1 2018-02-10 Jakub Jelinek <jakub@redhat.com>
3 PR rtl-optimization/84308
4 * shrink-wrap.c (spread_components): Release todo vector.
6 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
8 PR rtl-optimization/57193
9 * ira-color.c (struct allocno_color_data): Add member
10 conflict_allocno_hard_prefs.
11 (update_conflict_allocno_hard_prefs): New.
12 (bucket_allocno_compare_func): Add a preference based on
13 conflict_allocno_hard_prefs.
14 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
15 (color_allocnos): Remove a dead code. Initiate
16 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
18 2018-02-09 Jakub Jelinek <jakub@redhat.com>
21 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
22 constraint from =wa to wa. Avoid a subreg on the output operand,
23 instead use a pseudo and subreg it in a move.
24 (p9_xxbrd_<mode>): Changed to ...
25 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
26 (p9_xxbrd_v2df): New expander.
27 (p9_xxbrw_<mode>): Changed to ...
28 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
29 (p9_xxbrw_v4sf): New expander.
31 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
33 * config/rx.md: updated "movsicc" expand to be matched by GCC
34 * testsuite/gcc.target/rx/movsicc.c: new test case
36 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
39 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
40 multiply in 32-bit mode.
41 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
42 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
45 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
47 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
48 to allow or block "symbol_ref" depending on value of TARGET_JSR
49 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
50 call_value_internal insns
52 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
55 * dwarf2out.c (is_trivial_indirect_ref): New function.
56 (dwarf2out_late_global_decl): Do not generate a location
57 attribute for variables that have a non-trivial DECL_VALUE_EXPR
58 and that are not defined in the current unit.
60 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
62 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
63 instead of a libcall for UNORDERED.
65 2018-02-09 Tamar Christina <tamar.christina@arm.com>
68 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
69 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
71 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
74 * config/s390/s390.c (s390_set_current_function): Invoke
75 s390_indirect_branch_settings also if fndecl didn't change.
77 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
79 * config/rs6000/rs6000.md (blockage): Set length to zero.
81 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
83 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
85 2018-02-09 Jakub Jelinek <jakub@redhat.com>
88 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
89 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
93 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
94 PARALLEL incoming that failed vt_get_decl_and_offset check.
97 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
98 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
100 (get_variable_section): For decls in named .bss* sections pass true as
101 second argument to bss_initializer_p.
103 2018-02-09 Marek Polacek <polacek@redhat.com>
104 Jakub Jelinek <jakub@redhat.com>
107 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
108 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
109 Sync some changes from cxx_fold_indirect_ref.
111 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
113 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
115 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
116 (BLOCK_INLINE_ENTRY_LABEL): New.
117 (dwarf2out_var_location): Disregard inline entry markers.
118 (inline_entry_data): New struct.
119 (inline_entry_data_hasher): New hashtable type.
120 (inline_entry_data_hasher::hash): New.
121 (inline_entry_data_hasher::equal): New.
122 (inline_entry_data_table): New variable.
123 (add_high_low_attributes): Add DW_AT_entry_pc and
124 DW_AT_GNU_entry_view attributes if a pending entry is found
125 in inline_entry_data_table. Add old entry_pc attribute only
126 if debug nonbinding markers are disabled.
127 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
129 (block_within_block_p, dwarf2out_inline_entry): New.
130 (dwarf2out_finish): Check that no entries remained in
131 inline_entry_data_table.
132 * final.c (reemit_insn_block_notes): Handle inline entry notes.
133 (final_scan_insn, notice_source_line): Likewise.
134 (rest_of_clean_state): Skip inline entry markers.
135 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
137 * gimple.c (gimple_build_debug_inline_entry): New.
138 * gimple.h (enum gimple_debug_subcode): Add
139 GIMPLE_DEBUG_INLINE_ENTRY.
140 (gimple_build_debug_inline_entry): Declare.
141 (gimple_debug_inline_entry_p): New.
142 (gimple_debug_nonbind_marker_p): Adjust.
143 * insn-notes.def (INLINE_ENTRY): New.
144 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
145 inline entry marker notes.
146 (print_insn): Likewise.
147 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
148 (INSN_DEBUG_MARKER_KIND): Likewise.
149 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
150 * tree-inline.c (expand_call_inline): Build and insert
151 debug_inline_entry stmt.
152 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
153 inline entry blocks early, if nonbind markers are enabled.
154 (dump_scope_block): Dump fragment info.
155 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
156 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
157 (gimple_build_debug_inline_entry): New.
158 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
159 Enable/disable inline entry points too.
160 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
161 (DEBUG_INSN): Describe inline entry markers.
163 * common.opt (gvariable-location-views): New.
164 (gvariable-location-views=incompat5): New.
165 * config.in: Rebuilt.
166 * configure: Rebuilt.
167 * configure.ac: Test assembler for view support.
168 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
169 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
170 * dwarf2out.c (var_loc_view): New typedef.
171 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
172 (dwarf2out_locviews_in_attribute): New.
173 (dwarf2out_locviews_in_loclist): New.
174 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
175 (enum dw_line_info_opcode): Add LI_adv_address.
176 (struct dw_line_info_table): Add view.
177 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
178 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
179 (zero_view_p): New variable.
180 (ZERO_VIEW_P): New macro.
181 (output_asm_line_debug_info): New.
182 (struct var_loc_node): Add view.
183 (add_AT_view_list, AT_loc_list): New.
184 (add_var_loc_to_decl): Add view param. Test it against last.
185 (new_loc_list): Add view params. Record them.
186 (AT_loc_list_ptr): Handle loc and view lists.
187 (view_list_to_loc_list_val_node): New.
188 (print_dw_val): Handle dw_val_class_view_list.
189 (size_of_die): Likewise.
190 (value_format): Likewise.
191 (loc_list_has_views): New.
192 (gen_llsym): Set vl_symbol too.
193 (maybe_gen_llsym, skip_loc_list_entry): New.
194 (dwarf2out_maybe_output_loclist_view_pair): New.
195 (output_loc_list): Output view list or entries too.
196 (output_view_list_offset): New.
197 (output_die): Handle dw_val_class_view_list.
198 (output_dwarf_version): New.
199 (output_compilation_unit_header): Use it.
200 (output_skeleton_debug_sections): Likewise.
201 (output_rnglists, output_line_info): Likewise.
202 (output_pubnames, output_aranges): Update version comments.
203 (output_one_line_info_table): Output view numbers in asm comments.
204 (dw_loc_list): Determine current endview, pass it to new_loc_list.
205 Call maybe_gen_llsym.
206 (loc_list_from_tree_1): Adjust.
207 (add_AT_location_description): Create view list attribute if
208 needed, check it's absent otherwise.
209 (convert_cfa_to_fb_loc_list): Adjust.
210 (maybe_emit_file): Call output_asm_line_debug_info for test.
211 (dwarf2out_var_location): Reset views as needed. Precompute
212 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
214 (new_line_info_table): Reset next view.
215 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
216 (dwarf2out_source_line): Likewise. Output view resets and labels to
217 the assembler, or select appropriate line info opcodes.
218 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
219 (optimize_string_length): Catch it. Adjust.
220 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
221 dw_val_class_view_list, and remove it if no longer needed.
222 (hash_loc_list): Hash view numbers.
223 (loc_list_hasher::equal): Compare them.
224 (optimize_location_lists): Check whether a view list symbol is
225 needed, and whether the locview attribute is present, and
226 whether they match. Remove the locview attribute if no longer
228 (index_location_lists): Call skip_loc_list_entry for test.
229 (dwarf2out_finish): Call output_asm_line_debug_info for test.
230 Use output_dwarf_version.
231 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
232 (struct dw_val_node): Add val_view_list.
233 * final.c (SEEN_NEXT_VIEW): New.
234 (set_next_view_needed): New.
235 (clear_next_view_needed): New.
236 (maybe_output_next_view): New.
237 (final_start_function): Rename to...
238 (final_start_function_1): ... this. Take pointer to FIRST,
239 add SEEN parameter. Emit param bindings in the initial view.
240 (final_start_function): Reintroduce SEEN-less interface.
241 (final): Rename to...
242 (final_1): ... this. Take SEEN parameter. Output final pending
243 next view at the end.
244 (final): Reintroduce seen-less interface.
245 (final_scan_insn): Output pending next view before switching
246 sections or ending a block. Mark the next view as needed when
247 outputting variable locations. Notify debug backend of section
248 changes, and of location view changes.
249 (rest_of_handle_final): Adjust.
250 * toplev.c (process_options): Autodetect value for debug variable
251 location views option. Warn on incompat5 without -gdwarf-5.
252 * doc/invoke.texi (gvariable-location-views): New.
253 (gvariable-location-views=incompat5): New.
254 (gno-variable-location-views): New.
256 2018-02-08 David Malcolm <dmalcolm@redhat.com>
258 PR tree-optimization/84136
259 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
260 that the result of find_edge is non-NULL.
262 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
265 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
266 storing integer register in SImode. Fix cost of 256 and 512
267 byte aligned SSE register store.
269 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
271 * config/i386/i386.c (ix86_multiplication_cost): Fix
272 multiplication cost for TARGET_AVX512DQ.
274 2018-02-08 Marek Polacek <polacek@redhat.com>
276 PR tree-optimization/84238
277 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
280 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
282 PR tree-optimization/84265
283 * tree-vect-stmts.c (vectorizable_store): Don't treat
284 VMAT_CONTIGUOUS accesses as grouped.
285 (vectorizable_load): Likewise.
287 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
289 PR tree-optimization/81635
290 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
291 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
292 (test_round_for_mask): New functions.
293 (wide_int_cc_tests): Call test_round_for_mask.
294 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
295 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
296 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
297 range returned by get_range_info.
299 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
302 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
303 * symtab.c: Include builtins.h
304 (symtab_node::output_to_lto_symbol_table_p): Move here
305 from lto-streamer-out.c:output_symbol_p.
306 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
307 (output_symbol_p): Move all logic to symtab.c
308 (produce_symtab): Update.
310 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
312 * config/s390/s390-opts.h (enum indirect_branch): Define.
313 * config/s390/s390-protos.h (s390_return_addr_from_memory)
314 (s390_indirect_branch_via_thunk)
315 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
316 (enum s390_indirect_branch_type): Define.
317 * config/s390/s390.c (struct s390_frame_layout, struct
318 machine_function): Remove.
319 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
320 (indirect_branch_table_label_no, indirect_branch_table_name):
322 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
323 (enum s390_indirect_branch_option): Define.
324 (s390_return_addr_from_memory): New function.
325 (s390_handle_string_attribute): New function.
326 (s390_attribute_table): Add new attribute handler.
327 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
328 (s390_indirect_branch_via_thunk): New function.
329 (s390_indirect_branch_via_inline_thunk): New function.
330 (s390_function_ok_for_sibcall): When jumping via thunk disallow
331 sibling call optimization for non z10 compiles.
332 (s390_emit_call): Force indirect branch target to be a single
333 register. Add r1 clobber for non-z10 compiles.
334 (s390_emit_epilogue): Emit return jump via return_use expander.
335 (s390_reorg): Handle JUMP_INSNs as execute targets.
336 (s390_option_override_internal): Perform validity checks for the
337 new command line options.
338 (s390_indirect_branch_attrvalue): New function.
339 (s390_indirect_branch_settings): New function.
340 (s390_set_current_function): Invoke s390_indirect_branch_settings.
341 (s390_output_indirect_thunk_function): New function.
342 (s390_code_end): Implement target hook.
343 (s390_case_values_threshold): Implement target hook.
344 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
346 * config/s390/s390.h (struct s390_frame_layout)
347 (struct machine_function): Move here from s390.c.
348 (TARGET_INDIRECT_BRANCH_NOBP_RET)
349 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
350 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
351 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
352 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
353 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
354 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
355 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
356 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
357 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
358 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
359 (mnemonic attribute): Add values which aren't recognized
361 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
362 pattern for branch conversion. Fix mnemonic attribute.
363 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
364 indirect branch via thunk if requested.
365 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
366 ("*indirect_jump"): Disable for branch conversion using out of
368 ("indirect_jump_via_thunk<mode>_z10")
369 ("indirect_jump_via_thunk<mode>")
370 ("indirect_jump_via_inlinethunk<mode>_z10")
371 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
372 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
373 ("casesi_jump_via_inlinethunk<mode>_z10")
374 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
375 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
376 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
377 ("*indirect2_jump"): Disable for branch conversion.
378 ("casesi_jump"): Turn into expander and expand patterns for branch
380 ("return_use"): New expander.
381 ("*return"): Emit return via thunk and rename it to ...
382 ("*return<mode>"): ... this one.
383 * config/s390/s390.opt: Add new options and and enum for the
386 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
388 * lra-constraints.c (match_reload): Unconditionally use
389 gen_lowpart_SUBREG, rather than selecting between that
390 and equivalent gen_rtx_SUBREG code.
392 2018-02-08 Richard Biener <rguenther@suse.de>
394 PR tree-optimization/84233
395 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
396 changed flag instead of boguously re-using phi_inserted.
398 2018-02-08 Martin Jambor <mjambor@suse.cz>
400 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
401 static local variables.
403 2018-02-08 Richard Biener <rguenther@suse.de>
405 PR tree-optimization/84278
406 * tree-vect-stmts.c (vectorizable_store): When looking for
407 smaller vector types to perform grouped strided loads/stores
408 make sure the mode is supported by the target.
409 (vectorizable_load): Likewise.
411 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
413 * config/aarch64/aarch64.c (aarch64_components_for_bb):
414 Increase LDP/STP opportunities by adding adjacent callee-saves.
416 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
418 PR rtl-optimization/84068
419 PR rtl-optimization/83459
420 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
422 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
424 PR tree-optimization/84224
425 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
426 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
429 2018-02-07 Iain Sandoe <iain@codesourcery.com>
432 * config/rs6000/altivec.md (*restore_world): Remove LR use.
433 * config/rs6000/predicates.md (restore_world_operation): Adjust op
434 count, remove one USE.
436 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
438 * doc/install.texi (Configuration): Document the
439 --with-long-double-format={ibm,ieee} PowerPC configuration
443 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
444 Convert from define_expand to be define_insn_and_split. Rework
445 float/double/_Float128 conversions to QI/HI/SImode to work with
446 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
447 conversions to QI/HImode types did a store and then a load to
448 truncate the value. For conversions to VSX registers, don't split
449 the insn, instead emit the code directly. Use the code iterator
450 any_fix to combine signed and unsigned conversions.
451 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
452 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
453 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
454 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
455 (fix_<mode>di2_hw): Likewise.
456 (fixuns_<mode>di2_hw): Likewise.
457 (fix_<mode>si2_hw): Likewise.
458 (fixuns_<mode>si2_hw): Likewise.
459 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
460 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
461 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
462 fix<uns>_trunc<SFDF:mode>si2_p8.
463 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
465 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
466 (fix<uns>_<mode>_mem): Likewise.
467 (fctiw<u>z_<mode>_mem): Likewise.
468 (fix<uns>_<mode>_mem): Likewise.
469 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
470 the register allocator from doing a direct move to the GPRs to do
471 a store, and instead use the ISA 3.0 store byte/half-word from
472 vector register instruction. For IEEE 128-bit floating point,
473 also optimize stores of 32-bit ints.
474 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
476 2018-02-07 Alan Hayward <alan.hayward@arm.com>
478 * genextract.c (push_pathstr_operand): New function to support
480 (walk_rtx): Call push_pathstr_operand.
481 (print_path): Support [a-zA-Z].
483 2018-02-07 Richard Biener <rguenther@suse.de>
485 PR tree-optimization/84037
486 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
487 (cse_and_gimplify_to_preheader): Declare.
488 (vect_get_place_in_interleaving_chain): Likewise.
489 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
491 (_loop_vec_info::~_loop_vec_info): Delete it.
492 (cse_and_gimplify_to_preheader): New function.
493 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
494 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
495 (vectorizable_load): Likewise. For grouped stores always base
496 the IV on the first element.
497 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
498 condition before gimplifying.
500 2018-02-07 Jakub Jelinek <jakub@redhat.com>
502 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
503 *DIV_EXPR and *MOD_EXPR.
505 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
508 * config/i386/i386.c (ix86_option_override_internal): Mask out
509 the CF_SET bit when checking -fcf-protection.
511 2018-02-07 Tom de Vries <tom@codesourcery.com>
514 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
517 2018-02-07 Richard Biener <rguenther@suse.de>
519 PR tree-optimization/84204
520 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
523 PR tree-optimization/84205
524 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
525 special-case isl_ast_op_zdiv_r.
527 PR tree-optimization/84223
528 * graphite-scop-detection.c (gather_bbs::before_dom_children):
529 Only add conditions from within the region.
530 (gather_bbs::after_dom_children): Adjust.
532 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
535 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
536 * config/avr/avr.md: Only post-reload split REG-REG moves if
537 either register is REGERAL_REG_P.
539 2018-02-07 Jakub Jelinek <jakub@redhat.com>
541 PR tree-optimization/84235
542 * tree-ssa-scopedtables.c
543 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
544 if the subtraction is performed in floating point type where NaNs are
545 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
546 build 1. Formatting fix.
548 2018-02-06 Jakub Jelinek <jakub@redhat.com>
551 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
552 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
553 and skip it regardless of bb boundaries. Use CALL_P macro,
554 don't test INSN_P (insn) together with CALL_P or JUMP_P check
555 unnecessarily, formatting fix.
557 2018-02-06 Michael Collison <michael.collison@arm.com>
559 * config/arm/thumb2.md:
560 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
561 (*thumb_mov_notscc): Ditto.
563 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
566 * config/rs6000/rs6000.md (su code attribute): Use "u" for
567 unsigned_fix, not "s".
569 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
571 * configure.ac (gcc_fn_eh_frame_ro): New function.
572 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
573 correct .eh_frame permissions.
574 * configure: Regenerate.
576 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
578 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
581 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
583 * config/rs6000/rs6000.c (rs6000_option_override_internal):
584 Display warning message for -mno-speculate-indirect-jumps.
586 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
588 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
590 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
592 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
594 PR tree-optimization/84225
595 * tree-eh.c (find_trapping_overflow): Only call
596 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
598 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
601 * config/i386/i386.c: Reimplement the check of possible options
602 -mibt/-mshstk conbination. Change error messages.
603 * doc/invoke.texi: Fix a typo: remove extra '='.
605 2018-02-06 Marek Polacek <polacek@redhat.com>
607 PR tree-optimization/84228
608 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
610 2018-02-06 Tamar Christina <tamar.christina@arm.com>
613 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
614 emitted arch directives.
615 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
616 __ARM_FEATURE_COPROC before changing architectures.
618 2018-02-06 Richard Biener <rguenther@suse.de>
620 * config/i386/i386.c (print_reg): Fix typo.
621 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
623 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
625 * configure: Regenerate.
627 2018-02-05 Martin Sebor <msebor@redhat.com>
629 PR tree-optimization/83369
630 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
633 2018-02-05 Martin Liska <mliska@suse.cz>
635 * doc/invoke.texi: Cherry-pick upstream r323995.
637 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
639 * ira.c (ira_init_register_move_cost): Adjust comment.
641 2018-02-05 Martin Liska <mliska@suse.cz>
643 PR gcov-profile/84137
644 * doc/gcov.texi: Fix typo in documentation.
646 2018-02-05 Martin Liska <mliska@suse.cz>
648 PR gcov-profile/83879
649 * doc/gcov.texi: Document necessity of --dynamic-list-data when
650 using dlopen functionality.
652 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
654 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
655 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
656 _mm_maskz_range_ss, _mm_mask_range_round_ss,
657 _mm_maskz_range_round_ss): New intrinsics.
658 (__builtin_ia32_rangesd128_round)
659 (__builtin_ia32_rangess128_round): Remove.
660 (__builtin_ia32_rangesd128_mask_round,
661 __builtin_ia32_rangess128_mask_round): New builtins.
662 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
663 __builtin_ia32_rangess128_round): Remove.
664 (__builtin_ia32_rangesd128_mask_round,
665 __builtin_ia32_rangess128_mask_round): New builtins.
666 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
667 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
668 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
669 "<round_saeonly_constraint>")): Changed to ...
670 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
671 "<round_saeonly_scalar_constraint>")): ... this.
672 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
673 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
674 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
675 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
676 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
678 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
680 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
682 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
683 Remove all values except native, 8540 and 8548.
685 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
687 * config/i386/i386.c (ix86_output_function_return): Pass
688 INVALID_REGNUM, instead of -1, as invalid register number to
689 indirect_thunk_name and output_indirect_thunk.
691 2018-02-02 Julia Koval <julia.koval@intel.com>
693 * config.gcc: Add -march=icelake.
694 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
695 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
696 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
697 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
698 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
699 (processor_target_table): Add icelake.
700 (ix86_option_override_internal): Handle new PTAs.
701 (get_builtin_code_for_version): Handle icelake.
702 (M_INTEL_COREI7_ICELAKE): New.
703 (fold_builtin_cpu): Handle icelake.
704 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
705 * doc/invoke.texi: Add -march=icelake.
707 2018-02-02 Julia Koval <julia.koval@intel.com>
709 * config/i386/i386.c (ix86_option_override_internal): Change flags type
711 * wide-int-bitmask.h: New.
713 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
716 * config/i386/i386.md: Replace Pmode with word_mode in
717 builtin_setjmp_setup and builtin_longjmp to support x32.
719 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
723 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
725 (rs6000_supported_cpu_names): New static variable.
726 (linux_cpu_translation_table): Likewise.
727 (elf_platform) <cpu>: Define new static variable and use it.
728 Translate kernel AT_PLATFORM name to canonical name if needed.
729 Error if platform name is unknown.
731 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
734 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
736 2018-02-01 Jeff Law <law@redhat.com>
739 * config/i386/i386.c (release_scratch_register_on_entry): Add new
740 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
741 the scratch if RELEASE_VIA_POP is false.
742 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
743 If we have to save a temporary register, decrement SIZE appropriately.
744 Pass new arguments to release_scratch_register_on_entry.
745 (ix86_adjust_stack_and_probe): Likewise.
746 (ix86_emit_probe_stack_range): Pass new arguments to
747 release_scratch_register_on_entry.
749 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
751 PR rtl-optimization/84157
752 * combine.c (change_zero_ext): Use REG_P predicate in
753 front of HARD_REGISTER_P predicate.
755 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
757 * config/avr/avr.c (avr_option_override): Move disabling of
758 -fdelete-null-pointer-checks to...
759 * common/config/avr/avr-common.c (avr_option_optimization_table):
762 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
764 PR tree-optimization/81635
765 * tree-data-ref.c (split_constant_offset_1): For types that
766 wrap on overflow, try to use range info to prove that wrapping
769 2018-02-01 Renlin Li <renlin.li@arm.com>
772 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
774 (aarch64_register_move_cost): Likewise.
775 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
777 (REG_CLASS_NAMES): Likewise.
778 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
779 TAILCALL_ADDR_REGS. Remove IP registers.
780 * config/aarch64/aarch64.md (Ucs): Update register constraint.
782 2018-02-01 Richard Biener <rguenther@suse.de>
784 * domwalk.h (dom_walker::dom_walker): Add additional constructor
785 for specifying RPO order and allow NULL for that.
786 * domwalk.c (dom_walker::dom_walker): Likewise.
787 (dom_walker::walk): Handle NULL RPO order.
788 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
790 (rewrite_update_dom_walker): Likewise.
791 (mark_def_dom_walker): Likewise.
793 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
795 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
796 (aarch64_maybe_expand_sve_subreg_move): Declare.
797 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
798 * config/aarch64/predicates.md (aarch64_any_register_operand): New
800 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
801 that are semantically a reverse operation.
802 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
803 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
804 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
806 (aarch64_can_change_mode_class): For big-endian, forbid changes
807 between two SVE modes if they have different element sizes.
