (aarch64_print_address_internal): Use it to check for a zero offset.
3975 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3976 Alan Hayward <alan.hayward@arm.com>
3977 David Sherwood <david.sherwood@arm.com>
3979 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3980 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3981 Return a poly_int64 rather than a HOST_WIDE_INT.
3982 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3983 rather than a HOST_WIDE_INT.
3984 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3985 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3986 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3987 final_offset from HOST_WIDE_INT to poly_int64.
3988 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3989 to_constant when getting the number of units in an Advanced SIMD
3991 (aarch64_builtin_vectorized_function): Check for a constant number
3993 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3995 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3996 attribute instead of GET_MODE_NUNITS.
3997 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3998 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3999 GET_MODE_SIZE for fixed-size registers.
4000 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4001 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4002 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4003 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4004 (aarch64_print_operand, aarch64_print_address_internal)
4005 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4006 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4007 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4008 Handle polynomial GET_MODE_SIZE.
4009 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4010 wider than SImode without modification.
4011 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4012 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4013 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4014 passing and returning SVE modes.
4015 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4016 rather than GEN_INT.
4017 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4018 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4019 (aarch64_allocate_and_probe_stack_space): Likewise.
4020 (aarch64_layout_frame): Cope with polynomial offsets.
4021 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4022 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4024 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4025 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4026 poly_int64 rather than a HOST_WIDE_INT.
4027 (aarch64_get_separate_components, aarch64_process_components)
4028 (aarch64_expand_prologue, aarch64_expand_epilogue)
4029 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4030 (aarch64_anchor_offset): New function, split out from...
4031 (aarch64_legitimize_address): ...here.
4032 (aarch64_builtin_vectorization_cost): Handle polynomial
4033 TYPE_VECTOR_SUBPARTS.
4034 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4036 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4037 number of elements from the PARALLEL rather than the mode.
4038 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4039 rather than GET_MODE_BITSIZE.
4040 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4041 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4042 (aarch64_expand_vec_perm_const_1): Handle polynomial
4043 d->perm.length () and d->perm elements.
4044 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4045 Apply to_constant to d->perm elements.
4046 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4047 polynomial CONST_VECTOR_NUNITS.
4048 (aarch64_move_pointer): Take amount as a poly_int64 rather
4050 (aarch64_progress_pointer): Avoid temporary variable.
4051 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4052 the mode attribute instead of GET_MODE.
4054 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4055 Alan Hayward <alan.hayward@arm.com>
4056 David Sherwood <david.sherwood@arm.com>
4058 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4059 x exists before using it.
4060 (aarch64_add_constant_internal): Rename to...
4061 (aarch64_add_offset_1): ...this. Replace regnum with separate
4062 src and dest rtxes. Handle the case in which they're different,
4063 including when the offset is zero. Replace scratchreg with an rtx.
4064 Use 2 additions if there is no spare register into which we can
4065 move a 16-bit constant.
4066 (aarch64_add_constant): Delete.
4067 (aarch64_add_offset): Replace reg with separate src and dest
4068 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4069 Use aarch64_add_offset_1.
4070 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4071 an rtx rather than an int. Take the delta as a poly_int64
4072 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4073 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4074 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4075 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4076 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4078 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4079 aarch64_add_constant.
4081 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4083 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4084 Use scalar_float_mode.
4086 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4088 * config/aarch64/aarch64-simd.md
4089 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4090 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4091 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4092 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4093 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4094 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4095 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4096 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4097 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4098 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4100 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4103 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4104 targ_options->x_arm_arch_string is non NULL.
4106 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4108 * config/aarch64/aarch64.h
4109 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4111 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4114 * expmed.c (emit_store_flag_force): Swap if const op0
4115 and change VOIDmode to mode of op0.
4117 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4119 PR rtl-optimization/83761
4120 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4121 than bytes to mode_for_size.
4123 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4126 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4127 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4130 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4133 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4134 when in layout mode.
4135 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4136 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4139 2018-01-10 Michael Collison <michael.collison@arm.com>
4141 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4142 * config/aarch64/aarch64-option-extension.def: Add
4143 AARCH64_OPT_EXTENSION of 'fp16fml'.
4144 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4145 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4146 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4147 * config/aarch64/constraints.md (Ui7): New constraint.
4148 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4149 (VFMLA_SEL_W): Ditto.
4152 (VFMLA16_LOW): New int iterator.
4153 (VFMLA16_HIGH): Ditto.
4154 (UNSPEC_FMLAL): New unspec.
4155 (UNSPEC_FMLSL): Ditto.
4156 (UNSPEC_FMLAL2): Ditto.
4157 (UNSPEC_FMLSL2): Ditto.
4158 (f16mac): New code attribute.
4159 * config/aarch64/aarch64-simd-builtins.def
4160 (aarch64_fmlal_lowv2sf): Ditto.
4161 (aarch64_fmlsl_lowv2sf): Ditto.
4162 (aarch64_fmlalq_lowv4sf): Ditto.
4163 (aarch64_fmlslq_lowv4sf): Ditto.
4164 (aarch64_fmlal_highv2sf): Ditto.
4165 (aarch64_fmlsl_highv2sf): Ditto.
4166 (aarch64_fmlalq_highv4sf): Ditto.
4167 (aarch64_fmlslq_highv4sf): Ditto.
4168 (aarch64_fmlal_lane_lowv2sf): Ditto.
4169 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4170 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4171 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4172 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4173 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4174 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4175 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4176 (aarch64_fmlal_lane_highv2sf): Ditto.
4177 (aarch64_fmlsl_lane_highv2sf): Ditto.
4178 (aarch64_fmlal_laneq_highv2sf): Ditto.
4179 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4180 (aarch64_fmlalq_lane_highv4sf): Ditto.
4181 (aarch64_fmlsl_lane_highv4sf): Ditto.
4182 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4183 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4184 * config/aarch64/aarch64-simd.md:
4185 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4186 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4187 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4188 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4189 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4190 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4191 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4192 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4193 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4194 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4195 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4196 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4197 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4198 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4199 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4200 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4201 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4202 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4203 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4204 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4205 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4206 (vfmlsl_low_u32): Ditto.
4207 (vfmlalq_low_u32): Ditto.
4208 (vfmlslq_low_u32): Ditto.
4209 (vfmlal_high_u32): Ditto.
4210 (vfmlsl_high_u32): Ditto.
4211 (vfmlalq_high_u32): Ditto.
4212 (vfmlslq_high_u32): Ditto.
4213 (vfmlal_lane_low_u32): Ditto.
4214 (vfmlsl_lane_low_u32): Ditto.
4215 (vfmlal_laneq_low_u32): Ditto.
4216 (vfmlsl_laneq_low_u32): Ditto.
4217 (vfmlalq_lane_low_u32): Ditto.
4218 (vfmlslq_lane_low_u32): Ditto.
4219 (vfmlalq_laneq_low_u32): Ditto.
4220 (vfmlslq_laneq_low_u32): Ditto.
4221 (vfmlal_lane_high_u32): Ditto.
4222 (vfmlsl_lane_high_u32): Ditto.
4223 (vfmlal_laneq_high_u32): Ditto.
4224 (vfmlsl_laneq_high_u32): Ditto.
4225 (vfmlalq_lane_high_u32): Ditto.
4226 (vfmlslq_lane_high_u32): Ditto.
4227 (vfmlalq_laneq_high_u32): Ditto.
4228 (vfmlslq_laneq_high_u32): Ditto.
4229 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4230 (AARCH64_FL_FOR_ARCH8_4): New.
4231 (AARCH64_ISA_F16FML): New ISA flag.
4232 (TARGET_F16FML): New feature flag for fp16fml.
4233 (doc/invoke.texi): Document new fp16fml option.
4235 2018-01-10 Michael Collison <michael.collison@arm.com>
4237 * config/aarch64/aarch64-builtins.c:
4238 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4239 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4240 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4241 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4242 (AARCH64_ISA_SHA3): New ISA flag.
4243 (TARGET_SHA3): New feature flag for sha3.
4244 * config/aarch64/iterators.md (sha512_op): New int attribute.
4245 (CRYPTO_SHA512): New int iterator.
4246 (UNSPEC_SHA512H): New unspec.
4247 (UNSPEC_SHA512H2): Ditto.
4248 (UNSPEC_SHA512SU0): Ditto.
4249 (UNSPEC_SHA512SU1): Ditto.
4250 * config/aarch64/aarch64-simd-builtins.def
4251 (aarch64_crypto_sha512hqv2di): New builtin.
4252 (aarch64_crypto_sha512h2qv2di): Ditto.
4253 (aarch64_crypto_sha512su0qv2di): Ditto.
4254 (aarch64_crypto_sha512su1qv2di): Ditto.
4255 (aarch64_eor3qv8hi): Ditto.
4256 (aarch64_rax1qv2di): Ditto.
4257 (aarch64_xarqv2di): Ditto.
4258 (aarch64_bcaxqv8hi): Ditto.
4259 * config/aarch64/aarch64-simd.md:
4260 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4261 (aarch64_crypto_sha512su0qv2di): Ditto.
4262 (aarch64_crypto_sha512su1qv2di): Ditto.
4263 (aarch64_eor3qv8hi): Ditto.
4264 (aarch64_rax1qv2di): Ditto.
4265 (aarch64_xarqv2di): Ditto.
4266 (aarch64_bcaxqv8hi): Ditto.
4267 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4268 (vsha512h2q_u64): Ditto.
4269 (vsha512su0q_u64): Ditto.
4270 (vsha512su1q_u64): Ditto.
4271 (veor3q_u16): Ditto.
4272 (vrax1q_u64): Ditto.
4274 (vbcaxq_u16): Ditto.
4275 * config/arm/types.md (crypto_sha512): New type attribute.
4276 (crypto_sha3): Ditto.
4277 (doc/invoke.texi): Document new sha3 option.
4279 2018-01-10 Michael Collison <michael.collison@arm.com>
4281 * config/aarch64/aarch64-builtins.c:
4282 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4283 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4284 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4285 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4286 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4287 (AARCH64_ISA_SM4): New ISA flag.
