1 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
3 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
4 to allow or block "symbol_ref" depending on value of TARGET_JSR
5 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
6 call_value_internal insns
8 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
11 * dwarf2out.c (is_trivial_indirect_ref): New function.
12 (dwarf2out_late_global_decl): Do not generate a location
13 attribute for variables that have a non-trivial DECL_VALUE_EXPR
14 and that are not defined in the current unit.
16 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
18 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
19 instead of a libcall for UNORDERED.
21 2018-02-09 Tamar Christina <tamar.christina@arm.com>
24 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
25 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
27 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
30 * config/s390/s390.c (s390_set_current_function): Invoke
31 s390_indirect_branch_settings also if fndecl didn't change.
33 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
35 * config/rs6000/rs6000.md (blockage): Set length to zero.
37 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
39 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
41 2018-02-09 Jakub Jelinek <jakub@redhat.com>
44 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
45 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
49 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
50 PARALLEL incoming that failed vt_get_decl_and_offset check.
53 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
54 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
56 (get_variable_section): For decls in named .bss* sections pass true as
57 second argument to bss_initializer_p.
59 2018-02-09 Marek Polacek <polacek@redhat.com>
60 Jakub Jelinek <jakub@redhat.com>
63 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
64 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
65 Sync some changes from cxx_fold_indirect_ref.
67 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
69 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
71 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
72 (BLOCK_INLINE_ENTRY_LABEL): New.
73 (dwarf2out_var_location): Disregard inline entry markers.
74 (inline_entry_data): New struct.
75 (inline_entry_data_hasher): New hashtable type.
76 (inline_entry_data_hasher::hash): New.
77 (inline_entry_data_hasher::equal): New.
78 (inline_entry_data_table): New variable.
79 (add_high_low_attributes): Add DW_AT_entry_pc and
80 DW_AT_GNU_entry_view attributes if a pending entry is found
81 in inline_entry_data_table. Add old entry_pc attribute only
82 if debug nonbinding markers are disabled.
83 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
85 (block_within_block_p, dwarf2out_inline_entry): New.
86 (dwarf2out_finish): Check that no entries remained in
87 inline_entry_data_table.
88 * final.c (reemit_insn_block_notes): Handle inline entry notes.
89 (final_scan_insn, notice_source_line): Likewise.
90 (rest_of_clean_state): Skip inline entry markers.
91 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
93 * gimple.c (gimple_build_debug_inline_entry): New.
94 * gimple.h (enum gimple_debug_subcode): Add
95 GIMPLE_DEBUG_INLINE_ENTRY.
96 (gimple_build_debug_inline_entry): Declare.
97 (gimple_debug_inline_entry_p): New.
98 (gimple_debug_nonbind_marker_p): Adjust.
99 * insn-notes.def (INLINE_ENTRY): New.
100 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
101 inline entry marker notes.
102 (print_insn): Likewise.
103 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
104 (INSN_DEBUG_MARKER_KIND): Likewise.
105 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
106 * tree-inline.c (expand_call_inline): Build and insert
107 debug_inline_entry stmt.
108 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
109 inline entry blocks early, if nonbind markers are enabled.
110 (dump_scope_block): Dump fragment info.
111 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
112 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
113 (gimple_build_debug_inline_entry): New.
114 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
115 Enable/disable inline entry points too.
116 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
117 (DEBUG_INSN): Describe inline entry markers.
119 * common.opt (gvariable-location-views): New.
120 (gvariable-location-views=incompat5): New.
121 * config.in: Rebuilt.
122 * configure: Rebuilt.
123 * configure.ac: Test assembler for view support.
124 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
125 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
126 * dwarf2out.c (var_loc_view): New typedef.
127 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
128 (dwarf2out_locviews_in_attribute): New.
129 (dwarf2out_locviews_in_loclist): New.
130 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
131 (enum dw_line_info_opcode): Add LI_adv_address.
132 (struct dw_line_info_table): Add view.
133 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
134 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
135 (zero_view_p): New variable.
136 (ZERO_VIEW_P): New macro.
137 (output_asm_line_debug_info): New.
138 (struct var_loc_node): Add view.
139 (add_AT_view_list, AT_loc_list): New.
140 (add_var_loc_to_decl): Add view param. Test it against last.
141 (new_loc_list): Add view params. Record them.
142 (AT_loc_list_ptr): Handle loc and view lists.
143 (view_list_to_loc_list_val_node): New.
144 (print_dw_val): Handle dw_val_class_view_list.
145 (size_of_die): Likewise.
146 (value_format): Likewise.
147 (loc_list_has_views): New.
148 (gen_llsym): Set vl_symbol too.
149 (maybe_gen_llsym, skip_loc_list_entry): New.
150 (dwarf2out_maybe_output_loclist_view_pair): New.
151 (output_loc_list): Output view list or entries too.
152 (output_view_list_offset): New.
153 (output_die): Handle dw_val_class_view_list.
154 (output_dwarf_version): New.
155 (output_compilation_unit_header): Use it.
156 (output_skeleton_debug_sections): Likewise.
157 (output_rnglists, output_line_info): Likewise.
158 (output_pubnames, output_aranges): Update version comments.
159 (output_one_line_info_table): Output view numbers in asm comments.
160 (dw_loc_list): Determine current endview, pass it to new_loc_list.
161 Call maybe_gen_llsym.
162 (loc_list_from_tree_1): Adjust.
163 (add_AT_location_description): Create view list attribute if
164 needed, check it's absent otherwise.
165 (convert_cfa_to_fb_loc_list): Adjust.
166 (maybe_emit_file): Call output_asm_line_debug_info for test.
167 (dwarf2out_var_location): Reset views as needed. Precompute
168 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
170 (new_line_info_table): Reset next view.
171 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
172 (dwarf2out_source_line): Likewise. Output view resets and labels to
173 the assembler, or select appropriate line info opcodes.
174 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
175 (optimize_string_length): Catch it. Adjust.
176 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
177 dw_val_class_view_list, and remove it if no longer needed.
178 (hash_loc_list): Hash view numbers.
179 (loc_list_hasher::equal): Compare them.
180 (optimize_location_lists): Check whether a view list symbol is
181 needed, and whether the locview attribute is present, and
182 whether they match. Remove the locview attribute if no longer
184 (index_location_lists): Call skip_loc_list_entry for test.
185 (dwarf2out_finish): Call output_asm_line_debug_info for test.
186 Use output_dwarf_version.
187 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
188 (struct dw_val_node): Add val_view_list.
189 * final.c (SEEN_NEXT_VIEW): New.
190 (set_next_view_needed): New.
191 (clear_next_view_needed): New.
192 (maybe_output_next_view): New.
193 (final_start_function): Rename to...
194 (final_start_function_1): ... this. Take pointer to FIRST,
195 add SEEN parameter. Emit param bindings in the initial view.
196 (final_start_function): Reintroduce SEEN-less interface.
197 (final): Rename to...
198 (final_1): ... this. Take SEEN parameter. Output final pending
199 next view at the end.
200 (final): Reintroduce seen-less interface.
201 (final_scan_insn): Output pending next view before switching
202 sections or ending a block. Mark the next view as needed when
203 outputting variable locations. Notify debug backend of section
204 changes, and of location view changes.
205 (rest_of_handle_final): Adjust.
206 * toplev.c (process_options): Autodetect value for debug variable
207 location views option. Warn on incompat5 without -gdwarf-5.
208 * doc/invoke.texi (gvariable-location-views): New.
209 (gvariable-location-views=incompat5): New.
210 (gno-variable-location-views): New.
212 2018-02-08 David Malcolm <dmalcolm@redhat.com>
214 PR tree-optimization/84136
215 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
216 that the result of find_edge is non-NULL.
218 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
221 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
222 storing integer register in SImode. Fix cost of 256 and 512
223 byte aligned SSE register store.
225 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
227 * config/i386/i386.c (ix86_multiplication_cost): Fix
228 multiplication cost for TARGET_AVX512DQ.
230 2018-02-08 Marek Polacek <polacek@redhat.com>
232 PR tree-optimization/84238
233 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
236 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
238 PR tree-optimization/84265
239 * tree-vect-stmts.c (vectorizable_store): Don't treat
240 VMAT_CONTIGUOUS accesses as grouped.
241 (vectorizable_load): Likewise.
243 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
245 PR tree-optimization/81635
246 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
247 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
248 (test_round_for_mask): New functions.
249 (wide_int_cc_tests): Call test_round_for_mask.
250 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
251 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
252 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
253 range returned by get_range_info.
255 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
258 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
259 * symtab.c: Include builtins.h
260 (symtab_node::output_to_lto_symbol_table_p): Move here
261 from lto-streamer-out.c:output_symbol_p.
262 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
263 (output_symbol_p): Move all logic to symtab.c
264 (produce_symtab): Update.
266 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
268 * config/s390/s390-opts.h (enum indirect_branch): Define.
269 * config/s390/s390-protos.h (s390_return_addr_from_memory)
270 (s390_indirect_branch_via_thunk)
271 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
272 (enum s390_indirect_branch_type): Define.
273 * config/s390/s390.c (struct s390_frame_layout, struct
274 machine_function): Remove.
275 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
276 (indirect_branch_table_label_no, indirect_branch_table_name):
278 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
279 (enum s390_indirect_branch_option): Define.
280 (s390_return_addr_from_memory): New function.
281 (s390_handle_string_attribute): New function.
282 (s390_attribute_table): Add new attribute handler.
283 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
284 (s390_indirect_branch_via_thunk): New function.
285 (s390_indirect_branch_via_inline_thunk): New function.
286 (s390_function_ok_for_sibcall): When jumping via thunk disallow
287 sibling call optimization for non z10 compiles.
288 (s390_emit_call): Force indirect branch target to be a single
289 register. Add r1 clobber for non-z10 compiles.
290 (s390_emit_epilogue): Emit return jump via return_use expander.
291 (s390_reorg): Handle JUMP_INSNs as execute targets.
292 (s390_option_override_internal): Perform validity checks for the
293 new command line options.
294 (s390_indirect_branch_attrvalue): New function.
295 (s390_indirect_branch_settings): New function.
296 (s390_set_current_function): Invoke s390_indirect_branch_settings.
297 (s390_output_indirect_thunk_function): New function.
298 (s390_code_end): Implement target hook.
299 (s390_case_values_threshold): Implement target hook.
300 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
302 * config/s390/s390.h (struct s390_frame_layout)
303 (struct machine_function): Move here from s390.c.
304 (TARGET_INDIRECT_BRANCH_NOBP_RET)
305 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
306 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
307 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
308 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
309 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
310 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
311 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
312 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
313 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
314 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
315 (mnemonic attribute): Add values which aren't recognized
317 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
318 pattern for branch conversion. Fix mnemonic attribute.
319 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
320 indirect branch via thunk if requested.
321 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
322 ("*indirect_jump"): Disable for branch conversion using out of
324 ("indirect_jump_via_thunk<mode>_z10")
325 ("indirect_jump_via_thunk<mode>")
326 ("indirect_jump_via_inlinethunk<mode>_z10")
327 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
328 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
329 ("casesi_jump_via_inlinethunk<mode>_z10")
330 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
331 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
332 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
333 ("*indirect2_jump"): Disable for branch conversion.
334 ("casesi_jump"): Turn into expander and expand patterns for branch
336 ("return_use"): New expander.
337 ("*return"): Emit return via thunk and rename it to ...
338 ("*return<mode>"): ... this one.
339 * config/s390/s390.opt: Add new options and and enum for the
342 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
344 * lra-constraints.c (match_reload): Unconditionally use
345 gen_lowpart_SUBREG, rather than selecting between that
346 and equivalent gen_rtx_SUBREG code.
348 2018-02-08 Richard Biener <rguenther@suse.de>
350 PR tree-optimization/84233
351 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
352 changed flag instead of boguously re-using phi_inserted.
354 2018-02-08 Martin Jambor <mjambor@suse.cz>
356 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
357 static local variables.
359 2018-02-08 Richard Biener <rguenther@suse.de>
361 PR tree-optimization/84278
362 * tree-vect-stmts.c (vectorizable_store): When looking for
363 smaller vector types to perform grouped strided loads/stores
364 make sure the mode is supported by the target.
365 (vectorizable_load): Likewise.
367 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
369 * config/aarch64/aarch64.c (aarch64_components_for_bb):
370 Increase LDP/STP opportunities by adding adjacent callee-saves.
372 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
374 PR rtl-optimization/84068
375 PR rtl-optimization/83459
376 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
378 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
380 PR tree-optimization/84224
381 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
382 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
385 2018-02-07 Iain Sandoe <iain@codesourcery.com>
388 * config/rs6000/altivec.md (*restore_world): Remove LR use.
389 * config/rs6000/predicates.md (restore_world_operation): Adjust op
390 count, remove one USE.
392 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
394 * doc/install.texi (Configuration): Document the
395 --with-long-double-format={ibm,ieee} PowerPC configuration
399 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
400 Convert from define_expand to be define_insn_and_split. Rework
401 float/double/_Float128 conversions to QI/HI/SImode to work with
402 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
403 conversions to QI/HImode types did a store and then a load to
404 truncate the value. For conversions to VSX registers, don't split
405 the insn, instead emit the code directly. Use the code iterator
406 any_fix to combine signed and unsigned conversions.
407 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
408 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
409 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
410 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
411 (fix_<mode>di2_hw): Likewise.
412 (fixuns_<mode>di2_hw): Likewise.
413 (fix_<mode>si2_hw): Likewise.
414 (fixuns_<mode>si2_hw): Likewise.
415 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
416 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
417 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
418 fix<uns>_trunc<SFDF:mode>si2_p8.
419 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
421 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
422 (fix<uns>_<mode>_mem): Likewise.
423 (fctiw<u>z_<mode>_mem): Likewise.
424 (fix<uns>_<mode>_mem): Likewise.
425 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
426 the register allocator from doing a direct move to the GPRs to do
427 a store, and instead use the ISA 3.0 store byte/half-word from
428 vector register instruction. For IEEE 128-bit floating point,
429 also optimize stores of 32-bit ints.
430 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
432 2018-02-07 Alan Hayward <alan.hayward@arm.com>
434 * genextract.c (push_pathstr_operand): New function to support
436 (walk_rtx): Call push_pathstr_operand.
437 (print_path): Support [a-zA-Z].
439 2018-02-07 Richard Biener <rguenther@suse.de>
441 PR tree-optimization/84037
442 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
443 (cse_and_gimplify_to_preheader): Declare.
444 (vect_get_place_in_interleaving_chain): Likewise.
445 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
447 (_loop_vec_info::~_loop_vec_info): Delete it.
448 (cse_and_gimplify_to_preheader): New function.
449 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
450 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
451 (vectorizable_load): Likewise. For grouped stores always base
452 the IV on the first element.
453 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
454 condition before gimplifying.
456 2018-02-07 Jakub Jelinek <jakub@redhat.com>
458 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
459 *DIV_EXPR and *MOD_EXPR.
461 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
464 * config/i386/i386.c (ix86_option_override_internal): Mask out
465 the CF_SET bit when checking -fcf-protection.
467 2018-02-07 Tom de Vries <tom@codesourcery.com>
470 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
473 2018-02-07 Richard Biener <rguenther@suse.de>
475 PR tree-optimization/84204
476 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
479 PR tree-optimization/84205
480 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
481 special-case isl_ast_op_zdiv_r.
483 PR tree-optimization/84223
484 * graphite-scop-detection.c (gather_bbs::before_dom_children):
485 Only add conditions from within the region.
486 (gather_bbs::after_dom_children): Adjust.
488 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
491 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
492 * config/avr/avr.md: Only post-reload split REG-REG moves if
493 either register is REGERAL_REG_P.
495 2018-02-07 Jakub Jelinek <jakub@redhat.com>
497 PR tree-optimization/84235
498 * tree-ssa-scopedtables.c
499 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
500 if the subtraction is performed in floating point type where NaNs are
501 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
502 build 1. Formatting fix.
504 2018-02-06 Jakub Jelinek <jakub@redhat.com>
507 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
508 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
509 and skip it regardless of bb boundaries. Use CALL_P macro,
510 don't test INSN_P (insn) together with CALL_P or JUMP_P check
511 unnecessarily, formatting fix.
513 2018-02-06 Michael Collison <michael.collison@arm.com>
515 * config/arm/thumb2.md:
516 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
517 (*thumb_mov_notscc): Ditto.
519 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
522 * config/rs6000/rs6000.md (su code attribute): Use "u" for
523 unsigned_fix, not "s".
525 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
527 * configure.ac (gcc_fn_eh_frame_ro): New function.
528 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
529 correct .eh_frame permissions.
530 * configure: Regenerate.
532 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
534 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
537 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
539 * config/rs6000/rs6000.c (rs6000_option_override_internal):
540 Display warning message for -mno-speculate-indirect-jumps.
542 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
544 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
546 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
548 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
550 PR tree-optimization/84225
551 * tree-eh.c (find_trapping_overflow): Only call
552 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
554 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
557 * config/i386/i386.c: Reimplement the check of possible options
558 -mibt/-mshstk conbination. Change error messages.
559 * doc/invoke.texi: Fix a typo: remove extra '='.
561 2018-02-06 Marek Polacek <polacek@redhat.com>
563 PR tree-optimization/84228
564 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
566 2018-02-06 Tamar Christina <tamar.christina@arm.com>
569 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
570 emitted arch directives.
571 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
572 __ARM_FEATURE_COPROC before changing architectures.
574 2018-02-06 Richard Biener <rguenther@suse.de>
576 * config/i386/i386.c (print_reg): Fix typo.
577 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
579 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
581 * configure: Regenerate.
583 2018-02-05 Martin Sebor <msebor@redhat.com>
585 PR tree-optimization/83369
586 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
589 2018-02-05 Martin Liska <mliska@suse.cz>
591 * doc/invoke.texi: Cherry-pick upstream r323995.
593 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
595 * ira.c (ira_init_register_move_cost): Adjust comment.
597 2018-02-05 Martin Liska <mliska@suse.cz>
599 PR gcov-profile/84137
600 * doc/gcov.texi: Fix typo in documentation.
602 2018-02-05 Martin Liska <mliska@suse.cz>
604 PR gcov-profile/83879
605 * doc/gcov.texi: Document necessity of --dynamic-list-data when
606 using dlopen functionality.
608 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
610 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
611 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
612 _mm_maskz_range_ss, _mm_mask_range_round_ss,
613 _mm_maskz_range_round_ss): New intrinsics.
614 (__builtin_ia32_rangesd128_round)
615 (__builtin_ia32_rangess128_round): Remove.
616 (__builtin_ia32_rangesd128_mask_round,
617 __builtin_ia32_rangess128_mask_round): New builtins.
618 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
619 __builtin_ia32_rangess128_round): Remove.
620 (__builtin_ia32_rangesd128_mask_round,
621 __builtin_ia32_rangess128_mask_round): New builtins.
622 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
623 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
624 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
625 "<round_saeonly_constraint>")): Changed to ...
626 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
627 "<round_saeonly_scalar_constraint>")): ... this.
628 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
629 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
630 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
631 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
632 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
634 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
636 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
638 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
639 Remove all values except native, 8540 and 8548.
641 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
643 * config/i386/i386.c (ix86_output_function_return): Pass
644 INVALID_REGNUM, instead of -1, as invalid register number to
645 indirect_thunk_name and output_indirect_thunk.
647 2018-02-02 Julia Koval <julia.koval@intel.com>
649 * config.gcc: Add -march=icelake.
650 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
651 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
652 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
653 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
654 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
655 (processor_target_table): Add icelake.
656 (ix86_option_override_internal): Handle new PTAs.
657 (get_builtin_code_for_version): Handle icelake.
658 (M_INTEL_COREI7_ICELAKE): New.
659 (fold_builtin_cpu): Handle icelake.
660 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
661 * doc/invoke.texi: Add -march=icelake.
663 2018-02-02 Julia Koval <julia.koval@intel.com>
665 * config/i386/i386.c (ix86_option_override_internal): Change flags type
667 * wide-int-bitmask.h: New.
669 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
672 * config/i386/i386.md: Replace Pmode with word_mode in
673 builtin_setjmp_setup and builtin_longjmp to support x32.
675 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
679 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
681 (rs6000_supported_cpu_names): New static variable.
682 (linux_cpu_translation_table): Likewise.
683 (elf_platform) <cpu>: Define new static variable and use it.
684 Translate kernel AT_PLATFORM name to canonical name if needed.
685 Error if platform name is unknown.
687 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
690 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
692 2018-02-01 Jeff Law <law@redhat.com>
695 * config/i386/i386.c (release_scratch_register_on_entry): Add new
696 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
697 the scratch if RELEASE_VIA_POP is false.
698 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
699 If we have to save a temporary register, decrement SIZE appropriately.
700 Pass new arguments to release_scratch_register_on_entry.
701 (ix86_adjust_stack_and_probe): Likewise.
702 (ix86_emit_probe_stack_range): Pass new arguments to
703 release_scratch_register_on_entry.