809 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
811 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
812 the TImode handling for big-endian targets.
814 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
816 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
817 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
819 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
820 Remove BSWAP handing for big-endian targets and use the form of
821 LD1RQ appropariate for the mode.
823 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
825 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
826 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
829 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
832 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
833 check for operands that need to go through aarch64_sve_reload_be.
835 2018-02-01 Jakub Jelinek <jakub@redhat.com>
837 PR tree-optimization/81661
838 PR tree-optimization/84117
839 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
840 * tree-eh.c: Include gimplify.h.
841 (find_trapping_overflow, replace_trapping_overflow,
842 rewrite_to_non_trapping_overflow): New functions.
843 * tree-vect-loop.c: Include tree-eh.h.
844 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
845 * tree-data-ref.c: Include tree-eh.h.
846 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
848 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
850 PR rtl-optimization/84123
851 * combine.c (change_zero_ext): Check if hard register satisfies
852 can_change_dest_mode before calling gen_lowpart_SUBREG.
854 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
857 * ira.c (ira_init_register_move_cost): Remove assert.
859 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
861 PR rtl-optimization/84071
862 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
863 * doc/tm.texi: Regenerate.
865 2018-01-31 Richard Biener <rguenther@suse.de>
867 PR tree-optimization/84132
868 * tree-data-ref.c (analyze_miv_subscript): Properly
869 check whether evolution_function_is_affine_multivariate_p
870 before calling gcd_of_steps_may_divide_p.
872 2018-01-31 Julia Koval <julia.koval@intel.com>
875 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
876 * config/i386/i386.md (rdpid_rex64) New.
877 (rdpid): Make 32bit only.
879 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
882 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
883 an IDENTIFIER_NODE for FUNCTION_TYPE's.
885 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
888 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
890 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
892 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
894 PR rtl-optimization/84071
895 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
896 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
898 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
900 * config/arc/arc.c (arc_handle_aux_attribute): New function.
901 (arc_attribute_table): Add 'aux' attribute.
902 (arc_in_small_data_p): Consider aux like variables.
903 (arc_is_aux_reg_p): New function.
904 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
905 (arc_get_aux_arg): New function.
906 (prepare_move_operands): Handle aux-register access.
907 (arc_handle_aux_attribute): New function.
908 * doc/extend.texi (ARC Variable attributes): Add subsection.
910 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
912 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
913 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
914 (arc_attribute_table): Add 'uncached' attribute.
915 (arc_print_operand): Print '.di' flag for uncached memory
917 (arc_in_small_data_p): Do not consider for small data the uncached
919 (arc_is_uncached_mem_p): New function.
920 * config/arc/predicates.md (compact_store_memory_operand): Check
921 for uncached memory accesses.
922 (nonvol_nonimm_operand): Likewise.
923 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
925 2018-01-31 Jakub Jelinek <jakub@redhat.com>
928 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
929 falign-loops=): Add Optimization flag.
931 2018-01-30 Jeff Law <law@redhat.com>
934 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
935 INT_REGISTERS_SAVED. Check it prior to calling
936 get_scratch_register_on_entry.
937 (ix86_adjust_stack_and_probe): Similarly.
938 (ix86_emit_probe_stack_range): Similarly.
939 (ix86_expand_prologue): Corresponding changes.
941 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
944 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
945 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
947 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
950 * lra-constraints.c (curr_insn_transform): Process AND in the
953 2018-01-30 Jakub Jelinek <jakub@redhat.com>
955 PR rtl-optimization/83986
956 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
957 dependence against last_pending_memory_flush in addition to
960 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
962 PR tree-optimization/81611
963 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
966 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
968 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
971 2018-01-30 Richard Biener <rguenther@suse.de>
972 Jakub Jelinek <jakub@redhat.com>
974 PR tree-optimization/84111
975 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
976 inner loops added during recursion, as they don't have up-to-date
979 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
982 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
983 (can_inline_edge_by_limits_p): ... here.
984 (can_early_inline_edge_p, check_callers,
985 update_caller_keys, update_callee_keys, recursive_inlining,
986 add_new_edges_to_heap, speculation_useful_p,
987 inline_small_functions,
988 inline_small_functions, flatten_function,
989 inline_to_all_callers_1): Update.
991 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
993 * profile-count.c (profile_count::combine_with_ipa_count): Handle
996 2018-01-30 Richard Biener <rguenther@suse.de>
998 PR tree-optimization/83008
999 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1000 invariant and constant vector uses in stmts when they need
1003 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1006 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1007 * configure: Regenerate.
1009 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1011 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1013 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1014 Use gen_rtx_REG rather than gen_lowpart.
1016 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1018 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1019 rather than 0 when creating partial subregs.
1021 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1023 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1026 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1029 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1030 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1031 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1032 flags. This restores the settings used before the 2017-07-24.
1033 Turning off pre increment/decrement/modify allows IVOPTS to
1034 optimize DF/SF loops where the index is an int.
1036 2018-01-29 Richard Biener <rguenther@suse.de>
1037 Kelvin Nilsen <kelvin@gcc.gnu.org>
1040 * tree-vect-stmts.c (vectorizable_call): Don't call
1041 targetm.vectorize_builtin_md_vectorized_function if callee is
1044 2018-01-22 Carl Love <cel@us.ibm.com>
1046 * doc/extend.tex: Fix typo in second arg in
1047 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1049 2018-01-29 Richard Biener <rguenther@suse.de>
1051 PR tree-optimization/84086
1052 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1053 (flush_ssaname_freelist): When SSA names were released reset
1054 the SCEV hash table.
1056 2018-01-29 Richard Biener <rguenther@suse.de>
1058 PR tree-optimization/84057
1059 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1060 removed paths when removing edges.
1062 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1064 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1065 -mfunction-return=@var{choice}.
1067 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1070 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1072 (layout::print_source_line): Likewise.
1073 (test_get_line_width_without_trailing_whitespace): Add test cases.
1075 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1078 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1081 2018-01-26 Jim Wilson <jimw@sifive.com>
1083 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1085 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1088 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1090 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1091 and CMP + SUB-immediate -> SUBS.
1093 2018-01-26 Martin Sebor <msebor@redhat.com>
1095 PR tree-optimization/83896
1096 * tree-ssa-strlen.c (get_string_len): Rename...
1097 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1098 Avoid assuming length is constant.
1099 (handle_char_store): Use HOST_WIDE_INT for string length.
1101 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1104 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1105 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1107 2018-01-26 Richard Biener <rguenther@suse.de>
1109 PR rtl-optimization/84003
1110 * dse.c (record_store): Only record redundant stores when
1111 the earlier store aliases at least all accesses the later one does.
1113 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1115 PR rtl-optimization/83985
1116 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1117 REG_CFA_RESTORE insns.
1118 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1119 REG_CFA_RESTORE insns here.
1122 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1123 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1125 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1127 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1128 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1129 (arc_init): Likewise.
1130 (arc_override_options): Likewise.
1131 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1133 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1135 * config/arc/arc.h (TARGET_DBNZ): Define.
1136 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1137 properly set the tune attribute.
1138 (dbnz): Use TARGET_DBNZ guard.
1139 * config/arc/arc.opt (mtune): Add core3 option.
1141 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1143 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1144 recognize new pic like addresses.
1145 (arc_delegitimize_address): Clean up.
1147 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1149 * config/arc/arc-arches.def: Option mrf16 valid for all
1151 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1152 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1153 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1154 * config/arc/arc-tables.opt: Regenerate.
1155 * config/arc/arc.c (arc_conditional_register_usage): Handle
1156 reduced register file case.
1157 (arc_file_start): Set must have build attributes.
1158 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1160 * config/arc/arc.opt (mrf16): Add new option.
1161 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1162 * config/arc/genmultilib.awk: Handle new mrf16 option.
1163 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1164 * config/arc/t-multilib: Regenerate.
1165 * doc/invoke.texi (ARC Options): Document mrf16 option.
1167 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1169 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1170 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1171 (arc_attribute_table): Add 'secure_call' attribute.
1172 (arc_print_operand): Print secure call operand.
1173 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1175 (arc_is_secure_call_p): New function. * config/arc/arc.md
1176 (call_i): Add support for sjli instruction.
1177 (call_value_i): Likewise.
1178 * config/arc/constraints.md (Csc): New constraint.
1180 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1181 John Eric Martin <John.Martin@emmicro-us.com>
1183 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1184 * config/arc/arc.c (_arc_jli_section): New struct.
1185 (arc_jli_section): New type.
1186 (rc_jli_sections): New static variable.
1187 (arc_handle_jli_attribute): New function.
1188 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1189 (arc_file_end): New function.
1190 (TARGET_ASM_FILE_END): Define.
1191 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1192 (arc_add_jli_section): New function.
1193 (jli_call_scan): Likewise.
1194 (arc_reorg): Call jli_call_scan.
1195 (arc_output_addsi): Remove 'S' from printing asm operand.
1196 (arc_is_jli_call_p): New function.
1197 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1199 (movhi_insn): Likewise.
1200 (movsi_insn): Likewise.
1201 (movsi_set_cc_insn): Likewise.
1202 (loadqi_update): Likewise.
1203 (load_zeroextendqisi_update): Likewise.
1204 (load_signextendqisi_update): Likewise.
1205 (loadhi_update): Likewise.
1206 (load_zeroextendhisi_update): Likewise.
1207 (load_signextendhisi_update): Likewise.
1208 (loadsi_update): Likewise.
1209 (loadsf_update): Likewise.
1210 (movsicc_insn): Likewise.
1211 (bset_insn): Likewise.
1212 (bxor_insn): Likewise.
1213 (bclr_insn): Likewise.
1214 (bmsk_insn): Likewise.
1215 (bicsi3_insn): Likewise.
1216 (cmpsi_cc_c_insn): Likewise.
1217 (movsi_ne): Likewise.
1218 (movsi_cond_exec): Likewise.
1219 (clrsbsi2): Likewise.
1230 (call_i): Remove 'S' asm letter, add jli instruction.
1231 (call_value_i): Likewise.
1232 * config/arc/arc.op (mjli-always): New option.
1233 * config/arc/constraints.md (Cji): New constraint.
1234 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1236 (subsf3_fpx): Likewise.
1237 (mulsf3_fpx): Likewise.
1238 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1240 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1242 * doc/invoke.texi (ARC): Document mjli-always option.
1244 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1246 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1247 and use incw and decw where possible
1248 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1250 2018-01-26 Richard Biener <rguenther@suse.de>
1252 PR tree-optimization/81082
1253 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1254 association if it requires casting to unsigned.
1255 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1256 from fold_plusminus_mult_expr to catch important cases late when
1257 range info is available.
1259 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1261 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1262 * configure.ac (hidden_linkonce): New test.
1263 * configure: Regenerate.
1264 * config.in: Regenerate.
1266 2018-01-26 Julia Koval <julia.koval@intel.com>
1268 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1269 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1270 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1271 _mm_mask_bitshuffle_epi64_mask): Fix type.
1272 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1273 USI_FTYPE_V4DI_V4DI_USI): Remove.
1274 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1275 __builtin_ia32_vpshufbitqmb256_mask,
1276 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1277 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1278 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1280 2018-01-26 Alan Modra <amodra@gmail.com>
1283 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1284 UNSPEC_VBPERMQ. Sort other unspecs.
1286 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1288 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1290 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1293 * predict.c (drop_profile): Do not push/pop cfun; update also
1295 (handle_missing_profiles): Fix logic looking for zero profiles.
1297 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1300 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1301 on functions with #pragma omp declare simd or functions with simd
1303 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1304 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1305 Remove trailing \n from warning_at calls.
1307 2018-01-25 Tom de Vries <tom@codesourcery.com>
1310 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1311 for neutered workers.
1313 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1316 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1317 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1319 2018-01-24 Jeff Law <law@redhat.com>
1322 * i386.c (get_probe_interval): Move to earlier point.
1323 (ix86_compute_frame_layout): If -fstack-clash-protection and
1324 the frame is larger than the probe interval, then use pushes
1325 to save registers rather than reg->mem moves.
1326 (ix86_expand_prologue): Remove conditional for int_registers_saved
1329 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1332 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1333 min/max for never referenced object.
1335 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1338 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1340 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1341 attributes from DECL_ATTRIBUTES (decl) without affecting
1342 DECL_ATTRIBUTES (current_function_decl).
1343 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1344 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1346 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1348 PR tree-optimization/83979
1349 * fold-const.c (fold_comparison): Use constant_boolean_node
1350 instead of boolean_{true,false}_node.
1352 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1354 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1357 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1359 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1360 Simplify the clause that sets the length attribute.
1361 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1362 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1363 clause that sets the length attribute.
1364 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1366 2018-01-24 Tom de Vries <tom@codesourcery.com>
1369 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1370 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1371 Add strict parameter.
1372 (prevent_branch_around_nothing): Insert dummy insn between branch to
1373 label and label with no ptx insn inbetween.
1374 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1376 2018-01-24 Tom de Vries <tom@codesourcery.com>
1379 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1380 for neutered threads in warp.
1381 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1383 2018-01-24 Richard Biener <rguenther@suse.de>
1385 PR tree-optimization/83176
1386 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1389 2018-01-24 Richard Biener <rguenther@suse.de>
1391 PR tree-optimization/82819
1392 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1393 code generating pluses that are no-ops in the target precision.
1395 2018-01-24 Richard Biener <rguenther@suse.de>
1398 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1400 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1402 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1403 to merge probabilities.
1404 * predict.c (probably_never_executed): Also mark as cold functions
1405 with global 0 profile and guessed local profile.
1406 * profile-count.c (profile_probability::combine_with_count): New
1408 * profile-count.h (profile_probability::operator*,
1409 profile_probability::operator*=, profile_probability::operator/,
1410 profile_probability::operator/=): Reduce precision to adjusted
1411 and set value to guessed on contradictory divisions.
1412 (profile_probability::combine_with_freq): Remove.
1413 (profile_probability::combine_wiht_count): Declare.
1414 (profile_count::force_nonzero):: Set to adjusted.
1415 (profile_count::probability_in):: Set quality to adjusted.
1416 * tree-ssa-tail-merge.c (replace_block_by): Use
1419 2018-01-23 Andrew Waterman <andrew@sifive.com>
1420 Jim Wilson <jimw@sifive.com>
1422 * config/riscv/riscv.c (riscv_stack_boundary): New.
1423 (riscv_option_override): Set riscv_stack_boundary. Handle
1424 riscv_preferred_stack_boundary_arg.
1425 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1426 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1427 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1428 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1429 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1430 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1432 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1435 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1436 of struct ix86_frame.
1437 (ix86_expand_epilogue): Likewise. Add a local variable for
1438 the reg_save_offset field in struct ix86_frame.
1440 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1442 PR tree-optimization/82604
1443 * tree-loop-distribution.c (enum partition_kind): New enum item
1444 PKIND_PARTIAL_MEMSET.
1445 (partition_builtin_p): Support above new enum item.
1446 (generate_code_for_partition): Ditto.
1447 (compute_access_range): Differentiate cases that equality can be
1448 proven at all loops, the innermost loops or no loops.
1449 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1450 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1451 (finalize_partitions, distribute_loop): Don't fuse partition of
1452 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1453 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1456 2018-01-23 Martin Liska <mliska@suse.cz>
1458 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1459 order to ignore the predictor.
1460 (PRED_POLYMORPHIC_CALL): Likewise.
1461 (PRED_RECURSIVE_CALL): Likewise.
1463 2018-01-23 Martin Liska <mliska@suse.cz>
1465 * tree-profile.c (tree_profiling): Print function header to
1466 aware reader which function we are working on.
1467 * value-prof.c (gimple_find_values_to_profile): Do not print
1468 not interesting value histograms.
1470 2018-01-23 Martin Liska <mliska@suse.cz>
1472 * profile-count.h (enum profile_quality): Add
1473 profile_uninitialized as the first value. Do not number values
1474 as they are zero based.
1475 (profile_count::verify): Update sanity check.
1476 (profile_probability::verify): Likewise.
1478 2018-01-23 Nathan Sidwell <nathan@acm.org>
1480 * doc/invoke.texi (ffor-scope): Deprecate.
1482 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1484 PR tree-optimization/83510
1485 * domwalk.c (set_all_edges_as_executable): New function.
1486 (dom_walker::dom_walker): Convert bool param
1487 "skip_unreachable_blocks" to enum reachability. Move setup of
1488 edge flags to set_all_edges_as_executable and only do it when
1489 reachability is REACHABLE_BLOCKS.
1490 * domwalk.h (enum dom_walker::reachability): New enum.
1491 (dom_walker::dom_walker): Convert bool param
1492 "skip_unreachable_blocks" to enum reachability.
1493 (set_all_edges_as_executable): New decl.
1494 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1495 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1497 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1498 but converting true to REACHABLE_BLOCKS.
1499 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1501 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1502 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1503 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1505 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1506 if check_all_array_refs will be called.
1508 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1510 * tree.c (selftest::test_location_wrappers): Add more test
1513 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1515 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1516 (selftest::test_bit_in_range): Likewise.
1518 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1521 * doc/sourcebuild.texi (vect_float): Say that the selector
1522 only describes the situation when -funsafe-math-optimizations is on.
1523 (vect_float_strict): Document.
1525 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1527 PR tree-optimization/83965
1528 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1529 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1530 instead of checking only for a reduction.
1531 (vect_recog_widen_sum_pattern): Likewise.
1533 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1535 * predict.c (probably_never_executed): Only use precise profile info.
1536 (compute_function_frequency): Skip after inlining hack since we now
1537 have quality checking.
1539 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1541 * profile-count.h (profile_probability::very_unlikely,
1542 profile_probability::unlikely, profile_probability::even): Set
1543 precision to guessed.
1545 2018-01-23 Richard Biener <rguenther@suse.de>
1547 PR tree-optimization/83963
1548 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1549 Properly terminate dominator walk when crossing the exit edge not
1550 when visiting its source block.
1552 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1555 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1556 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1558 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1560 PR tree-optimization/83957
1561 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1562 semicolon after for body surrounded by braces.
1564 PR tree-optimization/83081
1565 * profile-count.h (profile_probability::split): New method.
1566 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1567 Use profile_probability::split.
1568 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1569 when splitting a single conditional jump into 2.
1571 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1573 PR tree-optimization/69452
1574 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1577 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1579 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1580 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1581 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1583 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1585 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1586 * config/rl78/rl78.md: New define_expand "movdi"
1587 * config/rl78/rl78.c: New function definition rl78_split_movdi
1589 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1592 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1594 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1595 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1596 128-bit to produce an UNSPEC move to get the double word with the
1597 signbit and then a shift directly to do signbit.
1598 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1599 implementation with a new version that just does either a direct
1600 move or a regular move. Move memory interface to separate insns.
1601 Move insns so they are next to the expander.
1602 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1603 with signbit move. Split big and little endian case.
1604 (signbit<mode>2_dm_mem_le): Likewise.
1605 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1606 (signbit<mode>2_dm2): Likewise.
1608 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1610 * config/rl78/rl78.md: New define_expand "anddi3".
1612 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1614 * config/rl78/rl78.md: New define_expand "umindi3".
1616 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1618 * config/rl78/rl78.md: New define_expand "smindi3".
1620 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1622 * config/rl78/rl78.md: New define_expand "smaxdi3".
1624 2018-01-22 Carl Love <cel@us.ibm.com>
1626 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1627 LVX_V1TI): Add macro expansion.
1628 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1629 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1630 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1631 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1632 Change check to determine if the instruction is a byte reversing
1633 entry. Fix typo in comment.
1634 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1635 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1636 Add def_builtin calls for new builtins.
1637 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1638 Add define_insn expansion.
1640 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1642 * config/rl78/rl78.md: New define_expand "umaxdi3".
1644 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1646 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1647 for non-QImode registers
1649 2018-01-22 Richard Biener <rguenther@suse.de>
1651 PR tree-optimization/83963
1652 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1653 including the loop exit block.
1654 (scop_detection::merge_sese): Likewise.
1655 (scop_detection::add_scop): Do it here instead.
1657 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1659 * doc/sourcebuild.texi (arm_softfloat): Document.
1661 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1664 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1665 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1666 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1668 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1669 David Edelsohn <dje.gcc@gmail.com>
1672 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1673 Change "crset eq" to "crset 2".
1674 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1675 (*call_indirect_aix<mode>_nospec): Likewise.
1676 (*call_value_indirect_aix<mode>_nospec): Likewise.
1677 (*call_indirect_elfv2<mode>_nospec): Likewise.
1678 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1679 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1680 change assembly output from . to $.
1681 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1682 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1683 (*tablejump<mode>_internal1_nospec): Likewise.
1685 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1688 * config/sh/sh_optimize_sett_clrt.cc:
1689 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1691 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1693 PR tree-optimization/83940
1694 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1695 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1696 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1697 use it to pass back the definition type.
1698 (vect_check_store_rhs): Likewise rhs_dt_out.
1699 (vect_build_gather_load_calls): Add a mask_dt argument and use
1700 it instead of a call to vect_is_simple_use.
1701 (vectorizable_store): Update calls to vect_check_load_store_mask
1702 and vect_check_store_rhs. Use the dt returned by the latter instead
1703 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1704 instead of calls to vect_is_simple_use. Pass the scalar rather
1705 than the vector operand to vect_is_simple_use when handling
1706 second and subsequent copies of an rhs value.
1707 (vectorizable_load): Update calls to vect_check_load_store_mask
1708 and vect_build_gather_load_calls. Use the cached mask_dt and
1709 gs_info.offset_dt instead of calls to vect_is_simple_use.
1711 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1714 * tree-emutls.c: Include gimplify.h.
1715 (lower_emutls_2): New function.
1716 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1717 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1718 it before further processing.
1721 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1722 UINTVAL (trueop1) instead of INTVAL (op1).
1724 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1728 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1729 INCOMING_FRAME_SP_OFFSET if not defined.
1730 (scan_trace): Add ENTRY argument. If true and
1731 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1732 emit a note to adjust the CFA offset.
1733 (create_cfi_notes): Adjust scan_trace callers.
1734 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1735 INCOMING_FRAME_SP_OFFSET in the CIE.
1736 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1737 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1739 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1740 * doc/tm.texi: Regenerated.
1742 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1744 PR rtl-optimization/83147
1745 * lra-constraints.c (remove_inheritance_pseudos): Use
1746 lra_substitute_pseudo_within_insn.
1748 2018-01-19 Tom de Vries <tom@codesourcery.com>
1749 Cesar Philippidis <cesar@codesourcery.com>
1752 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1754 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1757 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1758 spaces for function labels.
1760 2018-01-19 Martin Liska <mliska@suse.cz>
1762 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1763 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1764 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1765 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1766 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1767 (PRED_CONST_RETURN): Change from 69 to 65.
1768 (PRED_NULL_RETURN): Change from 91 to 71.
1769 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1770 (PRED_LOOP_GUARD): Change from 66 to 73.
1772 2018-01-19 Martin Liska <mliska@suse.cz>
1774 * predict.c (predict_insn_def): Add new assert.
1775 (struct branch_predictor): Change type to signed integer.
1776 (test_prediction_value_range): Amend test to cover
1778 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1779 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1780 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1781 (PRED_LOOP_IV_COMPARE): Likewise.
1782 * predict.h (PROB_UNINITIALIZED): Define new constant.
1784 2018-01-19 Martin Liska <mliska@suse.cz>
1786 * predict.c (dump_prediction): Add new format for
1787 analyze_brprob.py script which is enabled with -details
1789 * profile-count.h (precise_p): New function.
1791 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1793 PR tree-optimization/83922
1794 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1795 there are no statements that need masking.
1796 (vect_active_double_reduction_p): New function.
1797 (vect_analyze_loop_operations): Use it when handling phis that
1798 are not in the loop header.