4288 (TARGET_SM4): New feature flag for sm4.
4289 * config/aarch64/aarch64-simd-builtins.def
4290 (aarch64_sm3ss1qv4si): Ditto.
4291 (aarch64_sm3tt1aq4si): Ditto.
4292 (aarch64_sm3tt1bq4si): Ditto.
4293 (aarch64_sm3tt2aq4si): Ditto.
4294 (aarch64_sm3tt2bq4si): Ditto.
4295 (aarch64_sm3partw1qv4si): Ditto.
4296 (aarch64_sm3partw2qv4si): Ditto.
4297 (aarch64_sm4eqv4si): Ditto.
4298 (aarch64_sm4ekeyqv4si): Ditto.
4299 * config/aarch64/aarch64-simd.md:
4300 (aarch64_sm3ss1qv4si): Ditto.
4301 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4302 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4303 (aarch64_sm4eqv4si): Ditto.
4304 (aarch64_sm4ekeyqv4si): Ditto.
4305 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4306 (sm3part_op): Ditto.
4307 (CRYPTO_SM3TT): Ditto.
4308 (CRYPTO_SM3PART): Ditto.
4309 (UNSPEC_SM3SS1): New unspec.
4310 (UNSPEC_SM3TT1A): Ditto.
4311 (UNSPEC_SM3TT1B): Ditto.
4312 (UNSPEC_SM3TT2A): Ditto.
4313 (UNSPEC_SM3TT2B): Ditto.
4314 (UNSPEC_SM3PARTW1): Ditto.
4315 (UNSPEC_SM3PARTW2): Ditto.
4316 (UNSPEC_SM4E): Ditto.
4317 (UNSPEC_SM4EKEY): Ditto.
4318 * config/aarch64/constraints.md (Ui2): New constraint.
4319 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4320 * config/arm/types.md (crypto_sm3): New type attribute.
4321 (crypto_sm4): Ditto.
4322 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4323 (vsm3tt1aq_u32): Ditto.
4324 (vsm3tt1bq_u32): Ditto.
4325 (vsm3tt2aq_u32): Ditto.
4326 (vsm3tt2bq_u32): Ditto.
4327 (vsm3partw1q_u32): Ditto.
4328 (vsm3partw2q_u32): Ditto.
4329 (vsm4eq_u32): Ditto.
4330 (vsm4ekeyq_u32): Ditto.
4331 (doc/invoke.texi): Document new sm4 option.
4333 2018-01-10 Michael Collison <michael.collison@arm.com>
4335 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4336 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4337 (AARCH64_FL_FOR_ARCH8_4): New.
4338 (AARCH64_FL_V8_4): New flag.
4339 (doc/invoke.texi): Document new armv8.4-a option.
4341 2018-01-10 Michael Collison <michael.collison@arm.com>
4343 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4344 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4345 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4346 * config/aarch64/aarch64-option-extension.def: Add
4347 AARCH64_OPT_EXTENSION of 'sha2'.
4348 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4349 (crypto): Disable sha2 and aes if crypto disabled.
4350 (crypto): Enable aes and sha2 if enabled.
4351 (simd): Disable sha2 and aes if simd disabled.
4352 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4354 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4355 (TARGET_SHA2): New feature flag for sha2.
4356 (TARGET_AES): New feature flag for aes.
4357 * config/aarch64/aarch64-simd.md:
4358 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4359 conditional on TARGET_AES.
4360 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4361 (aarch64_crypto_sha1hsi): Make pattern conditional
4363 (aarch64_crypto_sha1hv4si): Ditto.
4364 (aarch64_be_crypto_sha1hv4si): Ditto.
4365 (aarch64_crypto_sha1su1v4si): Ditto.
4366 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4367 (aarch64_crypto_sha1su0v4si): Ditto.
4368 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4369 (aarch64_crypto_sha256su0v4si): Ditto.
4370 (aarch64_crypto_sha256su1v4si): Ditto.
4371 (doc/invoke.texi): Document new aes and sha2 options.
4373 2018-01-10 Martin Sebor <msebor@redhat.com>
4375 PR tree-optimization/83781
4376 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4379 2018-01-11 Martin Sebor <msebor@gmail.com>
4380 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4382 PR tree-optimization/83501
4383 PR tree-optimization/81703
4385 * tree-ssa-strlen.c (get_string_cst): Rename...
4386 (get_string_len): ...to this. Handle global constants.
4387 (handle_char_store): Adjust.
4389 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4390 Jim Wilson <jimw@sifive.com>
4392 * config/riscv/riscv-protos.h (riscv_output_return): New.
4393 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4394 (riscv_attribute_table, riscv_output_return),
4395 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4396 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4397 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4399 (riscv_expand_prologue): Add early return for naked function.
4400 (riscv_expand_epilogue): Likewise.
4401 (riscv_function_ok_for_sibcall): Return false for naked function.
4402 (riscv_set_current_function): New.
4403 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4404 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4405 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4406 * doc/extend.texi (RISC-V Function Attributes): New.
4408 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4410 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4411 check for 128-bit long double before checking TCmode.
4412 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4413 128-bit long doubles before checking TFmode or TCmode.
4414 (FLOAT128_IBM_P): Likewise.
4416 2018-01-10 Martin Sebor <msebor@redhat.com>
4418 PR tree-optimization/83671
4419 * builtins.c (c_strlen): Unconditionally return zero for the empty
4421 Use -Warray-bounds for warnings.
4422 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4423 for non-constant array indices with COMPONENT_REF, arrays of
4424 arrays, and pointers to arrays.
4425 (gimple_fold_builtin_strlen): Determine and set length range for
4426 non-constant character arrays.
4428 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4431 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4434 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4436 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4438 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4441 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4442 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4443 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4444 indexed_or_indirect_operand predicate.
4445 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4446 (*vsx_le_perm_load_v8hi): Likewise.
4447 (*vsx_le_perm_load_v16qi): Likewise.
4448 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4449 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4450 (*vsx_le_perm_store_v8hi): Likewise.
4451 (*vsx_le_perm_store_v16qi): Likewise.
4452 (eight unnamed splitters): Likewise.
4454 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4456 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4457 * config/rs6000/emmintrin.h: Likewise.
4458 * config/rs6000/mmintrin.h: Likewise.
4459 * config/rs6000/xmmintrin.h: Likewise.
4461 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4464 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4466 * tree.c (tree_nop_conversion): Return true for location wrapper
4468 (maybe_wrap_with_location): New function.
4469 (selftest::check_strip_nops): New function.
4470 (selftest::test_location_wrappers): New function.
4471 (selftest::tree_c_tests): Call it.
4472 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4473 (maybe_wrap_with_location): New decl.
4474 (EXPR_LOCATION_WRAPPER_P): New macro.
4475 (location_wrapper_p): New inline function.
4476 (tree_strip_any_location_wrapper): New inline function.
4478 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4481 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4482 stack_realign_offset for the largest alignment of stack slot
4484 (ix86_find_max_used_stack_alignment): New function.
4485 (ix86_finalize_stack_frame_flags): Use it. Set
4486 max_used_stack_alignment if we don't realign stack.
4487 * config/i386/i386.h (machine_function): Add
4488 max_used_stack_alignment.
4490 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4492 * config/arm/arm.opt (-mbranch-cost): New option.
4493 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4496 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4499 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4500 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4502 2018-01-10 Richard Biener <rguenther@suse.de>
4505 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4506 early out so it also covers the case where we have a non-NULL
4509 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4511 PR tree-optimization/83753
4512 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4513 for non-strided grouped accesses if the number of elements is 1.
4515 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4518 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4519 * i386.h (TARGET_USE_GATHER): Define.
4520 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4522 2018-01-10 Martin Liska <mliska@suse.cz>
4525 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4526 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4528 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4529 CLEANUP_NO_PARTITIONING is not set.
4531 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4533 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4534 for vectors, as a partial revert of r254296.
4535 * rtl.h (const_vec_p): Delete.
4536 (const_vec_duplicate_p): Don't test for vector CONSTs.
4537 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4538 * expmed.c (make_tree): Likewise.
4541 * common.md (E, F): Use CONSTANT_P instead of checking for
4543 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4544 checking for CONST_VECTOR.
4546 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4549 * predict.c (force_edge_cold): Handle in more sane way edges
4552 2018-01-09 Carl Love <cel@us.ibm.com>
4554 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4556 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4557 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4558 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4559 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4560 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4561 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4562 * config/rs6000/rs6000-protos.h: Add extern defition for
4563 rs6000_generate_float2_double_code.
4564 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4566 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4567 (float2_v2df): Add define_expand.
4569 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4572 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4573 op_mode in the force_to_mode call.
4575 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4577 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4578 instead of checking each element individually.
4579 (aarch64_evpc_uzp): Likewise.
4580 (aarch64_evpc_zip): Likewise.
4581 (aarch64_evpc_ext): Likewise.
4582 (aarch64_evpc_rev): Likewise.
4583 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4584 instead of checking each element individually. Return true without
4586 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4587 whether all selected elements come from the same input, instead of
4588 checking each element individually. Remove calls to gen_rtx_REG,
4589 start_sequence and end_sequence and instead assert that no rtl is
4592 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4594 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4595 order of HIGH and CONST checks.
4597 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4599 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4600 if the destination isn't an SSA_NAME.
4602 2018-01-09 Richard Biener <rguenther@suse.de>
4604 PR tree-optimization/83668
4605 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4607 (canonicalize_loop_form): ... here, renamed from ...
4608 (canonicalize_loop_closed_ssa_form): ... this and amended to
4609 swap successor edges for loop exit blocks to make us use
4610 the RPO order we need for initial schedule generation.
4612 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4614 PR tree-optimization/64811
4615 * match.pd: When optimizing comparisons with Inf, avoid
4616 introducing or losing exceptions from comparisons with NaN.
4618 2018-01-09 Martin Liska <mliska@suse.cz>
4621 * asan.c (shadow_mem_size): Add gcc_assert.