705 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
707 PR rtl-optimization/84157
708 * combine.c (change_zero_ext): Use REG_P predicate in
709 front of HARD_REGISTER_P predicate.
711 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
713 * config/avr/avr.c (avr_option_override): Move disabling of
714 -fdelete-null-pointer-checks to...
715 * common/config/avr/avr-common.c (avr_option_optimization_table):
718 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
720 PR tree-optimization/81635
721 * tree-data-ref.c (split_constant_offset_1): For types that
722 wrap on overflow, try to use range info to prove that wrapping
725 2018-02-01 Renlin Li <renlin.li@arm.com>
728 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
730 (aarch64_register_move_cost): Likewise.
731 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
733 (REG_CLASS_NAMES): Likewise.
734 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
735 TAILCALL_ADDR_REGS. Remove IP registers.
736 * config/aarch64/aarch64.md (Ucs): Update register constraint.
738 2018-02-01 Richard Biener <rguenther@suse.de>
740 * domwalk.h (dom_walker::dom_walker): Add additional constructor
741 for specifying RPO order and allow NULL for that.
742 * domwalk.c (dom_walker::dom_walker): Likewise.
743 (dom_walker::walk): Handle NULL RPO order.
744 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
746 (rewrite_update_dom_walker): Likewise.
747 (mark_def_dom_walker): Likewise.
749 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
751 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
752 (aarch64_maybe_expand_sve_subreg_move): Declare.
753 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
754 * config/aarch64/predicates.md (aarch64_any_register_operand): New
756 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
757 that are semantically a reverse operation.
758 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
759 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
760 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
762 (aarch64_can_change_mode_class): For big-endian, forbid changes
763 between two SVE modes if they have different element sizes.
765 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
767 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
768 the TImode handling for big-endian targets.
770 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
772 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
773 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
775 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
776 Remove BSWAP handing for big-endian targets and use the form of
777 LD1RQ appropariate for the mode.
779 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
781 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
782 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
785 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
788 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
789 check for operands that need to go through aarch64_sve_reload_be.
791 2018-02-01 Jakub Jelinek <jakub@redhat.com>
793 PR tree-optimization/81661
794 PR tree-optimization/84117
795 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
796 * tree-eh.c: Include gimplify.h.
797 (find_trapping_overflow, replace_trapping_overflow,
798 rewrite_to_non_trapping_overflow): New functions.
799 * tree-vect-loop.c: Include tree-eh.h.
800 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
801 * tree-data-ref.c: Include tree-eh.h.
802 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
804 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
806 PR rtl-optimization/84123
807 * combine.c (change_zero_ext): Check if hard register satisfies
808 can_change_dest_mode before calling gen_lowpart_SUBREG.
810 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
813 * ira.c (ira_init_register_move_cost): Remove assert.
815 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
817 PR rtl-optimization/84071
818 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
819 * doc/tm.texi: Regenerate.
821 2018-01-31 Richard Biener <rguenther@suse.de>
823 PR tree-optimization/84132
824 * tree-data-ref.c (analyze_miv_subscript): Properly
825 check whether evolution_function_is_affine_multivariate_p
826 before calling gcd_of_steps_may_divide_p.
828 2018-01-31 Julia Koval <julia.koval@intel.com>
831 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
832 * config/i386/i386.md (rdpid_rex64) New.
833 (rdpid): Make 32bit only.
835 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
838 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
839 an IDENTIFIER_NODE for FUNCTION_TYPE's.
841 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
844 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
846 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
848 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
850 PR rtl-optimization/84071
851 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
852 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
854 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
856 * config/arc/arc.c (arc_handle_aux_attribute): New function.
857 (arc_attribute_table): Add 'aux' attribute.
858 (arc_in_small_data_p): Consider aux like variables.
859 (arc_is_aux_reg_p): New function.
860 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
861 (arc_get_aux_arg): New function.
862 (prepare_move_operands): Handle aux-register access.
863 (arc_handle_aux_attribute): New function.
864 * doc/extend.texi (ARC Variable attributes): Add subsection.
866 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
868 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
869 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
870 (arc_attribute_table): Add 'uncached' attribute.
871 (arc_print_operand): Print '.di' flag for uncached memory
873 (arc_in_small_data_p): Do not consider for small data the uncached
875 (arc_is_uncached_mem_p): New function.
876 * config/arc/predicates.md (compact_store_memory_operand): Check
877 for uncached memory accesses.
878 (nonvol_nonimm_operand): Likewise.
879 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
881 2018-01-31 Jakub Jelinek <jakub@redhat.com>
884 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
885 falign-loops=): Add Optimization flag.
887 2018-01-30 Jeff Law <law@redhat.com>
890 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
891 INT_REGISTERS_SAVED. Check it prior to calling
892 get_scratch_register_on_entry.
893 (ix86_adjust_stack_and_probe): Similarly.
894 (ix86_emit_probe_stack_range): Similarly.
895 (ix86_expand_prologue): Corresponding changes.
897 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
900 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
901 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
903 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
906 * lra-constraints.c (curr_insn_transform): Process AND in the
909 2018-01-30 Jakub Jelinek <jakub@redhat.com>
911 PR rtl-optimization/83986
912 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
913 dependence against last_pending_memory_flush in addition to
916 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
918 PR tree-optimization/81611
919 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
922 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
924 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
927 2018-01-30 Richard Biener <rguenther@suse.de>
928 Jakub Jelinek <jakub@redhat.com>
930 PR tree-optimization/84111
931 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
932 inner loops added during recursion, as they don't have up-to-date
935 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
938 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
939 (can_inline_edge_by_limits_p): ... here.
940 (can_early_inline_edge_p, check_callers,
941 update_caller_keys, update_callee_keys, recursive_inlining,
942 add_new_edges_to_heap, speculation_useful_p,
943 inline_small_functions,
944 inline_small_functions, flatten_function,
945 inline_to_all_callers_1): Update.
947 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
949 * profile-count.c (profile_count::combine_with_ipa_count): Handle
952 2018-01-30 Richard Biener <rguenther@suse.de>
954 PR tree-optimization/83008
955 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
956 invariant and constant vector uses in stmts when they need
959 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
962 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
963 * configure: Regenerate.
965 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
967 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
969 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
970 Use gen_rtx_REG rather than gen_lowpart.
972 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
974 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
975 rather than 0 when creating partial subregs.
977 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
979 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
982 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
985 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
986 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
987 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
988 flags. This restores the settings used before the 2017-07-24.
989 Turning off pre increment/decrement/modify allows IVOPTS to
990 optimize DF/SF loops where the index is an int.
992 2018-01-29 Richard Biener <rguenther@suse.de>
993 Kelvin Nilsen <kelvin@gcc.gnu.org>
996 * tree-vect-stmts.c (vectorizable_call): Don't call
997 targetm.vectorize_builtin_md_vectorized_function if callee is
1000 2018-01-22 Carl Love <cel@us.ibm.com>
1002 * doc/extend.tex: Fix typo in second arg in
1003 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1005 2018-01-29 Richard Biener <rguenther@suse.de>
1007 PR tree-optimization/84086
1008 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1009 (flush_ssaname_freelist): When SSA names were released reset
1010 the SCEV hash table.
1012 2018-01-29 Richard Biener <rguenther@suse.de>
1014 PR tree-optimization/84057
1015 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1016 removed paths when removing edges.
1018 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1020 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1021 -mfunction-return=@var{choice}.
1023 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1026 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1028 (layout::print_source_line): Likewise.
1029 (test_get_line_width_without_trailing_whitespace): Add test cases.
1031 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1034 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1037 2018-01-26 Jim Wilson <jimw@sifive.com>
1039 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1041 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1044 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1046 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1047 and CMP + SUB-immediate -> SUBS.
1049 2018-01-26 Martin Sebor <msebor@redhat.com>
1051 PR tree-optimization/83896
1052 * tree-ssa-strlen.c (get_string_len): Rename...
1053 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1054 Avoid assuming length is constant.
1055 (handle_char_store): Use HOST_WIDE_INT for string length.
1057 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1060 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1061 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1063 2018-01-26 Richard Biener <rguenther@suse.de>
1065 PR rtl-optimization/84003
1066 * dse.c (record_store): Only record redundant stores when
1067 the earlier store aliases at least all accesses the later one does.
1069 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1071 PR rtl-optimization/83985
1072 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1073 REG_CFA_RESTORE insns.
1074 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1075 REG_CFA_RESTORE insns here.
1078 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1079 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1081 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1083 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1084 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1085 (arc_init): Likewise.
1086 (arc_override_options): Likewise.
1087 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1089 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1091 * config/arc/arc.h (TARGET_DBNZ): Define.
1092 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1093 properly set the tune attribute.
1094 (dbnz): Use TARGET_DBNZ guard.
1095 * config/arc/arc.opt (mtune): Add core3 option.
1097 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1099 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1100 recognize new pic like addresses.
1101 (arc_delegitimize_address): Clean up.
1103 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1105 * config/arc/arc-arches.def: Option mrf16 valid for all
1107 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1108 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1109 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1110 * config/arc/arc-tables.opt: Regenerate.
1111 * config/arc/arc.c (arc_conditional_register_usage): Handle
1112 reduced register file case.
1113 (arc_file_start): Set must have build attributes.
1114 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1116 * config/arc/arc.opt (mrf16): Add new option.
1117 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1118 * config/arc/genmultilib.awk: Handle new mrf16 option.
1119 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1120 * config/arc/t-multilib: Regenerate.
1121 * doc/invoke.texi (ARC Options): Document mrf16 option.
1123 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1125 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1126 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1127 (arc_attribute_table): Add 'secure_call' attribute.
1128 (arc_print_operand): Print secure call operand.
1129 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1131 (arc_is_secure_call_p): New function. * config/arc/arc.md
1132 (call_i): Add support for sjli instruction.
1133 (call_value_i): Likewise.
1134 * config/arc/constraints.md (Csc): New constraint.
1136 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1137 John Eric Martin <John.Martin@emmicro-us.com>
1139 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1140 * config/arc/arc.c (_arc_jli_section): New struct.
1141 (arc_jli_section): New type.
1142 (rc_jli_sections): New static variable.
1143 (arc_handle_jli_attribute): New function.
1144 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1145 (arc_file_end): New function.
1146 (TARGET_ASM_FILE_END): Define.
1147 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1148 (arc_add_jli_section): New function.
1149 (jli_call_scan): Likewise.
1150 (arc_reorg): Call jli_call_scan.
1151 (arc_output_addsi): Remove 'S' from printing asm operand.
1152 (arc_is_jli_call_p): New function.
1153 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1155 (movhi_insn): Likewise.
1156 (movsi_insn): Likewise.
1157 (movsi_set_cc_insn): Likewise.
1158 (loadqi_update): Likewise.
1159 (load_zeroextendqisi_update): Likewise.
1160 (load_signextendqisi_update): Likewise.
1161 (loadhi_update): Likewise.
1162 (load_zeroextendhisi_update): Likewise.
1163 (load_signextendhisi_update): Likewise.
1164 (loadsi_update): Likewise.
1165 (loadsf_update): Likewise.
1166 (movsicc_insn): Likewise.
1167 (bset_insn): Likewise.
1168 (bxor_insn): Likewise.
1169 (bclr_insn): Likewise.
1170 (bmsk_insn): Likewise.
1171 (bicsi3_insn): Likewise.
1172 (cmpsi_cc_c_insn): Likewise.
1173 (movsi_ne): Likewise.
1174 (movsi_cond_exec): Likewise.
1175 (clrsbsi2): Likewise.
1186 (call_i): Remove 'S' asm letter, add jli instruction.
1187 (call_value_i): Likewise.
1188 * config/arc/arc.op (mjli-always): New option.
1189 * config/arc/constraints.md (Cji): New constraint.
1190 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1192 (subsf3_fpx): Likewise.
1193 (mulsf3_fpx): Likewise.
1194 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1196 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1198 * doc/invoke.texi (ARC): Document mjli-always option.
1200 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1202 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1203 and use incw and decw where possible
1204 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1206 2018-01-26 Richard Biener <rguenther@suse.de>
1208 PR tree-optimization/81082
1209 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1210 association if it requires casting to unsigned.
1211 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1212 from fold_plusminus_mult_expr to catch important cases late when
1213 range info is available.
1215 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1217 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1218 * configure.ac (hidden_linkonce): New test.
1219 * configure: Regenerate.
1220 * config.in: Regenerate.
1222 2018-01-26 Julia Koval <julia.koval@intel.com>
1224 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1225 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1226 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1227 _mm_mask_bitshuffle_epi64_mask): Fix type.
1228 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1229 USI_FTYPE_V4DI_V4DI_USI): Remove.
1230 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1231 __builtin_ia32_vpshufbitqmb256_mask,
1232 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1233 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1234 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1236 2018-01-26 Alan Modra <amodra@gmail.com>
1239 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1240 UNSPEC_VBPERMQ. Sort other unspecs.
1242 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1244 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1246 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1249 * predict.c (drop_profile): Do not push/pop cfun; update also
1251 (handle_missing_profiles): Fix logic looking for zero profiles.
1253 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1256 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1257 on functions with #pragma omp declare simd or functions with simd
1259 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1260 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1261 Remove trailing \n from warning_at calls.
1263 2018-01-25 Tom de Vries <tom@codesourcery.com>
1266 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1267 for neutered workers.
1269 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1272 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1273 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1275 2018-01-24 Jeff Law <law@redhat.com>
1278 * i386.c (get_probe_interval): Move to earlier point.
1279 (ix86_compute_frame_layout): If -fstack-clash-protection and
1280 the frame is larger than the probe interval, then use pushes
1281 to save registers rather than reg->mem moves.
1282 (ix86_expand_prologue): Remove conditional for int_registers_saved
1285 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1288 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1289 min/max for never referenced object.
1291 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1294 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1296 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1297 attributes from DECL_ATTRIBUTES (decl) without affecting
1298 DECL_ATTRIBUTES (current_function_decl).
1299 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1300 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1302 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1304 PR tree-optimization/83979
1305 * fold-const.c (fold_comparison): Use constant_boolean_node
1306 instead of boolean_{true,false}_node.
1308 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1310 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1313 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1315 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1316 Simplify the clause that sets the length attribute.
1317 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1318 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1319 clause that sets the length attribute.
1320 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1322 2018-01-24 Tom de Vries <tom@codesourcery.com>
1325 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1326 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1327 Add strict parameter.
1328 (prevent_branch_around_nothing): Insert dummy insn between branch to
1329 label and label with no ptx insn inbetween.
1330 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1332 2018-01-24 Tom de Vries <tom@codesourcery.com>
1335 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1336 for neutered threads in warp.
1337 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1339 2018-01-24 Richard Biener <rguenther@suse.de>
1341 PR tree-optimization/83176
1342 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1345 2018-01-24 Richard Biener <rguenther@suse.de>
1347 PR tree-optimization/82819
1348 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1349 code generating pluses that are no-ops in the target precision.
1351 2018-01-24 Richard Biener <rguenther@suse.de>
1354 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1356 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1358 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1359 to merge probabilities.
1360 * predict.c (probably_never_executed): Also mark as cold functions
1361 with global 0 profile and guessed local profile.
1362 * profile-count.c (profile_probability::combine_with_count): New
1364 * profile-count.h (profile_probability::operator*,
1365 profile_probability::operator*=, profile_probability::operator/,
1366 profile_probability::operator/=): Reduce precision to adjusted
1367 and set value to guessed on contradictory divisions.
1368 (profile_probability::combine_with_freq): Remove.
1369 (profile_probability::combine_wiht_count): Declare.
1370 (profile_count::force_nonzero):: Set to adjusted.
1371 (profile_count::probability_in):: Set quality to adjusted.
1372 * tree-ssa-tail-merge.c (replace_block_by): Use
1375 2018-01-23 Andrew Waterman <andrew@sifive.com>
1376 Jim Wilson <jimw@sifive.com>
1378 * config/riscv/riscv.c (riscv_stack_boundary): New.
1379 (riscv_option_override): Set riscv_stack_boundary. Handle
1380 riscv_preferred_stack_boundary_arg.
1381 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1382 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1383 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1384 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1385 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1386 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1388 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1391 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1392 of struct ix86_frame.
1393 (ix86_expand_epilogue): Likewise. Add a local variable for
1394 the reg_save_offset field in struct ix86_frame.
1396 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1398 PR tree-optimization/82604
1399 * tree-loop-distribution.c (enum partition_kind): New enum item
1400 PKIND_PARTIAL_MEMSET.
1401 (partition_builtin_p): Support above new enum item.
1402 (generate_code_for_partition): Ditto.
1403 (compute_access_range): Differentiate cases that equality can be
1404 proven at all loops, the innermost loops or no loops.
1405 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1406 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1407 (finalize_partitions, distribute_loop): Don't fuse partition of
1408 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1409 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1412 2018-01-23 Martin Liska <mliska@suse.cz>
1414 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1415 order to ignore the predictor.
1416 (PRED_POLYMORPHIC_CALL): Likewise.
1417 (PRED_RECURSIVE_CALL): Likewise.
1419 2018-01-23 Martin Liska <mliska@suse.cz>
1421 * tree-profile.c (tree_profiling): Print function header to
1422 aware reader which function we are working on.
1423 * value-prof.c (gimple_find_values_to_profile): Do not print
1424 not interesting value histograms.
1426 2018-01-23 Martin Liska <mliska@suse.cz>
1428 * profile-count.h (enum profile_quality): Add
1429 profile_uninitialized as the first value. Do not number values
1430 as they are zero based.
1431 (profile_count::verify): Update sanity check.
1432 (profile_probability::verify): Likewise.
1434 2018-01-23 Nathan Sidwell <nathan@acm.org>
1436 * doc/invoke.texi (ffor-scope): Deprecate.
1438 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1440 PR tree-optimization/83510
1441 * domwalk.c (set_all_edges_as_executable): New function.
1442 (dom_walker::dom_walker): Convert bool param
1443 "skip_unreachable_blocks" to enum reachability. Move setup of
1444 edge flags to set_all_edges_as_executable and only do it when
1445 reachability is REACHABLE_BLOCKS.
1446 * domwalk.h (enum dom_walker::reachability): New enum.
1447 (dom_walker::dom_walker): Convert bool param
1448 "skip_unreachable_blocks" to enum reachability.
1449 (set_all_edges_as_executable): New decl.
1450 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1451 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1453 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1454 but converting true to REACHABLE_BLOCKS.
1455 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1457 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1458 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1459 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1461 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1462 if check_all_array_refs will be called.
1464 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1466 * tree.c (selftest::test_location_wrappers): Add more test
1469 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1471 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1472 (selftest::test_bit_in_range): Likewise.
1474 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1477 * doc/sourcebuild.texi (vect_float): Say that the selector
1478 only describes the situation when -funsafe-math-optimizations is on.
1479 (vect_float_strict): Document.
1481 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1483 PR tree-optimization/83965
1484 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1485 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1486 instead of checking only for a reduction.
1487 (vect_recog_widen_sum_pattern): Likewise.
1489 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1491 * predict.c (probably_never_executed): Only use precise profile info.
1492 (compute_function_frequency): Skip after inlining hack since we now
1493 have quality checking.
1495 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1497 * profile-count.h (profile_probability::very_unlikely,
1498 profile_probability::unlikely, profile_probability::even): Set
1499 precision to guessed.
1501 2018-01-23 Richard Biener <rguenther@suse.de>
1503 PR tree-optimization/83963
1504 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1505 Properly terminate dominator walk when crossing the exit edge not
1506 when visiting its source block.
1508 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1511 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1512 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1514 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1516 PR tree-optimization/83957
1517 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1518 semicolon after for body surrounded by braces.
1520 PR tree-optimization/83081
1521 * profile-count.h (profile_probability::split): New method.
1522 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1523 Use profile_probability::split.
1524 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1525 when splitting a single conditional jump into 2.
1527 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1529 PR tree-optimization/69452
1530 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1533 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1535 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1536 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1537 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1539 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1541 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1542 * config/rl78/rl78.md: New define_expand "movdi"
1543 * config/rl78/rl78.c: New function definition rl78_split_movdi
1545 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1548 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1550 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1551 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1552 128-bit to produce an UNSPEC move to get the double word with the
1553 signbit and then a shift directly to do signbit.
1554 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1555 implementation with a new version that just does either a direct
1556 move or a regular move. Move memory interface to separate insns.
1557 Move insns so they are next to the expander.
1558 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1559 with signbit move. Split big and little endian case.
1560 (signbit<mode>2_dm_mem_le): Likewise.
1561 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1562 (signbit<mode>2_dm2): Likewise.
1564 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1566 * config/rl78/rl78.md: New define_expand "anddi3".
1568 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1570 * config/rl78/rl78.md: New define_expand "umindi3".
1572 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1574 * config/rl78/rl78.md: New define_expand "smindi3".
1576 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1578 * config/rl78/rl78.md: New define_expand "smaxdi3".
1580 2018-01-22 Carl Love <cel@us.ibm.com>
1582 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1583 LVX_V1TI): Add macro expansion.
1584 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1585 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1586 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1587 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1588 Change check to determine if the instruction is a byte reversing
1589 entry. Fix typo in comment.