1800 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1802 PR tree-optimization/83914
1803 * tree-vect-loop.c (vectorizable_induction): Don't convert
1804 init_expr or apply the peeling adjustment for inductions
1805 that are nested within the vectorized loop.
1807 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1809 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1812 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1816 * function.h (gimplify_parameters): Add gimple_seq * argument.
1817 * function.c: Include gimple.h and options.h.
1818 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1819 for the added local temporaries if needed.
1820 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1821 if there are any parameter cleanups, wrap whole body into a
1822 try/finally with the cleanups.
1824 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1827 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1828 Use GET_MODE_CLASS for scalar floating point.
1830 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1834 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1835 Fix call of call_cgraph_insertion_hooks.
1837 2018-01-18 Martin Sebor <msebor@redhat.com>
1839 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1841 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1844 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1847 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1850 * common.opt: (-ffile-prefix-map): New option.
1851 * opts.c (common_handle_option): Defer it.
1852 * opts-global.c (handle_common_deferred_options): Handle it.
1853 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1854 * file-prefix-map.h: New file.
1855 (remap_debug_filename, add_debug_prefix_map): ...here.
1856 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1857 * final.c (debug_prefix_map, add_debug_prefix_map
1858 remap_debug_filename): Move to...
1859 * file-prefix-map.c: New file.
1860 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1861 generalize, get rid of alloca(), use strrchr() instead of strchr().
1862 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1863 Implement in terms of add_prefix_map().
1864 (remap_macro_filename, remap_debug_filename): Implement in term of
1866 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1867 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1868 * dbxout.c: Include file-prefix-map.h.
1869 * varasm.c: Likewise.
1870 * vmsdbgout.c: Likewise.
1871 * xcoffout.c: Likewise.
1872 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1873 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1874 * doc/invoke.texi (-ffile-prefix-map): Document.
1875 (-fdebug-prefix-map): Update description.
1877 2018-01-18 Martin Liska <mliska@suse.cz>
1879 * config/i386/i386.c (indirect_thunk_name): Document that also
1881 (output_indirect_thunk): Document why both instructions
1882 (pause and lfence) are generated.
1884 2018-01-18 Richard Biener <rguenther@suse.de>
1886 PR tree-optimization/83887
1887 * graphite-scop-detection.c
1888 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1889 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1890 (scop_detection::merge_sese): Re-implement with a flood-fill
1891 algorithm that properly finds a SESE region if it exists.
1893 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1896 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1897 pointer_diff optimizations use view_convert instead of convert.
1899 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1901 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1902 Generate different code for -mno-speculate-indirect-jumps.
1903 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1904 (*call_indirect_aix<mode>): Disable for
1905 -mno-speculate-indirect-jumps.
1906 (*call_indirect_aix<mode>_nospec): New define_insn.
1907 (*call_value_indirect_aix<mode>): Disable for
1908 -mno-speculate-indirect-jumps.
1909 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1910 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1911 -mno-speculate-indirect-jumps.
1912 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1914 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1916 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1917 long double type, set the flags for noting the default long double
1918 type, even if we don't pass or return a long double type.
1920 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1923 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1926 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1928 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1929 support for merge[hl].
1930 (fold_mergehl_helper): New helper function.
1931 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1932 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1933 (altivec_vmrglw_direct): Add xxmrglw insn.
1935 2018-01-17 Andrew Waterman <andrew@sifive.com>
1937 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1938 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1940 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1943 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1944 call the lto_location_cache before reading the
1945 DECL_SOURCE_LOCATION of the types.
1947 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1948 Richard Sandiford <richard.sandiford@linaro.org>
1950 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1951 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1952 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1953 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1954 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1956 * config/aarch64/constraints.md (aarch64_movti_operand):
1958 * config/aarch64/predicates.md (Uti): Add new constraint.
1960 2018-01-17 Carl Love <cel@us.ibm.com>
1961 * config/rs6000/vsx.md (define_expand xl_len_r,
1962 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1963 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1965 (define_expand, define_insn): Move the shift left from the
1966 define_insn to the define_expand for lxvl and stxvl instructions.
1967 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1968 and XL_LEN_R definitions to PURE.
1970 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1972 * config/i386/i386.c (indirect_thunk_name): Declare regno
1973 as unsigned int. Compare regno with INVALID_REGNUM.
1974 (output_indirect_thunk): Ditto.
1975 (output_indirect_thunk_function): Ditto.
1976 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1977 in the call to output_indirect_thunk_function.
1979 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1982 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1983 rather than the size of inner_type to determine the stack slot size
1984 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1986 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1989 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1992 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1994 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1995 endian Linux systems to optionally enable multilibs for selecting
1996 the long double type if the user configured an explicit type.
1997 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1998 have no long double multilibs if not defined.
1999 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2000 warn if the user used -mabi={ieee,ibm}longdouble and we built
2001 multilibs for long double.
2002 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2003 appropriate multilib option.
2004 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2006 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2007 for building long double multilibs.
2008 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2010 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2012 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2015 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2017 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2020 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2023 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2026 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2028 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2029 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2031 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2033 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2034 different rtl trees depending on TARGET_64BIT.
2035 (rs6000_gen_lvx): Likewise.
2037 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2039 * config/visium/visium.md (nop): Tweak comment.
2040 (hazard_nop): Likewise.
2042 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2044 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2045 -mspeculate-indirect-jumps.
2046 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2047 for -mno-speculate-indirect-jumps.
2048 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2049 (*call_value_indirect_elfv2<mode>): Disable for
2050 -mno-speculate-indirect-jumps.
2051 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2052 (indirect_jump): Emit different RTL for
2053 -mno-speculate-indirect-jumps.
2054 (*indirect_jump<mode>): Disable for
2055 -mno-speculate-indirect-jumps.
2056 (*indirect_jump<mode>_nospec): New define_insn.
2057 (tablejump): Emit different RTL for
2058 -mno-speculate-indirect-jumps.
2059 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2060 (tablejumpsi_nospec): New define_expand.
2061 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2062 (tablejumpdi_nospec): New define_expand.
2063 (*tablejump<mode>_internal1): Disable for
2064 -mno-speculate-indirect-jumps.
2065 (*tablejump<mode>_internal1_nospec): New define_insn.
2066 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2069 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2071 * caller-save.c (insert_save): Drop unnecessary parameter. All
2074 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2075 Richard Biener <rguenth@suse.de>
2078 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2079 return early, inline manually is_gimple_sizepos. Make sure if we
2080 call gimplify_expr we don't end up with a gimple constant.
2081 * tree.c (variably_modified_type_p): Don't return true for
2082 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2083 * gimplify.h (is_gimple_sizepos): Remove.
2085 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2087 PR tree-optimization/83857
2088 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2089 vectorizable_live_operation for pure SLP statements.
2090 (vectorizable_live_operation): Handle PHIs.
2092 2018-01-16 Richard Biener <rguenther@suse.de>
2094 PR tree-optimization/83867
2095 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2096 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2098 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2101 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2102 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2103 If off is not INTEGER_CST, issue a may not be aligned warning
2104 rather than isn't aligned. Use isn%'t rather than isn't.
2105 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2107 <case MULT_EXPR>: Improve the case when bottom and one of the
2108 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2109 operand, in that case check if the other operand is multiple of
2110 bottom divided by the INTEGER_CST operand.
2112 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2115 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2116 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2117 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2118 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2119 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2120 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2121 (pa_function_arg_size): New function.
2123 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2125 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2126 in a separate statement.
2128 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2130 PR tree-optimization/83847
2131 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2132 group gathers and scatters.
2134 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2136 PR rtl-optimization/86620
2137 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2139 PR rtl-optimization/83213
2140 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2141 to last if both are JUMP_INSNs.
2143 PR tree-optimization/83843
2144 * gimple-ssa-store-merging.c
2145 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2146 store_immediate_info for bswap/nop orig_stores.
2148 2018-01-15 Andrew Waterman <andrew@sifive.com>
2150 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2152 <UDIV>: Increase cost if !TARGET_DIV.
2154 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2156 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2157 (define_attr "cr_logical_3op"): New.
2158 (cceq_ior_compare): Adjust.
2159 (cceq_ior_compare_complement): Adjust.
2160 (*cceq_rev_compare): Adjust.
2161 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2162 (is_cracked_insn): Adjust.
2163 (insn_must_be_first_in_group): Adjust.
2164 * config/rs6000/40x.md: Adjust.
2165 * config/rs6000/440.md: Adjust.
2166 * config/rs6000/476.md: Adjust.
2167 * config/rs6000/601.md: Adjust.
2168 * config/rs6000/603.md: Adjust.
2169 * config/rs6000/6xx.md: Adjust.
2170 * config/rs6000/7450.md: Adjust.
2171 * config/rs6000/7xx.md: Adjust.
2172 * config/rs6000/8540.md: Adjust.
2173 * config/rs6000/cell.md: Adjust.
2174 * config/rs6000/e300c2c3.md: Adjust.
2175 * config/rs6000/e500mc.md: Adjust.
2176 * config/rs6000/e500mc64.md: Adjust.
2177 * config/rs6000/e5500.md: Adjust.
2178 * config/rs6000/e6500.md: Adjust.
2179 * config/rs6000/mpc.md: Adjust.
2180 * config/rs6000/power4.md: Adjust.
2181 * config/rs6000/power5.md: Adjust.
2182 * config/rs6000/power6.md: Adjust.
2183 * config/rs6000/power7.md: Adjust.
2184 * config/rs6000/power8.md: Adjust.
2185 * config/rs6000/power9.md: Adjust.
2186 * config/rs6000/rs64.md: Adjust.
2187 * config/rs6000/titan.md: Adjust.
2189 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2191 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2192 ix86_indirect_branch_register logic.
2194 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2196 * config/i386/constraints.md (Bs): Update
2197 ix86_indirect_branch_register check. Don't check
2198 ix86_indirect_branch_register with GOT_memory_operand.
2200 * config/i386/predicates.md (GOT_memory_operand): Don't check
2201 ix86_indirect_branch_register here.
2202 (GOT32_symbol_operand): Likewise.
2204 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2206 * config/i386/predicates.md (constant_call_address_operand):
2207 Rewrite ix86_indirect_branch_register logic.
2208 (sibcall_insn_operand): Likewise.
2210 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2212 * config/i386/constraints.md (Bs): Replace
2213 ix86_indirect_branch_thunk_register with
2214 ix86_indirect_branch_register.
2216 * config/i386/i386.md (indirect_jump): Likewise.
2217 (tablejump): Likewise.
2218 (*sibcall_memory): Likewise.
2219 (*sibcall_value_memory): Likewise.
2220 Peepholes of indirect call and jump via memory: Likewise.
2221 * config/i386/i386.opt: Likewise.
2222 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2223 (GOT_memory_operand): Likewise.
2224 (call_insn_operand): Likewise.
2225 (sibcall_insn_operand): Likewise.
2226 (GOT32_symbol_operand): Likewise.
2228 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2231 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2232 type rather than type addr's type points to.
2233 (expand_omp_atomic_mutex): Likewise.
2234 (expand_omp_atomic): Likewise.
2236 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2239 * config/i386/i386.c (output_indirect_thunk_function): Use
2240 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2241 for __x86_return_thunk.
2243 2018-01-15 Richard Biener <rguenther@suse.de>
2246 * expmed.c (extract_bit_field_1): Fix typo.
2248 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2251 * config/arm/iterators.md (VF): New mode iterator.
2252 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2253 Remove integer-related logic from pattern.
2254 (neon_vabd<mode>_3): Likewise.
2256 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2259 * common.opt (fstrict-overflow): No longer an alias.
2260 (fwrapv-pointer): New option.
2261 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2262 also for pointer types based on flag_wrapv_pointer.
2263 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2264 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2265 opts->x_flag_wrapv got set.
2266 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2267 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2268 POINTER_TYPE_OVERFLOW_UNDEFINED.
2269 * match.pd: Likewise in address comparison pattern.
2270 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2272 2018-01-15 Richard Biener <rguenther@suse.de>
2275 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2276 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2277 Reset type names to their identifier if their TYPE_DECL doesn't
2278 have linkage (and thus is used for ODR and devirt).
2279 (save_debug_info_for_decl): Remove.
2280 (save_debug_info_for_type): Likewise.
2281 (add_tree_to_fld_list): Adjust.
2282 * tree-pretty-print.c (dump_generic_node): Make dumping of
2283 type names more robust.
2285 2018-01-15 Richard Biener <rguenther@suse.de>
2287 * BASE-VER: Bump to 8.0.1.
2289 2018-01-14 Martin Sebor <msebor@redhat.com>
2292 * builtins.c (check_access): Avoid warning when the no-warning bit
2295 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2297 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2298 * ira-color (allocno_hard_regs_compare): Likewise.
2300 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2303 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2304 Use .pushsection/.popsection.
2306 2018-01-14 Martin Sebor <msebor@redhat.com>
2309 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2311 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2313 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2314 entry from extra_headers.
2315 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2316 extra_headers, make the list bitwise identical to the i?86-*-* one.
2318 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2320 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2321 -mcmodel=large with -mindirect-branch=thunk,
2322 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2323 -mfunction-return=thunk-extern.
2324 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2325 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2326 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2328 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2330 * config/i386/i386.c (print_reg): Print the name of the full
2331 integer register without '%'.
2332 (ix86_print_operand): Handle 'V'.
2333 * doc/extend.texi: Document 'V' modifier.
2335 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2337 * config/i386/constraints.md (Bs): Disallow memory operand for
2338 -mindirect-branch-register.
2340 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2341 (GOT_memory_operand): Likewise.
2342 (call_insn_operand): Likewise.
2343 (sibcall_insn_operand): Likewise.
2344 (GOT32_symbol_operand): Likewise.
2345 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2346 for -mindirect-branch-register.
2347 (tablejump): Likewise.
2348 (*sibcall_memory): Likewise.
2349 (*sibcall_value_memory): Likewise.
2350 Disallow peepholes of indirect call and jump via memory for
2351 -mindirect-branch-register.
2352 (*call_pop): Replace m with Bw.
2353 (*call_value_pop): Likewise.
2354 (*sibcall_pop_memory): Replace m with Bs.
2355 * config/i386/i386.opt (mindirect-branch-register): New option.
2356 * doc/invoke.texi: Document -mindirect-branch-register option.
2358 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2360 * config/i386/i386-protos.h (ix86_output_function_return): New.
2361 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2362 set function_return_type.
2363 (indirect_thunk_name): Add ret_p to indicate thunk for function
2365 (output_indirect_thunk_function): Pass false to
2366 indirect_thunk_name.
2367 (ix86_output_indirect_branch_via_reg): Likewise.
2368 (ix86_output_indirect_branch_via_push): Likewise.
2369 (output_indirect_thunk_function): Create alias for function
2370 return thunk if regno < 0.
2371 (ix86_output_function_return): New function.
2372 (ix86_handle_fndecl_attribute): Handle function_return.
2373 (ix86_attribute_table): Add function_return.
2374 * config/i386/i386.h (machine_function): Add
2375 function_return_type.
2376 * config/i386/i386.md (simple_return_internal): Use
2377 ix86_output_function_return.
2378 (simple_return_internal_long): Likewise.
2379 * config/i386/i386.opt (mfunction-return=): New option.
2380 (indirect_branch): Mention -mfunction-return=.
2381 * doc/extend.texi: Document function_return function attribute.
2382 * doc/invoke.texi: Document -mfunction-return= option.
2384 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2386 * config/i386/i386-opts.h (indirect_branch): New.
2387 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2388 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2389 with local indirect jump when converting indirect call and jump.
2390 (ix86_set_indirect_branch_type): New.
2391 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2392 (indirectlabelno): New.
2393 (indirect_thunk_needed): Likewise.
2394 (indirect_thunk_bnd_needed): Likewise.
2395 (indirect_thunks_used): Likewise.
2396 (indirect_thunks_bnd_used): Likewise.
2397 (INDIRECT_LABEL): Likewise.
2398 (indirect_thunk_name): Likewise.
2399 (output_indirect_thunk): Likewise.
2400 (output_indirect_thunk_function): Likewise.
2401 (ix86_output_indirect_branch_via_reg): Likewise.
2402 (ix86_output_indirect_branch_via_push): Likewise.
2403 (ix86_output_indirect_branch): Likewise.
2404 (ix86_output_indirect_jmp): Likewise.
2405 (ix86_code_end): Call output_indirect_thunk_function if needed.
2406 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2408 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2409 (ix86_attribute_table): Add indirect_branch.
2410 * config/i386/i386.h (machine_function): Add indirect_branch_type
2411 and has_local_indirect_jump.
2412 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2414 (tablejump): Likewise.
2415 (*indirect_jump): Use ix86_output_indirect_jmp.
2416 (*tablejump_1): Likewise.
2417 (simple_return_indirect_internal): Likewise.
2418 * config/i386/i386.opt (mindirect-branch=): New option.
2419 (indirect_branch): New.
2422 (thunk-inline): Likewise.
2423 (thunk-extern): Likewise.
2424 * doc/extend.texi: Document indirect_branch function attribute.
2425 * doc/invoke.texi: Document -mindirect-branch= option.
2427 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2430 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2432 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2434 * ipa-inline.c (want_inline_small_function_p): Return false if
2435 inlining has already failed with CIF_FINAL_ERROR.
2436 (update_caller_keys): Call want_inline_small_function_p before
2438 (update_callee_keys): Likewise.
2440 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2442 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2444 (rs6000_quadword_masked_address_p): Likewise.
2445 (quad_aligned_load_p): Likewise.
2446 (quad_aligned_store_p): Likewise.
2447 (const_load_sequence_p): Add comment to describe the outer-most loop.
2448 (mimic_memory_attributes_and_flags): New function.
2449 (rs6000_gen_stvx): Likewise.
2450 (replace_swapped_aligned_store): Likewise.
2451 (rs6000_gen_lvx): Likewise.
2452 (replace_swapped_aligned_load): Likewise.
2453 (replace_swapped_load_constant): Capitalize argument name in
2454 comment describing this function.
2455 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2456 and stores that access quad-word aligned addresses and replace
2457 with stvx or lvx instructions when appropriate.
2458 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2459 New function prototype.
2460 (rs6000_quadword_masked_address_p): Likewise.
2461 (rs6000_gen_lvx): Likewise.
2462 (rs6000_gen_stvx): Likewise.
2463 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2464 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2465 when memory address is aligned.
2466 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2467 this split to select lvx instruction when memory address is aligned.
2468 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2469 instruction when memory address is aligned.
2470 (*vsx_le_perm_load_v16qi): Likewise.
2471 (four unnamed splitters): Modify to select the stvx instruction
2472 when memory is aligned.
2474 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2476 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2477 which appears in the queue multiple times.
2479 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2480 Alan Hayward <alan.hayward@arm.com>
2481 David Sherwood <david.sherwood@arm.com>
2483 * tree-vectorizer.h (vec_lower_bound): New structure.
2484 (_loop_vec_info): Add check_nonzero and lower_bounds.
2485 (LOOP_VINFO_CHECK_NONZERO): New macro.
2486 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2487 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2488 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2489 fields. Make seg_len the distance travelled, not including the
2491 (dr_direction_indicator): Declare.
2492 (dr_zero_step_indicator): Likewise.
2493 (dr_known_forward_stride_p): Likewise.
2494 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2496 (runtime_alias_check_p): Allow runtime alias checks with
2498 (operator ==): Compare access_size and align.
2499 (prune_runtime_alias_test_list): Rework for new distinction between
2500 the access_size and seg_len.
2501 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2503 (get_segment_min_max): New function.
2504 (create_intersect_range_checks): Use it.
2505 (dr_step_indicator): New function.
2506 (dr_direction_indicator): Likewise.
2507 (dr_zero_step_indicator): Likewise.
2508 (dr_known_forward_stride_p): Likewise.
2509 * tree-loop-distribution.c (data_ref_segment_size): Return
2510 DR_STEP * (niters - 1).
2511 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2513 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2514 (vect_preserves_scalar_order_p): New function, split out from...
2515 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2516 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2517 (vect_vfa_access_size): New function.
2518 (vect_vfa_align): Likewise.
2519 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2520 (dump_lower_bound): New function.
2521 (vect_check_lower_bound): Likewise.
2522 (vect_small_gap_p): Likewise.
2523 (vectorizable_with_step_bound_p): Likewise.
2524 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2525 depencies if the vectorization factor is 1. Convert the checks
2526 for nonzero steps into checks on the bounds of DR_STEP. Try using
2527 a bunds check for variable steps if the minimum required step is
2528 relatively small. Update calls to the dr_with_seg_len
2529 constructor and to vect_compile_time_alias.
2530 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2532 (vect_loop_versioning): Call it.
2533 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2535 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2537 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2538 Alan Hayward <alan.hayward@arm.com>
2539 David Sherwood <david.sherwood@arm.com>
2541 * doc/sourcebuild.texi (vect_scatter_store): Document.
2542 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2544 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2546 * genopinit.c (main): Add supports_vec_scatter_store and
2547 supports_vec_scatter_store_cached to target_optabs.
2548 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2549 IFN_MASK_SCATTER_STORE.
2550 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2552 * internal-fn.h (internal_store_fn_p): Declare.
2553 (internal_fn_stored_value_index): Likewise.
2554 * internal-fn.c (scatter_store_direct): New macro.
2555 (expand_scatter_store_optab_fn): New function.
2556 (direct_scatter_store_optab_supported_p): New macro.
2557 (internal_store_fn_p): New function.
2558 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2559 IFN_MASK_SCATTER_STORE.
2560 (internal_fn_mask_index): Likewise.
2561 (internal_fn_stored_value_index): New function.
2562 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2564 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2565 * optabs-query.c (supports_vec_scatter_store_p): New function.
2566 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2567 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2568 true for scatter stores.
2569 (vect_gather_scatter_fn_p): Handle scatter stores too.
2570 (vect_check_gather_scatter): Consider using scatter stores if
2571 supports_vec_scatter_store_p.
2572 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2574 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2575 internal_fn_stored_value_index.
2576 (check_load_store_masking): Handle scatter stores too.
2577 (vect_get_store_rhs): Make public.
2578 (vectorizable_call): Use internal_store_fn_p.
2579 (vectorizable_store): Handle scatter store internal functions.
2580 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2581 when deciding whether the end of the group has been reached.
2582 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2583 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2584 (mask_scatter_store<mode>): New insns.
2586 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2587 Alan Hayward <alan.hayward@arm.com>
2588 David Sherwood <david.sherwood@arm.com>
2590 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2591 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2592 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2594 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2595 Use vect_truncate_gather_scatter_offset if we can't treat the
2596 operation as a normal gather load or scatter store.
2597 (get_group_load_store_type): Take the gather_scatter_info
2598 as argument. Try using a gather load or scatter store for
2599 single-element groups.
2600 (get_load_store_type): Update calls to get_group_load_store_type
2601 and vect_use_strided_gather_scatters_p.
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_create_data_ref_ptr): Take an extra
2608 optional tree argument.
2609 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2611 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2612 but continue to use the current value as a fallback.
2613 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2614 to compare the updates.
2615 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2616 (get_load_store_type): Use it when handling a strided access.
2617 (vect_get_strided_load_store_ops): New function.
2618 (vect_get_data_ptr_increment): Likewise.
2619 (vectorizable_load): Handle strided gather loads. Always pass
2620 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2622 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2623 Alan Hayward <alan.hayward@arm.com>
2624 David Sherwood <david.sherwood@arm.com>
2626 * doc/md.texi (gather_load@var{m}): Document.
2627 (mask_gather_load@var{m}): Likewise.
2628 * genopinit.c (main): Add supports_vec_gather_load and
2629 supports_vec_gather_load_cached to target_optabs.
2630 * optabs-tree.c (init_tree_optimization_optabs): Use
2631 ggc_cleared_alloc to allocate target_optabs.
2632 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2633 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2635 * internal-fn.h (internal_load_fn_p): Declare.