4623 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4625 Don't save registers in main().
4628 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4629 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4630 * config/avr/avr.c (avr_set_current_function): Don't error if
4631 naked, OS_task or OS_main are specified at the same time.
4632 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4634 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4636 * common/config/avr/avr-common.c (avr_option_optimization_table):
4637 Switch on -mmain-is-OS_task for optimizing compilations.
4639 2018-01-09 Richard Biener <rguenther@suse.de>
4641 PR tree-optimization/83572
4642 * graphite.c: Include cfganal.h.
4643 (graphite_transform_loops): Connect infinite loops to exit
4644 and remove fake edges at the end.
4646 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4648 * ipa-inline.c (edge_badness): Revert accidental checkin.
4650 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4653 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4654 symbols; not inline clones.
4656 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4659 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4660 hard registers. Formatting fixes.
4662 PR preprocessor/83722
4663 * gcc.c (try_generate_repro): Pass
4664 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4665 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4668 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4669 Kito Cheng <kito.cheng@gmail.com>
4671 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4672 (riscv_leaf_function_p): Delete.
4673 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4675 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4677 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4679 (do_ifelse): New function.
4680 (do_isel): New function.
4681 (do_sub3): New function.
4682 (do_add3): New function.
4683 (do_load_mask_compare): New function.
4684 (do_overlap_load_compare): New function.
4685 (expand_compare_loop): New function.
4686 (expand_block_compare): Call expand_compare_loop() when appropriate.
4687 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4689 (-mblock-compare-inline-loop-limit): New option.
4691 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4694 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4695 Reverse order of second and third operands in first alternative.
4696 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4697 of first and second elements in UNSPEC_VPERMR vector.
4698 (altivec_expand_vec_perm_le): Likewise.
4700 2018-01-08 Jeff Law <law@redhat.com>
4702 PR rtl-optimizatin/81308
4703 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4704 (process_switch): If group_case_labels makes a change, then set
4706 (pass_convert_switch::execute): If a switch is converted, then
4707 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4709 PR rtl-optimization/81308
4710 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4713 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4715 PR target/83663 - Revert r255946
4716 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4717 generation for cases where splatting a value is not useful.
4718 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4719 across a vec_duplicate and a paradoxical subreg forming a vector
4720 mode to a vec_concat.
4722 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4724 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4725 -march=armv8.3-a variants.
4726 * config/arm/t-multilib: Likewise.
4727 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4729 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4731 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4733 (cceq_ior_compare_complement): Give it a name so I can use it, and
4734 change boolean_or_operator predicate to boolean_operator so it can
4735 be used to generate a crand.
4736 (eqne): New code iterator.
4737 (bd/bd_neg): New code_attrs.
4738 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4739 a single define_insn.
4740 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4741 decrement (bdnzt/bdnzf/bdzt/bdzf).
4742 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4743 with the new names of the branch decrement patterns, and added the
4744 names of the branch decrement conditional patterns.
4746 2018-01-08 Richard Biener <rguenther@suse.de>
4748 PR tree-optimization/83563
4749 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4752 2018-01-08 Richard Biener <rguenther@suse.de>
4755 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4757 2018-01-08 Richard Biener <rguenther@suse.de>
4759 PR tree-optimization/83685
4760 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4761 references to abnormals.
4763 2018-01-08 Richard Biener <rguenther@suse.de>
4766 * dwarf2out.c (output_indirect_strings): Handle empty
4767 skeleton_debug_str_hash.
4768 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4770 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4772 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4773 (emit_store_direct): Likewise.
4774 (arc_trampoline_adjust_address): Likewise.
4775 (arc_asm_trampoline_template): New function.
4776 (arc_initialize_trampoline): Use asm_trampoline_template.
4777 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4778 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4779 * config/arc/arc.md (flush_icache): Delete pattern.
4781 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4783 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4784 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4787 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4790 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4791 by not USED_FOR_TARGET.
4792 (make_pass_resolve_sw_modes): Likewise.
4794 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4796 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4799 2018-01-08 Richard Biener <rguenther@suse.de>
4802 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4804 2018-01-08 Richard Biener <rguenther@suse.de>
4807 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4809 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4812 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4813 basic blocks with a small number of successors.
4814 (convert_control_dep_chain_into_preds): Improve handling of
4816 (dump_predicates): Split apart into...
4817 (dump_pred_chain): ...here...
4818 (dump_pred_info): ...and here.
4819 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4820 (can_chain_union_be_invalidated_p): Improve check for invalidation
4822 (uninit_uses_cannot_happen): Avoid unnecessary if
4823 convert_control_dep_chain_into_preds yielded nothing.
4825 2018-01-06 Martin Sebor <msebor@redhat.com>
4827 PR tree-optimization/83640
4828 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4829 subtracting negative offset from size.
4830 (builtin_access::overlap): Adjust offset bounds of the access to fall
4831 within the size of the object if possible.
4833 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4835 PR rtl-optimization/83699
4836 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4837 extract_bit_field_as_subreg to cases in which the extracted
4838 value is also a vector.
4840 * lra-constraints.c (process_alt_operands): Test for the equivalence
4841 substitutions when detecting a possible reload cycle.
4843 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4846 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4847 by default if flag_selective_schedling{,2}. Formatting fixes.
4849 PR rtl-optimization/83682
4850 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4851 if it has non-VECTOR_MODE element mode.
4852 (vec_duplicate_p): Likewise.
4855 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4856 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4858 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4861 * config/i386/i386-builtin.def
4862 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4863 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4864 Require also OPTION_MASK_ISA_AVX512F in addition to
4865 OPTION_MASK_ISA_GFNI.
4866 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4867 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4868 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4869 to OPTION_MASK_ISA_GFNI.
4870 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4871 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4872 OPTION_MASK_ISA_AVX512BW.
4873 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4874 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4875 addition to OPTION_MASK_ISA_GFNI.
4876 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4877 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4878 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4879 to OPTION_MASK_ISA_GFNI.
4880 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4881 a requirement for all ISAs rather than any of them with a few
4883 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4885 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4886 bitmasks to be enabled with 3 exceptions, instead of requiring any
4887 enabled ISA with lots of exceptions.
4888 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4889 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4890 Change avx512bw in isa attribute to avx512f.
4891 * config/i386/sgxintrin.h: Add license boilerplate.
4892 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4893 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4894 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4895 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4897 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4898 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4899 temporarily sse2 rather than sse if not enabled already.
4902 * config/i386/sse.md (VI248_VLBW): Rename to ...
4903 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4904 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4905 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4906 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4907 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4908 mode iterator instead of VI248_VLBW.
4910 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4912 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4913 (record_modified): Skip clobbers; add debug output.
4914 (param_change_prob): Use sreal frequencies.
4916 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4918 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4919 punt for user-aligned variables.
4921 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4923 * tree-chrec.c (chrec_contains_symbols): Return true for
4926 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4929 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4930 of (x|y) == x for BICS pattern.
4932 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4934 PR tree-optimization/83605
4935 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4936 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4939 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4941 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4942 * config/epiphany/rtems.h: New file.
4944 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4945 Uros Bizjak <ubizjak@gmail.com>
4948 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4949 QIreg_operand instead of register_operand predicate.
4950 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4951 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4952 comments instead of -fmitigate[-_]rop.
4954 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4957 * cgraphunit.c (symbol_table::compile): Switch to text_section
4958 before calling assembly_start debug hook.
4959 * run-rtl-passes.c (run_rtl_passes): Likewise.
4962 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4964 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4965 range_int_cst_p rather than !symbolic_range_p before calling
4966 extract_range_from_multiplicative_op_1.
4968 2018-01-04 Jeff Law <law@redhat.com>
4970 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4971 redundant test in assertion.
4973 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4975 * doc/rtl.texi: Document machine_mode wrapper classes.
4977 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4979 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4982 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4984 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4985 the VEC_PERM_EXPR fold to fail.
4987 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4990 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4991 to switched_sections.
4993 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4996 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4999 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5002 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5003 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5005 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5008 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5009 is BLKmode and bitpos not zero or mode change is needed.
5011 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5014 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5017 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5020 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5021 instead of MULT rtx. Update all corresponding splitters.
5023 (*ssub<modesuffix>): Ditto.
5025 (*cmp_sadd_di): Update split patterns.
5026 (*cmp_sadd_si): Ditto.
5027 (*cmp_sadd_sidi): Ditto.
5028 (*cmp_ssub_di): Ditto.
5029 (*cmp_ssub_si): Ditto.
5030 (*cmp_ssub_sidi): Ditto.
5031 * config/alpha/predicates.md (const23_operand): New predicate.
5032 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5033 Look for ASHIFT, not MULT inner operand.
5034 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5036 2018-01-04 Martin Liska <mliska@suse.cz>
5038 PR gcov-profile/83669
5039 * gcov.c (output_intermediate_file): Add version to intermediate
5041 * doc/gcov.texi: Document new field 'version' in intermediate
5042 file format. Fix location of '-k' option of gcov command.
5044 2018-01-04 Martin Liska <mliska@suse.cz>
5047 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5049 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5051 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5053 2018-01-03 Martin Sebor <msebor@redhat.com>
5055 PR tree-optimization/83655
5056 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5057 checking calls with invalid arguments.
5059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5061 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5062 (vectorizable_mask_load_store): Delete.
5063 (vectorizable_call): Return false for masked loads and stores.
5064 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5065 instead of gimple_assign_rhs1.
5066 (vectorizable_load): Handle IFN_MASK_LOAD.
5067 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5071 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5073 (vectorizable_mask_load_store): ...here.
5074 (vectorizable_load): ...and here.
5076 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5078 * tree-vect-stmts.c (vect_build_all_ones_mask)
5079 (vect_build_zero_merge_argument): New functions, split out from...
5080 (vectorizable_load): ...here.
5082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5084 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5086 (vectorizable_mask_load_store): ...here.
5087 (vectorizable_store): ...and here.
5089 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5091 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5093 (vectorizable_mask_load_store): ...here.