1590 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1591 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1592 Add def_builtin calls for new builtins.
1593 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1594 Add define_insn expansion.
1596 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1598 * config/rl78/rl78.md: New define_expand "umaxdi3".
1600 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1602 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1603 for non-QImode registers
1605 2018-01-22 Richard Biener <rguenther@suse.de>
1607 PR tree-optimization/83963
1608 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1609 including the loop exit block.
1610 (scop_detection::merge_sese): Likewise.
1611 (scop_detection::add_scop): Do it here instead.
1613 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1615 * doc/sourcebuild.texi (arm_softfloat): Document.
1617 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1620 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1621 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1622 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1624 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1625 David Edelsohn <dje.gcc@gmail.com>
1628 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1629 Change "crset eq" to "crset 2".
1630 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1631 (*call_indirect_aix<mode>_nospec): Likewise.
1632 (*call_value_indirect_aix<mode>_nospec): Likewise.
1633 (*call_indirect_elfv2<mode>_nospec): Likewise.
1634 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1635 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1636 change assembly output from . to $.
1637 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1638 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1639 (*tablejump<mode>_internal1_nospec): Likewise.
1641 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1644 * config/sh/sh_optimize_sett_clrt.cc:
1645 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1647 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1649 PR tree-optimization/83940
1650 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1651 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1652 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1653 use it to pass back the definition type.
1654 (vect_check_store_rhs): Likewise rhs_dt_out.
1655 (vect_build_gather_load_calls): Add a mask_dt argument and use
1656 it instead of a call to vect_is_simple_use.
1657 (vectorizable_store): Update calls to vect_check_load_store_mask
1658 and vect_check_store_rhs. Use the dt returned by the latter instead
1659 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1660 instead of calls to vect_is_simple_use. Pass the scalar rather
1661 than the vector operand to vect_is_simple_use when handling
1662 second and subsequent copies of an rhs value.
1663 (vectorizable_load): Update calls to vect_check_load_store_mask
1664 and vect_build_gather_load_calls. Use the cached mask_dt and
1665 gs_info.offset_dt instead of calls to vect_is_simple_use.
1667 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1670 * tree-emutls.c: Include gimplify.h.
1671 (lower_emutls_2): New function.
1672 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1673 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1674 it before further processing.
1677 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1678 UINTVAL (trueop1) instead of INTVAL (op1).
1680 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1684 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1685 INCOMING_FRAME_SP_OFFSET if not defined.
1686 (scan_trace): Add ENTRY argument. If true and
1687 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1688 emit a note to adjust the CFA offset.
1689 (create_cfi_notes): Adjust scan_trace callers.
1690 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1691 INCOMING_FRAME_SP_OFFSET in the CIE.
1692 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1693 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1695 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1696 * doc/tm.texi: Regenerated.
1698 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1700 PR rtl-optimization/83147
1701 * lra-constraints.c (remove_inheritance_pseudos): Use
1702 lra_substitute_pseudo_within_insn.
1704 2018-01-19 Tom de Vries <tom@codesourcery.com>
1705 Cesar Philippidis <cesar@codesourcery.com>
1708 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1710 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1713 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1714 spaces for function labels.
1716 2018-01-19 Martin Liska <mliska@suse.cz>
1718 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1719 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1720 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1721 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1722 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1723 (PRED_CONST_RETURN): Change from 69 to 65.
1724 (PRED_NULL_RETURN): Change from 91 to 71.
1725 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1726 (PRED_LOOP_GUARD): Change from 66 to 73.
1728 2018-01-19 Martin Liska <mliska@suse.cz>
1730 * predict.c (predict_insn_def): Add new assert.
1731 (struct branch_predictor): Change type to signed integer.
1732 (test_prediction_value_range): Amend test to cover
1734 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1735 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1736 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1737 (PRED_LOOP_IV_COMPARE): Likewise.
1738 * predict.h (PROB_UNINITIALIZED): Define new constant.
1740 2018-01-19 Martin Liska <mliska@suse.cz>
1742 * predict.c (dump_prediction): Add new format for
1743 analyze_brprob.py script which is enabled with -details
1745 * profile-count.h (precise_p): New function.
1747 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1749 PR tree-optimization/83922
1750 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1751 there are no statements that need masking.
1752 (vect_active_double_reduction_p): New function.
1753 (vect_analyze_loop_operations): Use it when handling phis that
1754 are not in the loop header.
1756 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1758 PR tree-optimization/83914
1759 * tree-vect-loop.c (vectorizable_induction): Don't convert
1760 init_expr or apply the peeling adjustment for inductions
1761 that are nested within the vectorized loop.
1763 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1765 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1768 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1772 * function.h (gimplify_parameters): Add gimple_seq * argument.
1773 * function.c: Include gimple.h and options.h.
1774 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1775 for the added local temporaries if needed.
1776 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1777 if there are any parameter cleanups, wrap whole body into a
1778 try/finally with the cleanups.
1780 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1783 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1784 Use GET_MODE_CLASS for scalar floating point.
1786 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1790 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1791 Fix call of call_cgraph_insertion_hooks.
1793 2018-01-18 Martin Sebor <msebor@redhat.com>
1795 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1797 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1800 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1803 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1806 * common.opt: (-ffile-prefix-map): New option.
1807 * opts.c (common_handle_option): Defer it.
1808 * opts-global.c (handle_common_deferred_options): Handle it.
1809 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1810 * file-prefix-map.h: New file.
1811 (remap_debug_filename, add_debug_prefix_map): ...here.
1812 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1813 * final.c (debug_prefix_map, add_debug_prefix_map
1814 remap_debug_filename): Move to...
1815 * file-prefix-map.c: New file.
1816 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1817 generalize, get rid of alloca(), use strrchr() instead of strchr().
1818 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1819 Implement in terms of add_prefix_map().
1820 (remap_macro_filename, remap_debug_filename): Implement in term of
1822 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1823 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1824 * dbxout.c: Include file-prefix-map.h.
1825 * varasm.c: Likewise.
1826 * vmsdbgout.c: Likewise.
1827 * xcoffout.c: Likewise.
1828 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1829 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1830 * doc/invoke.texi (-ffile-prefix-map): Document.
1831 (-fdebug-prefix-map): Update description.
1833 2018-01-18 Martin Liska <mliska@suse.cz>
1835 * config/i386/i386.c (indirect_thunk_name): Document that also
1837 (output_indirect_thunk): Document why both instructions
1838 (pause and lfence) are generated.
1840 2018-01-18 Richard Biener <rguenther@suse.de>
1842 PR tree-optimization/83887
1843 * graphite-scop-detection.c
1844 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1845 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1846 (scop_detection::merge_sese): Re-implement with a flood-fill
1847 algorithm that properly finds a SESE region if it exists.
1849 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1852 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1853 pointer_diff optimizations use view_convert instead of convert.
1855 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1857 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1858 Generate different code for -mno-speculate-indirect-jumps.
1859 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1860 (*call_indirect_aix<mode>): Disable for
1861 -mno-speculate-indirect-jumps.
1862 (*call_indirect_aix<mode>_nospec): New define_insn.
1863 (*call_value_indirect_aix<mode>): Disable for
1864 -mno-speculate-indirect-jumps.
1865 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1866 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1867 -mno-speculate-indirect-jumps.
1868 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1870 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1872 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1873 long double type, set the flags for noting the default long double
1874 type, even if we don't pass or return a long double type.
1876 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1879 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1882 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1884 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1885 support for merge[hl].
1886 (fold_mergehl_helper): New helper function.
1887 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1888 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1889 (altivec_vmrglw_direct): Add xxmrglw insn.
1891 2018-01-17 Andrew Waterman <andrew@sifive.com>
1893 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1894 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1896 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1899 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1900 call the lto_location_cache before reading the
1901 DECL_SOURCE_LOCATION of the types.
1903 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1904 Richard Sandiford <richard.sandiford@linaro.org>
1906 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1907 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1908 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1909 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1910 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1912 * config/aarch64/constraints.md (aarch64_movti_operand):
1914 * config/aarch64/predicates.md (Uti): Add new constraint.
1916 2018-01-17 Carl Love <cel@us.ibm.com>
1917 * config/rs6000/vsx.md (define_expand xl_len_r,
1918 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1919 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1921 (define_expand, define_insn): Move the shift left from the
1922 define_insn to the define_expand for lxvl and stxvl instructions.
1923 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1924 and XL_LEN_R definitions to PURE.
1926 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1928 * config/i386/i386.c (indirect_thunk_name): Declare regno
1929 as unsigned int. Compare regno with INVALID_REGNUM.
1930 (output_indirect_thunk): Ditto.
1931 (output_indirect_thunk_function): Ditto.
1932 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1933 in the call to output_indirect_thunk_function.
1935 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1938 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1939 rather than the size of inner_type to determine the stack slot size
1940 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1942 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1945 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1948 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1950 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1951 endian Linux systems to optionally enable multilibs for selecting
1952 the long double type if the user configured an explicit type.
1953 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1954 have no long double multilibs if not defined.
1955 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1956 warn if the user used -mabi={ieee,ibm}longdouble and we built
1957 multilibs for long double.
1958 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1959 appropriate multilib option.
1960 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1962 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1963 for building long double multilibs.
1964 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1966 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1968 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1971 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1973 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1976 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1979 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1982 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1984 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1985 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1987 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1989 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1990 different rtl trees depending on TARGET_64BIT.
1991 (rs6000_gen_lvx): Likewise.
1993 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1995 * config/visium/visium.md (nop): Tweak comment.
1996 (hazard_nop): Likewise.
1998 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2000 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2001 -mspeculate-indirect-jumps.
2002 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2003 for -mno-speculate-indirect-jumps.
2004 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2005 (*call_value_indirect_elfv2<mode>): Disable for
2006 -mno-speculate-indirect-jumps.
2007 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2008 (indirect_jump): Emit different RTL for
2009 -mno-speculate-indirect-jumps.
2010 (*indirect_jump<mode>): Disable for
2011 -mno-speculate-indirect-jumps.
2012 (*indirect_jump<mode>_nospec): New define_insn.
2013 (tablejump): Emit different RTL for
2014 -mno-speculate-indirect-jumps.
2015 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2016 (tablejumpsi_nospec): New define_expand.
2017 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2018 (tablejumpdi_nospec): New define_expand.
2019 (*tablejump<mode>_internal1): Disable for
2020 -mno-speculate-indirect-jumps.
2021 (*tablejump<mode>_internal1_nospec): New define_insn.
2022 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2025 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2027 * caller-save.c (insert_save): Drop unnecessary parameter. All
2030 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2031 Richard Biener <rguenth@suse.de>
2034 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2035 return early, inline manually is_gimple_sizepos. Make sure if we
2036 call gimplify_expr we don't end up with a gimple constant.
2037 * tree.c (variably_modified_type_p): Don't return true for
2038 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2039 * gimplify.h (is_gimple_sizepos): Remove.
2041 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2043 PR tree-optimization/83857
2044 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2045 vectorizable_live_operation for pure SLP statements.
2046 (vectorizable_live_operation): Handle PHIs.
2048 2018-01-16 Richard Biener <rguenther@suse.de>
2050 PR tree-optimization/83867
2051 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2052 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2054 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2057 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2058 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2059 If off is not INTEGER_CST, issue a may not be aligned warning
2060 rather than isn't aligned. Use isn%'t rather than isn't.
2061 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2063 <case MULT_EXPR>: Improve the case when bottom and one of the
2064 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2065 operand, in that case check if the other operand is multiple of
2066 bottom divided by the INTEGER_CST operand.
2068 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2071 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2072 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2073 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2074 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2075 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2076 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2077 (pa_function_arg_size): New function.
2079 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2081 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2082 in a separate statement.
2084 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2086 PR tree-optimization/83847
2087 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2088 group gathers and scatters.
2090 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2092 PR rtl-optimization/86620
2093 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2095 PR rtl-optimization/83213
2096 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2097 to last if both are JUMP_INSNs.
2099 PR tree-optimization/83843
2100 * gimple-ssa-store-merging.c
2101 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2102 store_immediate_info for bswap/nop orig_stores.
2104 2018-01-15 Andrew Waterman <andrew@sifive.com>
2106 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2108 <UDIV>: Increase cost if !TARGET_DIV.
2110 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2112 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2113 (define_attr "cr_logical_3op"): New.
2114 (cceq_ior_compare): Adjust.
2115 (cceq_ior_compare_complement): Adjust.
2116 (*cceq_rev_compare): Adjust.
2117 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2118 (is_cracked_insn): Adjust.
2119 (insn_must_be_first_in_group): Adjust.
2120 * config/rs6000/40x.md: Adjust.
2121 * config/rs6000/440.md: Adjust.
2122 * config/rs6000/476.md: Adjust.
2123 * config/rs6000/601.md: Adjust.
2124 * config/rs6000/603.md: Adjust.
2125 * config/rs6000/6xx.md: Adjust.
2126 * config/rs6000/7450.md: Adjust.
2127 * config/rs6000/7xx.md: Adjust.
2128 * config/rs6000/8540.md: Adjust.
2129 * config/rs6000/cell.md: Adjust.
2130 * config/rs6000/e300c2c3.md: Adjust.
2131 * config/rs6000/e500mc.md: Adjust.
2132 * config/rs6000/e500mc64.md: Adjust.
2133 * config/rs6000/e5500.md: Adjust.
2134 * config/rs6000/e6500.md: Adjust.
2135 * config/rs6000/mpc.md: Adjust.
2136 * config/rs6000/power4.md: Adjust.
2137 * config/rs6000/power5.md: Adjust.
2138 * config/rs6000/power6.md: Adjust.
2139 * config/rs6000/power7.md: Adjust.
2140 * config/rs6000/power8.md: Adjust.
2141 * config/rs6000/power9.md: Adjust.
2142 * config/rs6000/rs64.md: Adjust.
2143 * config/rs6000/titan.md: Adjust.
2145 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2147 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2148 ix86_indirect_branch_register logic.
2150 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2152 * config/i386/constraints.md (Bs): Update
2153 ix86_indirect_branch_register check. Don't check
2154 ix86_indirect_branch_register with GOT_memory_operand.
2156 * config/i386/predicates.md (GOT_memory_operand): Don't check
2157 ix86_indirect_branch_register here.
2158 (GOT32_symbol_operand): Likewise.
2160 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2162 * config/i386/predicates.md (constant_call_address_operand):
2163 Rewrite ix86_indirect_branch_register logic.
2164 (sibcall_insn_operand): Likewise.
2166 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2168 * config/i386/constraints.md (Bs): Replace
2169 ix86_indirect_branch_thunk_register with
2170 ix86_indirect_branch_register.
2172 * config/i386/i386.md (indirect_jump): Likewise.
2173 (tablejump): Likewise.
2174 (*sibcall_memory): Likewise.
2175 (*sibcall_value_memory): Likewise.
2176 Peepholes of indirect call and jump via memory: Likewise.
2177 * config/i386/i386.opt: Likewise.
2178 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2179 (GOT_memory_operand): Likewise.
2180 (call_insn_operand): Likewise.
2181 (sibcall_insn_operand): Likewise.
2182 (GOT32_symbol_operand): Likewise.
2184 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2187 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2188 type rather than type addr's type points to.
2189 (expand_omp_atomic_mutex): Likewise.
2190 (expand_omp_atomic): Likewise.
2192 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2195 * config/i386/i386.c (output_indirect_thunk_function): Use
2196 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2197 for __x86_return_thunk.
2199 2018-01-15 Richard Biener <rguenther@suse.de>
2202 * expmed.c (extract_bit_field_1): Fix typo.
2204 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2207 * config/arm/iterators.md (VF): New mode iterator.
2208 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2209 Remove integer-related logic from pattern.
2210 (neon_vabd<mode>_3): Likewise.
2212 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2215 * common.opt (fstrict-overflow): No longer an alias.
2216 (fwrapv-pointer): New option.
2217 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2218 also for pointer types based on flag_wrapv_pointer.
2219 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2220 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2221 opts->x_flag_wrapv got set.
2222 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2223 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2224 POINTER_TYPE_OVERFLOW_UNDEFINED.
2225 * match.pd: Likewise in address comparison pattern.
2226 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2228 2018-01-15 Richard Biener <rguenther@suse.de>
2231 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2232 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2233 Reset type names to their identifier if their TYPE_DECL doesn't
2234 have linkage (and thus is used for ODR and devirt).
2235 (save_debug_info_for_decl): Remove.
2236 (save_debug_info_for_type): Likewise.
2237 (add_tree_to_fld_list): Adjust.
2238 * tree-pretty-print.c (dump_generic_node): Make dumping of
2239 type names more robust.
2241 2018-01-15 Richard Biener <rguenther@suse.de>
2243 * BASE-VER: Bump to 8.0.1.
2245 2018-01-14 Martin Sebor <msebor@redhat.com>
2248 * builtins.c (check_access): Avoid warning when the no-warning bit
2251 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2253 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2254 * ira-color (allocno_hard_regs_compare): Likewise.
2256 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2259 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2260 Use .pushsection/.popsection.
2262 2018-01-14 Martin Sebor <msebor@redhat.com>
2265 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2267 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2269 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2270 entry from extra_headers.
2271 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2272 extra_headers, make the list bitwise identical to the i?86-*-* one.
2274 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2276 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2277 -mcmodel=large with -mindirect-branch=thunk,
2278 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2279 -mfunction-return=thunk-extern.
2280 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2281 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2282 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2284 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2286 * config/i386/i386.c (print_reg): Print the name of the full
2287 integer register without '%'.
2288 (ix86_print_operand): Handle 'V'.
2289 * doc/extend.texi: Document 'V' modifier.
2291 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2293 * config/i386/constraints.md (Bs): Disallow memory operand for
2294 -mindirect-branch-register.
2296 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2297 (GOT_memory_operand): Likewise.
2298 (call_insn_operand): Likewise.
2299 (sibcall_insn_operand): Likewise.
2300 (GOT32_symbol_operand): Likewise.
2301 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2302 for -mindirect-branch-register.
2303 (tablejump): Likewise.
2304 (*sibcall_memory): Likewise.
2305 (*sibcall_value_memory): Likewise.
2306 Disallow peepholes of indirect call and jump via memory for
2307 -mindirect-branch-register.
2308 (*call_pop): Replace m with Bw.
2309 (*call_value_pop): Likewise.
2310 (*sibcall_pop_memory): Replace m with Bs.
2311 * config/i386/i386.opt (mindirect-branch-register): New option.
2312 * doc/invoke.texi: Document -mindirect-branch-register option.
2314 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2316 * config/i386/i386-protos.h (ix86_output_function_return): New.
2317 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2318 set function_return_type.
2319 (indirect_thunk_name): Add ret_p to indicate thunk for function
2321 (output_indirect_thunk_function): Pass false to
2322 indirect_thunk_name.
2323 (ix86_output_indirect_branch_via_reg): Likewise.
2324 (ix86_output_indirect_branch_via_push): Likewise.
2325 (output_indirect_thunk_function): Create alias for function
2326 return thunk if regno < 0.
2327 (ix86_output_function_return): New function.
2328 (ix86_handle_fndecl_attribute): Handle function_return.
2329 (ix86_attribute_table): Add function_return.
2330 * config/i386/i386.h (machine_function): Add
2331 function_return_type.
2332 * config/i386/i386.md (simple_return_internal): Use
2333 ix86_output_function_return.
2334 (simple_return_internal_long): Likewise.
2335 * config/i386/i386.opt (mfunction-return=): New option.
2336 (indirect_branch): Mention -mfunction-return=.
2337 * doc/extend.texi: Document function_return function attribute.
2338 * doc/invoke.texi: Document -mfunction-return= option.
2340 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2342 * config/i386/i386-opts.h (indirect_branch): New.
2343 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2344 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2345 with local indirect jump when converting indirect call and jump.
2346 (ix86_set_indirect_branch_type): New.
2347 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2348 (indirectlabelno): New.
2349 (indirect_thunk_needed): Likewise.
2350 (indirect_thunk_bnd_needed): Likewise.
2351 (indirect_thunks_used): Likewise.
2352 (indirect_thunks_bnd_used): Likewise.
2353 (INDIRECT_LABEL): Likewise.
2354 (indirect_thunk_name): Likewise.
2355 (output_indirect_thunk): Likewise.
2356 (output_indirect_thunk_function): Likewise.
2357 (ix86_output_indirect_branch_via_reg): Likewise.
2358 (ix86_output_indirect_branch_via_push): Likewise.
2359 (ix86_output_indirect_branch): Likewise.
2360 (ix86_output_indirect_jmp): Likewise.
2361 (ix86_code_end): Call output_indirect_thunk_function if needed.
2362 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2364 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2365 (ix86_attribute_table): Add indirect_branch.
2366 * config/i386/i386.h (machine_function): Add indirect_branch_type
2367 and has_local_indirect_jump.
2368 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2370 (tablejump): Likewise.
2371 (*indirect_jump): Use ix86_output_indirect_jmp.
2372 (*tablejump_1): Likewise.
2373 (simple_return_indirect_internal): Likewise.