2636 (internal_gather_scatter_fn_p): Likewise.
2637 (internal_fn_mask_index): Likewise.
2638 (internal_gather_scatter_fn_supported_p): Likewise.
2639 * internal-fn.c (gather_load_direct): New macro.
2640 (expand_gather_load_optab_fn): New function.
2641 (direct_gather_load_optab_supported_p): New macro.
2642 (direct_internal_fn_optab): New function.
2643 (internal_load_fn_p): Likewise.
2644 (internal_gather_scatter_fn_p): Likewise.
2645 (internal_fn_mask_index): Likewise.
2646 (internal_gather_scatter_fn_supported_p): Likewise.
2647 * optabs-query.c (supports_at_least_one_mode_p): New function.
2648 (supports_vec_gather_load_p): Likewise.
2649 * optabs-query.h (supports_vec_gather_load_p): Declare.
2650 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2651 and memory_type field.
2652 (NUM_PATTERNS): Bump to 15.
2653 * tree-vect-data-refs.c: Include internal-fn.h.
2654 (vect_gather_scatter_fn_p): New function.
2655 (vect_describe_gather_scatter_call): Likewise.
2656 (vect_check_gather_scatter): Try using internal functions for
2657 gather loads. Recognize existing calls to a gather load function.
2658 (vect_analyze_data_refs): Consider using gather loads if
2659 supports_vec_gather_load_p.
2660 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2661 (vect_get_gather_scatter_offset_type): Likewise.
2662 (vect_convert_mask_for_vectype): Likewise.
2663 (vect_add_conversion_to_patterm): Likewise.
2664 (vect_try_gather_scatter_pattern): Likewise.
2665 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2666 (vect_vect_recog_func_ptrs): Add it.
2667 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2668 internal_fn_mask_index and internal_gather_scatter_fn_p.
2669 (check_load_store_masking): Take the gather_scatter_info as an
2670 argument and handle gather loads.
2671 (vect_get_gather_scatter_ops): New function.
2672 (vectorizable_call): Check internal_load_fn_p.
2673 (vectorizable_load): Likewise. Handle gather load internal
2675 (vectorizable_store): Update call to check_load_store_masking.
2676 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2677 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2678 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2679 (aarch64_gather_scale_operand_d): New predicates.
2680 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2681 (mask_gather_load<mode>): New insns.
2683 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2684 Alan Hayward <alan.hayward@arm.com>
2685 David Sherwood <david.sherwood@arm.com>
2687 * optabs.def (fold_left_plus_optab): New optab.
2688 * doc/md.texi (fold_left_plus_@var{m}): Document.
2689 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2690 * internal-fn.c (fold_left_direct): Define.
2691 (expand_fold_left_optab_fn): Likewise.
2692 (direct_fold_left_optab_supported_p): Likewise.
2693 * fold-const-call.c (fold_const_fold_left): New function.
2694 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2695 * tree-parloops.c (valid_reduction_p): New function.
2696 (gather_scalar_reductions): Use it.
2697 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2698 (vect_finish_replace_stmt): Declare.
2699 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2700 (needs_fold_left_reduction_p): New function, split out from...
2701 (vect_is_simple_reduction): ...here. Accept reductions that
2702 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2703 (vect_force_simple_reduction): Also store the reduction type in
2704 the assignment's STMT_VINFO_REDUC_TYPE.
2705 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2706 (merge_with_identity): New function.
2707 (vect_expand_fold_left): Likewise.
2708 (vectorize_fold_left_reduction): Likewise.
2709 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2710 scalar phi in place for it. Check for target support and reject
2711 cases that would reassociate the operation. Defer the transform
2712 phase to vectorize_fold_left_reduction.
2713 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2714 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2715 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2719 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2720 call to ifc_temp_var.
2722 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2723 Alan Hayward <alan.hayward@arm.com>
2724 David Sherwood <david.sherwood@arm.com>
2726 * target.def (legitimize_address_displacement): Take the original
2727 offset as a poly_int.
2728 * targhooks.h (default_legitimize_address_displacement): Update
2730 * targhooks.c (default_legitimize_address_displacement): Likewise.
2731 * doc/tm.texi: Regenerate.
2732 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2733 as an argument, moving assert of ad->disp == ad->disp_term to...
2734 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2735 Try calling targetm.legitimize_address_displacement before expanding
2736 the address rather than afterwards, and adjust for the new interface.
2737 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2738 Match the new hook interface. Handle SVE addresses.
2739 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2742 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2744 * Makefile.in (OBJS): Add early-remat.o.
2745 * target.def (select_early_remat_modes): New hook.
2746 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2747 * doc/tm.texi: Regenerate.
2748 * targhooks.h (default_select_early_remat_modes): Declare.
2749 * targhooks.c (default_select_early_remat_modes): New function.
2750 * timevar.def (TV_EARLY_REMAT): New timevar.
2751 * passes.def (pass_early_remat): New pass.
2752 * tree-pass.h (make_pass_early_remat): Declare.
2753 * early-remat.c: New file.
2754 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2756 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2758 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2759 Alan Hayward <alan.hayward@arm.com>
2760 David Sherwood <david.sherwood@arm.com>
2762 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2763 vfm1 with a bound_epilog parameter.
2764 (vect_do_peeling): Update calls accordingly, and move the prologue
2765 call earlier in the function. Treat the base bound_epilog as 0 for
2766 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2767 this base when peeling for gaps.
2768 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2769 with fully-masked loops.
2770 (vect_estimate_min_profitable_iters): Handle the single peeled
2771 iteration in that case.
2773 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2774 Alan Hayward <alan.hayward@arm.com>
2775 David Sherwood <david.sherwood@arm.com>
2777 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2778 single-element interleaving even if the size is not a power of 2.
2779 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2780 accesses for single-element interleaving if the group size is
2783 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2784 Alan Hayward <alan.hayward@arm.com>
2785 David Sherwood <david.sherwood@arm.com>
2787 * doc/md.texi (fold_extract_last_@var{m}): Document.
2788 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2789 * optabs.def (fold_extract_last_optab): New optab.
2790 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2791 * internal-fn.c (fold_extract_direct): New macro.
2792 (expand_fold_extract_optab_fn): Likewise.
2793 (direct_fold_extract_optab_supported_p): Likewise.
2794 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2795 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2796 EXTRACT_LAST_REDUCTION.
2797 (get_initial_def_for_reduction): Do not create an initial vector
2798 for EXTRACT_LAST_REDUCTION reductions.
2799 (vectorizable_reduction): Leave the scalar phi in place for
2800 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2801 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2802 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2803 transform phase to vectorizable_condition.
2804 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2806 (vect_finish_stmt_generation): ...here.
2807 (vect_finish_replace_stmt): New function.
2808 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2809 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2811 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2813 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2814 Alan Hayward <alan.hayward@arm.com>
2815 David Sherwood <david.sherwood@arm.com>
2817 * doc/md.texi (extract_last_@var{m}): Document.
2818 * optabs.def (extract_last_optab): New optab.
2819 * internal-fn.def (EXTRACT_LAST): New internal function.
2820 * internal-fn.c (cond_unary_direct): New macro.
2821 (expand_cond_unary_optab_fn): Likewise.
2822 (direct_cond_unary_optab_supported_p): Likewise.
2823 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2824 loops using EXTRACT_LAST.
2825 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2826 (extract_last_<mode>): ...this optab.
2827 (vec_extract<mode><Vel>): Update accordingly.
2829 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2830 Alan Hayward <alan.hayward@arm.com>
2831 David Sherwood <david.sherwood@arm.com>
2833 * target.def (empty_mask_is_expensive): New hook.
2834 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2835 * doc/tm.texi: Regenerate.
2836 * targhooks.h (default_empty_mask_is_expensive): Declare.
2837 * targhooks.c (default_empty_mask_is_expensive): New function.
2838 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2839 if the target says that empty masks are expensive.
2840 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2842 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2844 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2845 Alan Hayward <alan.hayward@arm.com>
2846 David Sherwood <david.sherwood@arm.com>
2848 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2849 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2850 (vect_use_loop_mask_for_alignment_p): New function.
2851 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2852 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2853 niters_skip argument. Make sure that the first niters_skip elements
2854 of the first iteration are inactive.
2855 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2856 Update call to vect_set_loop_masks_directly.
2857 (get_misalign_in_elems): New function, split out from...
2858 (vect_gen_prolog_loop_niters): ...here.
2859 (vect_update_init_of_dr): Take a code argument that specifies whether
2860 the adjustment should be added or subtracted.
2861 (vect_update_init_of_drs): Likewise.
2862 (vect_prepare_for_masked_peels): New function.
2863 (vect_do_peeling): Skip prologue peeling if we're using a mask
2864 instead. Update call to vect_update_inits_of_drs.
2865 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2867 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2868 alignment. Do not include the number of peeled iterations in
2869 the minimum threshold in that case.
2870 (vectorizable_induction): Adjust the start value down by
2871 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2872 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2873 Take the number of skipped iterations into account when calculating
2875 * tree-vect-stmts.c (vect_gen_while_not): New function.
2877 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2878 Alan Hayward <alan.hayward@arm.com>
2879 David Sherwood <david.sherwood@arm.com>
2881 * doc/sourcebuild.texi (vect_fully_masked): Document.
2882 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2884 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2886 (vect_analyze_loop_2): ...here. Don't check the vectorization
2887 factor against the number of loop iterations if the loop is
2890 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2891 Alan Hayward <alan.hayward@arm.com>
2892 David Sherwood <david.sherwood@arm.com>
2894 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2895 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2896 (dump_groups): Update accordingly.
2897 (iv_use::mem_type): New member variable.
2898 (address_p): New function.
2899 (record_use): Add a mem_type argument and initialize the new
2901 (record_group_use): Add a mem_type argument. Use address_p.
2902 Remove obsolete null checks of base_object. Update call to record_use.
2903 (find_interesting_uses_op): Update call to record_group_use.
2904 (find_interesting_uses_cond): Likewise.
2905 (find_interesting_uses_address): Likewise.
2906 (get_mem_type_for_internal_fn): New function.
2907 (find_address_like_use): Likewise.
2908 (find_interesting_uses_stmt): Try find_address_like_use before
2909 calling find_interesting_uses_op.
2910 (addr_offset_valid_p): Use the iv mem_type field as the type
2911 of the addressed memory.
2912 (add_autoinc_candidates): Likewise.
2913 (get_address_cost): Likewise.
2914 (split_small_address_groups_p): Use address_p.
2915 (split_address_groups): Likewise.
2916 (add_iv_candidate_for_use): Likewise.
2917 (autoinc_possible_for_pair): Likewise.
2918 (rewrite_groups): Likewise.
2919 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2920 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2921 (get_alias_ptr_type_for_ptr_address): New function.
2922 (rewrite_use_address): Rewrite address uses in calls that were
2923 identified by find_address_like_use.
2925 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2926 Alan Hayward <alan.hayward@arm.com>
2927 David Sherwood <david.sherwood@arm.com>
2929 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2931 * gimple-expr.h (is_gimple_addressable: Likewise.
2932 * gimple-expr.c (is_gimple_address): Likewise.
2933 * internal-fn.c (expand_call_mem_ref): New function.
2934 (expand_mask_load_optab_fn): Use it.
2935 (expand_mask_store_optab_fn): Likewise.
2937 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2938 Alan Hayward <alan.hayward@arm.com>
2939 David Sherwood <david.sherwood@arm.com>
2941 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2942 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2943 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2944 (cond_umax@var{mode}): Document.
2945 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2946 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2947 (cond_umin_optab, cond_umax_optab): New optabs.
2948 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2949 (COND_IOR, COND_XOR): New internal functions.
2950 * internal-fn.h (get_conditional_internal_fn): Declare.
2951 * internal-fn.c (cond_binary_direct): New macro.
2952 (expand_cond_binary_optab_fn): Likewise.
2953 (direct_cond_binary_optab_supported_p): Likewise.
2954 (get_conditional_internal_fn): New function.
2955 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2956 Cope with reduction statements that are vectorized as calls rather
2958 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2959 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2960 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2961 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2962 (UNSPEC_COND_EOR): New unspecs.
2963 (optab): Add mappings for them.
2964 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2965 (sve_int_op, sve_fp_op): New int attributes.
2967 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2968 Alan Hayward <alan.hayward@arm.com>
2969 David Sherwood <david.sherwood@arm.com>
2971 * optabs.def (while_ult_optab): New optab.
2972 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2973 * internal-fn.def (WHILE_ULT): New internal function.
2974 * internal-fn.h (direct_internal_fn_supported_p): New override
2975 that takes two types as argument.
2976 * internal-fn.c (while_direct): New macro.
2977 (expand_while_optab_fn): New function.
2978 (convert_optab_supported_p): Likewise.
2979 (direct_while_optab_supported_p): New macro.
2980 * wide-int.h (wi::udiv_ceil): New function.
2981 * tree-vectorizer.h (rgroup_masks): New structure.
2982 (vec_loop_masks): New typedef.
2983 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2985 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2986 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2987 (vect_max_vf): New function.
2988 (slpeel_make_loop_iterate_ntimes): Delete.
2989 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2990 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2991 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2992 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2993 internal-fn.h, stor-layout.h and optabs-query.h.
2994 (vect_set_loop_mask): New function.
2995 (add_preheader_seq): Likewise.
2996 (add_header_seq): Likewise.
2997 (interleave_supported_p): Likewise.
2998 (vect_maybe_permute_loop_masks): Likewise.
2999 (vect_set_loop_masks_directly): Likewise.
3000 (vect_set_loop_condition_masked): Likewise.
3001 (vect_set_loop_condition_unmasked): New function, split out from
3002 slpeel_make_loop_iterate_ntimes.
3003 (slpeel_make_loop_iterate_ntimes): Rename to..
3004 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3005 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3006 (vect_do_peeling): Update call accordingly.
3007 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3009 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3010 mask_compare_type, can_fully_mask_p and fully_masked_p.
3011 (release_vec_loop_masks): New function.
3012 (_loop_vec_info): Use it to free the loop masks.
3013 (can_produce_all_loop_masks_p): New function.
3014 (vect_get_max_nscalars_per_iter): Likewise.
3015 (vect_verify_full_masking): Likewise.
3016 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3017 retries, and free the mask rgroups before retrying. Check loop-wide
3018 reasons for disallowing fully-masked loops. Make the final decision
3019 about whether use a fully-masked loop or not.
3020 (vect_estimate_min_profitable_iters): Do not assume that peeling
3021 for the number of iterations will be needed for fully-masked loops.
3022 (vectorizable_reduction): Disable fully-masked loops.
3023 (vectorizable_live_operation): Likewise.
3024 (vect_halve_mask_nunits): New function.
3025 (vect_double_mask_nunits): Likewise.
3026 (vect_record_loop_mask): Likewise.
3027 (vect_get_loop_mask): Likewise.
3028 (vect_transform_loop): Handle the case in which the final loop
3029 iteration might handle a partial vector. Call vect_set_loop_condition
3030 instead of slpeel_make_loop_iterate_ntimes.
3031 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3032 (check_load_store_masking): New function.
3033 (prepare_load_store_mask): Likewise.
3034 (vectorizable_store): Handle fully-masked loops.
3035 (vectorizable_load): Likewise.
3036 (supportable_widening_operation): Use vect_halve_mask_nunits for
3038 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3039 (vect_gen_while): New function.
3040 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3041 (aarch64_uqdec<mode>): New insn.
3043 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3044 Alan Hayward <alan.hayward@arm.com>
3045 David Sherwood <david.sherwood@arm.com>
3047 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3048 (reduc_xor_scal_optab): New optabs.
3049 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3050 (reduc_xor_scal_@var{m}): Document.
3051 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3052 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3054 * fold-const-call.c (fold_const_call): Handle them.
3055 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3056 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3057 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3058 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3059 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3060 (UNSPEC_XORV): New unspecs.
3061 (optab): Add entries for them.
3062 (BITWISEV): New int iterator.
3063 (bit_reduc_op): New int attributes.
3065 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3066 Alan Hayward <alan.hayward@arm.com>
3067 David Sherwood <david.sherwood@arm.com>
3069 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3070 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3071 * optabs.def (vec_shl_insert_optab): New optab.
3072 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3073 (duplicate_and_interleave): Likewise.
3074 * tree-vect-loop.c: Include internal-fn.h.
3075 (neutral_op_for_slp_reduction): New function, split out from
3076 get_initial_defs_for_reduction.
3077 (get_initial_def_for_reduction): Handle option 2 for variable-length
3078 vectors by loading the neutral value into a vector and then shifting
3079 the initial value into element 0.
3080 (get_initial_defs_for_reduction): Replace the code argument with
3081 the neutral value calculated by neutral_op_for_slp_reduction.
3082 Use gimple_build_vector for constant-length vectors.
3083 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3084 but the first group_size elements have a neutral value.
3085 Use duplicate_and_interleave otherwise.
3086 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3087 Update call to get_initial_defs_for_reduction. Handle SLP
3088 reductions for variable-length vectors by creating one vector
3089 result for each scalar result, with the elements associated
3090 with other scalar results stubbed out with the neutral value.
3091 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3092 Require IFN_VEC_SHL_INSERT for double reductions on
3093 variable-length vectors, or SLP reductions that have
3094 a neutral value. Require can_duplicate_and_interleave_p
3095 support for variable-length unchained SLP reductions if there
3096 is no neutral value, such as for MIN/MAX reductions. Also require
3097 the number of vector elements to be a multiple of the number of
3098 SLP statements when doing variable-length unchained SLP reductions.
3099 Update call to vect_create_epilog_for_reduction.
3100 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3101 and remove initial values.
3102 (duplicate_and_interleave): Make public.
3103 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3104 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3106 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3107 Alan Hayward <alan.hayward@arm.com>
3108 David Sherwood <david.sherwood@arm.com>
3110 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3111 (can_duplicate_and_interleave_p): New function.
3112 (vect_get_and_check_slp_defs): Take the vector of statements
3113 rather than just the current one. Remove excess parentheses.
3114 Restriction rejectinon of vect_constant_def and vect_external_def
3115 for variable-length vectors to boolean types, or types for which
3116 can_duplicate_and_interleave_p is false.
3117 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3118 (duplicate_and_interleave): New function.
3119 (vect_get_constant_vectors): Use gimple_build_vector for
3120 constant-length vectors and suitable variable-length constant
3121 vectors. Use duplicate_and_interleave for other variable-length
3122 vectors. Don't defer the update when inserting new statements.
3124 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3125 Alan Hayward <alan.hayward@arm.com>
3126 David Sherwood <david.sherwood@arm.com>
3128 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3129 min_profitable_iters doesn't go negative.
3131 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3132 Alan Hayward <alan.hayward@arm.com>
3133 David Sherwood <david.sherwood@arm.com>
3135 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3136 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3137 * optabs.def (vec_mask_load_lanes_optab): New optab.
3138 (vec_mask_store_lanes_optab): Likewise.
3139 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3140 (MASK_STORE_LANES): Likewise.
3141 * internal-fn.c (mask_load_lanes_direct): New macro.
3142 (mask_store_lanes_direct): Likewise.
3143 (expand_mask_load_optab_fn): Handle masked operations.
3144 (expand_mask_load_lanes_optab_fn): New macro.
3145 (expand_mask_store_optab_fn): Handle masked operations.
3146 (expand_mask_store_lanes_optab_fn): New macro.
3147 (direct_mask_load_lanes_optab_supported_p): Likewise.
3148 (direct_mask_store_lanes_optab_supported_p): Likewise.
3149 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3151 (vect_load_lanes_supported): Likewise.
3152 * tree-vect-data-refs.c (strip_conversion): New function.
3153 (can_group_stmts_p): Likewise.
3154 (vect_analyze_data_ref_accesses): Use it instead of checking
3155 for a pair of assignments.
3156 (vect_store_lanes_supported): Take a masked_p parameter.
3157 (vect_load_lanes_supported): Likewise.
3158 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3159 vect_store_lanes_supported and vect_load_lanes_supported.
3160 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3161 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3162 parameter. Don't allow gaps for masked accesses.
3163 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3164 and vect_load_lanes_supported.
3165 (get_load_store_type): Take a masked_p parameter and update
3166 call to get_group_load_store_type.
3167 (vectorizable_store): Update call to get_load_store_type.
3168 Handle IFN_MASK_STORE_LANES.
3169 (vectorizable_load): Update call to get_load_store_type.
3170 Handle IFN_MASK_LOAD_LANES.
3172 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3173 Alan Hayward <alan.hayward@arm.com>
3174 David Sherwood <david.sherwood@arm.com>
3176 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3178 * config/aarch64/aarch64-protos.h
3179 (aarch64_sve_struct_memory_operand_p): Declare.
3180 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3181 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3182 (VPRED, vpred): Handle SVE structure modes.
3183 * config/aarch64/constraints.md (Utx): New constraint.
3184 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3185 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3186 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3187 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3188 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3189 structure modes. Split into pieces after RA.
3190 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3191 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3193 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3194 SVE structure modes.
3195 (aarch64_classify_address): Likewise.
3196 (sizetochar): Move earlier in file.
3197 (aarch64_print_operand): Handle SVE register lists.
3198 (aarch64_array_mode): New function.
3199 (aarch64_sve_struct_memory_operand_p): Likewise.
3200 (TARGET_ARRAY_MODE): Redefine.
3202 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3203 Alan Hayward <alan.hayward@arm.com>
3204 David Sherwood <david.sherwood@arm.com>
3206 * target.def (array_mode): New target hook.
3207 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3208 * doc/tm.texi: Regenerate.
3209 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3210 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3211 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3213 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3216 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3217 Alan Hayward <alan.hayward@arm.com>
3218 David Sherwood <david.sherwood@arm.com>
3220 * fold-const.c (fold_binary_loc): Check the argument types
3221 rather than the result type when testing for a vector operation.
3223 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3225 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3226 * doc/tm.texi: Regenerate.
3228 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3229 Alan Hayward <alan.hayward@arm.com>
3230 David Sherwood <david.sherwood@arm.com>
3232 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3233 (sve): Document new AArch64 extension.
3234 * doc/md.texi (w): Extend the description of the AArch64
3235 constraint to include SVE vectors.
3236 (Upl, Upa): Document new AArch64 predicate constraints.
3237 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3239 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3240 (msve-vector-bits=): New option.
3241 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3242 SVE when these are disabled.
3243 (sve): New extension.
3244 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3245 modes. Adjust their number of units based on aarch64_sve_vg.
3246 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3247 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3248 aarch64_addr_query_type.
3249 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3250 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3251 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3252 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3253 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3254 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3255 (aarch64_simd_imm_zero_p): Delete.
3256 (aarch64_check_zero_based_sve_index_immediate): Declare.
3257 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3258 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3259 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3260 (aarch64_sve_float_mul_immediate_p): Likewise.
3261 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3263 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3264 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3265 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3266 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3267 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3268 (aarch64_regmode_natural_size): Likewise.
3269 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3270 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3272 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3273 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3274 for VG and the SVE predicate registers.
3275 (V_ALIASES): Add a "z"-prefixed alias.
3276 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3277 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3278 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3279 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3280 (REG_CLASS_NAMES): Add entries for them.
3281 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3282 and the predicate registers.
3283 (aarch64_sve_vg): Declare.
3284 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3285 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3286 (REGMODE_NATURAL_SIZE): Define.
3287 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3289 * config/aarch64/aarch64.c: Include cfgrtl.h.
3290 (simd_immediate_info): Add a constructor for series vectors,
3291 and an associated step field.
3292 (aarch64_sve_vg): New variable.
3293 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3294 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3295 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3296 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3297 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3298 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3299 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3300 (aarch64_get_mask_mode): New functions.
3301 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3302 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3303 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3304 predicate modes and predicate registers. Explicitly restrict
3305 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3306 to store a vector mode if it is recognized by
3307 aarch64_classify_vector_mode.