5095 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5097 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5098 (vect_model_store_cost): Take a vec_load_store_type instead of a
5100 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5101 (vect_model_store_cost): Take a vec_load_store_type instead of a
5103 (vectorizable_mask_load_store): Update accordingly.
5104 (vectorizable_store): Likewise.
5105 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5109 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5110 IFN_MASK_LOAD calls here rather than...
5111 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5114 Alan Hayward <alan.hayward@arm.com>
5115 David Sherwood <david.sherwood@arm.com>
5117 * expmed.c (extract_bit_field_1): For vector extracts,
5118 fall back to extract_bit_field_as_subreg if vec_extract
5121 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5122 Alan Hayward <alan.hayward@arm.com>
5123 David Sherwood <david.sherwood@arm.com>
5125 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5126 they are variable or constant sized.
5127 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5128 slots for constant-sized data.
5130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5131 Alan Hayward <alan.hayward@arm.com>
5132 David Sherwood <david.sherwood@arm.com>
5134 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5135 handling COND_EXPRs with boolean comparisons, try to find a better
5136 basis for the mask type than the boolean itself.
5138 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5140 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5141 is calculated and how it can be overridden.
5142 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5143 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5145 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5149 Alan Hayward <alan.hayward@arm.com>
5150 David Sherwood <david.sherwood@arm.com>
5152 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5153 Remove the mode argument.
5154 (aarch64_simd_valid_immediate): Remove the mode and inverse
5156 * config/aarch64/iterators.md (bitsize): New iterator.
5157 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5158 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5159 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5160 aarch64_simd_valid_immediate.
5161 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5162 (aarch64_reg_or_bic_imm): Likewise.
5163 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5164 with an insn_type enum and msl with a modifier_type enum.
5165 Replace element_width with a scalar_mode. Change the shift
5166 to unsigned int. Add constructors for scalar_float_mode and
5167 scalar_int_mode elements.
5168 (aarch64_vect_float_const_representable_p): Delete.
5169 (aarch64_can_const_movi_rtx_p)
5170 (aarch64_simd_scalar_immediate_valid_for_move)
5171 (aarch64_simd_make_constant): Update call to
5172 aarch64_simd_valid_immediate.
5173 (aarch64_advsimd_valid_immediate_hs): New function.
5174 (aarch64_advsimd_valid_immediate): Likewise.
5175 (aarch64_simd_valid_immediate): Remove mode and inverse
5176 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5177 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5178 and aarch64_float_const_representable_p on the result.
5179 (aarch64_output_simd_mov_immediate): Remove mode argument.
5180 Update call to aarch64_simd_valid_immediate and use of
5181 simd_immediate_info.
5182 (aarch64_output_scalar_simd_mov_immediate): Update call
5185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5186 Alan Hayward <alan.hayward@arm.com>
5187 David Sherwood <david.sherwood@arm.com>
5189 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5190 (mode_nunits): Likewise CONST_MODE_NUNITS.
5191 * machmode.def (ADJUST_NUNITS): Document.
5192 * genmodes.c (mode_data::need_nunits_adj): New field.
5193 (blank_mode): Update accordingly.
5194 (adj_nunits): New variable.
5195 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5197 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5198 listed in adj_nunits.
5199 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5200 listed in adj_nunits. Don't emit case statements for such modes.
5201 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5202 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5203 nothing if adj_nunits is nonnull.
5204 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5205 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5206 (emit_mode_fbit): Update use of print_maybe_const_decl.
5207 (emit_move_size): Likewise. Treat the array as non-const
5209 (emit_mode_adjustments): Handle adj_nunits.
5211 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5213 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5214 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5215 (VECTOR_MODES): Use it.
5216 (make_vector_modes): Take the prefix as an argument.
5218 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5219 Alan Hayward <alan.hayward@arm.com>
5220 David Sherwood <david.sherwood@arm.com>
5222 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5223 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5224 for MODE_VECTOR_BOOL.
5225 * machmode.def (VECTOR_BOOL_MODE): Document.
5226 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5227 (make_vector_bool_mode): New function.
5228 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5230 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5231 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5233 * stor-layout.c (int_mode_for_mode): Likewise.
5234 * tree.c (build_vector_type_for_mode): Likewise.
5235 * varasm.c (output_constant_pool_2): Likewise.
5236 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5237 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5238 for MODE_VECTOR_BOOL.
5239 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5240 of mode class checks.
5241 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5242 instead of a list of mode class checks.
5243 (expand_vector_scalar_condition): Likewise.
5244 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5246 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5247 Alan Hayward <alan.hayward@arm.com>
5248 David Sherwood <david.sherwood@arm.com>
5250 * machmode.h (mode_size): Change from unsigned short to
5252 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5253 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5254 or if measurement_type is not polynomial.
5255 (fixed_size_mode::includes_p): Check for constant-sized modes.
5256 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5257 return a poly_uint16 rather than an unsigned short.
5258 (emit_mode_size): Change the type of mode_size from unsigned short
5259 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5260 (emit_mode_adjustments): Cope with polynomial vector sizes.
5261 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5263 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5265 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5266 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5267 * caller-save.c (setup_save_areas): Likewise.
5268 (replace_reg_with_saved_mem): Likewise.
5269 * calls.c (emit_library_call_value_1): Likewise.
5270 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5271 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5272 (gen_lowpart_for_combine): Likewise.
5273 * convert.c (convert_to_integer_1): Likewise.
5274 * cse.c (equiv_constant, cse_insn): Likewise.
5275 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5276 (cselib_subst_to_values): Likewise.
5277 * dce.c (word_dce_process_block): Likewise.
5278 * df-problems.c (df_word_lr_mark_ref): Likewise.
5279 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5280 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5281 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5282 (rtl_for_decl_location): Likewise.
5283 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5284 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5285 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5286 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5287 (expand_expr_real_1): Likewise.
5288 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5289 (pad_below): Likewise.
5290 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5291 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5292 * ira.c (get_subreg_tracking_sizes): Likewise.
5293 * ira-build.c (ira_create_allocno_objects): Likewise.
5294 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5295 (ira_sort_regnos_for_alter_reg): Likewise.
5296 * ira-costs.c (record_operand_costs): Likewise.
5297 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5298 (resolve_simple_move): Likewise.
5299 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5300 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5301 (lra_constraints): Likewise.
5302 (CONST_POOL_OK_P): Reject variable-sized modes.
5303 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5304 (add_pseudo_to_slot, lra_spill): Likewise.
5305 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5306 * optabs-query.c (get_best_extraction_insn): Likewise.
5307 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5308 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5309 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5310 * recog.c (offsettable_address_addr_space_p): Likewise.
5311 * regcprop.c (maybe_mode_change): Likewise.
5312 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5313 * regrename.c (build_def_use): Likewise.
5314 * regstat.c (dump_reg_info): Likewise.
5315 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5316 (find_reloads, find_reloads_subreg_address): Likewise.
5317 * reload1.c (eliminate_regs_1): Likewise.
5318 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5319 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5320 (simplify_binary_operation_1, simplify_subreg): Likewise.
5321 * targhooks.c (default_function_arg_padding): Likewise.
5322 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5323 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5324 (verify_gimple_assign_ternary): Likewise.
5325 * tree-inline.c (estimate_move_cost): Likewise.
5326 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5327 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5328 (get_address_cost_ainc): Likewise.
5329 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5330 (vect_supportable_dr_alignment): Likewise.
5331 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5332 (vectorizable_reduction): Likewise.
5333 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5334 (vectorizable_operation, vectorizable_load): Likewise.
5335 * tree.c (build_same_sized_truth_vector_type): Likewise.
5336 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5337 * var-tracking.c (emit_note_insn_var_location): Likewise.
5338 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5339 (ADDR_VEC_ALIGN): Likewise.
5341 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5342 Alan Hayward <alan.hayward@arm.com>
5343 David Sherwood <david.sherwood@arm.com>
5345 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5347 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5348 or if measurement_type is polynomial.
5349 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5350 * combine.c (make_extraction): Likewise.
5351 * dse.c (find_shift_sequence): Likewise.
5352 * dwarf2out.c (mem_loc_descriptor): Likewise.
5353 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5354 (extract_bit_field, extract_low_bits): Likewise.
5355 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5356 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5357 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5358 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5359 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5360 * reload.c (find_reloads): Likewise.
5361 * reload1.c (alter_reg): Likewise.
5362 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5363 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5364 * tree-if-conv.c (predicate_mem_writes): Likewise.
5365 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5366 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5367 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5368 * valtrack.c (dead_debug_insert_temp): Likewise.
5369 * varasm.c (mergeable_constant_section): Likewise.
5370 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5372 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5373 Alan Hayward <alan.hayward@arm.com>
5374 David Sherwood <david.sherwood@arm.com>
5376 * expr.c (expand_assignment): Cope with polynomial mode sizes
5377 when assigning to a CONCAT.
5379 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5380 Alan Hayward <alan.hayward@arm.com>
5381 David Sherwood <david.sherwood@arm.com>
5383 * machmode.h (mode_precision): Change from unsigned short to
5385 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5387 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5388 or if measurement_type is not polynomial.
5389 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5390 in which the mode is already known to be a scalar_int_mode.
5391 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5392 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5394 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5395 for GET_MODE_PRECISION.
5396 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5397 for GET_MODE_PRECISION.
5398 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5400 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5401 (expand_field_assignment, make_extraction): Likewise.
5402 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5403 (get_last_value): Likewise.
5404 * convert.c (convert_to_integer_1): Likewise.
5405 * cse.c (cse_insn): Likewise.
5406 * expr.c (expand_expr_real_1): Likewise.
5407 * lra-constraints.c (simplify_operand_subreg): Likewise.
5408 * optabs-query.c (can_atomic_load_p): Likewise.
5409 * optabs.c (expand_atomic_load): Likewise.
5410 (expand_atomic_store): Likewise.
5411 * ree.c (combine_reaching_defs): Likewise.
5412 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5413 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5414 * tree.h (type_has_mode_precision_p): Likewise.