2374 * config/i386/i386.opt (mindirect-branch=): New option.
2375 (indirect_branch): New.
2378 (thunk-inline): Likewise.
2379 (thunk-extern): Likewise.
2380 * doc/extend.texi: Document indirect_branch function attribute.
2381 * doc/invoke.texi: Document -mindirect-branch= option.
2383 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2386 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2388 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2390 * ipa-inline.c (want_inline_small_function_p): Return false if
2391 inlining has already failed with CIF_FINAL_ERROR.
2392 (update_caller_keys): Call want_inline_small_function_p before
2394 (update_callee_keys): Likewise.
2396 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2398 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2400 (rs6000_quadword_masked_address_p): Likewise.
2401 (quad_aligned_load_p): Likewise.
2402 (quad_aligned_store_p): Likewise.
2403 (const_load_sequence_p): Add comment to describe the outer-most loop.
2404 (mimic_memory_attributes_and_flags): New function.
2405 (rs6000_gen_stvx): Likewise.
2406 (replace_swapped_aligned_store): Likewise.
2407 (rs6000_gen_lvx): Likewise.
2408 (replace_swapped_aligned_load): Likewise.
2409 (replace_swapped_load_constant): Capitalize argument name in
2410 comment describing this function.
2411 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2412 and stores that access quad-word aligned addresses and replace
2413 with stvx or lvx instructions when appropriate.
2414 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2415 New function prototype.
2416 (rs6000_quadword_masked_address_p): Likewise.
2417 (rs6000_gen_lvx): Likewise.
2418 (rs6000_gen_stvx): Likewise.
2419 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2420 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2421 when memory address is aligned.
2422 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2423 this split to select lvx instruction when memory address is aligned.
2424 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2425 instruction when memory address is aligned.
2426 (*vsx_le_perm_load_v16qi): Likewise.
2427 (four unnamed splitters): Modify to select the stvx instruction
2428 when memory is aligned.
2430 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2432 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2433 which appears in the queue multiple times.
2435 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2436 Alan Hayward <alan.hayward@arm.com>
2437 David Sherwood <david.sherwood@arm.com>
2439 * tree-vectorizer.h (vec_lower_bound): New structure.
2440 (_loop_vec_info): Add check_nonzero and lower_bounds.
2441 (LOOP_VINFO_CHECK_NONZERO): New macro.
2442 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2443 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2444 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2445 fields. Make seg_len the distance travelled, not including the
2447 (dr_direction_indicator): Declare.
2448 (dr_zero_step_indicator): Likewise.
2449 (dr_known_forward_stride_p): Likewise.
2450 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2452 (runtime_alias_check_p): Allow runtime alias checks with
2454 (operator ==): Compare access_size and align.
2455 (prune_runtime_alias_test_list): Rework for new distinction between
2456 the access_size and seg_len.
2457 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2459 (get_segment_min_max): New function.
2460 (create_intersect_range_checks): Use it.
2461 (dr_step_indicator): New function.
2462 (dr_direction_indicator): Likewise.
2463 (dr_zero_step_indicator): Likewise.
2464 (dr_known_forward_stride_p): Likewise.
2465 * tree-loop-distribution.c (data_ref_segment_size): Return
2466 DR_STEP * (niters - 1).
2467 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2469 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2470 (vect_preserves_scalar_order_p): New function, split out from...
2471 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2472 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2473 (vect_vfa_access_size): New function.
2474 (vect_vfa_align): Likewise.
2475 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2476 (dump_lower_bound): New function.
2477 (vect_check_lower_bound): Likewise.
2478 (vect_small_gap_p): Likewise.
2479 (vectorizable_with_step_bound_p): Likewise.
2480 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2481 depencies if the vectorization factor is 1. Convert the checks
2482 for nonzero steps into checks on the bounds of DR_STEP. Try using
2483 a bunds check for variable steps if the minimum required step is
2484 relatively small. Update calls to the dr_with_seg_len
2485 constructor and to vect_compile_time_alias.
2486 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2488 (vect_loop_versioning): Call it.
2489 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2491 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2493 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2494 Alan Hayward <alan.hayward@arm.com>
2495 David Sherwood <david.sherwood@arm.com>
2497 * doc/sourcebuild.texi (vect_scatter_store): Document.
2498 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2500 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2502 * genopinit.c (main): Add supports_vec_scatter_store and
2503 supports_vec_scatter_store_cached to target_optabs.
2504 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2505 IFN_MASK_SCATTER_STORE.
2506 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2508 * internal-fn.h (internal_store_fn_p): Declare.
2509 (internal_fn_stored_value_index): Likewise.
2510 * internal-fn.c (scatter_store_direct): New macro.
2511 (expand_scatter_store_optab_fn): New function.
2512 (direct_scatter_store_optab_supported_p): New macro.
2513 (internal_store_fn_p): New function.
2514 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2515 IFN_MASK_SCATTER_STORE.
2516 (internal_fn_mask_index): Likewise.
2517 (internal_fn_stored_value_index): New function.
2518 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2520 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2521 * optabs-query.c (supports_vec_scatter_store_p): New function.
2522 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2523 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2524 true for scatter stores.
2525 (vect_gather_scatter_fn_p): Handle scatter stores too.
2526 (vect_check_gather_scatter): Consider using scatter stores if
2527 supports_vec_scatter_store_p.
2528 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2530 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2531 internal_fn_stored_value_index.
2532 (check_load_store_masking): Handle scatter stores too.
2533 (vect_get_store_rhs): Make public.
2534 (vectorizable_call): Use internal_store_fn_p.
2535 (vectorizable_store): Handle scatter store internal functions.
2536 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2537 when deciding whether the end of the group has been reached.
2538 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2539 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2540 (mask_scatter_store<mode>): New insns.
2542 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2543 Alan Hayward <alan.hayward@arm.com>
2544 David Sherwood <david.sherwood@arm.com>
2546 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2547 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2548 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2550 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2551 Use vect_truncate_gather_scatter_offset if we can't treat the
2552 operation as a normal gather load or scatter store.
2553 (get_group_load_store_type): Take the gather_scatter_info
2554 as argument. Try using a gather load or scatter store for
2555 single-element groups.
2556 (get_load_store_type): Update calls to get_group_load_store_type
2557 and vect_use_strided_gather_scatters_p.
2559 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2560 Alan Hayward <alan.hayward@arm.com>
2561 David Sherwood <david.sherwood@arm.com>
2563 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2564 optional tree argument.
2565 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2567 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2568 but continue to use the current value as a fallback.
2569 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2570 to compare the updates.
2571 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2572 (get_load_store_type): Use it when handling a strided access.
2573 (vect_get_strided_load_store_ops): New function.
2574 (vect_get_data_ptr_increment): Likewise.
2575 (vectorizable_load): Handle strided gather loads. Always pass
2576 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2578 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2579 Alan Hayward <alan.hayward@arm.com>
2580 David Sherwood <david.sherwood@arm.com>
2582 * doc/md.texi (gather_load@var{m}): Document.
2583 (mask_gather_load@var{m}): Likewise.
2584 * genopinit.c (main): Add supports_vec_gather_load and
2585 supports_vec_gather_load_cached to target_optabs.
2586 * optabs-tree.c (init_tree_optimization_optabs): Use
2587 ggc_cleared_alloc to allocate target_optabs.
2588 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2589 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2591 * internal-fn.h (internal_load_fn_p): Declare.
2592 (internal_gather_scatter_fn_p): Likewise.
2593 (internal_fn_mask_index): Likewise.
2594 (internal_gather_scatter_fn_supported_p): Likewise.
2595 * internal-fn.c (gather_load_direct): New macro.
2596 (expand_gather_load_optab_fn): New function.
2597 (direct_gather_load_optab_supported_p): New macro.
2598 (direct_internal_fn_optab): New function.
2599 (internal_load_fn_p): Likewise.
2600 (internal_gather_scatter_fn_p): Likewise.
2601 (internal_fn_mask_index): Likewise.
2602 (internal_gather_scatter_fn_supported_p): Likewise.
2603 * optabs-query.c (supports_at_least_one_mode_p): New function.
2604 (supports_vec_gather_load_p): Likewise.
2605 * optabs-query.h (supports_vec_gather_load_p): Declare.
2606 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2607 and memory_type field.
2608 (NUM_PATTERNS): Bump to 15.
2609 * tree-vect-data-refs.c: Include internal-fn.h.
2610 (vect_gather_scatter_fn_p): New function.
2611 (vect_describe_gather_scatter_call): Likewise.
2612 (vect_check_gather_scatter): Try using internal functions for
2613 gather loads. Recognize existing calls to a gather load function.
2614 (vect_analyze_data_refs): Consider using gather loads if
2615 supports_vec_gather_load_p.
2616 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2617 (vect_get_gather_scatter_offset_type): Likewise.
2618 (vect_convert_mask_for_vectype): Likewise.
2619 (vect_add_conversion_to_patterm): Likewise.
2620 (vect_try_gather_scatter_pattern): Likewise.
2621 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2622 (vect_vect_recog_func_ptrs): Add it.
2623 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2624 internal_fn_mask_index and internal_gather_scatter_fn_p.
2625 (check_load_store_masking): Take the gather_scatter_info as an
2626 argument and handle gather loads.
2627 (vect_get_gather_scatter_ops): New function.
2628 (vectorizable_call): Check internal_load_fn_p.
2629 (vectorizable_load): Likewise. Handle gather load internal
2631 (vectorizable_store): Update call to check_load_store_masking.
2632 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2633 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2634 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2635 (aarch64_gather_scale_operand_d): New predicates.
2636 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2637 (mask_gather_load<mode>): New insns.
2639 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2640 Alan Hayward <alan.hayward@arm.com>
2641 David Sherwood <david.sherwood@arm.com>
2643 * optabs.def (fold_left_plus_optab): New optab.
2644 * doc/md.texi (fold_left_plus_@var{m}): Document.
2645 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2646 * internal-fn.c (fold_left_direct): Define.
2647 (expand_fold_left_optab_fn): Likewise.
2648 (direct_fold_left_optab_supported_p): Likewise.
2649 * fold-const-call.c (fold_const_fold_left): New function.
2650 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2651 * tree-parloops.c (valid_reduction_p): New function.
2652 (gather_scalar_reductions): Use it.
2653 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2654 (vect_finish_replace_stmt): Declare.
2655 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2656 (needs_fold_left_reduction_p): New function, split out from...
2657 (vect_is_simple_reduction): ...here. Accept reductions that
2658 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2659 (vect_force_simple_reduction): Also store the reduction type in
2660 the assignment's STMT_VINFO_REDUC_TYPE.
2661 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2662 (merge_with_identity): New function.
2663 (vect_expand_fold_left): Likewise.
2664 (vectorize_fold_left_reduction): Likewise.
2665 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2666 scalar phi in place for it. Check for target support and reject
2667 cases that would reassociate the operation. Defer the transform
2668 phase to vectorize_fold_left_reduction.
2669 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2670 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2671 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2675 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2676 call to ifc_temp_var.
2678 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2679 Alan Hayward <alan.hayward@arm.com>
2680 David Sherwood <david.sherwood@arm.com>
2682 * target.def (legitimize_address_displacement): Take the original
2683 offset as a poly_int.
2684 * targhooks.h (default_legitimize_address_displacement): Update
2686 * targhooks.c (default_legitimize_address_displacement): Likewise.
2687 * doc/tm.texi: Regenerate.
2688 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2689 as an argument, moving assert of ad->disp == ad->disp_term to...
2690 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2691 Try calling targetm.legitimize_address_displacement before expanding
2692 the address rather than afterwards, and adjust for the new interface.
2693 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2694 Match the new hook interface. Handle SVE addresses.
2695 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2698 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2700 * Makefile.in (OBJS): Add early-remat.o.
2701 * target.def (select_early_remat_modes): New hook.
2702 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2703 * doc/tm.texi: Regenerate.
2704 * targhooks.h (default_select_early_remat_modes): Declare.
2705 * targhooks.c (default_select_early_remat_modes): New function.
2706 * timevar.def (TV_EARLY_REMAT): New timevar.
2707 * passes.def (pass_early_remat): New pass.
2708 * tree-pass.h (make_pass_early_remat): Declare.
2709 * early-remat.c: New file.
2710 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2712 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2714 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2715 Alan Hayward <alan.hayward@arm.com>
2716 David Sherwood <david.sherwood@arm.com>
2718 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2719 vfm1 with a bound_epilog parameter.
2720 (vect_do_peeling): Update calls accordingly, and move the prologue
2721 call earlier in the function. Treat the base bound_epilog as 0 for
2722 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2723 this base when peeling for gaps.
2724 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2725 with fully-masked loops.
2726 (vect_estimate_min_profitable_iters): Handle the single peeled
2727 iteration in that case.
2729 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2730 Alan Hayward <alan.hayward@arm.com>
2731 David Sherwood <david.sherwood@arm.com>
2733 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2734 single-element interleaving even if the size is not a power of 2.
2735 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2736 accesses for single-element interleaving if the group size is
2739 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2740 Alan Hayward <alan.hayward@arm.com>
2741 David Sherwood <david.sherwood@arm.com>
2743 * doc/md.texi (fold_extract_last_@var{m}): Document.
2744 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2745 * optabs.def (fold_extract_last_optab): New optab.
2746 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2747 * internal-fn.c (fold_extract_direct): New macro.
2748 (expand_fold_extract_optab_fn): Likewise.
2749 (direct_fold_extract_optab_supported_p): Likewise.
2750 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2751 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2752 EXTRACT_LAST_REDUCTION.
2753 (get_initial_def_for_reduction): Do not create an initial vector
2754 for EXTRACT_LAST_REDUCTION reductions.
2755 (vectorizable_reduction): Leave the scalar phi in place for
2756 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2757 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2758 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2759 transform phase to vectorizable_condition.
2760 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2762 (vect_finish_stmt_generation): ...here.
2763 (vect_finish_replace_stmt): New function.
2764 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2765 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2767 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2769 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2770 Alan Hayward <alan.hayward@arm.com>
2771 David Sherwood <david.sherwood@arm.com>
2773 * doc/md.texi (extract_last_@var{m}): Document.
2774 * optabs.def (extract_last_optab): New optab.
2775 * internal-fn.def (EXTRACT_LAST): New internal function.
2776 * internal-fn.c (cond_unary_direct): New macro.
2777 (expand_cond_unary_optab_fn): Likewise.
2778 (direct_cond_unary_optab_supported_p): Likewise.
2779 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2780 loops using EXTRACT_LAST.
2781 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2782 (extract_last_<mode>): ...this optab.
2783 (vec_extract<mode><Vel>): Update accordingly.
2785 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2786 Alan Hayward <alan.hayward@arm.com>
2787 David Sherwood <david.sherwood@arm.com>
2789 * target.def (empty_mask_is_expensive): New hook.
2790 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2791 * doc/tm.texi: Regenerate.
2792 * targhooks.h (default_empty_mask_is_expensive): Declare.
2793 * targhooks.c (default_empty_mask_is_expensive): New function.
2794 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2795 if the target says that empty masks are expensive.
2796 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2798 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2800 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2801 Alan Hayward <alan.hayward@arm.com>
2802 David Sherwood <david.sherwood@arm.com>
2804 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2805 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2806 (vect_use_loop_mask_for_alignment_p): New function.
2807 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2808 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2809 niters_skip argument. Make sure that the first niters_skip elements
2810 of the first iteration are inactive.
2811 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2812 Update call to vect_set_loop_masks_directly.
2813 (get_misalign_in_elems): New function, split out from...
2814 (vect_gen_prolog_loop_niters): ...here.
2815 (vect_update_init_of_dr): Take a code argument that specifies whether
2816 the adjustment should be added or subtracted.
2817 (vect_update_init_of_drs): Likewise.
2818 (vect_prepare_for_masked_peels): New function.
2819 (vect_do_peeling): Skip prologue peeling if we're using a mask
2820 instead. Update call to vect_update_inits_of_drs.
2821 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2823 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2824 alignment. Do not include the number of peeled iterations in
2825 the minimum threshold in that case.
2826 (vectorizable_induction): Adjust the start value down by
2827 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2828 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2829 Take the number of skipped iterations into account when calculating
2831 * tree-vect-stmts.c (vect_gen_while_not): New function.
2833 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2834 Alan Hayward <alan.hayward@arm.com>
2835 David Sherwood <david.sherwood@arm.com>
2837 * doc/sourcebuild.texi (vect_fully_masked): Document.
2838 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2840 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2842 (vect_analyze_loop_2): ...here. Don't check the vectorization
2843 factor against the number of loop iterations if the loop is
2846 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2847 Alan Hayward <alan.hayward@arm.com>
2848 David Sherwood <david.sherwood@arm.com>
2850 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2851 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2852 (dump_groups): Update accordingly.
2853 (iv_use::mem_type): New member variable.
2854 (address_p): New function.
2855 (record_use): Add a mem_type argument and initialize the new
2857 (record_group_use): Add a mem_type argument. Use address_p.
2858 Remove obsolete null checks of base_object. Update call to record_use.
2859 (find_interesting_uses_op): Update call to record_group_use.
2860 (find_interesting_uses_cond): Likewise.
2861 (find_interesting_uses_address): Likewise.
2862 (get_mem_type_for_internal_fn): New function.
2863 (find_address_like_use): Likewise.
2864 (find_interesting_uses_stmt): Try find_address_like_use before
2865 calling find_interesting_uses_op.
2866 (addr_offset_valid_p): Use the iv mem_type field as the type
2867 of the addressed memory.
2868 (add_autoinc_candidates): Likewise.
2869 (get_address_cost): Likewise.
2870 (split_small_address_groups_p): Use address_p.
2871 (split_address_groups): Likewise.
2872 (add_iv_candidate_for_use): Likewise.
2873 (autoinc_possible_for_pair): Likewise.
2874 (rewrite_groups): Likewise.
2875 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2876 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2877 (get_alias_ptr_type_for_ptr_address): New function.
2878 (rewrite_use_address): Rewrite address uses in calls that were
2879 identified by find_address_like_use.
2881 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2882 Alan Hayward <alan.hayward@arm.com>
2883 David Sherwood <david.sherwood@arm.com>
2885 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2887 * gimple-expr.h (is_gimple_addressable: Likewise.
2888 * gimple-expr.c (is_gimple_address): Likewise.
2889 * internal-fn.c (expand_call_mem_ref): New function.
2890 (expand_mask_load_optab_fn): Use it.
2891 (expand_mask_store_optab_fn): Likewise.
2893 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2894 Alan Hayward <alan.hayward@arm.com>
2895 David Sherwood <david.sherwood@arm.com>
2897 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2898 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2899 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2900 (cond_umax@var{mode}): Document.
2901 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2902 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2903 (cond_umin_optab, cond_umax_optab): New optabs.
2904 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2905 (COND_IOR, COND_XOR): New internal functions.
2906 * internal-fn.h (get_conditional_internal_fn): Declare.
2907 * internal-fn.c (cond_binary_direct): New macro.
2908 (expand_cond_binary_optab_fn): Likewise.
2909 (direct_cond_binary_optab_supported_p): Likewise.
2910 (get_conditional_internal_fn): New function.
2911 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2912 Cope with reduction statements that are vectorized as calls rather
2914 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2915 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2916 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2917 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2918 (UNSPEC_COND_EOR): New unspecs.
2919 (optab): Add mappings for them.
2920 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2921 (sve_int_op, sve_fp_op): New int attributes.
2923 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2924 Alan Hayward <alan.hayward@arm.com>
2925 David Sherwood <david.sherwood@arm.com>
2927 * optabs.def (while_ult_optab): New optab.
2928 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2929 * internal-fn.def (WHILE_ULT): New internal function.
2930 * internal-fn.h (direct_internal_fn_supported_p): New override
2931 that takes two types as argument.
2932 * internal-fn.c (while_direct): New macro.
2933 (expand_while_optab_fn): New function.
2934 (convert_optab_supported_p): Likewise.
2935 (direct_while_optab_supported_p): New macro.
2936 * wide-int.h (wi::udiv_ceil): New function.
2937 * tree-vectorizer.h (rgroup_masks): New structure.
2938 (vec_loop_masks): New typedef.
2939 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2941 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2942 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2943 (vect_max_vf): New function.
2944 (slpeel_make_loop_iterate_ntimes): Delete.
2945 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2946 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2947 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2948 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2949 internal-fn.h, stor-layout.h and optabs-query.h.
2950 (vect_set_loop_mask): New function.
2951 (add_preheader_seq): Likewise.
2952 (add_header_seq): Likewise.
2953 (interleave_supported_p): Likewise.
2954 (vect_maybe_permute_loop_masks): Likewise.
2955 (vect_set_loop_masks_directly): Likewise.
2956 (vect_set_loop_condition_masked): Likewise.
2957 (vect_set_loop_condition_unmasked): New function, split out from
2958 slpeel_make_loop_iterate_ntimes.
2959 (slpeel_make_loop_iterate_ntimes): Rename to..
2960 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2961 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2962 (vect_do_peeling): Update call accordingly.