3308 (aarch64_regmode_natural_size): New function.
3309 (aarch64_hard_regno_caller_save_mode): Return the original mode
3311 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3312 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3313 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3314 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3316 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3317 does not overlap dest if the function is frame-related. Handle
3319 (aarch64_split_add_offset): New function.
3320 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3321 them aarch64_add_offset.
3322 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3323 and update call to aarch64_sub_sp.
3324 (aarch64_add_cfa_expression): New function.
3325 (aarch64_expand_prologue): Pass extra temporary registers to the
3326 functions above. Handle the case in which we need to emit new
3327 DW_CFA_expressions for registers that were originally saved
3328 relative to the stack pointer, but now have to be expressed
3329 relative to the frame pointer.
3330 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3332 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3333 IP0 and IP1 values for SVE frames.
3334 (aarch64_expand_vec_series): New function.
3335 (aarch64_expand_sve_widened_duplicate): Likewise.
3336 (aarch64_expand_sve_const_vector): Likewise.
3337 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3338 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3339 into the register, rather than emitting a SET directly.
3340 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3341 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3342 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3343 (offset_9bit_signed_scaled_p): New functions.
3344 (aarch64_replicate_bitmask_imm): New function.
3345 (aarch64_bitmask_imm): Use it.
3346 (aarch64_cannot_force_const_mem): Reject expressions involving
3347 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3348 (aarch64_classify_index): Handle SVE indices, by requiring
3349 a plain register index with a scale that matches the element size.
3350 (aarch64_classify_address): Handle SVE addresses. Assert that
3351 the mode of the address is VOIDmode or an integer mode.
3352 Update call to aarch64_classify_symbol.
3353 (aarch64_classify_symbolic_expression): Update call to
3354 aarch64_classify_symbol.
3355 (aarch64_const_vec_all_in_range_p): New function.
3356 (aarch64_print_vector_float_operand): Likewise.
3357 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3358 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3359 and the FP immediates 1.0 and 0.5.
3360 (aarch64_print_address_internal): Handle SVE addresses.
3361 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3362 (aarch64_regno_regclass): Handle predicate registers.
3363 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3365 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3366 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3367 (aarch64_convert_sve_vector_bits): New function.
3368 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3369 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3371 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3372 Handle SVE vector and predicate modes. Accept VL-based constants
3373 that need only one temporary register, and VL offsets that require
3374 no temporary registers.
3375 (aarch64_conditional_register_usage): Mark the predicate registers
3376 as fixed if SVE isn't available.
3377 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3378 Return true for SVE vector and predicate modes.
3379 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3380 rather than an unsigned int. Handle SVE modes.
3381 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3383 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3385 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3386 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3387 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3388 (aarch64_sve_float_mul_immediate_p): New functions.
3389 (aarch64_sve_valid_immediate): New function.
3390 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3391 Explicitly reject structure modes. Check for INDEX constants.
3392 Handle PTRUE and PFALSE constants.
3393 (aarch64_check_zero_based_sve_index_immediate): New function.
3394 (aarch64_simd_imm_zero_p): Delete.
3395 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3396 vector modes. Accept constants in the range of CNT[BHWD].
3397 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3398 ask for an Advanced SIMD mode.
3399 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3400 (aarch64_simd_vector_alignment): Handle SVE predicates.
3401 (aarch64_vectorize_preferred_vector_alignment): New function.
3402 (aarch64_simd_vector_alignment_reachable): Use it instead of
3404 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3405 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3407 (MAX_VECT_LEN): Delete.
3408 (expand_vec_perm_d): Add a vec_flags field.
3409 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3410 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3411 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3413 (aarch64_evpc_rev): Rename to...
3414 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3415 (aarch64_evpc_rev_global): New function.
3416 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3417 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3419 (aarch64_evpc_sve_tbl): New function.
3420 (aarch64_expand_vec_perm_const_1): Update after rename of
3421 aarch64_evpc_rev. Handle SVE permutes too, trying
3422 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3423 than aarch64_evpc_tbl.
3424 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3425 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3426 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3427 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3428 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3429 (aarch64_expand_sve_vcond): New functions.
3430 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3431 of aarch64_vector_mode_p.
3432 (aarch64_dwarf_poly_indeterminate_value): New function.
3433 (aarch64_compute_pressure_classes): Likewise.
3434 (aarch64_can_change_mode_class): Likewise.
3435 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3436 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3437 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3438 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3439 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3440 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3441 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3442 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3444 (Dn, Dl, Dr): Accept const as well as const_vector.
3445 (Dz): Likewise. Compare against CONST0_RTX.
3446 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3447 of "vector" where appropriate.
3448 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3449 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3450 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3451 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3452 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3453 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3454 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3455 (v_int_equiv): Extend to SVE modes.
3456 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3458 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3459 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3460 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3461 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3462 (SVE_COND_FP_CMP): New int iterators.
3463 (perm_hilo): Handle the new unpack unspecs.
3464 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3466 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3467 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3468 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3469 (aarch64_equality_operator, aarch64_constant_vector_operand)
3470 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3471 (aarch64_sve_nonimmediate_operand): Likewise.
3472 (aarch64_sve_general_operand): Likewise.
3473 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3474 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3475 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3476 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3477 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3478 (aarch64_sve_float_arith_immediate): Likewise.
3479 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3480 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3481 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3482 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3483 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3484 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3485 (aarch64_sve_float_arith_operand): Likewise.
3486 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3487 (aarch64_sve_float_mul_operand): Likewise.
3488 (aarch64_sve_vec_perm_operand): Likewise.
3489 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3490 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3491 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3492 as well as const_vector.
3493 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3494 in file. Use CONST0_RTX and CONSTM1_RTX.
3495 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3496 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3497 Use aarch64_simd_imm_zero.
3498 * config/aarch64/aarch64-sve.md: New file.
3499 * config/aarch64/aarch64.md: Include it.
3500 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3501 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3502 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3503 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3504 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3505 (sve): New attribute.
3506 (enabled): Disable instructions with the sve attribute unless
3508 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3509 aarch64_expand_mov_immediate.
3510 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3511 CNT[BHSD] immediates.
3512 (movti): Split CONST_POLY_INT moves into two halves.
3513 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3514 Split additions that need a temporary here if the destination
3515 is the stack pointer.
3516 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3517 (*add<mode>3_poly_1): New instruction.
3518 (set_clobber_cc): New expander.
3520 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3522 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3523 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3524 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3525 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3526 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3527 Change innermode from fixed_mode_size to machine_mode.
3528 (simplify_subreg): Update call accordingly. Handle a constant-sized
3529 subreg of a variable-length CONST_VECTOR.
3531 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3532 Alan Hayward <alan.hayward@arm.com>
3533 David Sherwood <david.sherwood@arm.com>
3535 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3536 (add_offset_to_base): New function, split out from...
3537 (create_mem_ref): ...here. When handling a scale other than 1,
3538 check first whether the address is valid without the offset.
3539 Add it into the base if so, leaving the index and scale as-is.
3541 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3544 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3545 fold_for_warn before checking if arg2 is INTEGER_CST.
3547 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3549 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3550 (store_multiple_operation): Delete.
3551 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3552 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3553 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3554 guarded by TARGET_STRING.
3555 (rs6000_output_load_multiple): Delete.
3556 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3557 OPTION_MASK_STRING / TARGET_STRING handling.
3558 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3559 (const rs6000_opt_masks) <"string">: Change mask to 0.
3560 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3561 (MASK_STRING): Delete.
3562 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3564 (load_multiple): Delete.
3571 (store_multiple): Delete.
3578 (movmemsi_8reg): Delete.
3579 (corresponding unnamed define_insn): Delete.
3580 (movmemsi_6reg): Delete.
3581 (corresponding unnamed define_insn): Delete.
3582 (movmemsi_4reg): Delete.
3583 (corresponding unnamed define_insn): Delete.
3584 (movmemsi_2reg): Delete.
3585 (corresponding unnamed define_insn): Delete.
3586 (movmemsi_1reg): Delete.
3587 (corresponding unnamed define_insn): Delete.
3588 * config/rs6000/rs6000.opt (mno-string): New.
3589 (mstring): Replace by deprecation warning stub.
3590 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3592 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3594 * regrename.c (regrename_do_replace): If replacing the same
3595 reg multiple times, try to reuse last created gen_raw_REG.
3598 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3599 main to workaround a bug in GDB.
3601 2018-01-12 Tom de Vries <tom@codesourcery.com>
3604 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3606 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3608 PR rtl-optimization/80481
3609 * ira-color.c (get_cap_member): New function.
3610 (allocnos_conflict_by_live_ranges_p): Use it.
3611 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3612 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3614 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3617 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3618 (*saddl_se_1): Ditto.
3620 (*ssubl_se_1): Ditto.
3622 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3624 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3625 rather than wi::to_widest for DR_INITs.
3626 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3627 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3628 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3630 (vect_analyze_group_access_1): Note that here.
3632 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3634 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3635 polynomial type sizes.
3637 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3639 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3640 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3641 (gimple_add_tmp_var): Likewise.
3643 2018-01-12 Martin Liska <mliska@suse.cz>
3645 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3646 (gimple_alloc_sizes): Likewise.
3647 (dump_gimple_statistics): Use PRIu64 in printf format.
3648 * gimple.h: Change uint64_t to int.
3650 2018-01-12 Martin Liska <mliska@suse.cz>
3652 * tree-core.h: Use uint64_t instead of int.
3653 * tree.c (tree_node_counts): Likewise.
3654 (tree_node_sizes): Likewise.
3655 (dump_tree_statistics): Use PRIu64 in printf format.
3657 2018-01-12 Martin Liska <mliska@suse.cz>
3659 * Makefile.in: As qsort_chk is implemented in vec.c, add
3660 vec.o to linkage of gencfn-macros.
3661 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3662 passing the info to record_node_allocation_statistics.
3663 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3665 * ggc-common.c (struct ggc_usage): Add operator== and use
3666 it in operator< and compare function.
3667 * mem-stats.h (struct mem_usage): Likewise.
3668 * vec.c (struct vec_usage): Remove operator< and compare
3669 function. Can be simply inherited.
3671 2018-01-12 Martin Jambor <mjambor@suse.cz>
3674 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3675 * tree-ssa-math-opts.c: Include domwalk.h.
3676 (convert_mult_to_fma_1): New function.
3677 (fma_transformation_info): New type.
3678 (fma_deferring_state): Likewise.
3679 (cancel_fma_deferring): New function.
3680 (result_of_phi): Likewise.
3681 (last_fma_candidate_feeds_initial_phi): Likewise.
3682 (convert_mult_to_fma): Added deferring logic, split actual
3683 transformation to convert_mult_to_fma_1.
3684 (math_opts_dom_walker): New type.
3685 (math_opts_dom_walker::after_dom_children): New method, body moved
3686 here from pass_optimize_widening_mul::execute, added deferring logic
3688 (pass_optimize_widening_mul::execute): Moved most of code to
3689 math_opts_dom_walker::after_dom_children.
3690 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3691 * config/i386/i386.c (ix86_option_override_internal): Added
3692 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3694 2018-01-12 Richard Biener <rguenther@suse.de>
3697 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3698 inline instance vars.
3700 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3703 * config/rx/rx.c (rx_is_restricted_memory_address):
3706 2018-01-12 Richard Biener <rguenther@suse.de>
3708 PR tree-optimization/80846
3709 * target.def (split_reduction): New target hook.
3710 * targhooks.c (default_split_reduction): New function.
3711 * targhooks.h (default_split_reduction): Declare.
3712 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3713 target requests first reduce vectors by combining low and high
3715 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3716 (get_vectype_for_scalar_type_and_size): Export.
3717 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3718 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3719 * doc/tm.texi: Regenerate.
3720 * config/i386/i386.c (ix86_split_reduction): Implement
3721 TARGET_VECTORIZE_SPLIT_REDUCTION.
3723 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3726 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3727 in PIC mode except for TARGET_VXWORKS_RTP.
3728 * config/sparc/sparc.c: Include cfgrtl.h.
3729 (TARGET_INIT_PIC_REG): Define.
3730 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3731 (sparc_pic_register_p): New predicate.
3732 (sparc_legitimate_address_p): Use it.
3733 (sparc_legitimize_pic_address): Likewise.
3734 (sparc_delegitimize_address): Likewise.
3735 (sparc_mode_dependent_address_p): Likewise.
3736 (gen_load_pcrel_sym): Remove 4th parameter.
3737 (load_got_register): Adjust call to above. Remove obsolete stuff.
3738 (sparc_expand_prologue): Do not call load_got_register here.
3739 (sparc_flat_expand_prologue): Likewise.
3740 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3741 (sparc_use_pseudo_pic_reg): New function.
3742 (sparc_init_pic_reg): Likewise.
3743 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3744 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3746 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3748 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3749 Add item for branch_cost.
3751 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3753 PR rtl-optimization/83565
3754 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3755 not extend the result to a larger mode for rotate operations.
3756 (num_sign_bit_copies1): Likewise.
3758 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3761 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3763 Use values-Xc.o for -pedantic.
3764 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3766 2018-01-12 Martin Liska <mliska@suse.cz>
3769 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3771 (possible_polymorphic_call_targets): Use it.
3772 (ipa_devirt): Likewise.
3774 2018-01-12 Martin Liska <mliska@suse.cz>
3776 * profile-count.h (enum profile_quality): Use 0 as invalid
3777 enum value of profile_quality.
3779 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3781 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3782 -mext-string options.
3784 2018-01-12 Richard Biener <rguenther@suse.de>
3786 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3787 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3788 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3790 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3792 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3794 * configure.ac (--with-long-double-format): Add support for the
3795 configuration option to change the default long double format on
3797 * config.gcc (powerpc*-linux*-*): Likewise.
3798 * configure: Regenerate.
3799 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3800 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3801 used without modification.
3803 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3805 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3806 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3807 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3808 MISC_BUILTIN_SPEC_BARRIER.
3809 (rs6000_init_builtins): Likewise.
3810 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3812 (speculation_barrier): New define_insn.
3813 * doc/extend.texi: Document __builtin_speculation_barrier.
3815 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3818 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3819 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3820 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3822 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3823 integral modes instead of "ss" and "sd".
3824 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3825 vectors with 32-bit and 64-bit elements.
3826 (vecdupssescalarmodesuffix): New mode attribute.
3827 (vec_dup<mode>): Use it.
3829 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3832 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3833 frame if argument is passed on stack.
3835 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3838 * ree.c (combine_reaching_defs): Optimize also
3839 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3840 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3842 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3845 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3847 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3850 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3851 after they are computed.
3853 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3855 PR tree-optimization/83695
3856 * gimple-loop-linterchange.cc
3857 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3858 reset cached scev information after interchange.
3859 (pass_linterchange::execute): Remove call to scev_reset_htab.
3861 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3863 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3864 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3865 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3866 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3867 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3868 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3869 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3870 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3871 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3872 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3873 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3874 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3875 (V_lane_reg): Likewise.
3876 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3878 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3879 (vfmal_lane_low<mode>_intrinsic,
3880 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3881 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3882 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3883 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3884 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3885 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3887 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3889 * config/arm/arm-cpus.in (fp16fml): New feature.
3890 (ALL_SIMD): Add fp16fml.
3891 (armv8.2-a): Add fp16fml as an option.
3892 (armv8.3-a): Likewise.
3893 (armv8.4-a): Add fp16fml as part of fp16.
3894 * config/arm/arm.h (TARGET_FP16FML): Define.
3895 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3897 * config/arm/arm-modes.def (V2HF): Define.
3898 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3899 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3900 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3901 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3902 vfmsl_low, vfmsl_high): New set of builtins.
3903 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3904 (vfml_op): New code attribute.
3905 (VFMLHALVES): New int iterator.
3906 (VFML, VFMLSEL): New mode attributes.
3907 (V_reg): Define mapping for V2HF.
3908 (V_hi, V_lo): New mode attributes.
3909 (VF_constraint): Likewise.
3910 (vfml_half, vfml_half_selector): New int attributes.
3911 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3913 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3914 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3916 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3917 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3918 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3919 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3921 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3922 Document new effective target and option set.
3924 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3926 * config/arm/arm-cpus.in (armv8_4): New feature.
3927 (ARMv8_4a): New fgroup.
3928 (armv8.4-a): New arch.
3929 * config/arm/arm-tables.opt: Regenerate.
3930 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3931 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3932 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3933 Add matching rules for -march=armv8.4-a and extensions.
3934 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3936 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3939 * config/rx/rx.md (BW): New mode attribute.
3940 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3942 2018-01-11 Richard Biener <rguenther@suse.de>
3944 PR tree-optimization/83435
3945 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3946 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3947 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3949 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3950 Alan Hayward <alan.hayward@arm.com>
3951 David Sherwood <david.sherwood@arm.com>
3953 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3955 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3956 (aarch64_print_address_internal): Use it to check for a zero offset.
3958 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3959 Alan Hayward <alan.hayward@arm.com>
3960 David Sherwood <david.sherwood@arm.com>
3962 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3963 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3964 Return a poly_int64 rather than a HOST_WIDE_INT.
3965 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3966 rather than a HOST_WIDE_INT.
3967 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3968 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3969 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3970 final_offset from HOST_WIDE_INT to poly_int64.
3971 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3972 to_constant when getting the number of units in an Advanced SIMD
3974 (aarch64_builtin_vectorized_function): Check for a constant number
3976 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3978 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3979 attribute instead of GET_MODE_NUNITS.
3980 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3981 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3982 GET_MODE_SIZE for fixed-size registers.
3983 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3984 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3985 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3986 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3987 (aarch64_print_operand, aarch64_print_address_internal)
3988 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3989 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3990 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3991 Handle polynomial GET_MODE_SIZE.
3992 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3993 wider than SImode without modification.
3994 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3995 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3996 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3997 passing and returning SVE modes.
3998 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3999 rather than GEN_INT.
4000 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4001 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4002 (aarch64_allocate_and_probe_stack_space): Likewise.
4003 (aarch64_layout_frame): Cope with polynomial offsets.
4004 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4005 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4007 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4008 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4009 poly_int64 rather than a HOST_WIDE_INT.
4010 (aarch64_get_separate_components, aarch64_process_components)
4011 (aarch64_expand_prologue, aarch64_expand_epilogue)
4012 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4013 (aarch64_anchor_offset): New function, split out from...
4014 (aarch64_legitimize_address): ...here.
4015 (aarch64_builtin_vectorization_cost): Handle polynomial
4016 TYPE_VECTOR_SUBPARTS.
4017 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4019 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4020 number of elements from the PARALLEL rather than the mode.
4021 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4022 rather than GET_MODE_BITSIZE.
4023 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4024 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4025 (aarch64_expand_vec_perm_const_1): Handle polynomial
4026 d->perm.length () and d->perm elements.
4027 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4028 Apply to_constant to d->perm elements.
4029 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4030 polynomial CONST_VECTOR_NUNITS.
4031 (aarch64_move_pointer): Take amount as a poly_int64 rather
4033 (aarch64_progress_pointer): Avoid temporary variable.
4034 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4035 the mode attribute instead of GET_MODE.
4037 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4038 Alan Hayward <alan.hayward@arm.com>
4039 David Sherwood <david.sherwood@arm.com>
4041 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4042 x exists before using it.
4043 (aarch64_add_constant_internal): Rename to...
4044 (aarch64_add_offset_1): ...this. Replace regnum with separate
4045 src and dest rtxes. Handle the case in which they're different,
4046 including when the offset is zero. Replace scratchreg with an rtx.
4047 Use 2 additions if there is no spare register into which we can
4048 move a 16-bit constant.
4049 (aarch64_add_constant): Delete.
4050 (aarch64_add_offset): Replace reg with separate src and dest
4051 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4052 Use aarch64_add_offset_1.
4053 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4054 an rtx rather than an int. Take the delta as a poly_int64
4055 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4056 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4057 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4058 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4059 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4061 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4062 aarch64_add_constant.
4064 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4066 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4067 Use scalar_float_mode.
4069 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4071 * config/aarch64/aarch64-simd.md
4072 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4073 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4074 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4075 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4076 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4077 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4078 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4079 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4080 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4081 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4083 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4086 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4087 targ_options->x_arm_arch_string is non NULL.
4089 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4091 * config/aarch64/aarch64.h
4092 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4094 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4097 * expmed.c (emit_store_flag_force): Swap if const op0
4098 and change VOIDmode to mode of op0.
4100 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4102 PR rtl-optimization/83761
4103 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4104 than bytes to mode_for_size.
4106 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4109 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4110 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4113 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4116 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4117 when in layout mode.
4118 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4119 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4122 2018-01-10 Michael Collison <michael.collison@arm.com>
4124 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4125 * config/aarch64/aarch64-option-extension.def: Add
4126 AARCH64_OPT_EXTENSION of 'fp16fml'.
4127 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4128 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4129 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4130 * config/aarch64/constraints.md (Ui7): New constraint.
4131 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4132 (VFMLA_SEL_W): Ditto.
4135 (VFMLA16_LOW): New int iterator.
4136 (VFMLA16_HIGH): Ditto.
4137 (UNSPEC_FMLAL): New unspec.
4138 (UNSPEC_FMLSL): Ditto.
4139 (UNSPEC_FMLAL2): Ditto.
4140 (UNSPEC_FMLSL2): Ditto.
4141 (f16mac): New code attribute.
4142 * config/aarch64/aarch64-simd-builtins.def
4143 (aarch64_fmlal_lowv2sf): Ditto.
4144 (aarch64_fmlsl_lowv2sf): Ditto.
4145 (aarch64_fmlalq_lowv4sf): Ditto.
4146 (aarch64_fmlslq_lowv4sf): Ditto.
4147 (aarch64_fmlal_highv2sf): Ditto.
4148 (aarch64_fmlsl_highv2sf): Ditto.
4149 (aarch64_fmlalq_highv4sf): Ditto.
4150 (aarch64_fmlslq_highv4sf): Ditto.
4151 (aarch64_fmlal_lane_lowv2sf): Ditto.
4152 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4153 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4154 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4155 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4156 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4157 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4158 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4159 (aarch64_fmlal_lane_highv2sf): Ditto.
4160 (aarch64_fmlsl_lane_highv2sf): Ditto.
4161 (aarch64_fmlal_laneq_highv2sf): Ditto.
4162 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4163 (aarch64_fmlalq_lane_highv4sf): Ditto.
4164 (aarch64_fmlsl_lane_highv4sf): Ditto.
4165 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4166 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4167 * config/aarch64/aarch64-simd.md:
4168 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4169 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4170 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4171 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4172 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4173 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4174 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4175 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4176 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4177 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4178 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4179 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4180 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4181 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4182 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4183 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4184 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4185 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4186 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4187 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4188 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4189 (vfmlsl_low_u32): Ditto.
4190 (vfmlalq_low_u32): Ditto.
4191 (vfmlslq_low_u32): Ditto.
4192 (vfmlal_high_u32): Ditto.
4193 (vfmlsl_high_u32): Ditto.
4194 (vfmlalq_high_u32): Ditto.
4195 (vfmlslq_high_u32): Ditto.
4196 (vfmlal_lane_low_u32): Ditto.
4197 (vfmlsl_lane_low_u32): Ditto.
4198 (vfmlal_laneq_low_u32): Ditto.
4199 (vfmlsl_laneq_low_u32): Ditto.
4200 (vfmlalq_lane_low_u32): Ditto.
4201 (vfmlslq_lane_low_u32): Ditto.
4202 (vfmlalq_laneq_low_u32): Ditto.
4203 (vfmlslq_laneq_low_u32): Ditto.
4204 (vfmlal_lane_high_u32): Ditto.
4205 (vfmlsl_lane_high_u32): Ditto.
4206 (vfmlal_laneq_high_u32): Ditto.
4207 (vfmlsl_laneq_high_u32): Ditto.