5415 * ubsan.c (instrument_si_overflow): Likewise.
5417 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5418 Alan Hayward <alan.hayward@arm.com>
5419 David Sherwood <david.sherwood@arm.com>
5421 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5422 polynomial numbers of units.
5423 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5424 (valid_vector_subparts_p): New function.
5425 (build_vector_type): Remove temporary shim and take the number
5426 of units as a poly_uint64 rather than an int.
5427 (build_opaque_vector_type): Take the number of units as a
5428 poly_uint64 rather than an int.
5429 * tree.c (build_vector_from_ctor): Handle polynomial
5430 TYPE_VECTOR_SUBPARTS.
5431 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5432 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5433 (build_vector_from_val): If the number of units is variable,
5434 use build_vec_duplicate_cst for constant operands and
5435 VEC_DUPLICATE_EXPR otherwise.
5436 (make_vector_type): Remove temporary is_constant ().
5437 (build_vector_type, build_opaque_vector_type): Take the number of
5438 units as a poly_uint64 rather than an int.
5439 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5441 * cfgexpand.c (expand_debug_expr): Likewise.
5442 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5443 (store_constructor, expand_expr_real_1): Likewise.
5444 (const_scalar_mask_from_tree): Likewise.
5445 * fold-const-call.c (fold_const_reduction): Likewise.
5446 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5447 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5448 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5449 (fold_relational_const): Likewise.
5450 (native_interpret_vector): Likewise. Change the size from an
5451 int to an unsigned int.
5452 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5453 TYPE_VECTOR_SUBPARTS.
5454 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5455 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5456 duplicating a non-constant operand into a variable-length vector.
5457 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5458 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5459 * ipa-icf.c (sem_variable::equals): Likewise.
5460 * match.pd: Likewise.
5461 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5462 * print-tree.c (print_node): Likewise.
5463 * stor-layout.c (layout_type): Likewise.
5464 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5465 * tree-cfg.c (verify_gimple_comparison): Likewise.
5466 (verify_gimple_assign_binary): Likewise.
5467 (verify_gimple_assign_ternary): Likewise.
5468 (verify_gimple_assign_single): Likewise.
5469 * tree-pretty-print.c (dump_generic_node): Likewise.
5470 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5471 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5472 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5473 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5474 (vect_shift_permute_load_chain): Likewise.
5475 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5476 (expand_vector_condition, optimize_vector_constructor): Likewise.
5477 (lower_vec_perm, get_compute_type): Likewise.
5478 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5479 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5480 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5481 (vect_recog_mask_conversion_pattern): Likewise.
5482 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5483 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5484 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5485 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5486 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5487 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5488 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5489 (supportable_widening_operation): Likewise.
5490 (supportable_narrowing_operation): Likewise.
5491 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5493 * varasm.c (output_constant): Likewise.
5495 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5496 Alan Hayward <alan.hayward@arm.com>
5497 David Sherwood <david.sherwood@arm.com>
5499 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5500 so that both the length == 3 and length != 3 cases set up their
5501 own permute vectors. Add comments explaining why we know the
5502 number of elements is constant.
5503 (vect_permute_load_chain): Likewise.
5505 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5506 Alan Hayward <alan.hayward@arm.com>
5507 David Sherwood <david.sherwood@arm.com>
5509 * machmode.h (mode_nunits): Change from unsigned char to
5511 (ONLY_FIXED_SIZE_MODES): New macro.
5512 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5513 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5514 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5516 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5517 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5518 or if measurement_type is not polynomial.
5519 * genmodes.c (ZERO_COEFFS): New macro.
5520 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5522 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5523 Use ZERO_COEFFS when emitting initializers.
5524 * data-streamer.h (bp_pack_poly_value): New function.
5525 (bp_unpack_poly_value): Likewise.
5526 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5527 for GET_MODE_NUNITS.
5528 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5529 for GET_MODE_NUNITS.
5530 * tree.c (make_vector_type): Remove temporary shim and make
5531 the real function take the number of units as a poly_uint64
5533 (build_vector_type_for_mode): Handle polynomial nunits.
5534 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5535 * emit-rtl.c (const_vec_series_p_1): Likewise.
5536 (gen_rtx_CONST_VECTOR): Likewise.
5537 * fold-const.c (test_vec_duplicate_folding): Likewise.
5538 * genrecog.c (validate_pattern): Likewise.
5539 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5540 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5541 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5542 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5543 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5544 * rtlanal.c (subreg_get_info): Likewise.
5545 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5546 (vect_grouped_load_supported): Likewise.
5547 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5548 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5549 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5550 (simplify_const_unary_operation, simplify_binary_operation_1)
5551 (simplify_const_binary_operation, simplify_ternary_operation)
5552 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5553 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5554 instead of CONST_VECTOR_NUNITS.
5555 * varasm.c (output_constant_pool_2): Likewise.
5556 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5557 explicit-encoded elements in the XVEC for variable-length vectors.
5559 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5561 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5563 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5564 Alan Hayward <alan.hayward@arm.com>
5565 David Sherwood <david.sherwood@arm.com>
5567 * coretypes.h (fixed_size_mode): Declare.
5568 (fixed_size_mode_pod): New typedef.
5569 * builtins.h (target_builtins::x_apply_args_mode)
5570 (target_builtins::x_apply_result_mode): Change type to
5571 fixed_size_mode_pod.
5572 * builtins.c (apply_args_size, apply_result_size, result_vector)
5573 (expand_builtin_apply_args_1, expand_builtin_apply)
5574 (expand_builtin_return): Update accordingly.
5576 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5578 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5579 * cselib.c (cselib_hash_rtx): Likewise.
5580 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5581 CONST_VECTOR encoding.
5583 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5584 Jeff Law <law@redhat.com>
5587 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5588 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5589 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5590 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5593 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5594 explicitly probe *sp in a noreturn function if there were any callee
5595 register saves or frame pointer is needed.
5597 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5600 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5601 BLKmode for ternary, binary or unary expressions.
5604 * var-tracking.c (delete_vta_debug_insn): New inline function.
5605 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5606 insns from get_insns () to NULL instead of each bb separately.
5607 Use delete_vta_debug_insn. No longer static.
5608 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5609 delete_vta_debug_insns callers.
5610 * rtl.h (delete_vta_debug_insns): Declare.
5611 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5612 instead of variable_tracking_main.
5614 2018-01-03 Martin Sebor <msebor@redhat.com>
5616 PR tree-optimization/83603
5617 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5618 arguments past the endof the argument list in functions declared
5619 without a prototype.
5620 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5621 Avoid checking when arguments are null.
5623 2018-01-03 Martin Sebor <msebor@redhat.com>
5626 * doc/extend.texi (attribute const): Fix a typo.
5627 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5628 issuing -Wsuggest-attribute for void functions.
5630 2018-01-03 Martin Sebor <msebor@redhat.com>
5632 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5633 offset_int::from instead of wide_int::to_shwi.
5634 (maybe_diag_overlap): Remove assertion.
5635 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5636 * gimple-ssa-sprintf.c (format_directive): Same.
5637 (parse_directive): Same.
5638 (sprintf_dom_walker::compute_format_length): Same.
5639 (try_substitute_return_value): Same.
5641 2018-01-03 Jeff Law <law@redhat.com>
5644 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5645 non-constant residual for zero at runtime and avoid probing in
5646 that case. Reorganize code for trailing problem to mirror handling
5649 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5651 PR tree-optimization/83501
5652 * tree-ssa-strlen.c (get_string_cst): New.
5653 (handle_char_store): Call get_string_cst.
5655 2018-01-03 Martin Liska <mliska@suse.cz>
5657 PR tree-optimization/83593
5658 * tree-ssa-strlen.c: Include tree-cfg.h.
5659 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5660 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5661 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5663 (strlen_dom_walker::before_dom_children): Call
5664 gimple_purge_dead_eh_edges. Dump tranformation with details
5666 (strlen_dom_walker::before_dom_children): Update call by adding
5667 new argument cleanup_eh.
5668 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5670 2018-01-03 Martin Liska <mliska@suse.cz>
5673 * cif-code.def (VARIADIC_THUNK): New enum value.
5674 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5677 2018-01-03 Jan Beulich <jbeulich@suse.com>
5679 * sse.md (mov<mode>_internal): Tighten condition for when to use
5680 vmovdqu<ssescalarsize> for TI and OI modes.
5682 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5684 Update copyright years.
5686 2018-01-03 Martin Liska <mliska@suse.cz>
5689 * ipa-visibility.c (function_and_variable_visibility): Skip
5690 functions with noipa attribure.
5692 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5694 * gcc.c (process_command): Update copyright notice dates.
5695 * gcov-dump.c (print_version): Ditto.
5696 * gcov.c (print_version): Ditto.
5697 * gcov-tool.c (print_version): Ditto.
5698 * gengtype.c (create_file): Ditto.
5699 * doc/cpp.texi: Bump @copying's copyright year.
5700 * doc/cppinternals.texi: Ditto.
5701 * doc/gcc.texi: Ditto.
5702 * doc/gccint.texi: Ditto.
5703 * doc/gcov.texi: Ditto.
5704 * doc/install.texi: Ditto.
5705 * doc/invoke.texi: Ditto.
5707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5709 * vector-builder.h (vector_builder::m_full_nelts): Change from
5710 unsigned int to poly_uint64.
5711 (vector_builder::full_nelts): Update prototype accordingly.
5712 (vector_builder::new_vector): Likewise.
5713 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5714 (vector_builder::operator ==): Likewise.
5715 (vector_builder::finalize): Likewise.
5716 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5717 Take the number of elements as a poly_uint64 rather than an
5719 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5720 from unsigned int to poly_uint64.
5721 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5722 (vec_perm_indices::new_vector): Likewise.
5723 (vec_perm_indices::length): Likewise.
5724 (vec_perm_indices::nelts_per_input): Likewise.
5725 (vec_perm_indices::input_nelts): Likewise.