2963 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2965 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2966 mask_compare_type, can_fully_mask_p and fully_masked_p.
2967 (release_vec_loop_masks): New function.
2968 (_loop_vec_info): Use it to free the loop masks.
2969 (can_produce_all_loop_masks_p): New function.
2970 (vect_get_max_nscalars_per_iter): Likewise.
2971 (vect_verify_full_masking): Likewise.
2972 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2973 retries, and free the mask rgroups before retrying. Check loop-wide
2974 reasons for disallowing fully-masked loops. Make the final decision
2975 about whether use a fully-masked loop or not.
2976 (vect_estimate_min_profitable_iters): Do not assume that peeling
2977 for the number of iterations will be needed for fully-masked loops.
2978 (vectorizable_reduction): Disable fully-masked loops.
2979 (vectorizable_live_operation): Likewise.
2980 (vect_halve_mask_nunits): New function.
2981 (vect_double_mask_nunits): Likewise.
2982 (vect_record_loop_mask): Likewise.
2983 (vect_get_loop_mask): Likewise.
2984 (vect_transform_loop): Handle the case in which the final loop
2985 iteration might handle a partial vector. Call vect_set_loop_condition
2986 instead of slpeel_make_loop_iterate_ntimes.
2987 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2988 (check_load_store_masking): New function.
2989 (prepare_load_store_mask): Likewise.
2990 (vectorizable_store): Handle fully-masked loops.
2991 (vectorizable_load): Likewise.
2992 (supportable_widening_operation): Use vect_halve_mask_nunits for
2994 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2995 (vect_gen_while): New function.
2996 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2997 (aarch64_uqdec<mode>): New insn.
2999 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3000 Alan Hayward <alan.hayward@arm.com>
3001 David Sherwood <david.sherwood@arm.com>
3003 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3004 (reduc_xor_scal_optab): New optabs.
3005 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3006 (reduc_xor_scal_@var{m}): Document.
3007 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3008 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3010 * fold-const-call.c (fold_const_call): Handle them.
3011 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3012 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3013 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3014 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3015 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3016 (UNSPEC_XORV): New unspecs.
3017 (optab): Add entries for them.
3018 (BITWISEV): New int iterator.
3019 (bit_reduc_op): New int attributes.
3021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3022 Alan Hayward <alan.hayward@arm.com>
3023 David Sherwood <david.sherwood@arm.com>
3025 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3026 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3027 * optabs.def (vec_shl_insert_optab): New optab.
3028 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3029 (duplicate_and_interleave): Likewise.
3030 * tree-vect-loop.c: Include internal-fn.h.
3031 (neutral_op_for_slp_reduction): New function, split out from
3032 get_initial_defs_for_reduction.
3033 (get_initial_def_for_reduction): Handle option 2 for variable-length
3034 vectors by loading the neutral value into a vector and then shifting
3035 the initial value into element 0.
3036 (get_initial_defs_for_reduction): Replace the code argument with
3037 the neutral value calculated by neutral_op_for_slp_reduction.
3038 Use gimple_build_vector for constant-length vectors.
3039 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3040 but the first group_size elements have a neutral value.
3041 Use duplicate_and_interleave otherwise.
3042 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3043 Update call to get_initial_defs_for_reduction. Handle SLP
3044 reductions for variable-length vectors by creating one vector
3045 result for each scalar result, with the elements associated
3046 with other scalar results stubbed out with the neutral value.
3047 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3048 Require IFN_VEC_SHL_INSERT for double reductions on
3049 variable-length vectors, or SLP reductions that have
3050 a neutral value. Require can_duplicate_and_interleave_p
3051 support for variable-length unchained SLP reductions if there
3052 is no neutral value, such as for MIN/MAX reductions. Also require
3053 the number of vector elements to be a multiple of the number of
3054 SLP statements when doing variable-length unchained SLP reductions.
3055 Update call to vect_create_epilog_for_reduction.
3056 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3057 and remove initial values.
3058 (duplicate_and_interleave): Make public.
3059 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3060 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3062 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3063 Alan Hayward <alan.hayward@arm.com>
3064 David Sherwood <david.sherwood@arm.com>
3066 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3067 (can_duplicate_and_interleave_p): New function.
3068 (vect_get_and_check_slp_defs): Take the vector of statements
3069 rather than just the current one. Remove excess parentheses.
3070 Restriction rejectinon of vect_constant_def and vect_external_def
3071 for variable-length vectors to boolean types, or types for which
3072 can_duplicate_and_interleave_p is false.
3073 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3074 (duplicate_and_interleave): New function.
3075 (vect_get_constant_vectors): Use gimple_build_vector for
3076 constant-length vectors and suitable variable-length constant
3077 vectors. Use duplicate_and_interleave for other variable-length
3078 vectors. Don't defer the update when inserting new statements.
3080 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3081 Alan Hayward <alan.hayward@arm.com>
3082 David Sherwood <david.sherwood@arm.com>
3084 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3085 min_profitable_iters doesn't go negative.
3087 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3088 Alan Hayward <alan.hayward@arm.com>
3089 David Sherwood <david.sherwood@arm.com>
3091 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3092 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3093 * optabs.def (vec_mask_load_lanes_optab): New optab.
3094 (vec_mask_store_lanes_optab): Likewise.
3095 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3096 (MASK_STORE_LANES): Likewise.
3097 * internal-fn.c (mask_load_lanes_direct): New macro.
3098 (mask_store_lanes_direct): Likewise.
3099 (expand_mask_load_optab_fn): Handle masked operations.
3100 (expand_mask_load_lanes_optab_fn): New macro.
3101 (expand_mask_store_optab_fn): Handle masked operations.
3102 (expand_mask_store_lanes_optab_fn): New macro.
3103 (direct_mask_load_lanes_optab_supported_p): Likewise.
3104 (direct_mask_store_lanes_optab_supported_p): Likewise.
3105 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3107 (vect_load_lanes_supported): Likewise.
3108 * tree-vect-data-refs.c (strip_conversion): New function.
3109 (can_group_stmts_p): Likewise.
3110 (vect_analyze_data_ref_accesses): Use it instead of checking
3111 for a pair of assignments.
3112 (vect_store_lanes_supported): Take a masked_p parameter.
3113 (vect_load_lanes_supported): Likewise.
3114 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3115 vect_store_lanes_supported and vect_load_lanes_supported.
3116 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3117 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3118 parameter. Don't allow gaps for masked accesses.
3119 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3120 and vect_load_lanes_supported.
3121 (get_load_store_type): Take a masked_p parameter and update
3122 call to get_group_load_store_type.
3123 (vectorizable_store): Update call to get_load_store_type.
3124 Handle IFN_MASK_STORE_LANES.
3125 (vectorizable_load): Update call to get_load_store_type.
3126 Handle IFN_MASK_LOAD_LANES.
3128 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3129 Alan Hayward <alan.hayward@arm.com>
3130 David Sherwood <david.sherwood@arm.com>
3132 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3134 * config/aarch64/aarch64-protos.h
3135 (aarch64_sve_struct_memory_operand_p): Declare.
3136 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3137 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3138 (VPRED, vpred): Handle SVE structure modes.
3139 * config/aarch64/constraints.md (Utx): New constraint.
3140 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3141 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3142 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3143 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3144 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3145 structure modes. Split into pieces after RA.
3146 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3147 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3149 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3150 SVE structure modes.
3151 (aarch64_classify_address): Likewise.
3152 (sizetochar): Move earlier in file.
3153 (aarch64_print_operand): Handle SVE register lists.
3154 (aarch64_array_mode): New function.
3155 (aarch64_sve_struct_memory_operand_p): Likewise.
3156 (TARGET_ARRAY_MODE): Redefine.
3158 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3159 Alan Hayward <alan.hayward@arm.com>
3160 David Sherwood <david.sherwood@arm.com>
3162 * target.def (array_mode): New target hook.
3163 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3164 * doc/tm.texi: Regenerate.
3165 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3166 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3167 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3169 * stor-layout.c (mode_for_array): Likewise. Support polynomial
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 * fold-const.c (fold_binary_loc): Check the argument types
3177 rather than the result type when testing for a vector operation.
3179 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3181 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3182 * doc/tm.texi: Regenerate.
3184 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3185 Alan Hayward <alan.hayward@arm.com>
3186 David Sherwood <david.sherwood@arm.com>
3188 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3189 (sve): Document new AArch64 extension.
3190 * doc/md.texi (w): Extend the description of the AArch64
3191 constraint to include SVE vectors.
3192 (Upl, Upa): Document new AArch64 predicate constraints.
3193 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3195 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3196 (msve-vector-bits=): New option.
3197 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3198 SVE when these are disabled.
3199 (sve): New extension.
3200 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3201 modes. Adjust their number of units based on aarch64_sve_vg.
3202 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3203 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3204 aarch64_addr_query_type.
3205 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3206 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3207 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3208 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3209 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3210 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3211 (aarch64_simd_imm_zero_p): Delete.
3212 (aarch64_check_zero_based_sve_index_immediate): Declare.
3213 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3214 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3215 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3216 (aarch64_sve_float_mul_immediate_p): Likewise.
3217 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3219 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3220 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3221 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3222 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3223 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3224 (aarch64_regmode_natural_size): Likewise.
3225 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3226 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3228 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3229 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3230 for VG and the SVE predicate registers.
3231 (V_ALIASES): Add a "z"-prefixed alias.
3232 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3233 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3234 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3235 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3236 (REG_CLASS_NAMES): Add entries for them.
3237 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3238 and the predicate registers.
3239 (aarch64_sve_vg): Declare.
3240 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3241 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3242 (REGMODE_NATURAL_SIZE): Define.
3243 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3245 * config/aarch64/aarch64.c: Include cfgrtl.h.
3246 (simd_immediate_info): Add a constructor for series vectors,
3247 and an associated step field.
3248 (aarch64_sve_vg): New variable.
3249 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3250 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3251 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3252 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3253 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3254 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3255 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3256 (aarch64_get_mask_mode): New functions.
3257 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3258 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3259 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3260 predicate modes and predicate registers. Explicitly restrict
3261 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3262 to store a vector mode if it is recognized by
3263 aarch64_classify_vector_mode.
3264 (aarch64_regmode_natural_size): New function.
3265 (aarch64_hard_regno_caller_save_mode): Return the original mode
3267 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3268 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3269 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3270 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3272 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3273 does not overlap dest if the function is frame-related. Handle
3275 (aarch64_split_add_offset): New function.
3276 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3277 them aarch64_add_offset.
3278 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3279 and update call to aarch64_sub_sp.
3280 (aarch64_add_cfa_expression): New function.
3281 (aarch64_expand_prologue): Pass extra temporary registers to the
3282 functions above. Handle the case in which we need to emit new
3283 DW_CFA_expressions for registers that were originally saved
3284 relative to the stack pointer, but now have to be expressed
3285 relative to the frame pointer.
3286 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3288 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3289 IP0 and IP1 values for SVE frames.
3290 (aarch64_expand_vec_series): New function.
3291 (aarch64_expand_sve_widened_duplicate): Likewise.
3292 (aarch64_expand_sve_const_vector): Likewise.
3293 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3294 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3295 into the register, rather than emitting a SET directly.
3296 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3297 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3298 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3299 (offset_9bit_signed_scaled_p): New functions.
3300 (aarch64_replicate_bitmask_imm): New function.
3301 (aarch64_bitmask_imm): Use it.
3302 (aarch64_cannot_force_const_mem): Reject expressions involving
3303 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3304 (aarch64_classify_index): Handle SVE indices, by requiring
3305 a plain register index with a scale that matches the element size.
3306 (aarch64_classify_address): Handle SVE addresses. Assert that
3307 the mode of the address is VOIDmode or an integer mode.
3308 Update call to aarch64_classify_symbol.
3309 (aarch64_classify_symbolic_expression): Update call to
3310 aarch64_classify_symbol.
3311 (aarch64_const_vec_all_in_range_p): New function.
3312 (aarch64_print_vector_float_operand): Likewise.
3313 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3314 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3315 and the FP immediates 1.0 and 0.5.
3316 (aarch64_print_address_internal): Handle SVE addresses.
3317 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3318 (aarch64_regno_regclass): Handle predicate registers.
3319 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3321 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3322 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3323 (aarch64_convert_sve_vector_bits): New function.
3324 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3325 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3327 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3328 Handle SVE vector and predicate modes. Accept VL-based constants
3329 that need only one temporary register, and VL offsets that require
3330 no temporary registers.
3331 (aarch64_conditional_register_usage): Mark the predicate registers
3332 as fixed if SVE isn't available.
3333 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3334 Return true for SVE vector and predicate modes.
3335 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3336 rather than an unsigned int. Handle SVE modes.
3337 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3339 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3341 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3342 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3343 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3344 (aarch64_sve_float_mul_immediate_p): New functions.
3345 (aarch64_sve_valid_immediate): New function.
3346 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3347 Explicitly reject structure modes. Check for INDEX constants.
3348 Handle PTRUE and PFALSE constants.
3349 (aarch64_check_zero_based_sve_index_immediate): New function.
3350 (aarch64_simd_imm_zero_p): Delete.
3351 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3352 vector modes. Accept constants in the range of CNT[BHWD].
3353 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3354 ask for an Advanced SIMD mode.
3355 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3356 (aarch64_simd_vector_alignment): Handle SVE predicates.
3357 (aarch64_vectorize_preferred_vector_alignment): New function.
3358 (aarch64_simd_vector_alignment_reachable): Use it instead of
3360 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3361 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3363 (MAX_VECT_LEN): Delete.
3364 (expand_vec_perm_d): Add a vec_flags field.
3365 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3366 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3367 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3369 (aarch64_evpc_rev): Rename to...
3370 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3371 (aarch64_evpc_rev_global): New function.
3372 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3373 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3375 (aarch64_evpc_sve_tbl): New function.
3376 (aarch64_expand_vec_perm_const_1): Update after rename of
3377 aarch64_evpc_rev. Handle SVE permutes too, trying
3378 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3379 than aarch64_evpc_tbl.
3380 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3381 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3382 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3383 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3384 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3385 (aarch64_expand_sve_vcond): New functions.
3386 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3387 of aarch64_vector_mode_p.
3388 (aarch64_dwarf_poly_indeterminate_value): New function.
3389 (aarch64_compute_pressure_classes): Likewise.
3390 (aarch64_can_change_mode_class): Likewise.
3391 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3392 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3393 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3394 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3395 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3396 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3397 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3398 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3400 (Dn, Dl, Dr): Accept const as well as const_vector.
3401 (Dz): Likewise. Compare against CONST0_RTX.
3402 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3403 of "vector" where appropriate.
3404 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3405 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3406 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3407 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3408 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3409 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3410 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3411 (v_int_equiv): Extend to SVE modes.
3412 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3414 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3415 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3416 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3417 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3418 (SVE_COND_FP_CMP): New int iterators.
3419 (perm_hilo): Handle the new unpack unspecs.
3420 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3422 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3423 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3424 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3425 (aarch64_equality_operator, aarch64_constant_vector_operand)
3426 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3427 (aarch64_sve_nonimmediate_operand): Likewise.
3428 (aarch64_sve_general_operand): Likewise.
3429 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3430 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3431 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3432 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3433 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3434 (aarch64_sve_float_arith_immediate): Likewise.
3435 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3436 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3437 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3438 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3439 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3440 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3441 (aarch64_sve_float_arith_operand): Likewise.
3442 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3443 (aarch64_sve_float_mul_operand): Likewise.
3444 (aarch64_sve_vec_perm_operand): Likewise.
3445 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3446 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3447 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3448 as well as const_vector.
3449 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3450 in file. Use CONST0_RTX and CONSTM1_RTX.
3451 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3452 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3453 Use aarch64_simd_imm_zero.
3454 * config/aarch64/aarch64-sve.md: New file.
3455 * config/aarch64/aarch64.md: Include it.
3456 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3457 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3458 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3459 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3460 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3461 (sve): New attribute.
3462 (enabled): Disable instructions with the sve attribute unless
3464 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3465 aarch64_expand_mov_immediate.
3466 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3467 CNT[BHSD] immediates.
3468 (movti): Split CONST_POLY_INT moves into two halves.
3469 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3470 Split additions that need a temporary here if the destination
3471 is the stack pointer.
3472 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3473 (*add<mode>3_poly_1): New instruction.
3474 (set_clobber_cc): New expander.
3476 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3478 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3479 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3480 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3481 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3482 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3483 Change innermode from fixed_mode_size to machine_mode.
3484 (simplify_subreg): Update call accordingly. Handle a constant-sized
3485 subreg of a variable-length CONST_VECTOR.
3487 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3488 Alan Hayward <alan.hayward@arm.com>
3489 David Sherwood <david.sherwood@arm.com>
3491 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3492 (add_offset_to_base): New function, split out from...
3493 (create_mem_ref): ...here. When handling a scale other than 1,
3494 check first whether the address is valid without the offset.
3495 Add it into the base if so, leaving the index and scale as-is.
3497 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3500 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3501 fold_for_warn before checking if arg2 is INTEGER_CST.
3503 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3505 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3506 (store_multiple_operation): Delete.
3507 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3508 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3509 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3510 guarded by TARGET_STRING.
3511 (rs6000_output_load_multiple): Delete.
3512 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3513 OPTION_MASK_STRING / TARGET_STRING handling.
3514 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3515 (const rs6000_opt_masks) <"string">: Change mask to 0.
3516 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3517 (MASK_STRING): Delete.
3518 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3520 (load_multiple): Delete.
3527 (store_multiple): Delete.
3534 (movmemsi_8reg): Delete.
3535 (corresponding unnamed define_insn): Delete.
3536 (movmemsi_6reg): Delete.
3537 (corresponding unnamed define_insn): Delete.
3538 (movmemsi_4reg): Delete.
3539 (corresponding unnamed define_insn): Delete.
3540 (movmemsi_2reg): Delete.
3541 (corresponding unnamed define_insn): Delete.
3542 (movmemsi_1reg): Delete.
3543 (corresponding unnamed define_insn): Delete.
3544 * config/rs6000/rs6000.opt (mno-string): New.
3545 (mstring): Replace by deprecation warning stub.
3546 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3548 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3550 * regrename.c (regrename_do_replace): If replacing the same
3551 reg multiple times, try to reuse last created gen_raw_REG.
3554 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3555 main to workaround a bug in GDB.
3557 2018-01-12 Tom de Vries <tom@codesourcery.com>
3560 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3562 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3564 PR rtl-optimization/80481
3565 * ira-color.c (get_cap_member): New function.
3566 (allocnos_conflict_by_live_ranges_p): Use it.
3567 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3568 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3570 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3573 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3574 (*saddl_se_1): Ditto.
3576 (*ssubl_se_1): Ditto.
3578 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3580 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3581 rather than wi::to_widest for DR_INITs.
3582 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3583 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3584 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3586 (vect_analyze_group_access_1): Note that here.
3588 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3590 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3591 polynomial type sizes.
3593 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3595 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3596 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3597 (gimple_add_tmp_var): Likewise.
3599 2018-01-12 Martin Liska <mliska@suse.cz>
3601 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3602 (gimple_alloc_sizes): Likewise.
3603 (dump_gimple_statistics): Use PRIu64 in printf format.
3604 * gimple.h: Change uint64_t to int.
3606 2018-01-12 Martin Liska <mliska@suse.cz>
3608 * tree-core.h: Use uint64_t instead of int.
3609 * tree.c (tree_node_counts): Likewise.
3610 (tree_node_sizes): Likewise.
3611 (dump_tree_statistics): Use PRIu64 in printf format.
3613 2018-01-12 Martin Liska <mliska@suse.cz>
3615 * Makefile.in: As qsort_chk is implemented in vec.c, add
3616 vec.o to linkage of gencfn-macros.
3617 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3618 passing the info to record_node_allocation_statistics.
3619 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3621 * ggc-common.c (struct ggc_usage): Add operator== and use
3622 it in operator< and compare function.
3623 * mem-stats.h (struct mem_usage): Likewise.
3624 * vec.c (struct vec_usage): Remove operator< and compare
3625 function. Can be simply inherited.
3627 2018-01-12 Martin Jambor <mjambor@suse.cz>
3630 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3631 * tree-ssa-math-opts.c: Include domwalk.h.
3632 (convert_mult_to_fma_1): New function.
3633 (fma_transformation_info): New type.
3634 (fma_deferring_state): Likewise.
3635 (cancel_fma_deferring): New function.
3636 (result_of_phi): Likewise.
3637 (last_fma_candidate_feeds_initial_phi): Likewise.
3638 (convert_mult_to_fma): Added deferring logic, split actual
3639 transformation to convert_mult_to_fma_1.
3640 (math_opts_dom_walker): New type.
3641 (math_opts_dom_walker::after_dom_children): New method, body moved
3642 here from pass_optimize_widening_mul::execute, added deferring logic
3644 (pass_optimize_widening_mul::execute): Moved most of code to
3645 math_opts_dom_walker::after_dom_children.
3646 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3647 * config/i386/i386.c (ix86_option_override_internal): Added
3648 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3650 2018-01-12 Richard Biener <rguenther@suse.de>
3653 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3654 inline instance vars.