4208 (vfmlalq_lane_high_u32): Ditto.
4209 (vfmlslq_lane_high_u32): Ditto.
4210 (vfmlalq_laneq_high_u32): Ditto.
4211 (vfmlslq_laneq_high_u32): Ditto.
4212 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4213 (AARCH64_FL_FOR_ARCH8_4): New.
4214 (AARCH64_ISA_F16FML): New ISA flag.
4215 (TARGET_F16FML): New feature flag for fp16fml.
4216 (doc/invoke.texi): Document new fp16fml option.
4218 2018-01-10 Michael Collison <michael.collison@arm.com>
4220 * config/aarch64/aarch64-builtins.c:
4221 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4222 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4223 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4224 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4225 (AARCH64_ISA_SHA3): New ISA flag.
4226 (TARGET_SHA3): New feature flag for sha3.
4227 * config/aarch64/iterators.md (sha512_op): New int attribute.
4228 (CRYPTO_SHA512): New int iterator.
4229 (UNSPEC_SHA512H): New unspec.
4230 (UNSPEC_SHA512H2): Ditto.
4231 (UNSPEC_SHA512SU0): Ditto.
4232 (UNSPEC_SHA512SU1): Ditto.
4233 * config/aarch64/aarch64-simd-builtins.def
4234 (aarch64_crypto_sha512hqv2di): New builtin.
4235 (aarch64_crypto_sha512h2qv2di): Ditto.
4236 (aarch64_crypto_sha512su0qv2di): Ditto.
4237 (aarch64_crypto_sha512su1qv2di): Ditto.
4238 (aarch64_eor3qv8hi): Ditto.
4239 (aarch64_rax1qv2di): Ditto.
4240 (aarch64_xarqv2di): Ditto.
4241 (aarch64_bcaxqv8hi): Ditto.
4242 * config/aarch64/aarch64-simd.md:
4243 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4244 (aarch64_crypto_sha512su0qv2di): Ditto.
4245 (aarch64_crypto_sha512su1qv2di): Ditto.
4246 (aarch64_eor3qv8hi): Ditto.
4247 (aarch64_rax1qv2di): Ditto.
4248 (aarch64_xarqv2di): Ditto.
4249 (aarch64_bcaxqv8hi): Ditto.
4250 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4251 (vsha512h2q_u64): Ditto.
4252 (vsha512su0q_u64): Ditto.
4253 (vsha512su1q_u64): Ditto.
4254 (veor3q_u16): Ditto.
4255 (vrax1q_u64): Ditto.
4257 (vbcaxq_u16): Ditto.
4258 * config/arm/types.md (crypto_sha512): New type attribute.
4259 (crypto_sha3): Ditto.
4260 (doc/invoke.texi): Document new sha3 option.
4262 2018-01-10 Michael Collison <michael.collison@arm.com>
4264 * config/aarch64/aarch64-builtins.c:
4265 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4266 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4267 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4268 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4269 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4270 (AARCH64_ISA_SM4): New ISA flag.
4271 (TARGET_SM4): New feature flag for sm4.
4272 * config/aarch64/aarch64-simd-builtins.def
4273 (aarch64_sm3ss1qv4si): Ditto.
4274 (aarch64_sm3tt1aq4si): Ditto.
4275 (aarch64_sm3tt1bq4si): Ditto.
4276 (aarch64_sm3tt2aq4si): Ditto.
4277 (aarch64_sm3tt2bq4si): Ditto.
4278 (aarch64_sm3partw1qv4si): Ditto.
4279 (aarch64_sm3partw2qv4si): Ditto.
4280 (aarch64_sm4eqv4si): Ditto.
4281 (aarch64_sm4ekeyqv4si): Ditto.
4282 * config/aarch64/aarch64-simd.md:
4283 (aarch64_sm3ss1qv4si): Ditto.
4284 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4285 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4286 (aarch64_sm4eqv4si): Ditto.
4287 (aarch64_sm4ekeyqv4si): Ditto.
4288 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4289 (sm3part_op): Ditto.
4290 (CRYPTO_SM3TT): Ditto.
4291 (CRYPTO_SM3PART): Ditto.
4292 (UNSPEC_SM3SS1): New unspec.
4293 (UNSPEC_SM3TT1A): Ditto.
4294 (UNSPEC_SM3TT1B): Ditto.
4295 (UNSPEC_SM3TT2A): Ditto.
4296 (UNSPEC_SM3TT2B): Ditto.
4297 (UNSPEC_SM3PARTW1): Ditto.
4298 (UNSPEC_SM3PARTW2): Ditto.
4299 (UNSPEC_SM4E): Ditto.
4300 (UNSPEC_SM4EKEY): Ditto.
4301 * config/aarch64/constraints.md (Ui2): New constraint.
4302 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4303 * config/arm/types.md (crypto_sm3): New type attribute.
4304 (crypto_sm4): Ditto.
4305 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4306 (vsm3tt1aq_u32): Ditto.
4307 (vsm3tt1bq_u32): Ditto.
4308 (vsm3tt2aq_u32): Ditto.
4309 (vsm3tt2bq_u32): Ditto.
4310 (vsm3partw1q_u32): Ditto.
4311 (vsm3partw2q_u32): Ditto.
4312 (vsm4eq_u32): Ditto.
4313 (vsm4ekeyq_u32): Ditto.
4314 (doc/invoke.texi): Document new sm4 option.
4316 2018-01-10 Michael Collison <michael.collison@arm.com>
4318 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4319 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4320 (AARCH64_FL_FOR_ARCH8_4): New.
4321 (AARCH64_FL_V8_4): New flag.
4322 (doc/invoke.texi): Document new armv8.4-a option.
4324 2018-01-10 Michael Collison <michael.collison@arm.com>
4326 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4327 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4328 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4329 * config/aarch64/aarch64-option-extension.def: Add
4330 AARCH64_OPT_EXTENSION of 'sha2'.
4331 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4332 (crypto): Disable sha2 and aes if crypto disabled.
4333 (crypto): Enable aes and sha2 if enabled.
4334 (simd): Disable sha2 and aes if simd disabled.
4335 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4337 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4338 (TARGET_SHA2): New feature flag for sha2.
4339 (TARGET_AES): New feature flag for aes.
4340 * config/aarch64/aarch64-simd.md:
4341 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4342 conditional on TARGET_AES.
4343 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4344 (aarch64_crypto_sha1hsi): Make pattern conditional
4346 (aarch64_crypto_sha1hv4si): Ditto.
4347 (aarch64_be_crypto_sha1hv4si): Ditto.
4348 (aarch64_crypto_sha1su1v4si): Ditto.
4349 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4350 (aarch64_crypto_sha1su0v4si): Ditto.
4351 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4352 (aarch64_crypto_sha256su0v4si): Ditto.
4353 (aarch64_crypto_sha256su1v4si): Ditto.
4354 (doc/invoke.texi): Document new aes and sha2 options.
4356 2018-01-10 Martin Sebor <msebor@redhat.com>
4358 PR tree-optimization/83781
4359 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4362 2018-01-11 Martin Sebor <msebor@gmail.com>
4363 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4365 PR tree-optimization/83501
4366 PR tree-optimization/81703
4368 * tree-ssa-strlen.c (get_string_cst): Rename...
4369 (get_string_len): ...to this. Handle global constants.
4370 (handle_char_store): Adjust.
4372 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4373 Jim Wilson <jimw@sifive.com>
4375 * config/riscv/riscv-protos.h (riscv_output_return): New.
4376 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4377 (riscv_attribute_table, riscv_output_return),
4378 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4379 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4380 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4382 (riscv_expand_prologue): Add early return for naked function.
4383 (riscv_expand_epilogue): Likewise.
4384 (riscv_function_ok_for_sibcall): Return false for naked function.
4385 (riscv_set_current_function): New.
4386 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4387 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4388 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4389 * doc/extend.texi (RISC-V Function Attributes): New.
4391 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4393 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4394 check for 128-bit long double before checking TCmode.
4395 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4396 128-bit long doubles before checking TFmode or TCmode.
4397 (FLOAT128_IBM_P): Likewise.
4399 2018-01-10 Martin Sebor <msebor@redhat.com>
4401 PR tree-optimization/83671
4402 * builtins.c (c_strlen): Unconditionally return zero for the empty
4404 Use -Warray-bounds for warnings.
4405 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4406 for non-constant array indices with COMPONENT_REF, arrays of
4407 arrays, and pointers to arrays.
4408 (gimple_fold_builtin_strlen): Determine and set length range for
4409 non-constant character arrays.
4411 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4414 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4417 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4419 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4421 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4424 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4425 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4426 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4427 indexed_or_indirect_operand predicate.
4428 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4429 (*vsx_le_perm_load_v8hi): Likewise.
4430 (*vsx_le_perm_load_v16qi): Likewise.
4431 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4432 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4433 (*vsx_le_perm_store_v8hi): Likewise.
4434 (*vsx_le_perm_store_v16qi): Likewise.
4435 (eight unnamed splitters): Likewise.
4437 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4439 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4440 * config/rs6000/emmintrin.h: Likewise.
4441 * config/rs6000/mmintrin.h: Likewise.
4442 * config/rs6000/xmmintrin.h: Likewise.
4444 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4447 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4449 * tree.c (tree_nop_conversion): Return true for location wrapper
4451 (maybe_wrap_with_location): New function.
4452 (selftest::check_strip_nops): New function.
4453 (selftest::test_location_wrappers): New function.
4454 (selftest::tree_c_tests): Call it.
4455 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4456 (maybe_wrap_with_location): New decl.
4457 (EXPR_LOCATION_WRAPPER_P): New macro.
4458 (location_wrapper_p): New inline function.
4459 (tree_strip_any_location_wrapper): New inline function.
4461 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4464 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4465 stack_realign_offset for the largest alignment of stack slot
4467 (ix86_find_max_used_stack_alignment): New function.
4468 (ix86_finalize_stack_frame_flags): Use it. Set
4469 max_used_stack_alignment if we don't realign stack.
4470 * config/i386/i386.h (machine_function): Add
4471 max_used_stack_alignment.
4473 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4475 * config/arm/arm.opt (-mbranch-cost): New option.
4476 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4479 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4482 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4483 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4485 2018-01-10 Richard Biener <rguenther@suse.de>
4488 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4489 early out so it also covers the case where we have a non-NULL
4492 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4494 PR tree-optimization/83753
4495 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4496 for non-strided grouped accesses if the number of elements is 1.
4498 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4501 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4502 * i386.h (TARGET_USE_GATHER): Define.
4503 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4505 2018-01-10 Martin Liska <mliska@suse.cz>
4508 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4509 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4511 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4512 CLEANUP_NO_PARTITIONING is not set.
4514 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4516 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4517 for vectors, as a partial revert of r254296.
4518 * rtl.h (const_vec_p): Delete.
4519 (const_vec_duplicate_p): Don't test for vector CONSTs.
4520 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4521 * expmed.c (make_tree): Likewise.
4524 * common.md (E, F): Use CONSTANT_P instead of checking for
4526 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4527 checking for CONST_VECTOR.
4529 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4532 * predict.c (force_edge_cold): Handle in more sane way edges
4535 2018-01-09 Carl Love <cel@us.ibm.com>
4537 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4539 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4540 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4541 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4542 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4543 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4544 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4545 * config/rs6000/rs6000-protos.h: Add extern defition for
4546 rs6000_generate_float2_double_code.
4547 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4549 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4550 (float2_v2df): Add define_expand.
4552 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4555 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4556 op_mode in the force_to_mode call.
4558 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4560 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4561 instead of checking each element individually.
4562 (aarch64_evpc_uzp): Likewise.
4563 (aarch64_evpc_zip): Likewise.
4564 (aarch64_evpc_ext): Likewise.
4565 (aarch64_evpc_rev): Likewise.
4566 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4567 instead of checking each element individually. Return true without
4569 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4570 whether all selected elements come from the same input, instead of
4571 checking each element individually. Remove calls to gen_rtx_REG,
4572 start_sequence and end_sequence and instead assert that no rtl is
4575 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4577 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4578 order of HIGH and CONST checks.
4580 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4582 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4583 if the destination isn't an SSA_NAME.
4585 2018-01-09 Richard Biener <rguenther@suse.de>
4587 PR tree-optimization/83668
4588 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4590 (canonicalize_loop_form): ... here, renamed from ...
4591 (canonicalize_loop_closed_ssa_form): ... this and amended to
4592 swap successor edges for loop exit blocks to make us use
4593 the RPO order we need for initial schedule generation.
4595 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4597 PR tree-optimization/64811
4598 * match.pd: When optimizing comparisons with Inf, avoid
4599 introducing or losing exceptions from comparisons with NaN.
4601 2018-01-09 Martin Liska <mliska@suse.cz>
4604 * asan.c (shadow_mem_size): Add gcc_assert.
4606 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4608 Don't save registers in main().
4611 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4612 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4613 * config/avr/avr.c (avr_set_current_function): Don't error if
4614 naked, OS_task or OS_main are specified at the same time.
4615 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4617 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4619 * common/config/avr/avr-common.c (avr_option_optimization_table):
4620 Switch on -mmain-is-OS_task for optimizing compilations.
4622 2018-01-09 Richard Biener <rguenther@suse.de>
4624 PR tree-optimization/83572
4625 * graphite.c: Include cfganal.h.
4626 (graphite_transform_loops): Connect infinite loops to exit
4627 and remove fake edges at the end.
4629 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4631 * ipa-inline.c (edge_badness): Revert accidental checkin.
4633 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4636 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4637 symbols; not inline clones.
4639 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4642 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4643 hard registers. Formatting fixes.
4645 PR preprocessor/83722
4646 * gcc.c (try_generate_repro): Pass
4647 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4648 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4651 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4652 Kito Cheng <kito.cheng@gmail.com>
4654 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4655 (riscv_leaf_function_p): Delete.
4656 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4658 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4660 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4662 (do_ifelse): New function.
4663 (do_isel): New function.
4664 (do_sub3): New function.
4665 (do_add3): New function.
4666 (do_load_mask_compare): New function.
4667 (do_overlap_load_compare): New function.
4668 (expand_compare_loop): New function.
4669 (expand_block_compare): Call expand_compare_loop() when appropriate.
4670 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4672 (-mblock-compare-inline-loop-limit): New option.
4674 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4677 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4678 Reverse order of second and third operands in first alternative.
4679 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4680 of first and second elements in UNSPEC_VPERMR vector.
4681 (altivec_expand_vec_perm_le): Likewise.
4683 2018-01-08 Jeff Law <law@redhat.com>
4685 PR rtl-optimizatin/81308
4686 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4687 (process_switch): If group_case_labels makes a change, then set
4689 (pass_convert_switch::execute): If a switch is converted, then
4690 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4692 PR rtl-optimization/81308
4693 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4696 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4698 PR target/83663 - Revert r255946
4699 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4700 generation for cases where splatting a value is not useful.
4701 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4702 across a vec_duplicate and a paradoxical subreg forming a vector
4703 mode to a vec_concat.
4705 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4707 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4708 -march=armv8.3-a variants.
4709 * config/arm/t-multilib: Likewise.
4710 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4712 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4714 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4716 (cceq_ior_compare_complement): Give it a name so I can use it, and
4717 change boolean_or_operator predicate to boolean_operator so it can
4718 be used to generate a crand.
4719 (eqne): New code iterator.
4720 (bd/bd_neg): New code_attrs.
4721 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4722 a single define_insn.
4723 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4724 decrement (bdnzt/bdnzf/bdzt/bdzf).
4725 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4726 with the new names of the branch decrement patterns, and added the
4727 names of the branch decrement conditional patterns.
4729 2018-01-08 Richard Biener <rguenther@suse.de>
4731 PR tree-optimization/83563
4732 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4735 2018-01-08 Richard Biener <rguenther@suse.de>
4738 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4740 2018-01-08 Richard Biener <rguenther@suse.de>
4742 PR tree-optimization/83685
4743 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4744 references to abnormals.
4746 2018-01-08 Richard Biener <rguenther@suse.de>
4749 * dwarf2out.c (output_indirect_strings): Handle empty
4750 skeleton_debug_str_hash.
4751 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4753 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4755 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4756 (emit_store_direct): Likewise.
4757 (arc_trampoline_adjust_address): Likewise.
4758 (arc_asm_trampoline_template): New function.
4759 (arc_initialize_trampoline): Use asm_trampoline_template.
4760 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4761 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4762 * config/arc/arc.md (flush_icache): Delete pattern.
4764 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4766 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4767 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4770 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4773 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4774 by not USED_FOR_TARGET.
4775 (make_pass_resolve_sw_modes): Likewise.
4777 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4779 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4782 2018-01-08 Richard Biener <rguenther@suse.de>
4785 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4787 2018-01-08 Richard Biener <rguenther@suse.de>
4790 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4792 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4795 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4796 basic blocks with a small number of successors.
4797 (convert_control_dep_chain_into_preds): Improve handling of
4799 (dump_predicates): Split apart into...
4800 (dump_pred_chain): ...here...
4801 (dump_pred_info): ...and here.
4802 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4803 (can_chain_union_be_invalidated_p): Improve check for invalidation
4805 (uninit_uses_cannot_happen): Avoid unnecessary if
4806 convert_control_dep_chain_into_preds yielded nothing.
4808 2018-01-06 Martin Sebor <msebor@redhat.com>
4810 PR tree-optimization/83640
4811 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4812 subtracting negative offset from size.
4813 (builtin_access::overlap): Adjust offset bounds of the access to fall
4814 within the size of the object if possible.
4816 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4818 PR rtl-optimization/83699
4819 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4820 extract_bit_field_as_subreg to cases in which the extracted
4821 value is also a vector.
4823 * lra-constraints.c (process_alt_operands): Test for the equivalence
4824 substitutions when detecting a possible reload cycle.
4826 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4829 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4830 by default if flag_selective_schedling{,2}. Formatting fixes.
4832 PR rtl-optimization/83682
4833 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4834 if it has non-VECTOR_MODE element mode.
4835 (vec_duplicate_p): Likewise.
4838 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4839 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4841 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4844 * config/i386/i386-builtin.def
4845 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4846 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4847 Require also OPTION_MASK_ISA_AVX512F in addition to
4848 OPTION_MASK_ISA_GFNI.
4849 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4850 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4851 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4852 to OPTION_MASK_ISA_GFNI.
4853 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4854 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4855 OPTION_MASK_ISA_AVX512BW.
4856 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4857 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4858 addition to OPTION_MASK_ISA_GFNI.
4859 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4860 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4861 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4862 to OPTION_MASK_ISA_GFNI.
4863 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4864 a requirement for all ISAs rather than any of them with a few
4866 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4868 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4869 bitmasks to be enabled with 3 exceptions, instead of requiring any
4870 enabled ISA with lots of exceptions.
4871 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4872 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4873 Change avx512bw in isa attribute to avx512f.
4874 * config/i386/sgxintrin.h: Add license boilerplate.
4875 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4876 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4877 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4878 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4880 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4881 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4882 temporarily sse2 rather than sse if not enabled already.
4885 * config/i386/sse.md (VI248_VLBW): Rename to ...
4886 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4887 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4888 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4889 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4890 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4891 mode iterator instead of VI248_VLBW.
4893 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4895 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4896 (record_modified): Skip clobbers; add debug output.
4897 (param_change_prob): Use sreal frequencies.
4899 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4901 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4902 punt for user-aligned variables.
4904 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4906 * tree-chrec.c (chrec_contains_symbols): Return true for
4909 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4912 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4913 of (x|y) == x for BICS pattern.
4915 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4917 PR tree-optimization/83605
4918 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4919 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4922 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4924 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4925 * config/epiphany/rtems.h: New file.
4927 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4928 Uros Bizjak <ubizjak@gmail.com>
4931 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4932 QIreg_operand instead of register_operand predicate.
4933 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4934 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4935 comments instead of -fmitigate[-_]rop.
4937 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4940 * cgraphunit.c (symbol_table::compile): Switch to text_section
4941 before calling assembly_start debug hook.
4942 * run-rtl-passes.c (run_rtl_passes): Likewise.
4945 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4947 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4948 range_int_cst_p rather than !symbolic_range_p before calling
4949 extract_range_from_multiplicative_op_1.
4951 2018-01-04 Jeff Law <law@redhat.com>
4953 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4954 redundant test in assertion.
4956 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4958 * doc/rtl.texi: Document machine_mode wrapper classes.
4960 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4962 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4965 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4967 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4968 the VEC_PERM_EXPR fold to fail.
4970 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4973 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4974 to switched_sections.
4976 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4979 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4982 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4985 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4986 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4988 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4991 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4992 is BLKmode and bitpos not zero or mode change is needed.
4994 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4997 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5000 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5003 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5004 instead of MULT rtx. Update all corresponding splitters.
5006 (*ssub<modesuffix>): Ditto.
5008 (*cmp_sadd_di): Update split patterns.
5009 (*cmp_sadd_si): Ditto.
5010 (*cmp_sadd_sidi): Ditto.
5011 (*cmp_ssub_di): Ditto.
5012 (*cmp_ssub_si): Ditto.
5013 (*cmp_ssub_sidi): Ditto.
5014 * config/alpha/predicates.md (const23_operand): New predicate.
5015 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5016 Look for ASHIFT, not MULT inner operand.
5017 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5019 2018-01-04 Martin Liska <mliska@suse.cz>
5021 PR gcov-profile/83669
5022 * gcov.c (output_intermediate_file): Add version to intermediate
5024 * doc/gcov.texi: Document new field 'version' in intermediate
5025 file format. Fix location of '-k' option of gcov command.
5027 2018-01-04 Martin Liska <mliska@suse.cz>
5030 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5032 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5034 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5036 2018-01-03 Martin Sebor <msebor@redhat.com>
5038 PR tree-optimization/83655
5039 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5040 checking calls with invalid arguments.
5042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5044 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5045 (vectorizable_mask_load_store): Delete.
5046 (vectorizable_call): Return false for masked loads and stores.
5047 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5048 instead of gimple_assign_rhs1.
5049 (vectorizable_load): Handle IFN_MASK_LOAD.
5050 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5054 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5056 (vectorizable_mask_load_store): ...here.
5057 (vectorizable_load): ...and here.
5059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5061 * tree-vect-stmts.c (vect_build_all_ones_mask)
5062 (vect_build_zero_merge_argument): New functions, split out from...
5063 (vectorizable_load): ...here.
5065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5067 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5069 (vectorizable_mask_load_store): ...here.
5070 (vectorizable_store): ...and here.
5072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5074 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5076 (vectorizable_mask_load_store): ...here.
5078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5080 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5081 (vect_model_store_cost): Take a vec_load_store_type instead of a
5083 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5084 (vect_model_store_cost): Take a vec_load_store_type instead of a
5086 (vectorizable_mask_load_store): Update accordingly.
5087 (vectorizable_store): Likewise.
5088 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5090 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5092 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5093 IFN_MASK_LOAD calls here rather than...
5094 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5097 Alan Hayward <alan.hayward@arm.com>
5098 David Sherwood <david.sherwood@arm.com>
5100 * expmed.c (extract_bit_field_1): For vector extracts,
5101 fall back to extract_bit_field_as_subreg if vec_extract
5104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5105 Alan Hayward <alan.hayward@arm.com>
5106 David Sherwood <david.sherwood@arm.com>
5108 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5109 they are variable or constant sized.
5110 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5111 slots for constant-sized data.
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 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5118 handling COND_EXPRs with boolean comparisons, try to find a better
5119 basis for the mask type than the boolean itself.
5121 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5123 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5124 is calculated and how it can be overridden.
5125 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5126 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5128 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5131 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5132 Alan Hayward <alan.hayward@arm.com>
5133 David Sherwood <david.sherwood@arm.com>
5135 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5136 Remove the mode argument.
5137 (aarch64_simd_valid_immediate): Remove the mode and inverse
5139 * config/aarch64/iterators.md (bitsize): New iterator.