5726 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5727 number of elements per input as a poly_uint64 rather than an
5728 unsigned int. Use the original encoding for variable-length
5729 vectors, rather than clamping each individual element.
5730 For the second and subsequent elements in each pattern,
5731 clamp the step and base before clamping their sum.
5732 (vec_perm_indices::series_p): Handle polynomial element counts.
5733 (vec_perm_indices::all_in_range_p): Likewise.
5734 (vec_perm_indices_to_tree): Likewise.
5735 (vec_perm_indices_to_rtx): Likewise.
5736 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5737 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5738 (tree_vector_builder::new_binary_operation): Handle polynomial
5739 element counts. Return false if we need to know the number
5740 of elements at compile time.
5741 * fold-const.c (fold_vec_perm): Punt if the number of elements
5742 isn't known at compile time.
5744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5746 * vec-perm-indices.h (vec_perm_builder): Change element type
5747 from HOST_WIDE_INT to poly_int64.
5748 (vec_perm_indices::element_type): Update accordingly.
5749 (vec_perm_indices::clamp): Handle polynomial element_types.
5750 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5751 (vec_perm_indices::all_in_range_p): Likewise.
5752 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5754 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5755 polynomial vec_perm_indices element types.
5756 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5757 * fold-const.c (fold_vec_perm): Likewise.
5758 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5759 * tree-vect-generic.c (lower_vec_perm): Likewise.
5760 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5761 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5762 element type to HOST_WIDE_INT.
5764 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5765 Alan Hayward <alan.hayward@arm.com>
5766 David Sherwood <david.sherwood@arm.com>
5768 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5769 rather than an int. Use plus_constant.
5770 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5771 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5774 Alan Hayward <alan.hayward@arm.com>
5775 David Sherwood <david.sherwood@arm.com>
5777 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5778 a HOST_WIDE_INT to a poly_int64.
5780 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5781 Alan Hayward <alan.hayward@arm.com>
5782 David Sherwood <david.sherwood@arm.com>
5784 * calls.c (load_register_parameters): Cope with polynomial
5785 mode sizes. Require a constant size for BLKmode parameters
5786 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5787 forces a parameter to be padded at the lsb end in order to
5788 fill a complete number of words, require the parameter size
5789 to be ordered wrt UNITS_PER_WORD.
5791 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5792 Alan Hayward <alan.hayward@arm.com>
5793 David Sherwood <david.sherwood@arm.com>
5795 * reload1.c (spill_stack_slot_width): Change element type
5796 from unsigned int to poly_uint64_pod.
5797 (alter_reg): Treat mode sizes as polynomial.
5799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5800 Alan Hayward <alan.hayward@arm.com>
5801 David Sherwood <david.sherwood@arm.com>
5803 * reload.c (complex_word_subreg_p): New function.
5804 (reload_inner_reg_of_subreg, push_reload): Use it.
5806 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5807 Alan Hayward <alan.hayward@arm.com>
5808 David Sherwood <david.sherwood@arm.com>
5810 * lra-constraints.c (process_alt_operands): Reject matched
5811 operands whose sizes aren't ordered.
5812 (match_reload): Refer to this check here.
5814 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5815 Alan Hayward <alan.hayward@arm.com>
5816 David Sherwood <david.sherwood@arm.com>
5818 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5819 that the mode size is in the set {1, 2, 4, 8, 16}.
5821 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5822 Alan Hayward <alan.hayward@arm.com>
5823 David Sherwood <david.sherwood@arm.com>
5825 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5826 Use plus_constant instead of gen_rtx_PLUS.
5828 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5829 Alan Hayward <alan.hayward@arm.com>
5830 David Sherwood <david.sherwood@arm.com>
5832 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5833 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5834 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5835 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5836 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5837 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5838 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5839 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5840 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5841 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5843 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5844 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5845 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5846 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5847 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5848 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5849 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5850 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5851 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5852 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5854 * expr.c (emit_move_resolve_push): Treat the input and result
5855 of PUSH_ROUNDING as a poly_int64.
5856 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5857 (emit_push_insn): Likewise.
5858 * lra-eliminations.c (mark_not_eliminable): Likewise.
5859 * recog.c (push_operand): Likewise.
5860 * reload1.c (elimination_effects): Likewise.
5861 * rtlanal.c (nonzero_bits1): Likewise.
5862 * calls.c (store_one_arg): Likewise. Require the padding to be
5863 known at compile time.
5865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5866 Alan Hayward <alan.hayward@arm.com>
5867 David Sherwood <david.sherwood@arm.com>
5869 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5870 Use plus_constant instead of gen_rtx_PLUS.
5872 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5873 Alan Hayward <alan.hayward@arm.com>
5874 David Sherwood <david.sherwood@arm.com>
5876 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5879 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5880 Alan Hayward <alan.hayward@arm.com>
5881 David Sherwood <david.sherwood@arm.com>
5883 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5884 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5885 via stack temporaries. Treat the mode size as polynomial too.
5887 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5888 Alan Hayward <alan.hayward@arm.com>
5889 David Sherwood <david.sherwood@arm.com>
5891 * expr.c (expand_expr_real_2): When handling conversions involving
5892 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5893 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5894 as a poly_uint64 too.
5896 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5897 Alan Hayward <alan.hayward@arm.com>
5898 David Sherwood <david.sherwood@arm.com>
5900 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5903 Alan Hayward <alan.hayward@arm.com>
5904 David Sherwood <david.sherwood@arm.com>
5906 * combine.c (can_change_dest_mode): Handle polynomial
5907 REGMODE_NATURAL_SIZE.
5908 * expmed.c (store_bit_field_1): Likewise.
5909 * expr.c (store_constructor): Likewise.
5910 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5911 and polynomial REGMODE_NATURAL_SIZE.
5912 (gen_lowpart_common): Likewise.
5913 * reginfo.c (record_subregs_of_mode): Likewise.
5914 * rtlanal.c (read_modify_subreg_p): Likewise.
5916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5917 Alan Hayward <alan.hayward@arm.com>
5918 David Sherwood <david.sherwood@arm.com>
5920 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5921 numbers of elements.
5923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5924 Alan Hayward <alan.hayward@arm.com>
5925 David Sherwood <david.sherwood@arm.com>
5927 * match.pd: Cope with polynomial numbers of vector elements.
5929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5930 Alan Hayward <alan.hayward@arm.com>
5931 David Sherwood <david.sherwood@arm.com>
5933 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5934 in a POINTER_PLUS_EXPR.
5936 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5937 Alan Hayward <alan.hayward@arm.com>
5938 David Sherwood <david.sherwood@arm.com>
5940 * omp-simd-clone.c (simd_clone_subparts): New function.
5941 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5942 (ipa_simd_modify_function_body): Likewise.
5944 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5945 Alan Hayward <alan.hayward@arm.com>
5946 David Sherwood <david.sherwood@arm.com>
5948 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5949 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5950 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5951 (expand_vector_condition, vector_element): Likewise.
5952 (subparts_gt): New function.
5953 (get_compute_type): Use subparts_gt.
5954 (count_type_subparts): Delete.
5955 (expand_vector_operations_1): Use subparts_gt instead of
5956 count_type_subparts.
5958 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5959 Alan Hayward <alan.hayward@arm.com>
5960 David Sherwood <david.sherwood@arm.com>
5962 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5963 (vect_compile_time_alias): ...this new function. Do the calculation
5964 on poly_ints rather than trees.
5965 (vect_prune_runtime_alias_test_list): Update call accordingly.
5967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5968 Alan Hayward <alan.hayward@arm.com>
5969 David Sherwood <david.sherwood@arm.com>
5971 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5973 (vect_schedule_slp_instance): Likewise.
5975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5976 Alan Hayward <alan.hayward@arm.com>
5977 David Sherwood <david.sherwood@arm.com>
5979 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5980 constant and extern definitions for variable-length vectors.
5981 (vect_get_constant_vectors): Note that the number of units
5982 is known to be constant.
5984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5985 Alan Hayward <alan.hayward@arm.com>
5986 David Sherwood <david.sherwood@arm.com>
5988 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5989 of units as polynomial. Choose between WIDE and NARROW based
5992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5993 Alan Hayward <alan.hayward@arm.com>
5994 David Sherwood <david.sherwood@arm.com>
5996 * tree-vect-stmts.c (simd_clone_subparts): New function.
5997 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5999 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6000 Alan Hayward <alan.hayward@arm.com>
6001 David Sherwood <david.sherwood@arm.com>
6003 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6004 vectors as polynomial. Use build_index_vector for
6007 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6008 Alan Hayward <alan.hayward@arm.com>
6009 David Sherwood <david.sherwood@arm.com>
6011 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6012 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6013 for variable-length vectors.
6014 (vectorizable_mask_load_store): Treat the number of units as
6015 polynomial, asserting that it is constant if the condition has
6016 already been enforced.
6017 (vectorizable_store, vectorizable_load): Likewise.
6019 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6020 Alan Hayward <alan.hayward@arm.com>
6021 David Sherwood <david.sherwood@arm.com>
6023 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6024 of units as polynomial. Punt if we can't tell at compile time
6025 which vector contains the final result.
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 * tree-vect-loop.c (vectorizable_induction): Treat the number
6032 of units as polynomial. Punt on SLP inductions. Use an integer
6033 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6034 cast of such a series for variable-length floating-point
6037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6038 Alan Hayward <alan.hayward@arm.com>
6039 David Sherwood <david.sherwood@arm.com>
6041 * tree.h (build_index_vector): Declare.
6042 * tree.c (build_index_vector): New function.
6043 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6044 of units as polynomial, forcibly converting it to a constant if
6045 vectorizable_reduction has already enforced the condition.
6046 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6047 to create a {1,2,3,...} vector.
6048 (vectorizable_reduction): Treat the number of units as polynomial.
6049 Choose vectype_in based on the largest scalar element size rather
6050 than the smallest number of units. Enforce the restrictions
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 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6058 number of units as polynomial.