3656 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3659 * config/rx/rx.c (rx_is_restricted_memory_address):
3662 2018-01-12 Richard Biener <rguenther@suse.de>
3664 PR tree-optimization/80846
3665 * target.def (split_reduction): New target hook.
3666 * targhooks.c (default_split_reduction): New function.
3667 * targhooks.h (default_split_reduction): Declare.
3668 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3669 target requests first reduce vectors by combining low and high
3671 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3672 (get_vectype_for_scalar_type_and_size): Export.
3673 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3674 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3675 * doc/tm.texi: Regenerate.
3676 * config/i386/i386.c (ix86_split_reduction): Implement
3677 TARGET_VECTORIZE_SPLIT_REDUCTION.
3679 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3682 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3683 in PIC mode except for TARGET_VXWORKS_RTP.
3684 * config/sparc/sparc.c: Include cfgrtl.h.
3685 (TARGET_INIT_PIC_REG): Define.
3686 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3687 (sparc_pic_register_p): New predicate.
3688 (sparc_legitimate_address_p): Use it.
3689 (sparc_legitimize_pic_address): Likewise.
3690 (sparc_delegitimize_address): Likewise.
3691 (sparc_mode_dependent_address_p): Likewise.
3692 (gen_load_pcrel_sym): Remove 4th parameter.
3693 (load_got_register): Adjust call to above. Remove obsolete stuff.
3694 (sparc_expand_prologue): Do not call load_got_register here.
3695 (sparc_flat_expand_prologue): Likewise.
3696 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3697 (sparc_use_pseudo_pic_reg): New function.
3698 (sparc_init_pic_reg): Likewise.
3699 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3700 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3702 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3704 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3705 Add item for branch_cost.
3707 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3709 PR rtl-optimization/83565
3710 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3711 not extend the result to a larger mode for rotate operations.
3712 (num_sign_bit_copies1): Likewise.
3714 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3717 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3719 Use values-Xc.o for -pedantic.
3720 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3722 2018-01-12 Martin Liska <mliska@suse.cz>
3725 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3727 (possible_polymorphic_call_targets): Use it.
3728 (ipa_devirt): Likewise.
3730 2018-01-12 Martin Liska <mliska@suse.cz>
3732 * profile-count.h (enum profile_quality): Use 0 as invalid
3733 enum value of profile_quality.
3735 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3737 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3738 -mext-string options.
3740 2018-01-12 Richard Biener <rguenther@suse.de>
3742 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3743 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3744 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3746 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3748 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3750 * configure.ac (--with-long-double-format): Add support for the
3751 configuration option to change the default long double format on
3753 * config.gcc (powerpc*-linux*-*): Likewise.
3754 * configure: Regenerate.
3755 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3756 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3757 used without modification.
3759 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3761 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3762 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3763 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3764 MISC_BUILTIN_SPEC_BARRIER.
3765 (rs6000_init_builtins): Likewise.
3766 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3768 (speculation_barrier): New define_insn.
3769 * doc/extend.texi: Document __builtin_speculation_barrier.
3771 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3774 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3775 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3776 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3778 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3779 integral modes instead of "ss" and "sd".
3780 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3781 vectors with 32-bit and 64-bit elements.
3782 (vecdupssescalarmodesuffix): New mode attribute.
3783 (vec_dup<mode>): Use it.
3785 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3788 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3789 frame if argument is passed on stack.
3791 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3794 * ree.c (combine_reaching_defs): Optimize also
3795 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3796 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3798 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3801 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3803 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3806 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3807 after they are computed.
3809 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3811 PR tree-optimization/83695
3812 * gimple-loop-linterchange.cc
3813 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3814 reset cached scev information after interchange.
3815 (pass_linterchange::execute): Remove call to scev_reset_htab.
3817 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3819 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3820 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3821 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3822 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3823 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3824 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3825 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3826 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3827 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3828 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3829 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3830 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3831 (V_lane_reg): Likewise.
3832 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3834 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3835 (vfmal_lane_low<mode>_intrinsic,
3836 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3837 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3838 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3839 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3840 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3841 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3843 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3845 * config/arm/arm-cpus.in (fp16fml): New feature.
3846 (ALL_SIMD): Add fp16fml.
3847 (armv8.2-a): Add fp16fml as an option.
3848 (armv8.3-a): Likewise.
3849 (armv8.4-a): Add fp16fml as part of fp16.
3850 * config/arm/arm.h (TARGET_FP16FML): Define.
3851 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3853 * config/arm/arm-modes.def (V2HF): Define.
3854 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3855 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3856 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3857 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3858 vfmsl_low, vfmsl_high): New set of builtins.
3859 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3860 (vfml_op): New code attribute.
3861 (VFMLHALVES): New int iterator.
3862 (VFML, VFMLSEL): New mode attributes.
3863 (V_reg): Define mapping for V2HF.
3864 (V_hi, V_lo): New mode attributes.
3865 (VF_constraint): Likewise.
3866 (vfml_half, vfml_half_selector): New int attributes.
3867 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3869 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3870 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3872 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3873 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3874 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3875 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3877 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3878 Document new effective target and option set.
3880 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3882 * config/arm/arm-cpus.in (armv8_4): New feature.
3883 (ARMv8_4a): New fgroup.
3884 (armv8.4-a): New arch.
3885 * config/arm/arm-tables.opt: Regenerate.
3886 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3887 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3888 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3889 Add matching rules for -march=armv8.4-a and extensions.
3890 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3892 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3895 * config/rx/rx.md (BW): New mode attribute.
3896 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3898 2018-01-11 Richard Biener <rguenther@suse.de>
3900 PR tree-optimization/83435
3901 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3902 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3903 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3905 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3906 Alan Hayward <alan.hayward@arm.com>
3907 David Sherwood <david.sherwood@arm.com>
3909 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3911 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3912 (aarch64_print_address_internal): Use it to check for a zero offset.
3914 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3915 Alan Hayward <alan.hayward@arm.com>
3916 David Sherwood <david.sherwood@arm.com>
3918 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3919 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3920 Return a poly_int64 rather than a HOST_WIDE_INT.
3921 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3922 rather than a HOST_WIDE_INT.
3923 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3924 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3925 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3926 final_offset from HOST_WIDE_INT to poly_int64.
3927 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3928 to_constant when getting the number of units in an Advanced SIMD
3930 (aarch64_builtin_vectorized_function): Check for a constant number
3932 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3934 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3935 attribute instead of GET_MODE_NUNITS.
3936 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3937 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3938 GET_MODE_SIZE for fixed-size registers.
3939 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3940 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3941 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3942 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3943 (aarch64_print_operand, aarch64_print_address_internal)
3944 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3945 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3946 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3947 Handle polynomial GET_MODE_SIZE.
3948 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3949 wider than SImode without modification.
3950 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3951 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3952 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3953 passing and returning SVE modes.
3954 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3955 rather than GEN_INT.
3956 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3957 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3958 (aarch64_allocate_and_probe_stack_space): Likewise.
3959 (aarch64_layout_frame): Cope with polynomial offsets.
3960 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3961 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3963 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3964 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3965 poly_int64 rather than a HOST_WIDE_INT.
3966 (aarch64_get_separate_components, aarch64_process_components)
3967 (aarch64_expand_prologue, aarch64_expand_epilogue)
3968 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3969 (aarch64_anchor_offset): New function, split out from...
3970 (aarch64_legitimize_address): ...here.
3971 (aarch64_builtin_vectorization_cost): Handle polynomial
3972 TYPE_VECTOR_SUBPARTS.
3973 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3975 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3976 number of elements from the PARALLEL rather than the mode.
3977 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3978 rather than GET_MODE_BITSIZE.
3979 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3980 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3981 (aarch64_expand_vec_perm_const_1): Handle polynomial
3982 d->perm.length () and d->perm elements.
3983 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3984 Apply to_constant to d->perm elements.
3985 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3986 polynomial CONST_VECTOR_NUNITS.
3987 (aarch64_move_pointer): Take amount as a poly_int64 rather
3989 (aarch64_progress_pointer): Avoid temporary variable.
3990 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3991 the mode attribute instead of GET_MODE.
3993 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3994 Alan Hayward <alan.hayward@arm.com>
3995 David Sherwood <david.sherwood@arm.com>
3997 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3998 x exists before using it.
3999 (aarch64_add_constant_internal): Rename to...
4000 (aarch64_add_offset_1): ...this. Replace regnum with separate
4001 src and dest rtxes. Handle the case in which they're different,
4002 including when the offset is zero. Replace scratchreg with an rtx.
4003 Use 2 additions if there is no spare register into which we can
4004 move a 16-bit constant.
4005 (aarch64_add_constant): Delete.
4006 (aarch64_add_offset): Replace reg with separate src and dest
4007 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4008 Use aarch64_add_offset_1.
4009 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4010 an rtx rather than an int. Take the delta as a poly_int64
4011 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4012 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4013 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4014 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4015 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4017 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4018 aarch64_add_constant.
4020 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4022 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4023 Use scalar_float_mode.
4025 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4027 * config/aarch64/aarch64-simd.md
4028 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4029 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4030 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4031 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4032 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4033 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4034 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4035 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4036 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4037 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4039 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4042 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4043 targ_options->x_arm_arch_string is non NULL.
4045 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4047 * config/aarch64/aarch64.h
4048 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4050 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4053 * expmed.c (emit_store_flag_force): Swap if const op0
4054 and change VOIDmode to mode of op0.
4056 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4058 PR rtl-optimization/83761
4059 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4060 than bytes to mode_for_size.
4062 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4065 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4066 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4069 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4072 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4073 when in layout mode.
4074 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4075 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4078 2018-01-10 Michael Collison <michael.collison@arm.com>
4080 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4081 * config/aarch64/aarch64-option-extension.def: Add
4082 AARCH64_OPT_EXTENSION of 'fp16fml'.
4083 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4084 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4085 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4086 * config/aarch64/constraints.md (Ui7): New constraint.
4087 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4088 (VFMLA_SEL_W): Ditto.
4091 (VFMLA16_LOW): New int iterator.
4092 (VFMLA16_HIGH): Ditto.
4093 (UNSPEC_FMLAL): New unspec.
4094 (UNSPEC_FMLSL): Ditto.
4095 (UNSPEC_FMLAL2): Ditto.
4096 (UNSPEC_FMLSL2): Ditto.
4097 (f16mac): New code attribute.
4098 * config/aarch64/aarch64-simd-builtins.def
4099 (aarch64_fmlal_lowv2sf): Ditto.
4100 (aarch64_fmlsl_lowv2sf): Ditto.
4101 (aarch64_fmlalq_lowv4sf): Ditto.
4102 (aarch64_fmlslq_lowv4sf): Ditto.
4103 (aarch64_fmlal_highv2sf): Ditto.
4104 (aarch64_fmlsl_highv2sf): Ditto.
4105 (aarch64_fmlalq_highv4sf): Ditto.
4106 (aarch64_fmlslq_highv4sf): Ditto.
4107 (aarch64_fmlal_lane_lowv2sf): Ditto.
4108 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4109 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4110 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4111 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4112 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4113 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4114 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4115 (aarch64_fmlal_lane_highv2sf): Ditto.
4116 (aarch64_fmlsl_lane_highv2sf): Ditto.
4117 (aarch64_fmlal_laneq_highv2sf): Ditto.
4118 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4119 (aarch64_fmlalq_lane_highv4sf): Ditto.
4120 (aarch64_fmlsl_lane_highv4sf): Ditto.
4121 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4122 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4123 * config/aarch64/aarch64-simd.md:
4124 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4125 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4126 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4127 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4128 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4129 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4130 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4131 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4132 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4133 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4134 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4135 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4136 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4137 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4138 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4139 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4140 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4141 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4142 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4143 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4144 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4145 (vfmlsl_low_u32): Ditto.
4146 (vfmlalq_low_u32): Ditto.
4147 (vfmlslq_low_u32): Ditto.
4148 (vfmlal_high_u32): Ditto.
4149 (vfmlsl_high_u32): Ditto.
4150 (vfmlalq_high_u32): Ditto.
4151 (vfmlslq_high_u32): Ditto.
4152 (vfmlal_lane_low_u32): Ditto.
4153 (vfmlsl_lane_low_u32): Ditto.
4154 (vfmlal_laneq_low_u32): Ditto.
4155 (vfmlsl_laneq_low_u32): Ditto.
4156 (vfmlalq_lane_low_u32): Ditto.
4157 (vfmlslq_lane_low_u32): Ditto.
4158 (vfmlalq_laneq_low_u32): Ditto.
4159 (vfmlslq_laneq_low_u32): Ditto.
4160 (vfmlal_lane_high_u32): Ditto.
4161 (vfmlsl_lane_high_u32): Ditto.
4162 (vfmlal_laneq_high_u32): Ditto.
4163 (vfmlsl_laneq_high_u32): Ditto.
4164 (vfmlalq_lane_high_u32): Ditto.
4165 (vfmlslq_lane_high_u32): Ditto.
4166 (vfmlalq_laneq_high_u32): Ditto.
4167 (vfmlslq_laneq_high_u32): Ditto.
4168 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4169 (AARCH64_FL_FOR_ARCH8_4): New.
4170 (AARCH64_ISA_F16FML): New ISA flag.
4171 (TARGET_F16FML): New feature flag for fp16fml.
4172 (doc/invoke.texi): Document new fp16fml option.
4174 2018-01-10 Michael Collison <michael.collison@arm.com>
4176 * config/aarch64/aarch64-builtins.c:
4177 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4178 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4179 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4180 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4181 (AARCH64_ISA_SHA3): New ISA flag.
4182 (TARGET_SHA3): New feature flag for sha3.
4183 * config/aarch64/iterators.md (sha512_op): New int attribute.
4184 (CRYPTO_SHA512): New int iterator.
4185 (UNSPEC_SHA512H): New unspec.
4186 (UNSPEC_SHA512H2): Ditto.
4187 (UNSPEC_SHA512SU0): Ditto.
4188 (UNSPEC_SHA512SU1): Ditto.
4189 * config/aarch64/aarch64-simd-builtins.def
4190 (aarch64_crypto_sha512hqv2di): New builtin.
4191 (aarch64_crypto_sha512h2qv2di): Ditto.
4192 (aarch64_crypto_sha512su0qv2di): Ditto.
4193 (aarch64_crypto_sha512su1qv2di): Ditto.
4194 (aarch64_eor3qv8hi): Ditto.
4195 (aarch64_rax1qv2di): Ditto.
4196 (aarch64_xarqv2di): Ditto.
4197 (aarch64_bcaxqv8hi): Ditto.
4198 * config/aarch64/aarch64-simd.md:
4199 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4200 (aarch64_crypto_sha512su0qv2di): Ditto.
4201 (aarch64_crypto_sha512su1qv2di): Ditto.
4202 (aarch64_eor3qv8hi): Ditto.
4203 (aarch64_rax1qv2di): Ditto.
4204 (aarch64_xarqv2di): Ditto.
4205 (aarch64_bcaxqv8hi): Ditto.
4206 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4207 (vsha512h2q_u64): Ditto.
4208 (vsha512su0q_u64): Ditto.
4209 (vsha512su1q_u64): Ditto.
4210 (veor3q_u16): Ditto.
4211 (vrax1q_u64): Ditto.
4213 (vbcaxq_u16): Ditto.
4214 * config/arm/types.md (crypto_sha512): New type attribute.
4215 (crypto_sha3): Ditto.
4216 (doc/invoke.texi): Document new sha3 option.
4218 2018-01-10 Michael Collison <michael.collison@arm.com>
4220 * config/aarch64/aarch64-builtins.c:
4221 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4222 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4223 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4224 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4225 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4226 (AARCH64_ISA_SM4): New ISA flag.
4227 (TARGET_SM4): New feature flag for sm4.
4228 * config/aarch64/aarch64-simd-builtins.def
4229 (aarch64_sm3ss1qv4si): Ditto.
4230 (aarch64_sm3tt1aq4si): Ditto.
4231 (aarch64_sm3tt1bq4si): Ditto.
4232 (aarch64_sm3tt2aq4si): Ditto.
4233 (aarch64_sm3tt2bq4si): Ditto.
4234 (aarch64_sm3partw1qv4si): Ditto.
4235 (aarch64_sm3partw2qv4si): Ditto.
4236 (aarch64_sm4eqv4si): Ditto.
4237 (aarch64_sm4ekeyqv4si): Ditto.
4238 * config/aarch64/aarch64-simd.md:
4239 (aarch64_sm3ss1qv4si): Ditto.
4240 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4241 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4242 (aarch64_sm4eqv4si): Ditto.
4243 (aarch64_sm4ekeyqv4si): Ditto.
4244 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4245 (sm3part_op): Ditto.
4246 (CRYPTO_SM3TT): Ditto.
4247 (CRYPTO_SM3PART): Ditto.
4248 (UNSPEC_SM3SS1): New unspec.
4249 (UNSPEC_SM3TT1A): Ditto.
4250 (UNSPEC_SM3TT1B): Ditto.
4251 (UNSPEC_SM3TT2A): Ditto.
4252 (UNSPEC_SM3TT2B): Ditto.
4253 (UNSPEC_SM3PARTW1): Ditto.
4254 (UNSPEC_SM3PARTW2): Ditto.
4255 (UNSPEC_SM4E): Ditto.
4256 (UNSPEC_SM4EKEY): Ditto.
4257 * config/aarch64/constraints.md (Ui2): New constraint.
4258 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4259 * config/arm/types.md (crypto_sm3): New type attribute.
4260 (crypto_sm4): Ditto.
4261 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4262 (vsm3tt1aq_u32): Ditto.
4263 (vsm3tt1bq_u32): Ditto.
4264 (vsm3tt2aq_u32): Ditto.
4265 (vsm3tt2bq_u32): Ditto.
4266 (vsm3partw1q_u32): Ditto.
4267 (vsm3partw2q_u32): Ditto.
4268 (vsm4eq_u32): Ditto.
4269 (vsm4ekeyq_u32): Ditto.
4270 (doc/invoke.texi): Document new sm4 option.
4272 2018-01-10 Michael Collison <michael.collison@arm.com>
4274 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4275 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4276 (AARCH64_FL_FOR_ARCH8_4): New.
4277 (AARCH64_FL_V8_4): New flag.
4278 (doc/invoke.texi): Document new armv8.4-a option.
4280 2018-01-10 Michael Collison <michael.collison@arm.com>
4282 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4283 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4284 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4285 * config/aarch64/aarch64-option-extension.def: Add
4286 AARCH64_OPT_EXTENSION of 'sha2'.
4287 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4288 (crypto): Disable sha2 and aes if crypto disabled.
4289 (crypto): Enable aes and sha2 if enabled.
4290 (simd): Disable sha2 and aes if simd disabled.
4291 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4293 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4294 (TARGET_SHA2): New feature flag for sha2.
4295 (TARGET_AES): New feature flag for aes.
4296 * config/aarch64/aarch64-simd.md:
4297 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4298 conditional on TARGET_AES.
4299 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4300 (aarch64_crypto_sha1hsi): Make pattern conditional
4302 (aarch64_crypto_sha1hv4si): Ditto.
4303 (aarch64_be_crypto_sha1hv4si): Ditto.
4304 (aarch64_crypto_sha1su1v4si): Ditto.
4305 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4306 (aarch64_crypto_sha1su0v4si): Ditto.
4307 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4308 (aarch64_crypto_sha256su0v4si): Ditto.
4309 (aarch64_crypto_sha256su1v4si): Ditto.
4310 (doc/invoke.texi): Document new aes and sha2 options.
4312 2018-01-10 Martin Sebor <msebor@redhat.com>
4314 PR tree-optimization/83781
4315 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4318 2018-01-11 Martin Sebor <msebor@gmail.com>
4319 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4321 PR tree-optimization/83501
4322 PR tree-optimization/81703
4324 * tree-ssa-strlen.c (get_string_cst): Rename...
4325 (get_string_len): ...to this. Handle global constants.
4326 (handle_char_store): Adjust.
4328 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4329 Jim Wilson <jimw@sifive.com>
4331 * config/riscv/riscv-protos.h (riscv_output_return): New.
4332 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4333 (riscv_attribute_table, riscv_output_return),
4334 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4335 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4336 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4338 (riscv_expand_prologue): Add early return for naked function.
4339 (riscv_expand_epilogue): Likewise.
4340 (riscv_function_ok_for_sibcall): Return false for naked function.
4341 (riscv_set_current_function): New.
4342 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4343 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4344 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4345 * doc/extend.texi (RISC-V Function Attributes): New.
4347 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4349 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4350 check for 128-bit long double before checking TCmode.
4351 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4352 128-bit long doubles before checking TFmode or TCmode.
4353 (FLOAT128_IBM_P): Likewise.
4355 2018-01-10 Martin Sebor <msebor@redhat.com>
4357 PR tree-optimization/83671
4358 * builtins.c (c_strlen): Unconditionally return zero for the empty
4360 Use -Warray-bounds for warnings.
4361 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4362 for non-constant array indices with COMPONENT_REF, arrays of
4363 arrays, and pointers to arrays.