5140 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5141 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5142 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5143 aarch64_simd_valid_immediate.
5144 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5145 (aarch64_reg_or_bic_imm): Likewise.
5146 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5147 with an insn_type enum and msl with a modifier_type enum.
5148 Replace element_width with a scalar_mode. Change the shift
5149 to unsigned int. Add constructors for scalar_float_mode and
5150 scalar_int_mode elements.
5151 (aarch64_vect_float_const_representable_p): Delete.
5152 (aarch64_can_const_movi_rtx_p)
5153 (aarch64_simd_scalar_immediate_valid_for_move)
5154 (aarch64_simd_make_constant): Update call to
5155 aarch64_simd_valid_immediate.
5156 (aarch64_advsimd_valid_immediate_hs): New function.
5157 (aarch64_advsimd_valid_immediate): Likewise.
5158 (aarch64_simd_valid_immediate): Remove mode and inverse
5159 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5160 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5161 and aarch64_float_const_representable_p on the result.
5162 (aarch64_output_simd_mov_immediate): Remove mode argument.
5163 Update call to aarch64_simd_valid_immediate and use of
5164 simd_immediate_info.
5165 (aarch64_output_scalar_simd_mov_immediate): Update call
5168 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5169 Alan Hayward <alan.hayward@arm.com>
5170 David Sherwood <david.sherwood@arm.com>
5172 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5173 (mode_nunits): Likewise CONST_MODE_NUNITS.
5174 * machmode.def (ADJUST_NUNITS): Document.
5175 * genmodes.c (mode_data::need_nunits_adj): New field.
5176 (blank_mode): Update accordingly.
5177 (adj_nunits): New variable.
5178 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5180 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5181 listed in adj_nunits.
5182 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5183 listed in adj_nunits. Don't emit case statements for such modes.
5184 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5185 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5186 nothing if adj_nunits is nonnull.
5187 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5188 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5189 (emit_mode_fbit): Update use of print_maybe_const_decl.
5190 (emit_move_size): Likewise. Treat the array as non-const
5192 (emit_mode_adjustments): Handle adj_nunits.
5194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5196 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5197 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5198 (VECTOR_MODES): Use it.
5199 (make_vector_modes): Take the prefix as an argument.
5201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5202 Alan Hayward <alan.hayward@arm.com>
5203 David Sherwood <david.sherwood@arm.com>
5205 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5206 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5207 for MODE_VECTOR_BOOL.
5208 * machmode.def (VECTOR_BOOL_MODE): Document.
5209 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5210 (make_vector_bool_mode): New function.
5211 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5213 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5214 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5216 * stor-layout.c (int_mode_for_mode): Likewise.
5217 * tree.c (build_vector_type_for_mode): Likewise.
5218 * varasm.c (output_constant_pool_2): Likewise.
5219 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5220 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5221 for MODE_VECTOR_BOOL.
5222 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5223 of mode class checks.
5224 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5225 instead of a list of mode class checks.
5226 (expand_vector_scalar_condition): Likewise.
5227 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5230 Alan Hayward <alan.hayward@arm.com>
5231 David Sherwood <david.sherwood@arm.com>
5233 * machmode.h (mode_size): Change from unsigned short to
5235 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5236 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5237 or if measurement_type is not polynomial.
5238 (fixed_size_mode::includes_p): Check for constant-sized modes.
5239 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5240 return a poly_uint16 rather than an unsigned short.
5241 (emit_mode_size): Change the type of mode_size from unsigned short
5242 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5243 (emit_mode_adjustments): Cope with polynomial vector sizes.
5244 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5246 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5248 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5249 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5250 * caller-save.c (setup_save_areas): Likewise.
5251 (replace_reg_with_saved_mem): Likewise.
5252 * calls.c (emit_library_call_value_1): Likewise.
5253 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5254 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5255 (gen_lowpart_for_combine): Likewise.
5256 * convert.c (convert_to_integer_1): Likewise.
5257 * cse.c (equiv_constant, cse_insn): Likewise.
5258 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5259 (cselib_subst_to_values): Likewise.
5260 * dce.c (word_dce_process_block): Likewise.
5261 * df-problems.c (df_word_lr_mark_ref): Likewise.
5262 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5263 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5264 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5265 (rtl_for_decl_location): Likewise.
5266 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5267 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5268 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5269 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5270 (expand_expr_real_1): Likewise.
5271 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5272 (pad_below): Likewise.
5273 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5274 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5275 * ira.c (get_subreg_tracking_sizes): Likewise.
5276 * ira-build.c (ira_create_allocno_objects): Likewise.
5277 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5278 (ira_sort_regnos_for_alter_reg): Likewise.
5279 * ira-costs.c (record_operand_costs): Likewise.
5280 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5281 (resolve_simple_move): Likewise.
5282 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5283 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5284 (lra_constraints): Likewise.
5285 (CONST_POOL_OK_P): Reject variable-sized modes.
5286 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5287 (add_pseudo_to_slot, lra_spill): Likewise.
5288 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5289 * optabs-query.c (get_best_extraction_insn): Likewise.
5290 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5291 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5292 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5293 * recog.c (offsettable_address_addr_space_p): Likewise.
5294 * regcprop.c (maybe_mode_change): Likewise.
5295 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5296 * regrename.c (build_def_use): Likewise.
5297 * regstat.c (dump_reg_info): Likewise.
5298 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5299 (find_reloads, find_reloads_subreg_address): Likewise.
5300 * reload1.c (eliminate_regs_1): Likewise.
5301 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5302 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5303 (simplify_binary_operation_1, simplify_subreg): Likewise.
5304 * targhooks.c (default_function_arg_padding): Likewise.
5305 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5306 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5307 (verify_gimple_assign_ternary): Likewise.
5308 * tree-inline.c (estimate_move_cost): Likewise.
5309 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5310 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5311 (get_address_cost_ainc): Likewise.
5312 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5313 (vect_supportable_dr_alignment): Likewise.
5314 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5315 (vectorizable_reduction): Likewise.
5316 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5317 (vectorizable_operation, vectorizable_load): Likewise.
5318 * tree.c (build_same_sized_truth_vector_type): Likewise.
5319 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5320 * var-tracking.c (emit_note_insn_var_location): Likewise.
5321 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5322 (ADDR_VEC_ALIGN): Likewise.
5324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5325 Alan Hayward <alan.hayward@arm.com>
5326 David Sherwood <david.sherwood@arm.com>
5328 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5330 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5331 or if measurement_type is polynomial.
5332 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5333 * combine.c (make_extraction): Likewise.
5334 * dse.c (find_shift_sequence): Likewise.
5335 * dwarf2out.c (mem_loc_descriptor): Likewise.
5336 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5337 (extract_bit_field, extract_low_bits): Likewise.
5338 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5339 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5340 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5341 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5342 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5343 * reload.c (find_reloads): Likewise.
5344 * reload1.c (alter_reg): Likewise.
5345 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5346 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5347 * tree-if-conv.c (predicate_mem_writes): Likewise.
5348 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5349 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5350 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5351 * valtrack.c (dead_debug_insert_temp): Likewise.
5352 * varasm.c (mergeable_constant_section): Likewise.
5353 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5355 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5356 Alan Hayward <alan.hayward@arm.com>
5357 David Sherwood <david.sherwood@arm.com>
5359 * expr.c (expand_assignment): Cope with polynomial mode sizes
5360 when assigning to a CONCAT.
5362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5363 Alan Hayward <alan.hayward@arm.com>
5364 David Sherwood <david.sherwood@arm.com>
5366 * machmode.h (mode_precision): Change from unsigned short to
5368 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5370 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5371 or if measurement_type is not polynomial.
5372 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5373 in which the mode is already known to be a scalar_int_mode.
5374 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5375 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5377 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5378 for GET_MODE_PRECISION.
5379 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5380 for GET_MODE_PRECISION.
5381 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5383 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5384 (expand_field_assignment, make_extraction): Likewise.
5385 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5386 (get_last_value): Likewise.
5387 * convert.c (convert_to_integer_1): Likewise.
5388 * cse.c (cse_insn): Likewise.
5389 * expr.c (expand_expr_real_1): Likewise.
5390 * lra-constraints.c (simplify_operand_subreg): Likewise.
5391 * optabs-query.c (can_atomic_load_p): Likewise.
5392 * optabs.c (expand_atomic_load): Likewise.
5393 (expand_atomic_store): Likewise.
5394 * ree.c (combine_reaching_defs): Likewise.
5395 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5396 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5397 * tree.h (type_has_mode_precision_p): Likewise.
5398 * ubsan.c (instrument_si_overflow): Likewise.
5400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5401 Alan Hayward <alan.hayward@arm.com>
5402 David Sherwood <david.sherwood@arm.com>
5404 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5405 polynomial numbers of units.
5406 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5407 (valid_vector_subparts_p): New function.
5408 (build_vector_type): Remove temporary shim and take the number
5409 of units as a poly_uint64 rather than an int.
5410 (build_opaque_vector_type): Take the number of units as a
5411 poly_uint64 rather than an int.
5412 * tree.c (build_vector_from_ctor): Handle polynomial
5413 TYPE_VECTOR_SUBPARTS.
5414 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5415 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5416 (build_vector_from_val): If the number of units is variable,
5417 use build_vec_duplicate_cst for constant operands and
5418 VEC_DUPLICATE_EXPR otherwise.
5419 (make_vector_type): Remove temporary is_constant ().
5420 (build_vector_type, build_opaque_vector_type): Take the number of
5421 units as a poly_uint64 rather than an int.
5422 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5424 * cfgexpand.c (expand_debug_expr): Likewise.
5425 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5426 (store_constructor, expand_expr_real_1): Likewise.
5427 (const_scalar_mask_from_tree): Likewise.
5428 * fold-const-call.c (fold_const_reduction): Likewise.
5429 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5430 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5431 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5432 (fold_relational_const): Likewise.
5433 (native_interpret_vector): Likewise. Change the size from an
5434 int to an unsigned int.
5435 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5436 TYPE_VECTOR_SUBPARTS.
5437 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5438 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5439 duplicating a non-constant operand into a variable-length vector.
5440 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5441 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5442 * ipa-icf.c (sem_variable::equals): Likewise.
5443 * match.pd: Likewise.
5444 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5445 * print-tree.c (print_node): Likewise.
5446 * stor-layout.c (layout_type): Likewise.
5447 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5448 * tree-cfg.c (verify_gimple_comparison): Likewise.
5449 (verify_gimple_assign_binary): Likewise.
5450 (verify_gimple_assign_ternary): Likewise.
5451 (verify_gimple_assign_single): Likewise.
5452 * tree-pretty-print.c (dump_generic_node): Likewise.
5453 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5454 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5455 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5456 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5457 (vect_shift_permute_load_chain): Likewise.
5458 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5459 (expand_vector_condition, optimize_vector_constructor): Likewise.
5460 (lower_vec_perm, get_compute_type): Likewise.
5461 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5462 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5463 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5464 (vect_recog_mask_conversion_pattern): Likewise.
5465 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5466 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5467 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5468 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5469 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5470 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5471 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5472 (supportable_widening_operation): Likewise.
5473 (supportable_narrowing_operation): Likewise.
5474 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5476 * varasm.c (output_constant): Likewise.
5478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5479 Alan Hayward <alan.hayward@arm.com>
5480 David Sherwood <david.sherwood@arm.com>
5482 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5483 so that both the length == 3 and length != 3 cases set up their
5484 own permute vectors. Add comments explaining why we know the
5485 number of elements is constant.
5486 (vect_permute_load_chain): Likewise.
5488 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5489 Alan Hayward <alan.hayward@arm.com>
5490 David Sherwood <david.sherwood@arm.com>
5492 * machmode.h (mode_nunits): Change from unsigned char to
5494 (ONLY_FIXED_SIZE_MODES): New macro.
5495 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5496 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5497 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5499 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5500 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5501 or if measurement_type is not polynomial.
5502 * genmodes.c (ZERO_COEFFS): New macro.
5503 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5505 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5506 Use ZERO_COEFFS when emitting initializers.
5507 * data-streamer.h (bp_pack_poly_value): New function.
5508 (bp_unpack_poly_value): Likewise.
5509 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5510 for GET_MODE_NUNITS.
5511 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5512 for GET_MODE_NUNITS.
5513 * tree.c (make_vector_type): Remove temporary shim and make
5514 the real function take the number of units as a poly_uint64
5516 (build_vector_type_for_mode): Handle polynomial nunits.
5517 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5518 * emit-rtl.c (const_vec_series_p_1): Likewise.
5519 (gen_rtx_CONST_VECTOR): Likewise.
5520 * fold-const.c (test_vec_duplicate_folding): Likewise.
5521 * genrecog.c (validate_pattern): Likewise.
5522 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5523 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5524 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5525 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5526 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5527 * rtlanal.c (subreg_get_info): Likewise.
5528 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5529 (vect_grouped_load_supported): Likewise.
5530 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5531 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5532 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5533 (simplify_const_unary_operation, simplify_binary_operation_1)
5534 (simplify_const_binary_operation, simplify_ternary_operation)
5535 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5536 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5537 instead of CONST_VECTOR_NUNITS.
5538 * varasm.c (output_constant_pool_2): Likewise.
5539 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5540 explicit-encoded elements in the XVEC for variable-length vectors.
5542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5544 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5546 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5547 Alan Hayward <alan.hayward@arm.com>
5548 David Sherwood <david.sherwood@arm.com>
5550 * coretypes.h (fixed_size_mode): Declare.
5551 (fixed_size_mode_pod): New typedef.
5552 * builtins.h (target_builtins::x_apply_args_mode)
5553 (target_builtins::x_apply_result_mode): Change type to
5554 fixed_size_mode_pod.
5555 * builtins.c (apply_args_size, apply_result_size, result_vector)
5556 (expand_builtin_apply_args_1, expand_builtin_apply)
5557 (expand_builtin_return): Update accordingly.
5559 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5561 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5562 * cselib.c (cselib_hash_rtx): Likewise.
5563 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5564 CONST_VECTOR encoding.
5566 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5567 Jeff Law <law@redhat.com>
5570 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5571 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5572 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5573 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5576 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5577 explicitly probe *sp in a noreturn function if there were any callee
5578 register saves or frame pointer is needed.
5580 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5583 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5584 BLKmode for ternary, binary or unary expressions.
5587 * var-tracking.c (delete_vta_debug_insn): New inline function.
5588 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5589 insns from get_insns () to NULL instead of each bb separately.
5590 Use delete_vta_debug_insn. No longer static.
5591 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5592 delete_vta_debug_insns callers.
5593 * rtl.h (delete_vta_debug_insns): Declare.
5594 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5595 instead of variable_tracking_main.
5597 2018-01-03 Martin Sebor <msebor@redhat.com>
5599 PR tree-optimization/83603
5600 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5601 arguments past the endof the argument list in functions declared
5602 without a prototype.
5603 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5604 Avoid checking when arguments are null.
5606 2018-01-03 Martin Sebor <msebor@redhat.com>
5609 * doc/extend.texi (attribute const): Fix a typo.
5610 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5611 issuing -Wsuggest-attribute for void functions.
5613 2018-01-03 Martin Sebor <msebor@redhat.com>
5615 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5616 offset_int::from instead of wide_int::to_shwi.
5617 (maybe_diag_overlap): Remove assertion.
5618 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5619 * gimple-ssa-sprintf.c (format_directive): Same.
5620 (parse_directive): Same.
5621 (sprintf_dom_walker::compute_format_length): Same.
5622 (try_substitute_return_value): Same.
5624 2018-01-03 Jeff Law <law@redhat.com>
5627 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5628 non-constant residual for zero at runtime and avoid probing in
5629 that case. Reorganize code for trailing problem to mirror handling
5632 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5634 PR tree-optimization/83501
5635 * tree-ssa-strlen.c (get_string_cst): New.
5636 (handle_char_store): Call get_string_cst.
5638 2018-01-03 Martin Liska <mliska@suse.cz>
5640 PR tree-optimization/83593
5641 * tree-ssa-strlen.c: Include tree-cfg.h.
5642 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5643 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5644 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5646 (strlen_dom_walker::before_dom_children): Call
5647 gimple_purge_dead_eh_edges. Dump tranformation with details
5649 (strlen_dom_walker::before_dom_children): Update call by adding
5650 new argument cleanup_eh.
5651 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5653 2018-01-03 Martin Liska <mliska@suse.cz>
5656 * cif-code.def (VARIADIC_THUNK): New enum value.
5657 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5660 2018-01-03 Jan Beulich <jbeulich@suse.com>
5662 * sse.md (mov<mode>_internal): Tighten condition for when to use
5663 vmovdqu<ssescalarsize> for TI and OI modes.
5665 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5667 Update copyright years.
5669 2018-01-03 Martin Liska <mliska@suse.cz>
5672 * ipa-visibility.c (function_and_variable_visibility): Skip
5673 functions with noipa attribure.
5675 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5677 * gcc.c (process_command): Update copyright notice dates.
5678 * gcov-dump.c (print_version): Ditto.
5679 * gcov.c (print_version): Ditto.
5680 * gcov-tool.c (print_version): Ditto.
5681 * gengtype.c (create_file): Ditto.
5682 * doc/cpp.texi: Bump @copying's copyright year.
5683 * doc/cppinternals.texi: Ditto.
5684 * doc/gcc.texi: Ditto.
5685 * doc/gccint.texi: Ditto.
5686 * doc/gcov.texi: Ditto.
5687 * doc/install.texi: Ditto.
5688 * doc/invoke.texi: Ditto.
5690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5692 * vector-builder.h (vector_builder::m_full_nelts): Change from
5693 unsigned int to poly_uint64.
5694 (vector_builder::full_nelts): Update prototype accordingly.
5695 (vector_builder::new_vector): Likewise.
5696 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5697 (vector_builder::operator ==): Likewise.
5698 (vector_builder::finalize): Likewise.
5699 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5700 Take the number of elements as a poly_uint64 rather than an
5702 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5703 from unsigned int to poly_uint64.
5704 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5705 (vec_perm_indices::new_vector): Likewise.
5706 (vec_perm_indices::length): Likewise.
5707 (vec_perm_indices::nelts_per_input): Likewise.
5708 (vec_perm_indices::input_nelts): Likewise.
5709 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5710 number of elements per input as a poly_uint64 rather than an
5711 unsigned int. Use the original encoding for variable-length
5712 vectors, rather than clamping each individual element.
5713 For the second and subsequent elements in each pattern,
5714 clamp the step and base before clamping their sum.
5715 (vec_perm_indices::series_p): Handle polynomial element counts.
5716 (vec_perm_indices::all_in_range_p): Likewise.
5717 (vec_perm_indices_to_tree): Likewise.
5718 (vec_perm_indices_to_rtx): Likewise.
5719 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5720 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5721 (tree_vector_builder::new_binary_operation): Handle polynomial
5722 element counts. Return false if we need to know the number
5723 of elements at compile time.
5724 * fold-const.c (fold_vec_perm): Punt if the number of elements
5725 isn't known at compile time.
5727 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5729 * vec-perm-indices.h (vec_perm_builder): Change element type
5730 from HOST_WIDE_INT to poly_int64.
5731 (vec_perm_indices::element_type): Update accordingly.
5732 (vec_perm_indices::clamp): Handle polynomial element_types.
5733 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5734 (vec_perm_indices::all_in_range_p): Likewise.
5735 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5737 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5738 polynomial vec_perm_indices element types.
5739 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5740 * fold-const.c (fold_vec_perm): Likewise.
5741 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5742 * tree-vect-generic.c (lower_vec_perm): Likewise.
5743 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5744 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5745 element type to HOST_WIDE_INT.
5747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5748 Alan Hayward <alan.hayward@arm.com>
5749 David Sherwood <david.sherwood@arm.com>
5751 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5752 rather than an int. Use plus_constant.
5753 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5754 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5757 Alan Hayward <alan.hayward@arm.com>
5758 David Sherwood <david.sherwood@arm.com>
5760 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5761 a HOST_WIDE_INT to a poly_int64.
5763 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5764 Alan Hayward <alan.hayward@arm.com>
5765 David Sherwood <david.sherwood@arm.com>
5767 * calls.c (load_register_parameters): Cope with polynomial
5768 mode sizes. Require a constant size for BLKmode parameters
5769 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5770 forces a parameter to be padded at the lsb end in order to
5771 fill a complete number of words, require the parameter size
5772 to be ordered wrt UNITS_PER_WORD.
5774 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5775 Alan Hayward <alan.hayward@arm.com>
5776 David Sherwood <david.sherwood@arm.com>
5778 * reload1.c (spill_stack_slot_width): Change element type
5779 from unsigned int to poly_uint64_pod.
5780 (alter_reg): Treat mode sizes as polynomial.
5782 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5783 Alan Hayward <alan.hayward@arm.com>
5784 David Sherwood <david.sherwood@arm.com>
5786 * reload.c (complex_word_subreg_p): New function.
5787 (reload_inner_reg_of_subreg, push_reload): Use it.
5789 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5790 Alan Hayward <alan.hayward@arm.com>
5791 David Sherwood <david.sherwood@arm.com>
5793 * lra-constraints.c (process_alt_operands): Reject matched
5794 operands whose sizes aren't ordered.
5795 (match_reload): Refer to this check here.
5797 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5798 Alan Hayward <alan.hayward@arm.com>
5799 David Sherwood <david.sherwood@arm.com>
5801 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5802 that the mode size is in the set {1, 2, 4, 8, 16}.
5804 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5805 Alan Hayward <alan.hayward@arm.com>
5806 David Sherwood <david.sherwood@arm.com>
5808 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5809 Use plus_constant instead of gen_rtx_PLUS.
5811 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5812 Alan Hayward <alan.hayward@arm.com>
5813 David Sherwood <david.sherwood@arm.com>
5815 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5816 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5817 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5818 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5819 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5820 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5821 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5822 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5823 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5824 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5826 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5827 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5828 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5829 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5830 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5831 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5832 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5833 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5834 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5835 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5837 * expr.c (emit_move_resolve_push): Treat the input and result
5838 of PUSH_ROUNDING as a poly_int64.
5839 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5840 (emit_push_insn): Likewise.
5841 * lra-eliminations.c (mark_not_eliminable): Likewise.
5842 * recog.c (push_operand): Likewise.
5843 * reload1.c (elimination_effects): Likewise.
5844 * rtlanal.c (nonzero_bits1): Likewise.
5845 * calls.c (store_one_arg): Likewise. Require the padding to be
5846 known at compile time.
5848 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5849 Alan Hayward <alan.hayward@arm.com>
5850 David Sherwood <david.sherwood@arm.com>
5852 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5853 Use plus_constant instead of gen_rtx_PLUS.
5855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5856 Alan Hayward <alan.hayward@arm.com>
5857 David Sherwood <david.sherwood@arm.com>
5859 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5862 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5863 Alan Hayward <alan.hayward@arm.com>
5864 David Sherwood <david.sherwood@arm.com>
5866 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5867 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5868 via stack temporaries. Treat the mode size as polynomial too.
5870 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5871 Alan Hayward <alan.hayward@arm.com>
5872 David Sherwood <david.sherwood@arm.com>
5874 * expr.c (expand_expr_real_2): When handling conversions involving
5875 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5876 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5877 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5885 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5886 Alan Hayward <alan.hayward@arm.com>
5887 David Sherwood <david.sherwood@arm.com>
5889 * combine.c (can_change_dest_mode): Handle polynomial
5890 REGMODE_NATURAL_SIZE.
5891 * expmed.c (store_bit_field_1): Likewise.
5892 * expr.c (store_constructor): Likewise.
5893 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5894 and polynomial REGMODE_NATURAL_SIZE.
5895 (gen_lowpart_common): Likewise.
5896 * reginfo.c (record_subregs_of_mode): Likewise.