6060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6061 Alan Hayward <alan.hayward@arm.com>
6062 David Sherwood <david.sherwood@arm.com>
6064 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6065 * target.def (autovectorize_vector_sizes): Return the vector sizes
6066 by pointer, using vector_sizes rather than a bitmask.
6067 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6068 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6069 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6071 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6072 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6073 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6074 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6075 * omp-general.c (omp_max_vf): Likewise.
6076 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6077 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6078 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6079 * tree-vect-slp.c (vect_slp_bb): Likewise.
6080 * doc/tm.texi: Regenerate.
6081 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6083 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6084 the vector size as a poly_uint64 rather than an unsigned int.
6085 (current_vector_size): Change from an unsigned int to a poly_uint64.
6086 (get_vectype_for_scalar_type): Update accordingly.
6087 * tree.h (build_truth_vector_type): Take the size and number of
6088 units as a poly_uint64 rather than an unsigned int.
6089 (build_vector_type): Add a temporary overload that takes
6090 the number of units as a poly_uint64 rather than an unsigned int.
6091 * tree.c (make_vector_type): Likewise.
6092 (build_truth_vector_type): Take the number of units as a poly_uint64
6093 rather than an unsigned int.
6095 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6096 Alan Hayward <alan.hayward@arm.com>
6097 David Sherwood <david.sherwood@arm.com>
6099 * target.def (get_mask_mode): Take the number of units and length
6100 as poly_uint64s rather than unsigned ints.
6101 * targhooks.h (default_get_mask_mode): Update accordingly.
6102 * targhooks.c (default_get_mask_mode): Likewise.
6103 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6104 * doc/tm.texi: Regenerate.
6106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6107 Alan Hayward <alan.hayward@arm.com>
6108 David Sherwood <david.sherwood@arm.com>
6110 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6111 * omp-general.c (omp_max_vf): Likewise.
6112 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6113 (expand_omp_simd): Handle polynomial safelen.
6114 * omp-low.c (omplow_simd_context): Add a default constructor.
6115 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6116 (lower_rec_simd_input_clauses): Update accordingly.
6117 (lower_rec_input_clauses): Likewise.
6119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6120 Alan Hayward <alan.hayward@arm.com>
6121 David Sherwood <david.sherwood@arm.com>
6123 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6124 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6125 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6126 (vect_analyze_slp_cost): Likewise.
6127 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6128 (vect_model_load_cost): Likewise.
6130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6131 Alan Hayward <alan.hayward@arm.com>
6132 David Sherwood <david.sherwood@arm.com>
6134 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6135 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6136 from an unsigned int * to a poly_uint64_pod *.
6137 (calculate_unrolling_factor): New function.
6138 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6140 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6141 Alan Hayward <alan.hayward@arm.com>
6142 David Sherwood <david.sherwood@arm.com>
6144 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6145 from an unsigned int to a poly_uint64.
6146 (_loop_vec_info::slp_unrolling_factor): Likewise.
6147 (_loop_vec_info::vectorization_factor): Change from an int
6149 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6150 (vect_get_num_vectors): New function.
6151 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6152 (vect_get_num_copies): Use vect_get_num_vectors.
6153 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6154 to an unsigned int *.
6155 (vect_analyze_data_refs): Change min_vf from an int * to a
6157 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6158 than an unsigned HOST_WIDE_INT.
6159 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6160 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6161 to an unsigned int *.
6162 (vect_analyze_data_ref_dependences): Likewise.
6163 (vect_compute_data_ref_alignment): Handle polynomial vf.
6164 (vect_enhance_data_refs_alignment): Likewise.
6165 (vect_prune_runtime_alias_test_list): Likewise.
6166 (vect_shift_permute_load_chain): Likewise.
6167 (vect_supportable_dr_alignment): Likewise.
6168 (dependence_distance_ge_vf): Take the vectorization factor as a
6169 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6170 (vect_analyze_data_refs): Change min_vf from an int * to a
6172 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6173 vfm1 as a poly_uint64 rather than an int. Make the same change
6174 for the returned bound_scalar.
6175 (vect_gen_vector_loop_niters): Handle polynomial vf.
6176 (vect_do_peeling): Likewise. Update call to
6177 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6178 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6180 * tree-vect-loop.c (vect_determine_vectorization_factor)
6181 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6182 (vect_get_known_peeling_cost): Likewise.
6183 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6184 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6185 (vect_transform_loop): Likewise. Use the lowest possible VF when
6186 updating the upper bounds of the loop.
6187 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6189 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6190 polynomial unroll factors.
6191 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6192 (vect_make_slp_decision): Likewise.
6193 (vect_supported_load_permutation_p): Likewise, and polynomial
6195 (vect_analyze_slp_cost): Handle polynomial vf.
6196 (vect_slp_analyze_node_operations): Likewise.
6197 (vect_slp_analyze_bb_1): Likewise.
6198 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6199 than an unsigned HOST_WIDE_INT.
6200 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6201 (vectorizable_load): Handle polynomial vf.
6202 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6204 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6206 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6207 Alan Hayward <alan.hayward@arm.com>
6208 David Sherwood <david.sherwood@arm.com>
6210 * match.pd: Handle bit operations involving three constants
6211 and try to fold one pair.
6213 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6215 * tree-vect-loop-manip.c: Include gimple-fold.h.
6216 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6217 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6218 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6219 Add a path that uses a step of VF instead of 1, but disable it
6221 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6222 and niters_no_overflow parameters. Update calls to
6223 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6224 Create a new SSA name if the latter choses to use a ste other
6225 than zero, and return it via niters_vector_mult_vf_var.
6226 * tree-vect-loop.c (vect_transform_loop): Update calls to
6227 vect_do_peeling, vect_gen_vector_loop_niters and
6228 slpeel_make_loop_iterate_ntimes.
6229 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6230 (vect_gen_vector_loop_niters): Update declarations after above changes.
6232 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6234 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6235 128-bit round to integer instructions.
6236 (ceil<mode>2): Likewise.
6237 (btrunc<mode>2): Likewise.
6238 (round<mode>2): Likewise.
6240 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6242 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6243 unaligned VSX load/store on P8/P9.
6244 (expand_block_clear): Allow the use of unaligned VSX
6245 load/store on P8/P9.
6247 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6249 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6251 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6252 swap associated with both a load and a store.
6254 2018-01-02 Andrew Waterman <andrew@sifive.com>
6256 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6257 * config/riscv/riscv.md (clear_cache): Use it.
6259 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6261 * web.c: Remove out-of-date comment.
6263 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6265 * expr.c (fixup_args_size_notes): Check that any existing
6266 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6267 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6268 (emit_single_push_insn): ...here.
6270 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6272 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6273 (const_vector_encoded_nelts): New function.
6274 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6275 (const_vector_int_elt, const_vector_elt): Declare.
6276 * emit-rtl.c (const_vector_int_elt_1): New function.
6277 (const_vector_elt): Likewise.
6278 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6279 of CONST_VECTOR_ELT.
6281 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6283 * expr.c: Include rtx-vector-builder.h.
6284 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6285 directly on the tree encoding.
6286 (const_vector_from_tree): Likewise.
6287 * optabs.c: Include rtx-vector-builder.h.
6288 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6289 sequence of "u" values.
6290 * vec-perm-indices.c: Include rtx-vector-builder.h.
6291 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6292 directly on the vec_perm_indices encoding.
6294 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6296 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6297 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6298 * rtx-vector-builder.h: New file.
6299 * rtx-vector-builder.c: Likewise.
6300 * rtl.h (rtx_def::u2): Add a const_vector field.
6301 (CONST_VECTOR_NPATTERNS): New macro.
6302 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6303 (CONST_VECTOR_DUPLICATE_P): Likewise.
6304 (CONST_VECTOR_STEPPED_P): Likewise.
6305 (CONST_VECTOR_ENCODED_ELT): Likewise.
6306 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6307 (unwrap_const_vec_duplicate): Likewise.
6308 (const_vec_series_p): Check for a non-duplicated vector encoding.
6309 Say that the function only returns true for integer vectors.
6310 * emit-rtl.c: Include rtx-vector-builder.h.
6311 (gen_const_vec_duplicate_1): Delete.
6312 (gen_const_vector): Call gen_const_vec_duplicate instead of
6313 gen_const_vec_duplicate_1.
6314 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6315 (gen_const_vec_duplicate): Use rtx_vector_builder.
6316 (gen_const_vec_series): Likewise.
6317 (gen_rtx_CONST_VECTOR): Likewise.
6318 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6319 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6320 Build a new vector rather than modifying a CONST_VECTOR in-place.
6321 (handle_special_swappables): Update call accordingly.
6322 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6323 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6324 Build a new vector rather than modifying a CONST_VECTOR in-place.
6325 (handle_special_swappables): Update call accordingly.
6327 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6329 * simplify-rtx.c (simplify_const_binary_operation): Use
6330 CONST_VECTOR_ELT instead of XVECEXP.
6332 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6334 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6335 the selector elements to be different from the data elements
6336 if the selector is a VECTOR_CST.
6337 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6338 ssizetype for the selector.
6340 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6342 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6343 before testing each element individually.
6344 * tree-vect-generic.c (lower_vec_perm): Likewise.
6346 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6348 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6349 * selftest-run-tests.c (selftest::run_tests): Call it.
6350 * vector-builder.h (vector_builder::operator ==): New function.
6351 (vector_builder::operator !=): Likewise.
6352 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6353 (vec_perm_indices::all_from_input_p): New function.
6354 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6355 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6356 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6357 instead of reading the VECTOR_CST directly. Detect whether both
6358 vector inputs are the same before constructing the vec_perm_indices,
6359 and update the number of inputs argument accordingly. Use the
6360 utility functions added above. Only construct sel2 if we need to.
6362 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6364 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6365 the broadcast of the low byte.
6366 (expand_mult_highpart): Use an explicit encoding for the permutes.
6367 * optabs-query.c (can_mult_highpart_p): Likewise.
6368 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6369 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6370 (vectorizable_bswap): Likewise.
6371 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6372 explicit encoding for the power-of-2 permutes.
6373 (vect_permute_store_chain): Likewise.