4364 (gimple_fold_builtin_strlen): Determine and set length range for
4365 non-constant character arrays.
4367 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4370 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4373 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4375 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4377 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4380 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4381 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4382 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4383 indexed_or_indirect_operand predicate.
4384 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4385 (*vsx_le_perm_load_v8hi): Likewise.
4386 (*vsx_le_perm_load_v16qi): Likewise.
4387 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4388 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4389 (*vsx_le_perm_store_v8hi): Likewise.
4390 (*vsx_le_perm_store_v16qi): Likewise.
4391 (eight unnamed splitters): Likewise.
4393 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4395 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4396 * config/rs6000/emmintrin.h: Likewise.
4397 * config/rs6000/mmintrin.h: Likewise.
4398 * config/rs6000/xmmintrin.h: Likewise.
4400 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4403 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4405 * tree.c (tree_nop_conversion): Return true for location wrapper
4407 (maybe_wrap_with_location): New function.
4408 (selftest::check_strip_nops): New function.
4409 (selftest::test_location_wrappers): New function.
4410 (selftest::tree_c_tests): Call it.
4411 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4412 (maybe_wrap_with_location): New decl.
4413 (EXPR_LOCATION_WRAPPER_P): New macro.
4414 (location_wrapper_p): New inline function.
4415 (tree_strip_any_location_wrapper): New inline function.
4417 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4420 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4421 stack_realign_offset for the largest alignment of stack slot
4423 (ix86_find_max_used_stack_alignment): New function.
4424 (ix86_finalize_stack_frame_flags): Use it. Set
4425 max_used_stack_alignment if we don't realign stack.
4426 * config/i386/i386.h (machine_function): Add
4427 max_used_stack_alignment.
4429 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4431 * config/arm/arm.opt (-mbranch-cost): New option.
4432 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4435 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4438 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4439 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4441 2018-01-10 Richard Biener <rguenther@suse.de>
4444 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4445 early out so it also covers the case where we have a non-NULL
4448 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4450 PR tree-optimization/83753
4451 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4452 for non-strided grouped accesses if the number of elements is 1.
4454 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4457 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4458 * i386.h (TARGET_USE_GATHER): Define.
4459 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4461 2018-01-10 Martin Liska <mliska@suse.cz>
4464 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4465 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4467 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4468 CLEANUP_NO_PARTITIONING is not set.
4470 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4472 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4473 for vectors, as a partial revert of r254296.
4474 * rtl.h (const_vec_p): Delete.
4475 (const_vec_duplicate_p): Don't test for vector CONSTs.
4476 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4477 * expmed.c (make_tree): Likewise.
4480 * common.md (E, F): Use CONSTANT_P instead of checking for
4482 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4483 checking for CONST_VECTOR.
4485 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4488 * predict.c (force_edge_cold): Handle in more sane way edges
4491 2018-01-09 Carl Love <cel@us.ibm.com>
4493 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4495 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4496 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4497 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4498 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4499 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4500 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4501 * config/rs6000/rs6000-protos.h: Add extern defition for
4502 rs6000_generate_float2_double_code.
4503 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4505 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4506 (float2_v2df): Add define_expand.
4508 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4511 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4512 op_mode in the force_to_mode call.
4514 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4516 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4517 instead of checking each element individually.
4518 (aarch64_evpc_uzp): Likewise.
4519 (aarch64_evpc_zip): Likewise.
4520 (aarch64_evpc_ext): Likewise.
4521 (aarch64_evpc_rev): Likewise.
4522 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4523 instead of checking each element individually. Return true without
4525 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4526 whether all selected elements come from the same input, instead of
4527 checking each element individually. Remove calls to gen_rtx_REG,
4528 start_sequence and end_sequence and instead assert that no rtl is
4531 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4533 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4534 order of HIGH and CONST checks.
4536 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4538 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4539 if the destination isn't an SSA_NAME.
4541 2018-01-09 Richard Biener <rguenther@suse.de>
4543 PR tree-optimization/83668
4544 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4546 (canonicalize_loop_form): ... here, renamed from ...
4547 (canonicalize_loop_closed_ssa_form): ... this and amended to
4548 swap successor edges for loop exit blocks to make us use
4549 the RPO order we need for initial schedule generation.
4551 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4553 PR tree-optimization/64811
4554 * match.pd: When optimizing comparisons with Inf, avoid
4555 introducing or losing exceptions from comparisons with NaN.
4557 2018-01-09 Martin Liska <mliska@suse.cz>
4560 * asan.c (shadow_mem_size): Add gcc_assert.
4562 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4564 Don't save registers in main().
4567 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4568 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4569 * config/avr/avr.c (avr_set_current_function): Don't error if
4570 naked, OS_task or OS_main are specified at the same time.
4571 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4573 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4575 * common/config/avr/avr-common.c (avr_option_optimization_table):
4576 Switch on -mmain-is-OS_task for optimizing compilations.
4578 2018-01-09 Richard Biener <rguenther@suse.de>
4580 PR tree-optimization/83572
4581 * graphite.c: Include cfganal.h.
4582 (graphite_transform_loops): Connect infinite loops to exit
4583 and remove fake edges at the end.
4585 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4587 * ipa-inline.c (edge_badness): Revert accidental checkin.
4589 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4592 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4593 symbols; not inline clones.
4595 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4598 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4599 hard registers. Formatting fixes.
4601 PR preprocessor/83722
4602 * gcc.c (try_generate_repro): Pass
4603 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4604 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4607 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4608 Kito Cheng <kito.cheng@gmail.com>
4610 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4611 (riscv_leaf_function_p): Delete.
4612 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4614 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4616 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4618 (do_ifelse): New function.
4619 (do_isel): New function.
4620 (do_sub3): New function.
4621 (do_add3): New function.
4622 (do_load_mask_compare): New function.
4623 (do_overlap_load_compare): New function.
4624 (expand_compare_loop): New function.
4625 (expand_block_compare): Call expand_compare_loop() when appropriate.
4626 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4628 (-mblock-compare-inline-loop-limit): New option.
4630 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4633 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4634 Reverse order of second and third operands in first alternative.
4635 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4636 of first and second elements in UNSPEC_VPERMR vector.
4637 (altivec_expand_vec_perm_le): Likewise.
4639 2018-01-08 Jeff Law <law@redhat.com>
4641 PR rtl-optimizatin/81308
4642 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4643 (process_switch): If group_case_labels makes a change, then set
4645 (pass_convert_switch::execute): If a switch is converted, then
4646 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4648 PR rtl-optimization/81308
4649 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4652 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4654 PR target/83663 - Revert r255946
4655 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4656 generation for cases where splatting a value is not useful.
4657 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4658 across a vec_duplicate and a paradoxical subreg forming a vector
4659 mode to a vec_concat.
4661 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4663 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4664 -march=armv8.3-a variants.
4665 * config/arm/t-multilib: Likewise.
4666 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4668 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4670 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4672 (cceq_ior_compare_complement): Give it a name so I can use it, and
4673 change boolean_or_operator predicate to boolean_operator so it can
4674 be used to generate a crand.
4675 (eqne): New code iterator.
4676 (bd/bd_neg): New code_attrs.
4677 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4678 a single define_insn.
4679 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4680 decrement (bdnzt/bdnzf/bdzt/bdzf).
4681 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4682 with the new names of the branch decrement patterns, and added the
4683 names of the branch decrement conditional patterns.
4685 2018-01-08 Richard Biener <rguenther@suse.de>
4687 PR tree-optimization/83563
4688 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4691 2018-01-08 Richard Biener <rguenther@suse.de>
4694 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4696 2018-01-08 Richard Biener <rguenther@suse.de>
4698 PR tree-optimization/83685
4699 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4700 references to abnormals.
4702 2018-01-08 Richard Biener <rguenther@suse.de>
4705 * dwarf2out.c (output_indirect_strings): Handle empty
4706 skeleton_debug_str_hash.
4707 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4709 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4711 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4712 (emit_store_direct): Likewise.
4713 (arc_trampoline_adjust_address): Likewise.
4714 (arc_asm_trampoline_template): New function.
4715 (arc_initialize_trampoline): Use asm_trampoline_template.
4716 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4717 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4718 * config/arc/arc.md (flush_icache): Delete pattern.
4720 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4722 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4723 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4726 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4729 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4730 by not USED_FOR_TARGET.
4731 (make_pass_resolve_sw_modes): Likewise.
4733 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4735 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4738 2018-01-08 Richard Biener <rguenther@suse.de>
4741 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4743 2018-01-08 Richard Biener <rguenther@suse.de>
4746 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4748 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4751 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4752 basic blocks with a small number of successors.
4753 (convert_control_dep_chain_into_preds): Improve handling of
4755 (dump_predicates): Split apart into...
4756 (dump_pred_chain): ...here...
4757 (dump_pred_info): ...and here.
4758 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4759 (can_chain_union_be_invalidated_p): Improve check for invalidation
4761 (uninit_uses_cannot_happen): Avoid unnecessary if
4762 convert_control_dep_chain_into_preds yielded nothing.
4764 2018-01-06 Martin Sebor <msebor@redhat.com>
4766 PR tree-optimization/83640
4767 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4768 subtracting negative offset from size.
4769 (builtin_access::overlap): Adjust offset bounds of the access to fall
4770 within the size of the object if possible.
4772 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4774 PR rtl-optimization/83699
4775 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4776 extract_bit_field_as_subreg to cases in which the extracted
4777 value is also a vector.
4779 * lra-constraints.c (process_alt_operands): Test for the equivalence
4780 substitutions when detecting a possible reload cycle.
4782 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4785 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4786 by default if flag_selective_schedling{,2}. Formatting fixes.
4788 PR rtl-optimization/83682
4789 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4790 if it has non-VECTOR_MODE element mode.
4791 (vec_duplicate_p): Likewise.
4794 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4795 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4797 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4800 * config/i386/i386-builtin.def
4801 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4802 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4803 Require also OPTION_MASK_ISA_AVX512F in addition to
4804 OPTION_MASK_ISA_GFNI.
4805 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4806 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4807 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4808 to OPTION_MASK_ISA_GFNI.
4809 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4810 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4811 OPTION_MASK_ISA_AVX512BW.
4812 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4813 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4814 addition to OPTION_MASK_ISA_GFNI.
4815 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4816 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4817 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4818 to OPTION_MASK_ISA_GFNI.
4819 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4820 a requirement for all ISAs rather than any of them with a few
4822 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4824 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4825 bitmasks to be enabled with 3 exceptions, instead of requiring any
4826 enabled ISA with lots of exceptions.
4827 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4828 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4829 Change avx512bw in isa attribute to avx512f.
4830 * config/i386/sgxintrin.h: Add license boilerplate.
4831 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4832 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4833 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4834 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4836 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4837 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4838 temporarily sse2 rather than sse if not enabled already.
4841 * config/i386/sse.md (VI248_VLBW): Rename to ...
4842 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4843 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4844 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4845 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4846 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4847 mode iterator instead of VI248_VLBW.
4849 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4851 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4852 (record_modified): Skip clobbers; add debug output.
4853 (param_change_prob): Use sreal frequencies.
4855 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4857 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4858 punt for user-aligned variables.
4860 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4862 * tree-chrec.c (chrec_contains_symbols): Return true for
4865 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4868 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4869 of (x|y) == x for BICS pattern.
4871 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4873 PR tree-optimization/83605
4874 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4875 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4878 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4880 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4881 * config/epiphany/rtems.h: New file.
4883 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4884 Uros Bizjak <ubizjak@gmail.com>
4887 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4888 QIreg_operand instead of register_operand predicate.
4889 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4890 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4891 comments instead of -fmitigate[-_]rop.
4893 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4896 * cgraphunit.c (symbol_table::compile): Switch to text_section
4897 before calling assembly_start debug hook.
4898 * run-rtl-passes.c (run_rtl_passes): Likewise.
4901 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4903 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4904 range_int_cst_p rather than !symbolic_range_p before calling
4905 extract_range_from_multiplicative_op_1.
4907 2018-01-04 Jeff Law <law@redhat.com>
4909 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4910 redundant test in assertion.
4912 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4914 * doc/rtl.texi: Document machine_mode wrapper classes.
4916 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4918 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4921 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4923 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4924 the VEC_PERM_EXPR fold to fail.
4926 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4929 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4930 to switched_sections.
4932 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4935 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4938 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4941 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4942 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4944 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4947 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4948 is BLKmode and bitpos not zero or mode change is needed.
4950 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4953 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4956 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4959 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4960 instead of MULT rtx. Update all corresponding splitters.
4962 (*ssub<modesuffix>): Ditto.
4964 (*cmp_sadd_di): Update split patterns.
4965 (*cmp_sadd_si): Ditto.
4966 (*cmp_sadd_sidi): Ditto.
4967 (*cmp_ssub_di): Ditto.
4968 (*cmp_ssub_si): Ditto.
4969 (*cmp_ssub_sidi): Ditto.
4970 * config/alpha/predicates.md (const23_operand): New predicate.
4971 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4972 Look for ASHIFT, not MULT inner operand.
4973 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4975 2018-01-04 Martin Liska <mliska@suse.cz>
4977 PR gcov-profile/83669
4978 * gcov.c (output_intermediate_file): Add version to intermediate
4980 * doc/gcov.texi: Document new field 'version' in intermediate
4981 file format. Fix location of '-k' option of gcov command.
4983 2018-01-04 Martin Liska <mliska@suse.cz>
4986 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4988 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4990 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4992 2018-01-03 Martin Sebor <msebor@redhat.com>
4994 PR tree-optimization/83655
4995 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4996 checking calls with invalid arguments.
4998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5000 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5001 (vectorizable_mask_load_store): Delete.
5002 (vectorizable_call): Return false for masked loads and stores.
5003 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5004 instead of gimple_assign_rhs1.
5005 (vectorizable_load): Handle IFN_MASK_LOAD.
5006 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5010 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5012 (vectorizable_mask_load_store): ...here.
5013 (vectorizable_load): ...and here.
5015 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5017 * tree-vect-stmts.c (vect_build_all_ones_mask)
5018 (vect_build_zero_merge_argument): New functions, split out from...
5019 (vectorizable_load): ...here.
5021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5023 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5025 (vectorizable_mask_load_store): ...here.
5026 (vectorizable_store): ...and here.
5028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5030 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5032 (vectorizable_mask_load_store): ...here.
5034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5036 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5037 (vect_model_store_cost): Take a vec_load_store_type instead of a
5039 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5040 (vect_model_store_cost): Take a vec_load_store_type instead of a
5042 (vectorizable_mask_load_store): Update accordingly.
5043 (vectorizable_store): Likewise.
5044 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5046 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5048 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5049 IFN_MASK_LOAD calls here rather than...
5050 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5053 Alan Hayward <alan.hayward@arm.com>
5054 David Sherwood <david.sherwood@arm.com>
5056 * expmed.c (extract_bit_field_1): For vector extracts,
5057 fall back to extract_bit_field_as_subreg if vec_extract
5060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5061 Alan Hayward <alan.hayward@arm.com>
5062 David Sherwood <david.sherwood@arm.com>
5064 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5065 they are variable or constant sized.
5066 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5067 slots for constant-sized data.
5069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5070 Alan Hayward <alan.hayward@arm.com>
5071 David Sherwood <david.sherwood@arm.com>
5073 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5074 handling COND_EXPRs with boolean comparisons, try to find a better
5075 basis for the mask type than the boolean itself.
5077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5079 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5080 is calculated and how it can be overridden.
5081 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5082 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5084 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5087 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5088 Alan Hayward <alan.hayward@arm.com>
5089 David Sherwood <david.sherwood@arm.com>
5091 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5092 Remove the mode argument.
5093 (aarch64_simd_valid_immediate): Remove the mode and inverse
5095 * config/aarch64/iterators.md (bitsize): New iterator.
5096 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5097 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5098 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5099 aarch64_simd_valid_immediate.
5100 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5101 (aarch64_reg_or_bic_imm): Likewise.
5102 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5103 with an insn_type enum and msl with a modifier_type enum.
5104 Replace element_width with a scalar_mode. Change the shift
5105 to unsigned int. Add constructors for scalar_float_mode and
5106 scalar_int_mode elements.
5107 (aarch64_vect_float_const_representable_p): Delete.
5108 (aarch64_can_const_movi_rtx_p)
5109 (aarch64_simd_scalar_immediate_valid_for_move)
5110 (aarch64_simd_make_constant): Update call to
5111 aarch64_simd_valid_immediate.
5112 (aarch64_advsimd_valid_immediate_hs): New function.
5113 (aarch64_advsimd_valid_immediate): Likewise.
5114 (aarch64_simd_valid_immediate): Remove mode and inverse
5115 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5116 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5117 and aarch64_float_const_representable_p on the result.
5118 (aarch64_output_simd_mov_immediate): Remove mode argument.
5119 Update call to aarch64_simd_valid_immediate and use of
5120 simd_immediate_info.
5121 (aarch64_output_scalar_simd_mov_immediate): Update call
5124 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5125 Alan Hayward <alan.hayward@arm.com>
5126 David Sherwood <david.sherwood@arm.com>
5128 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5129 (mode_nunits): Likewise CONST_MODE_NUNITS.
5130 * machmode.def (ADJUST_NUNITS): Document.
5131 * genmodes.c (mode_data::need_nunits_adj): New field.
5132 (blank_mode): Update accordingly.
5133 (adj_nunits): New variable.
5134 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5136 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5137 listed in adj_nunits.
5138 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5139 listed in adj_nunits. Don't emit case statements for such modes.
5140 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5141 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5142 nothing if adj_nunits is nonnull.
5143 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5144 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5145 (emit_mode_fbit): Update use of print_maybe_const_decl.
5146 (emit_move_size): Likewise. Treat the array as non-const
5148 (emit_mode_adjustments): Handle adj_nunits.
5150 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5152 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5153 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5154 (VECTOR_MODES): Use it.
5155 (make_vector_modes): Take the prefix as an argument.
5157 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5158 Alan Hayward <alan.hayward@arm.com>
5159 David Sherwood <david.sherwood@arm.com>
5161 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5162 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5163 for MODE_VECTOR_BOOL.
5164 * machmode.def (VECTOR_BOOL_MODE): Document.
5165 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5166 (make_vector_bool_mode): New function.
5167 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5169 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5170 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5172 * stor-layout.c (int_mode_for_mode): Likewise.
5173 * tree.c (build_vector_type_for_mode): Likewise.
5174 * varasm.c (output_constant_pool_2): Likewise.
5175 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5176 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5177 for MODE_VECTOR_BOOL.
5178 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5179 of mode class checks.
5180 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5181 instead of a list of mode class checks.
5182 (expand_vector_scalar_condition): Likewise.
5183 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5186 Alan Hayward <alan.hayward@arm.com>
5187 David Sherwood <david.sherwood@arm.com>
5189 * machmode.h (mode_size): Change from unsigned short to
5191 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5192 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5193 or if measurement_type is not polynomial.
5194 (fixed_size_mode::includes_p): Check for constant-sized modes.
5195 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5196 return a poly_uint16 rather than an unsigned short.
5197 (emit_mode_size): Change the type of mode_size from unsigned short
5198 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5199 (emit_mode_adjustments): Cope with polynomial vector sizes.
5200 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5202 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5204 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5205 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5206 * caller-save.c (setup_save_areas): Likewise.
5207 (replace_reg_with_saved_mem): Likewise.
5208 * calls.c (emit_library_call_value_1): Likewise.
5209 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5210 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5211 (gen_lowpart_for_combine): Likewise.
5212 * convert.c (convert_to_integer_1): Likewise.
5213 * cse.c (equiv_constant, cse_insn): Likewise.
5214 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5215 (cselib_subst_to_values): Likewise.
5216 * dce.c (word_dce_process_block): Likewise.
5217 * df-problems.c (df_word_lr_mark_ref): Likewise.
5218 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5219 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5220 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5221 (rtl_for_decl_location): Likewise.
5222 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5223 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5224 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5225 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5226 (expand_expr_real_1): Likewise.
5227 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5228 (pad_below): Likewise.
5229 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5230 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5231 * ira.c (get_subreg_tracking_sizes): Likewise.
5232 * ira-build.c (ira_create_allocno_objects): Likewise.
5233 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5234 (ira_sort_regnos_for_alter_reg): Likewise.
5235 * ira-costs.c (record_operand_costs): Likewise.
5236 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5237 (resolve_simple_move): Likewise.
5238 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5239 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5240 (lra_constraints): Likewise.
5241 (CONST_POOL_OK_P): Reject variable-sized modes.
5242 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5243 (add_pseudo_to_slot, lra_spill): Likewise.
5244 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5245 * optabs-query.c (get_best_extraction_insn): Likewise.
5246 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5247 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5248 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5249 * recog.c (offsettable_address_addr_space_p): Likewise.
5250 * regcprop.c (maybe_mode_change): Likewise.
5251 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5252 * regrename.c (build_def_use): Likewise.
5253 * regstat.c (dump_reg_info): Likewise.
5254 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5255 (find_reloads, find_reloads_subreg_address): Likewise.