5897 * rtlanal.c (read_modify_subreg_p): Likewise.
5899 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5900 Alan Hayward <alan.hayward@arm.com>
5901 David Sherwood <david.sherwood@arm.com>
5903 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5904 numbers of elements.
5906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5907 Alan Hayward <alan.hayward@arm.com>
5908 David Sherwood <david.sherwood@arm.com>
5910 * match.pd: Cope with polynomial numbers of vector elements.
5912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5913 Alan Hayward <alan.hayward@arm.com>
5914 David Sherwood <david.sherwood@arm.com>
5916 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5917 in a POINTER_PLUS_EXPR.
5919 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5920 Alan Hayward <alan.hayward@arm.com>
5921 David Sherwood <david.sherwood@arm.com>
5923 * omp-simd-clone.c (simd_clone_subparts): New function.
5924 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5925 (ipa_simd_modify_function_body): Likewise.
5927 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5928 Alan Hayward <alan.hayward@arm.com>
5929 David Sherwood <david.sherwood@arm.com>
5931 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5932 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5933 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5934 (expand_vector_condition, vector_element): Likewise.
5935 (subparts_gt): New function.
5936 (get_compute_type): Use subparts_gt.
5937 (count_type_subparts): Delete.
5938 (expand_vector_operations_1): Use subparts_gt instead of
5939 count_type_subparts.
5941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5942 Alan Hayward <alan.hayward@arm.com>
5943 David Sherwood <david.sherwood@arm.com>
5945 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5946 (vect_compile_time_alias): ...this new function. Do the calculation
5947 on poly_ints rather than trees.
5948 (vect_prune_runtime_alias_test_list): Update call accordingly.
5950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5951 Alan Hayward <alan.hayward@arm.com>
5952 David Sherwood <david.sherwood@arm.com>
5954 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5956 (vect_schedule_slp_instance): Likewise.
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-slp.c (vect_get_and_check_slp_defs): Reject
5963 constant and extern definitions for variable-length vectors.
5964 (vect_get_constant_vectors): Note that the number of units
5965 is known to be constant.
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-stmts.c (vectorizable_conversion): Treat the number
5972 of units as polynomial. Choose between WIDE and NARROW based
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-stmts.c (simd_clone_subparts): New function.
5980 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5983 Alan Hayward <alan.hayward@arm.com>
5984 David Sherwood <david.sherwood@arm.com>
5986 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5987 vectors as polynomial. Use build_index_vector for
5990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5991 Alan Hayward <alan.hayward@arm.com>
5992 David Sherwood <david.sherwood@arm.com>
5994 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5995 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5996 for variable-length vectors.
5997 (vectorizable_mask_load_store): Treat the number of units as
5998 polynomial, asserting that it is constant if the condition has
5999 already been enforced.
6000 (vectorizable_store, vectorizable_load): Likewise.
6002 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6003 Alan Hayward <alan.hayward@arm.com>
6004 David Sherwood <david.sherwood@arm.com>
6006 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6007 of units as polynomial. Punt if we can't tell at compile time
6008 which vector contains the final result.
6010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6011 Alan Hayward <alan.hayward@arm.com>
6012 David Sherwood <david.sherwood@arm.com>
6014 * tree-vect-loop.c (vectorizable_induction): Treat the number
6015 of units as polynomial. Punt on SLP inductions. Use an integer
6016 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6017 cast of such a series for variable-length floating-point
6020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6021 Alan Hayward <alan.hayward@arm.com>
6022 David Sherwood <david.sherwood@arm.com>
6024 * tree.h (build_index_vector): Declare.
6025 * tree.c (build_index_vector): New function.
6026 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6027 of units as polynomial, forcibly converting it to a constant if
6028 vectorizable_reduction has already enforced the condition.
6029 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6030 to create a {1,2,3,...} vector.
6031 (vectorizable_reduction): Treat the number of units as polynomial.
6032 Choose vectype_in based on the largest scalar element size rather
6033 than the smallest number of units. Enforce the restrictions
6036 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6037 Alan Hayward <alan.hayward@arm.com>
6038 David Sherwood <david.sherwood@arm.com>
6040 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6041 number of units as polynomial.
6043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6044 Alan Hayward <alan.hayward@arm.com>
6045 David Sherwood <david.sherwood@arm.com>
6047 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6048 * target.def (autovectorize_vector_sizes): Return the vector sizes
6049 by pointer, using vector_sizes rather than a bitmask.
6050 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6051 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6052 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6054 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6055 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6056 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6057 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6058 * omp-general.c (omp_max_vf): Likewise.
6059 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6060 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6061 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6062 * tree-vect-slp.c (vect_slp_bb): Likewise.
6063 * doc/tm.texi: Regenerate.
6064 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6066 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6067 the vector size as a poly_uint64 rather than an unsigned int.
6068 (current_vector_size): Change from an unsigned int to a poly_uint64.
6069 (get_vectype_for_scalar_type): Update accordingly.
6070 * tree.h (build_truth_vector_type): Take the size and number of
6071 units as a poly_uint64 rather than an unsigned int.
6072 (build_vector_type): Add a temporary overload that takes
6073 the number of units as a poly_uint64 rather than an unsigned int.
6074 * tree.c (make_vector_type): Likewise.
6075 (build_truth_vector_type): Take the number of units as a poly_uint64
6076 rather than an unsigned int.
6078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6079 Alan Hayward <alan.hayward@arm.com>
6080 David Sherwood <david.sherwood@arm.com>
6082 * target.def (get_mask_mode): Take the number of units and length
6083 as poly_uint64s rather than unsigned ints.
6084 * targhooks.h (default_get_mask_mode): Update accordingly.
6085 * targhooks.c (default_get_mask_mode): Likewise.
6086 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6087 * doc/tm.texi: Regenerate.
6089 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6090 Alan Hayward <alan.hayward@arm.com>
6091 David Sherwood <david.sherwood@arm.com>
6093 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6094 * omp-general.c (omp_max_vf): Likewise.
6095 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6096 (expand_omp_simd): Handle polynomial safelen.
6097 * omp-low.c (omplow_simd_context): Add a default constructor.
6098 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6099 (lower_rec_simd_input_clauses): Update accordingly.
6100 (lower_rec_input_clauses): Likewise.
6102 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6103 Alan Hayward <alan.hayward@arm.com>
6104 David Sherwood <david.sherwood@arm.com>
6106 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6107 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6108 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6109 (vect_analyze_slp_cost): Likewise.
6110 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6111 (vect_model_load_cost): Likewise.
6113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6114 Alan Hayward <alan.hayward@arm.com>
6115 David Sherwood <david.sherwood@arm.com>
6117 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6118 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6119 from an unsigned int * to a poly_uint64_pod *.
6120 (calculate_unrolling_factor): New function.
6121 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6124 Alan Hayward <alan.hayward@arm.com>
6125 David Sherwood <david.sherwood@arm.com>
6127 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6128 from an unsigned int to a poly_uint64.
6129 (_loop_vec_info::slp_unrolling_factor): Likewise.
6130 (_loop_vec_info::vectorization_factor): Change from an int
6132 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6133 (vect_get_num_vectors): New function.
6134 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6135 (vect_get_num_copies): Use vect_get_num_vectors.
6136 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6137 to an unsigned int *.
6138 (vect_analyze_data_refs): Change min_vf from an int * to a
6140 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6141 than an unsigned HOST_WIDE_INT.
6142 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6143 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6144 to an unsigned int *.
6145 (vect_analyze_data_ref_dependences): Likewise.
6146 (vect_compute_data_ref_alignment): Handle polynomial vf.
6147 (vect_enhance_data_refs_alignment): Likewise.
6148 (vect_prune_runtime_alias_test_list): Likewise.
6149 (vect_shift_permute_load_chain): Likewise.
6150 (vect_supportable_dr_alignment): Likewise.
6151 (dependence_distance_ge_vf): Take the vectorization factor as a
6152 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6153 (vect_analyze_data_refs): Change min_vf from an int * to a
6155 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6156 vfm1 as a poly_uint64 rather than an int. Make the same change
6157 for the returned bound_scalar.
6158 (vect_gen_vector_loop_niters): Handle polynomial vf.
6159 (vect_do_peeling): Likewise. Update call to
6160 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6161 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6163 * tree-vect-loop.c (vect_determine_vectorization_factor)
6164 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6165 (vect_get_known_peeling_cost): Likewise.
6166 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6167 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6168 (vect_transform_loop): Likewise. Use the lowest possible VF when
6169 updating the upper bounds of the loop.
6170 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6172 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6173 polynomial unroll factors.
6174 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6175 (vect_make_slp_decision): Likewise.
6176 (vect_supported_load_permutation_p): Likewise, and polynomial
6178 (vect_analyze_slp_cost): Handle polynomial vf.
6179 (vect_slp_analyze_node_operations): Likewise.
6180 (vect_slp_analyze_bb_1): Likewise.
6181 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6182 than an unsigned HOST_WIDE_INT.
6183 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6184 (vectorizable_load): Handle polynomial vf.
6185 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6187 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6190 Alan Hayward <alan.hayward@arm.com>
6191 David Sherwood <david.sherwood@arm.com>
6193 * match.pd: Handle bit operations involving three constants
6194 and try to fold one pair.
6196 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6198 * tree-vect-loop-manip.c: Include gimple-fold.h.
6199 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6200 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6201 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6202 Add a path that uses a step of VF instead of 1, but disable it
6204 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6205 and niters_no_overflow parameters. Update calls to
6206 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6207 Create a new SSA name if the latter choses to use a ste other
6208 than zero, and return it via niters_vector_mult_vf_var.
6209 * tree-vect-loop.c (vect_transform_loop): Update calls to
6210 vect_do_peeling, vect_gen_vector_loop_niters and
6211 slpeel_make_loop_iterate_ntimes.
6212 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6213 (vect_gen_vector_loop_niters): Update declarations after above changes.
6215 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6217 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6218 128-bit round to integer instructions.
6219 (ceil<mode>2): Likewise.
6220 (btrunc<mode>2): Likewise.
6221 (round<mode>2): Likewise.
6223 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6225 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6226 unaligned VSX load/store on P8/P9.
6227 (expand_block_clear): Allow the use of unaligned VSX
6228 load/store on P8/P9.
6230 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6232 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6234 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6235 swap associated with both a load and a store.
6237 2018-01-02 Andrew Waterman <andrew@sifive.com>
6239 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6240 * config/riscv/riscv.md (clear_cache): Use it.
6242 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6244 * web.c: Remove out-of-date comment.
6246 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6248 * expr.c (fixup_args_size_notes): Check that any existing
6249 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6250 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6251 (emit_single_push_insn): ...here.
6253 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6255 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6256 (const_vector_encoded_nelts): New function.
6257 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6258 (const_vector_int_elt, const_vector_elt): Declare.
6259 * emit-rtl.c (const_vector_int_elt_1): New function.
6260 (const_vector_elt): Likewise.
6261 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6262 of CONST_VECTOR_ELT.
6264 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6266 * expr.c: Include rtx-vector-builder.h.
6267 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6268 directly on the tree encoding.
6269 (const_vector_from_tree): Likewise.
6270 * optabs.c: Include rtx-vector-builder.h.
6271 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6272 sequence of "u" values.
6273 * vec-perm-indices.c: Include rtx-vector-builder.h.
6274 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6275 directly on the vec_perm_indices encoding.
6277 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6279 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6280 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6281 * rtx-vector-builder.h: New file.
6282 * rtx-vector-builder.c: Likewise.
6283 * rtl.h (rtx_def::u2): Add a const_vector field.
6284 (CONST_VECTOR_NPATTERNS): New macro.
6285 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6286 (CONST_VECTOR_DUPLICATE_P): Likewise.
6287 (CONST_VECTOR_STEPPED_P): Likewise.
6288 (CONST_VECTOR_ENCODED_ELT): Likewise.
6289 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6290 (unwrap_const_vec_duplicate): Likewise.
6291 (const_vec_series_p): Check for a non-duplicated vector encoding.
6292 Say that the function only returns true for integer vectors.
6293 * emit-rtl.c: Include rtx-vector-builder.h.
6294 (gen_const_vec_duplicate_1): Delete.
6295 (gen_const_vector): Call gen_const_vec_duplicate instead of
6296 gen_const_vec_duplicate_1.
6297 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6298 (gen_const_vec_duplicate): Use rtx_vector_builder.
6299 (gen_const_vec_series): Likewise.
6300 (gen_rtx_CONST_VECTOR): Likewise.
6301 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6302 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6303 Build a new vector rather than modifying a CONST_VECTOR in-place.
6304 (handle_special_swappables): Update call accordingly.
6305 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6306 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6307 Build a new vector rather than modifying a CONST_VECTOR in-place.
6308 (handle_special_swappables): Update call accordingly.
6310 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6312 * simplify-rtx.c (simplify_const_binary_operation): Use
6313 CONST_VECTOR_ELT instead of XVECEXP.
6315 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6317 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6318 the selector elements to be different from the data elements
6319 if the selector is a VECTOR_CST.
6320 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6321 ssizetype for the selector.
6323 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6325 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6326 before testing each element individually.
6327 * tree-vect-generic.c (lower_vec_perm): Likewise.
6329 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6331 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6332 * selftest-run-tests.c (selftest::run_tests): Call it.
6333 * vector-builder.h (vector_builder::operator ==): New function.
6334 (vector_builder::operator !=): Likewise.
6335 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6336 (vec_perm_indices::all_from_input_p): New function.
6337 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6338 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6339 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6340 instead of reading the VECTOR_CST directly. Detect whether both
6341 vector inputs are the same before constructing the vec_perm_indices,
6342 and update the number of inputs argument accordingly. Use the
6343 utility functions added above. Only construct sel2 if we need to.
6345 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6347 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6348 the broadcast of the low byte.
6349 (expand_mult_highpart): Use an explicit encoding for the permutes.
6350 * optabs-query.c (can_mult_highpart_p): Likewise.
6351 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6352 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6353 (vectorizable_bswap): Likewise.
6354 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6355 explicit encoding for the power-of-2 permutes.
6356 (vect_permute_store_chain): Likewise.
6357 (vect_grouped_load_supported): Likewise.
6358 (vect_permute_load_chain): Likewise.
6360 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6362 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6363 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6364 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6365 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6366 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6367 (vect_gen_perm_mask_any): Likewise.
6369 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6371 * int-vector-builder.h: New file.
6372 * vec-perm-indices.h: Include int-vector-builder.h.
6373 (vec_perm_indices): Redefine as an int_vector_builder.
6374 (auto_vec_perm_indices): Delete.
6375 (vec_perm_builder): Redefine as a stand-alone class.
6376 (vec_perm_indices::vec_perm_indices): New function.
6377 (vec_perm_indices::clamp): Likewise.
6378 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6379 (vec_perm_indices::new_vector): New function.
6380 (vec_perm_indices::new_expanded_vector): Update for new
6381 vec_perm_indices class.
6382 (vec_perm_indices::rotate_inputs): New function.
6383 (vec_perm_indices::all_in_range_p): Operate directly on the
6384 encoded form, without computing elided elements.
6385 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6386 encoding. Update for new vec_perm_indices class.
6387 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6388 the given vec_perm_builder.
6389 (expand_vec_perm_var): Update vec_perm_builder constructor.
6390 (expand_mult_highpart): Use vec_perm_builder instead of
6391 auto_vec_perm_indices.
6392 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6393 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6394 or double series encoding as appropriate.
6395 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6396 vec_perm_indices instead of auto_vec_perm_indices.
6397 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6398 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6399 (vect_permute_store_chain): Likewise.
6400 (vect_grouped_load_supported): Likewise.
6401 (vect_permute_load_chain): Likewise.
6402 (vect_shift_permute_load_chain): Likewise.
6403 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6404 (vect_transform_slp_perm_load): Likewise.
6405 (vect_schedule_slp_instance): Likewise.
6406 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6407 (vectorizable_mask_load_store): Likewise.
6408 (vectorizable_bswap): Likewise.
6409 (vectorizable_store): Likewise.
6410 (vectorizable_load): Likewise.
6411 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6412 vec_perm_indices instead of auto_vec_perm_indices. Use
6413 tree_to_vec_perm_builder to read the vector from a tree.
6414 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6415 vec_perm_builder instead of a vec_perm_indices.
6416 (have_whole_vector_shift): Use vec_perm_builder and
6417 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6418 truncation to calc_vec_perm_mask_for_shift.
6419 (vect_create_epilog_for_reduction): Likewise.
6420 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6421 from auto_vec_perm_indices to vec_perm_indices.
6422 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6423 instead of changing individual elements.
6424 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6425 the vector in d.perm.
6426 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6427 from auto_vec_perm_indices to vec_perm_indices.
6428 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6429 instead of changing individual elements.
6430 (arm_vectorize_vec_perm_const): Use new_vector to install
6431 the vector in d.perm.
6432 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6433 Update vec_perm_builder constructor.
6434 (rs6000_expand_interleave): Likewise.
6435 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6436 (rs6000_expand_interleave): Likewise.
6438 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6440 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6441 to qimode could truncate the indices.
6442 * optabs.c (expand_vec_perm_var): Likewise.
6444 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6446 * Makefile.in (OBJS): Add vec-perm-indices.o.
6447 * vec-perm-indices.h: New file.
6448 * vec-perm-indices.c: Likewise.
6449 * target.h (vec_perm_indices): Replace with a forward class
6451 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6452 * optabs.h: Include vec-perm-indices.h.
6453 (expand_vec_perm): Delete.
6454 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6455 (expand_vec_perm_const): Declare.
6456 * target.def (vec_perm_const_ok): Replace with...
6457 (vec_perm_const): ...this new hook.
6458 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6459 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6460 * doc/tm.texi: Regenerate.
6461 * optabs.def (vec_perm_const): Delete.
6462 * doc/md.texi (vec_perm_const): Likewise.
6463 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6464 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6465 expand_vec_perm for constant permutation vectors. Assert that
6466 the mode of variable permutation vectors is the integer equivalent
6467 of the mode that is being permuted.
6468 * optabs-query.h (selector_fits_mode_p): Declare.
6469 * optabs-query.c: Include vec-perm-indices.h.
6470 (selector_fits_mode_p): New function.
6471 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6472 is defined, instead of checking whether the vec_perm_const_optab
6473 exists. Use targetm.vectorize.vec_perm_const instead of
6474 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6475 fit in the vector mode before using a variable permute.
6476 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6477 vec_perm_indices instead of an rtx.
6478 (expand_vec_perm): Replace with...
6479 (expand_vec_perm_const): ...this new function. Take the selector
6480 as a vec_perm_indices rather than an rtx. Also take the mode of
6481 the selector. Update call to shift_amt_for_vec_perm_mask.
6482 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6483 Use vec_perm_indices::new_expanded_vector to expand the original
6484 selector into bytes. Check whether the indices fit in the vector
6485 mode before using a variable permute.
6486 (expand_vec_perm_var): Make global.
6487 (expand_mult_highpart): Use expand_vec_perm_const.
6488 * fold-const.c: Includes vec-perm-indices.h.
6489 * tree-ssa-forwprop.c: Likewise.
6490 * tree-vect-data-refs.c: Likewise.
6491 * tree-vect-generic.c: Likewise.
6492 * tree-vect-loop.c: Likewise.
6493 * tree-vect-slp.c: Likewise.
6494 * tree-vect-stmts.c: Likewise.
6495 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6497 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6498 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6499 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6500 (aarch64_vectorize_vec_perm_const): ...this new function.
6501 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6502 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6503 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6504 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6505 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6506 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6507 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6509 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6510 check for NEON modes.
6511 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6512 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6513 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6514 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6516 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6517 the old VEC_PERM_CONST conditions.
6518 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6519 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6520 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6521 (ia64_vectorize_vec_perm_const_ok): Merge into...
6522 (ia64_vectorize_vec_perm_const): ...this new function.
6523 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6524 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6525 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6526 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6527 * config/mips/mips.c (mips_expand_vec_perm_const)
6528 (mips_vectorize_vec_perm_const_ok): Merge into...
6529 (mips_vectorize_vec_perm_const): ...this new function.
6530 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6531 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6532 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6533 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6534 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6535 (rs6000_expand_vec_perm_const): Delete.
6536 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6538 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6539 (altivec_expand_vec_perm_const_le): Take each operand individually.
6540 Operate on constant selectors rather than rtxes.
6541 (altivec_expand_vec_perm_const): Likewise. Update call to
6542 altivec_expand_vec_perm_const_le.
6543 (rs6000_expand_vec_perm_const): Delete.
6544 (rs6000_vectorize_vec_perm_const_ok): Delete.
6545 (rs6000_vectorize_vec_perm_const): New function.
6546 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6547 an element count and rtx array.
6548 (rs6000_expand_extract_even): Update call accordingly.
6549 (rs6000_expand_interleave): Likewise.
6550 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6551 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6552 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6553 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6554 (rs6000_expand_vec_perm_const): Delete.
6555 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6556 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6557 (altivec_expand_vec_perm_const_le): Take each operand individually.
6558 Operate on constant selectors rather than rtxes.
6559 (altivec_expand_vec_perm_const): Likewise. Update call to
6560 altivec_expand_vec_perm_const_le.
6561 (rs6000_expand_vec_perm_const): Delete.
6562 (rs6000_vectorize_vec_perm_const_ok): Delete.
6563 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6564 reference to the SPE evmerge intructions.
6565 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6566 an element count and rtx array.
6567 (rs6000_expand_extract_even): Update call accordingly.
6568 (rs6000_expand_interleave): Likewise.
6569 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6570 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6572 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6574 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6576 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6577 vector mode and that that mode matches the mode of the data
6579 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6580 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6581 directly using expand_vec_perm_1 when forcing selectors into
6583 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6585 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6587 * optabs-query.h (can_vec_perm_p): Delete.
6588 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6589 * optabs-query.c (can_vec_perm_p): Split into...
6590 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6591 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6592 particular selector is valid.
6593 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6594 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6595 (vect_grouped_load_supported): Likewise.
6596 (vect_shift_permute_load_chain): Likewise.
6597 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6598 (vect_transform_slp_perm_load): Likewise.
6599 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6600 (vectorizable_bswap): Likewise.
6601 (vect_gen_perm_mask_checked): Likewise.
6602 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6603 implementations of variable permutation vectors into account
6604 when deciding which selector to use.
6605 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6606 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6607 with a false third argument.
6608 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6609 to test whether the constant selector is valid and can_vec_perm_var_p
6610 to test whether a variable selector is valid.
6612 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6614 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6615 * optabs-query.c (can_vec_perm_p): Likewise.
6616 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6617 instead of vec_perm_indices.
6618 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6619 (vect_gen_perm_mask_checked): Likewise,
6620 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6621 (vect_gen_perm_mask_checked): Likewise,
6623 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6625 * optabs-query.h (qimode_for_vec_perm): Declare.
6626 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6627 (qimode_for_vec_perm): ...this new function.
6628 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6630 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6632 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6633 does not have a conditional at the top.
6635 2018-01-02 Richard Biener <rguenther@suse.de>
6637 * ipa-inline.c (big_speedup_p): Fix expression.
6639 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6642 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6645 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6649 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6650 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6651 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6652 cond_taken_branch_cost 3->4.
6654 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6656 PR tree-optimization/83581
6657 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6658 TODO_cleanup_cfg if any changes have been made.
6661 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6662 convert_modes if target mode has the right side, but different mode
6666 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6667 last argument when extracting from CONCAT. If either from_real or
6668 from_imag is NULL, use expansion through memory. If result is not
6669 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6670 the parts directly to inner mode, if even that fails, use expansion
6674 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6675 check for bswap in mode rather than HImode and use that in expand_unop
6678 Copyright (C) 2018 Free Software Foundation, Inc.
6680 Copying and distribution of this file, with or without modification,
6681 are permitted in any medium without royalty provided the copyright
6682 notice and this notice are preserved.