6374 (vect_grouped_load_supported): Likewise.
6375 (vect_permute_load_chain): Likewise.
6377 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6379 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6380 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6381 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6382 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6383 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6384 (vect_gen_perm_mask_any): Likewise.
6386 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6388 * int-vector-builder.h: New file.
6389 * vec-perm-indices.h: Include int-vector-builder.h.
6390 (vec_perm_indices): Redefine as an int_vector_builder.
6391 (auto_vec_perm_indices): Delete.
6392 (vec_perm_builder): Redefine as a stand-alone class.
6393 (vec_perm_indices::vec_perm_indices): New function.
6394 (vec_perm_indices::clamp): Likewise.
6395 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6396 (vec_perm_indices::new_vector): New function.
6397 (vec_perm_indices::new_expanded_vector): Update for new
6398 vec_perm_indices class.
6399 (vec_perm_indices::rotate_inputs): New function.
6400 (vec_perm_indices::all_in_range_p): Operate directly on the
6401 encoded form, without computing elided elements.
6402 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6403 encoding. Update for new vec_perm_indices class.
6404 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6405 the given vec_perm_builder.
6406 (expand_vec_perm_var): Update vec_perm_builder constructor.
6407 (expand_mult_highpart): Use vec_perm_builder instead of
6408 auto_vec_perm_indices.
6409 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6410 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6411 or double series encoding as appropriate.
6412 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6413 vec_perm_indices instead of auto_vec_perm_indices.
6414 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6415 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6416 (vect_permute_store_chain): Likewise.
6417 (vect_grouped_load_supported): Likewise.
6418 (vect_permute_load_chain): Likewise.
6419 (vect_shift_permute_load_chain): Likewise.
6420 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6421 (vect_transform_slp_perm_load): Likewise.
6422 (vect_schedule_slp_instance): Likewise.
6423 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6424 (vectorizable_mask_load_store): Likewise.
6425 (vectorizable_bswap): Likewise.
6426 (vectorizable_store): Likewise.
6427 (vectorizable_load): Likewise.
6428 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6429 vec_perm_indices instead of auto_vec_perm_indices. Use
6430 tree_to_vec_perm_builder to read the vector from a tree.
6431 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6432 vec_perm_builder instead of a vec_perm_indices.
6433 (have_whole_vector_shift): Use vec_perm_builder and
6434 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6435 truncation to calc_vec_perm_mask_for_shift.
6436 (vect_create_epilog_for_reduction): Likewise.
6437 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6438 from auto_vec_perm_indices to vec_perm_indices.
6439 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6440 instead of changing individual elements.
6441 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6442 the vector in d.perm.
6443 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6444 from auto_vec_perm_indices to vec_perm_indices.
6445 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6446 instead of changing individual elements.
6447 (arm_vectorize_vec_perm_const): Use new_vector to install
6448 the vector in d.perm.
6449 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6450 Update vec_perm_builder constructor.
6451 (rs6000_expand_interleave): Likewise.
6452 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6453 (rs6000_expand_interleave): Likewise.
6455 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6457 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6458 to qimode could truncate the indices.
6459 * optabs.c (expand_vec_perm_var): Likewise.
6461 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6463 * Makefile.in (OBJS): Add vec-perm-indices.o.
6464 * vec-perm-indices.h: New file.
6465 * vec-perm-indices.c: Likewise.
6466 * target.h (vec_perm_indices): Replace with a forward class
6468 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6469 * optabs.h: Include vec-perm-indices.h.
6470 (expand_vec_perm): Delete.
6471 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6472 (expand_vec_perm_const): Declare.
6473 * target.def (vec_perm_const_ok): Replace with...
6474 (vec_perm_const): ...this new hook.
6475 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6476 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6477 * doc/tm.texi: Regenerate.
6478 * optabs.def (vec_perm_const): Delete.
6479 * doc/md.texi (vec_perm_const): Likewise.
6480 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6481 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6482 expand_vec_perm for constant permutation vectors. Assert that
6483 the mode of variable permutation vectors is the integer equivalent
6484 of the mode that is being permuted.
6485 * optabs-query.h (selector_fits_mode_p): Declare.
6486 * optabs-query.c: Include vec-perm-indices.h.
6487 (selector_fits_mode_p): New function.
6488 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6489 is defined, instead of checking whether the vec_perm_const_optab
6490 exists. Use targetm.vectorize.vec_perm_const instead of
6491 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6492 fit in the vector mode before using a variable permute.
6493 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6494 vec_perm_indices instead of an rtx.
6495 (expand_vec_perm): Replace with...
6496 (expand_vec_perm_const): ...this new function. Take the selector
6497 as a vec_perm_indices rather than an rtx. Also take the mode of
6498 the selector. Update call to shift_amt_for_vec_perm_mask.
6499 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6500 Use vec_perm_indices::new_expanded_vector to expand the original
6501 selector into bytes. Check whether the indices fit in the vector
6502 mode before using a variable permute.
6503 (expand_vec_perm_var): Make global.
6504 (expand_mult_highpart): Use expand_vec_perm_const.
6505 * fold-const.c: Includes vec-perm-indices.h.
6506 * tree-ssa-forwprop.c: Likewise.
6507 * tree-vect-data-refs.c: Likewise.
6508 * tree-vect-generic.c: Likewise.
6509 * tree-vect-loop.c: Likewise.
6510 * tree-vect-slp.c: Likewise.
6511 * tree-vect-stmts.c: Likewise.
6512 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6514 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6515 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6516 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6517 (aarch64_vectorize_vec_perm_const): ...this new function.
6518 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6519 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6520 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6521 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6522 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6523 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6524 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6526 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6527 check for NEON modes.
6528 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6529 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6530 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6531 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6533 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6534 the old VEC_PERM_CONST conditions.
6535 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6536 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6537 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6538 (ia64_vectorize_vec_perm_const_ok): Merge into...
6539 (ia64_vectorize_vec_perm_const): ...this new function.
6540 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6541 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6542 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6543 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6544 * config/mips/mips.c (mips_expand_vec_perm_const)
6545 (mips_vectorize_vec_perm_const_ok): Merge into...
6546 (mips_vectorize_vec_perm_const): ...this new function.
6547 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6548 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6549 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6550 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6551 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6552 (rs6000_expand_vec_perm_const): Delete.
6553 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6555 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6556 (altivec_expand_vec_perm_const_le): Take each operand individually.
6557 Operate on constant selectors rather than rtxes.
6558 (altivec_expand_vec_perm_const): Likewise. Update call to
6559 altivec_expand_vec_perm_const_le.
6560 (rs6000_expand_vec_perm_const): Delete.
6561 (rs6000_vectorize_vec_perm_const_ok): Delete.
6562 (rs6000_vectorize_vec_perm_const): New function.
6563 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6564 an element count and rtx array.
6565 (rs6000_expand_extract_even): Update call accordingly.
6566 (rs6000_expand_interleave): Likewise.
6567 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6568 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6569 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6570 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6571 (rs6000_expand_vec_perm_const): Delete.
6572 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6573 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6574 (altivec_expand_vec_perm_const_le): Take each operand individually.
6575 Operate on constant selectors rather than rtxes.
6576 (altivec_expand_vec_perm_const): Likewise. Update call to
6577 altivec_expand_vec_perm_const_le.
6578 (rs6000_expand_vec_perm_const): Delete.
6579 (rs6000_vectorize_vec_perm_const_ok): Delete.
6580 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6581 reference to the SPE evmerge intructions.
6582 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6583 an element count and rtx array.
6584 (rs6000_expand_extract_even): Update call accordingly.
6585 (rs6000_expand_interleave): Likewise.
6586 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6587 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6589 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6591 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6593 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6594 vector mode and that that mode matches the mode of the data
6596 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6597 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6598 directly using expand_vec_perm_1 when forcing selectors into
6600 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6602 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6604 * optabs-query.h (can_vec_perm_p): Delete.
6605 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6606 * optabs-query.c (can_vec_perm_p): Split into...
6607 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6608 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6609 particular selector is valid.
6610 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6611 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6612 (vect_grouped_load_supported): Likewise.
6613 (vect_shift_permute_load_chain): Likewise.
6614 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6615 (vect_transform_slp_perm_load): Likewise.
6616 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6617 (vectorizable_bswap): Likewise.
6618 (vect_gen_perm_mask_checked): Likewise.
6619 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6620 implementations of variable permutation vectors into account
6621 when deciding which selector to use.
6622 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6623 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6624 with a false third argument.
6625 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6626 to test whether the constant selector is valid and can_vec_perm_var_p
6627 to test whether a variable selector is valid.
6629 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6631 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6632 * optabs-query.c (can_vec_perm_p): Likewise.
6633 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6634 instead of vec_perm_indices.
6635 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6636 (vect_gen_perm_mask_checked): Likewise,
6637 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6638 (vect_gen_perm_mask_checked): Likewise,
6640 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6642 * optabs-query.h (qimode_for_vec_perm): Declare.
6643 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6644 (qimode_for_vec_perm): ...this new function.
6645 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6647 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6649 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6650 does not have a conditional at the top.
6652 2018-01-02 Richard Biener <rguenther@suse.de>
6654 * ipa-inline.c (big_speedup_p): Fix expression.
6656 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6659 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6662 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6666 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6667 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6668 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6669 cond_taken_branch_cost 3->4.
6671 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6673 PR tree-optimization/83581
6674 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6675 TODO_cleanup_cfg if any changes have been made.
6678 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6679 convert_modes if target mode has the right side, but different mode
6683 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6684 last argument when extracting from CONCAT. If either from_real or
6685 from_imag is NULL, use expansion through memory. If result is not
6686 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6687 the parts directly to inner mode, if even that fails, use expansion
6691 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6692 check for bswap in mode rather than HImode and use that in expand_unop
6695 Copyright (C) 2018 Free Software Foundation, Inc.
6697 Copying and distribution of this file, with or without modification,
6698 are permitted in any medium without royalty provided the copyright
6699 notice and this notice are preserved.