5256 * reload1.c (eliminate_regs_1): Likewise.
5257 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5258 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5259 (simplify_binary_operation_1, simplify_subreg): Likewise.
5260 * targhooks.c (default_function_arg_padding): Likewise.
5261 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5262 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5263 (verify_gimple_assign_ternary): Likewise.
5264 * tree-inline.c (estimate_move_cost): Likewise.
5265 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5266 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5267 (get_address_cost_ainc): Likewise.
5268 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5269 (vect_supportable_dr_alignment): Likewise.
5270 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5271 (vectorizable_reduction): Likewise.
5272 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5273 (vectorizable_operation, vectorizable_load): Likewise.
5274 * tree.c (build_same_sized_truth_vector_type): Likewise.
5275 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5276 * var-tracking.c (emit_note_insn_var_location): Likewise.
5277 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5278 (ADDR_VEC_ALIGN): Likewise.
5280 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5281 Alan Hayward <alan.hayward@arm.com>
5282 David Sherwood <david.sherwood@arm.com>
5284 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5286 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5287 or if measurement_type is polynomial.
5288 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5289 * combine.c (make_extraction): Likewise.
5290 * dse.c (find_shift_sequence): Likewise.
5291 * dwarf2out.c (mem_loc_descriptor): Likewise.
5292 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5293 (extract_bit_field, extract_low_bits): Likewise.
5294 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5295 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5296 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5297 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5298 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5299 * reload.c (find_reloads): Likewise.
5300 * reload1.c (alter_reg): Likewise.
5301 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5302 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5303 * tree-if-conv.c (predicate_mem_writes): Likewise.
5304 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5305 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5306 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5307 * valtrack.c (dead_debug_insert_temp): Likewise.
5308 * varasm.c (mergeable_constant_section): Likewise.
5309 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5311 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5312 Alan Hayward <alan.hayward@arm.com>
5313 David Sherwood <david.sherwood@arm.com>
5315 * expr.c (expand_assignment): Cope with polynomial mode sizes
5316 when assigning to a CONCAT.
5318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5319 Alan Hayward <alan.hayward@arm.com>
5320 David Sherwood <david.sherwood@arm.com>
5322 * machmode.h (mode_precision): Change from unsigned short to
5324 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5326 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5327 or if measurement_type is not polynomial.
5328 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5329 in which the mode is already known to be a scalar_int_mode.
5330 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5331 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5333 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5334 for GET_MODE_PRECISION.
5335 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5336 for GET_MODE_PRECISION.
5337 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5339 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5340 (expand_field_assignment, make_extraction): Likewise.
5341 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5342 (get_last_value): Likewise.
5343 * convert.c (convert_to_integer_1): Likewise.
5344 * cse.c (cse_insn): Likewise.
5345 * expr.c (expand_expr_real_1): Likewise.
5346 * lra-constraints.c (simplify_operand_subreg): Likewise.
5347 * optabs-query.c (can_atomic_load_p): Likewise.
5348 * optabs.c (expand_atomic_load): Likewise.
5349 (expand_atomic_store): Likewise.
5350 * ree.c (combine_reaching_defs): Likewise.
5351 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5352 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5353 * tree.h (type_has_mode_precision_p): Likewise.
5354 * ubsan.c (instrument_si_overflow): Likewise.
5356 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5357 Alan Hayward <alan.hayward@arm.com>
5358 David Sherwood <david.sherwood@arm.com>
5360 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5361 polynomial numbers of units.
5362 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5363 (valid_vector_subparts_p): New function.
5364 (build_vector_type): Remove temporary shim and take the number
5365 of units as a poly_uint64 rather than an int.
5366 (build_opaque_vector_type): Take the number of units as a
5367 poly_uint64 rather than an int.
5368 * tree.c (build_vector_from_ctor): Handle polynomial
5369 TYPE_VECTOR_SUBPARTS.
5370 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5371 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5372 (build_vector_from_val): If the number of units is variable,
5373 use build_vec_duplicate_cst for constant operands and
5374 VEC_DUPLICATE_EXPR otherwise.
5375 (make_vector_type): Remove temporary is_constant ().
5376 (build_vector_type, build_opaque_vector_type): Take the number of
5377 units as a poly_uint64 rather than an int.
5378 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5380 * cfgexpand.c (expand_debug_expr): Likewise.
5381 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5382 (store_constructor, expand_expr_real_1): Likewise.
5383 (const_scalar_mask_from_tree): Likewise.
5384 * fold-const-call.c (fold_const_reduction): Likewise.
5385 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5386 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5387 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5388 (fold_relational_const): Likewise.
5389 (native_interpret_vector): Likewise. Change the size from an
5390 int to an unsigned int.
5391 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5392 TYPE_VECTOR_SUBPARTS.
5393 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5394 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5395 duplicating a non-constant operand into a variable-length vector.
5396 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5397 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5398 * ipa-icf.c (sem_variable::equals): Likewise.
5399 * match.pd: Likewise.
5400 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5401 * print-tree.c (print_node): Likewise.
5402 * stor-layout.c (layout_type): Likewise.
5403 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5404 * tree-cfg.c (verify_gimple_comparison): Likewise.
5405 (verify_gimple_assign_binary): Likewise.
5406 (verify_gimple_assign_ternary): Likewise.
5407 (verify_gimple_assign_single): Likewise.
5408 * tree-pretty-print.c (dump_generic_node): Likewise.
5409 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5410 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5411 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5412 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5413 (vect_shift_permute_load_chain): Likewise.
5414 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5415 (expand_vector_condition, optimize_vector_constructor): Likewise.
5416 (lower_vec_perm, get_compute_type): Likewise.
5417 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5418 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5419 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5420 (vect_recog_mask_conversion_pattern): Likewise.
5421 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5422 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5423 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5424 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5425 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5426 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5427 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5428 (supportable_widening_operation): Likewise.
5429 (supportable_narrowing_operation): Likewise.
5430 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5432 * varasm.c (output_constant): Likewise.
5434 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5435 Alan Hayward <alan.hayward@arm.com>
5436 David Sherwood <david.sherwood@arm.com>
5438 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5439 so that both the length == 3 and length != 3 cases set up their
5440 own permute vectors. Add comments explaining why we know the
5441 number of elements is constant.
5442 (vect_permute_load_chain): Likewise.
5444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5445 Alan Hayward <alan.hayward@arm.com>
5446 David Sherwood <david.sherwood@arm.com>
5448 * machmode.h (mode_nunits): Change from unsigned char to
5450 (ONLY_FIXED_SIZE_MODES): New macro.
5451 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5452 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5453 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5455 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5456 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5457 or if measurement_type is not polynomial.
5458 * genmodes.c (ZERO_COEFFS): New macro.
5459 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5461 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5462 Use ZERO_COEFFS when emitting initializers.
5463 * data-streamer.h (bp_pack_poly_value): New function.
5464 (bp_unpack_poly_value): Likewise.
5465 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5466 for GET_MODE_NUNITS.
5467 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5468 for GET_MODE_NUNITS.
5469 * tree.c (make_vector_type): Remove temporary shim and make
5470 the real function take the number of units as a poly_uint64
5472 (build_vector_type_for_mode): Handle polynomial nunits.
5473 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5474 * emit-rtl.c (const_vec_series_p_1): Likewise.
5475 (gen_rtx_CONST_VECTOR): Likewise.
5476 * fold-const.c (test_vec_duplicate_folding): Likewise.
5477 * genrecog.c (validate_pattern): Likewise.
5478 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5479 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5480 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5481 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5482 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5483 * rtlanal.c (subreg_get_info): Likewise.
5484 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5485 (vect_grouped_load_supported): Likewise.
5486 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5487 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5488 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5489 (simplify_const_unary_operation, simplify_binary_operation_1)
5490 (simplify_const_binary_operation, simplify_ternary_operation)
5491 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5492 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5493 instead of CONST_VECTOR_NUNITS.
5494 * varasm.c (output_constant_pool_2): Likewise.
5495 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5496 explicit-encoded elements in the XVEC for variable-length vectors.
5498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5500 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5503 Alan Hayward <alan.hayward@arm.com>
5504 David Sherwood <david.sherwood@arm.com>
5506 * coretypes.h (fixed_size_mode): Declare.
5507 (fixed_size_mode_pod): New typedef.
5508 * builtins.h (target_builtins::x_apply_args_mode)
5509 (target_builtins::x_apply_result_mode): Change type to
5510 fixed_size_mode_pod.
5511 * builtins.c (apply_args_size, apply_result_size, result_vector)
5512 (expand_builtin_apply_args_1, expand_builtin_apply)
5513 (expand_builtin_return): Update accordingly.
5515 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5517 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5518 * cselib.c (cselib_hash_rtx): Likewise.
5519 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5520 CONST_VECTOR encoding.
5522 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5523 Jeff Law <law@redhat.com>
5526 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5527 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5528 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5529 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5532 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5533 explicitly probe *sp in a noreturn function if there were any callee
5534 register saves or frame pointer is needed.
5536 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5539 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5540 BLKmode for ternary, binary or unary expressions.
5543 * var-tracking.c (delete_vta_debug_insn): New inline function.
5544 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5545 insns from get_insns () to NULL instead of each bb separately.
5546 Use delete_vta_debug_insn. No longer static.
5547 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5548 delete_vta_debug_insns callers.
5549 * rtl.h (delete_vta_debug_insns): Declare.
5550 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5551 instead of variable_tracking_main.
5553 2018-01-03 Martin Sebor <msebor@redhat.com>
5555 PR tree-optimization/83603
5556 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5557 arguments past the endof the argument list in functions declared
5558 without a prototype.
5559 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5560 Avoid checking when arguments are null.
5562 2018-01-03 Martin Sebor <msebor@redhat.com>
5565 * doc/extend.texi (attribute const): Fix a typo.
5566 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5567 issuing -Wsuggest-attribute for void functions.
5569 2018-01-03 Martin Sebor <msebor@redhat.com>
5571 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5572 offset_int::from instead of wide_int::to_shwi.
5573 (maybe_diag_overlap): Remove assertion.
5574 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5575 * gimple-ssa-sprintf.c (format_directive): Same.
5576 (parse_directive): Same.
5577 (sprintf_dom_walker::compute_format_length): Same.
5578 (try_substitute_return_value): Same.
5580 2018-01-03 Jeff Law <law@redhat.com>
5583 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5584 non-constant residual for zero at runtime and avoid probing in
5585 that case. Reorganize code for trailing problem to mirror handling
5588 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5590 PR tree-optimization/83501
5591 * tree-ssa-strlen.c (get_string_cst): New.
5592 (handle_char_store): Call get_string_cst.
5594 2018-01-03 Martin Liska <mliska@suse.cz>
5596 PR tree-optimization/83593
5597 * tree-ssa-strlen.c: Include tree-cfg.h.
5598 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5599 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5600 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5602 (strlen_dom_walker::before_dom_children): Call
5603 gimple_purge_dead_eh_edges. Dump tranformation with details
5605 (strlen_dom_walker::before_dom_children): Update call by adding
5606 new argument cleanup_eh.
5607 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5609 2018-01-03 Martin Liska <mliska@suse.cz>
5612 * cif-code.def (VARIADIC_THUNK): New enum value.
5613 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5616 2018-01-03 Jan Beulich <jbeulich@suse.com>
5618 * sse.md (mov<mode>_internal): Tighten condition for when to use
5619 vmovdqu<ssescalarsize> for TI and OI modes.
5621 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5623 Update copyright years.
5625 2018-01-03 Martin Liska <mliska@suse.cz>
5628 * ipa-visibility.c (function_and_variable_visibility): Skip
5629 functions with noipa attribure.
5631 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5633 * gcc.c (process_command): Update copyright notice dates.
5634 * gcov-dump.c (print_version): Ditto.
5635 * gcov.c (print_version): Ditto.
5636 * gcov-tool.c (print_version): Ditto.
5637 * gengtype.c (create_file): Ditto.
5638 * doc/cpp.texi: Bump @copying's copyright year.
5639 * doc/cppinternals.texi: Ditto.
5640 * doc/gcc.texi: Ditto.
5641 * doc/gccint.texi: Ditto.
5642 * doc/gcov.texi: Ditto.
5643 * doc/install.texi: Ditto.
5644 * doc/invoke.texi: Ditto.
5646 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5648 * vector-builder.h (vector_builder::m_full_nelts): Change from
5649 unsigned int to poly_uint64.
5650 (vector_builder::full_nelts): Update prototype accordingly.
5651 (vector_builder::new_vector): Likewise.
5652 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5653 (vector_builder::operator ==): Likewise.
5654 (vector_builder::finalize): Likewise.
5655 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5656 Take the number of elements as a poly_uint64 rather than an
5658 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5659 from unsigned int to poly_uint64.
5660 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5661 (vec_perm_indices::new_vector): Likewise.
5662 (vec_perm_indices::length): Likewise.
5663 (vec_perm_indices::nelts_per_input): Likewise.
5664 (vec_perm_indices::input_nelts): Likewise.
5665 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5666 number of elements per input as a poly_uint64 rather than an
5667 unsigned int. Use the original encoding for variable-length
5668 vectors, rather than clamping each individual element.
5669 For the second and subsequent elements in each pattern,
5670 clamp the step and base before clamping their sum.
5671 (vec_perm_indices::series_p): Handle polynomial element counts.
5672 (vec_perm_indices::all_in_range_p): Likewise.
5673 (vec_perm_indices_to_tree): Likewise.
5674 (vec_perm_indices_to_rtx): Likewise.
5675 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5676 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5677 (tree_vector_builder::new_binary_operation): Handle polynomial
5678 element counts. Return false if we need to know the number
5679 of elements at compile time.
5680 * fold-const.c (fold_vec_perm): Punt if the number of elements
5681 isn't known at compile time.
5683 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5685 * vec-perm-indices.h (vec_perm_builder): Change element type
5686 from HOST_WIDE_INT to poly_int64.
5687 (vec_perm_indices::element_type): Update accordingly.
5688 (vec_perm_indices::clamp): Handle polynomial element_types.
5689 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5690 (vec_perm_indices::all_in_range_p): Likewise.
5691 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5693 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5694 polynomial vec_perm_indices element types.
5695 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5696 * fold-const.c (fold_vec_perm): Likewise.
5697 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5698 * tree-vect-generic.c (lower_vec_perm): Likewise.
5699 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5700 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5701 element type to HOST_WIDE_INT.
5703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5704 Alan Hayward <alan.hayward@arm.com>
5705 David Sherwood <david.sherwood@arm.com>
5707 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5708 rather than an int. Use plus_constant.
5709 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5710 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5713 Alan Hayward <alan.hayward@arm.com>
5714 David Sherwood <david.sherwood@arm.com>
5716 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5717 a HOST_WIDE_INT to a poly_int64.
5719 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5720 Alan Hayward <alan.hayward@arm.com>
5721 David Sherwood <david.sherwood@arm.com>
5723 * calls.c (load_register_parameters): Cope with polynomial
5724 mode sizes. Require a constant size for BLKmode parameters
5725 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5726 forces a parameter to be padded at the lsb end in order to
5727 fill a complete number of words, require the parameter size
5728 to be ordered wrt UNITS_PER_WORD.
5730 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5731 Alan Hayward <alan.hayward@arm.com>
5732 David Sherwood <david.sherwood@arm.com>
5734 * reload1.c (spill_stack_slot_width): Change element type
5735 from unsigned int to poly_uint64_pod.
5736 (alter_reg): Treat mode sizes as polynomial.
5738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5739 Alan Hayward <alan.hayward@arm.com>
5740 David Sherwood <david.sherwood@arm.com>
5742 * reload.c (complex_word_subreg_p): New function.
5743 (reload_inner_reg_of_subreg, push_reload): Use it.
5745 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5746 Alan Hayward <alan.hayward@arm.com>
5747 David Sherwood <david.sherwood@arm.com>
5749 * lra-constraints.c (process_alt_operands): Reject matched
5750 operands whose sizes aren't ordered.
5751 (match_reload): Refer to this check here.
5753 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5754 Alan Hayward <alan.hayward@arm.com>
5755 David Sherwood <david.sherwood@arm.com>
5757 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5758 that the mode size is in the set {1, 2, 4, 8, 16}.
5760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5761 Alan Hayward <alan.hayward@arm.com>
5762 David Sherwood <david.sherwood@arm.com>
5764 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5765 Use plus_constant instead of gen_rtx_PLUS.
5767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5768 Alan Hayward <alan.hayward@arm.com>
5769 David Sherwood <david.sherwood@arm.com>
5771 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5772 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5773 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5774 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5775 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5776 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5777 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5778 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5779 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5780 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5782 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5783 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5784 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5785 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5786 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5787 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5788 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5789 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5790 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5791 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5793 * expr.c (emit_move_resolve_push): Treat the input and result
5794 of PUSH_ROUNDING as a poly_int64.
5795 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5796 (emit_push_insn): Likewise.
5797 * lra-eliminations.c (mark_not_eliminable): Likewise.
5798 * recog.c (push_operand): Likewise.
5799 * reload1.c (elimination_effects): Likewise.
5800 * rtlanal.c (nonzero_bits1): Likewise.
5801 * calls.c (store_one_arg): Likewise. Require the padding to be
5802 known at compile time.
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 * expr.c (emit_single_push_insn_1): 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 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5818 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5819 Alan Hayward <alan.hayward@arm.com>
5820 David Sherwood <david.sherwood@arm.com>
5822 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5823 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5824 via stack temporaries. Treat the mode size as polynomial too.
5826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5827 Alan Hayward <alan.hayward@arm.com>
5828 David Sherwood <david.sherwood@arm.com>
5830 * expr.c (expand_expr_real_2): When handling conversions involving
5831 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5832 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5833 as a poly_uint64 too.
5835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5836 Alan Hayward <alan.hayward@arm.com>
5837 David Sherwood <david.sherwood@arm.com>
5839 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5842 Alan Hayward <alan.hayward@arm.com>
5843 David Sherwood <david.sherwood@arm.com>
5845 * combine.c (can_change_dest_mode): Handle polynomial
5846 REGMODE_NATURAL_SIZE.
5847 * expmed.c (store_bit_field_1): Likewise.
5848 * expr.c (store_constructor): Likewise.
5849 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5850 and polynomial REGMODE_NATURAL_SIZE.
5851 (gen_lowpart_common): Likewise.
5852 * reginfo.c (record_subregs_of_mode): Likewise.
5853 * rtlanal.c (read_modify_subreg_p): Likewise.
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 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5860 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
5868 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5869 Alan Hayward <alan.hayward@arm.com>
5870 David Sherwood <david.sherwood@arm.com>
5872 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5873 in a POINTER_PLUS_EXPR.
5875 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5876 Alan Hayward <alan.hayward@arm.com>
5877 David Sherwood <david.sherwood@arm.com>
5879 * omp-simd-clone.c (simd_clone_subparts): New function.
5880 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5881 (ipa_simd_modify_function_body): Likewise.
5883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5884 Alan Hayward <alan.hayward@arm.com>
5885 David Sherwood <david.sherwood@arm.com>
5887 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5888 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5889 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5890 (expand_vector_condition, vector_element): Likewise.
5891 (subparts_gt): New function.
5892 (get_compute_type): Use subparts_gt.
5893 (count_type_subparts): Delete.
5894 (expand_vector_operations_1): Use subparts_gt instead of
5895 count_type_subparts.
5897 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5898 Alan Hayward <alan.hayward@arm.com>
5899 David Sherwood <david.sherwood@arm.com>
5901 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5902 (vect_compile_time_alias): ...this new function. Do the calculation
5903 on poly_ints rather than trees.
5904 (vect_prune_runtime_alias_test_list): Update call accordingly.
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 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5912 (vect_schedule_slp_instance): Likewise.
5914 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5915 Alan Hayward <alan.hayward@arm.com>
5916 David Sherwood <david.sherwood@arm.com>
5918 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5919 constant and extern definitions for variable-length vectors.
5920 (vect_get_constant_vectors): Note that the number of units
5921 is known to be constant.
5923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5924 Alan Hayward <alan.hayward@arm.com>
5925 David Sherwood <david.sherwood@arm.com>
5927 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5928 of units as polynomial. Choose between WIDE and NARROW based
5931 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5932 Alan Hayward <alan.hayward@arm.com>
5933 David Sherwood <david.sherwood@arm.com>
5935 * tree-vect-stmts.c (simd_clone_subparts): New function.
5936 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5938 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5939 Alan Hayward <alan.hayward@arm.com>
5940 David Sherwood <david.sherwood@arm.com>
5942 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5943 vectors as polynomial. Use build_index_vector for
5946 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5947 Alan Hayward <alan.hayward@arm.com>
5948 David Sherwood <david.sherwood@arm.com>
5950 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5951 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5952 for variable-length vectors.
5953 (vectorizable_mask_load_store): Treat the number of units as
5954 polynomial, asserting that it is constant if the condition has
5955 already been enforced.
5956 (vectorizable_store, vectorizable_load): 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-loop.c (vectorizable_live_operation): Treat the number
5963 of units as polynomial. Punt if we can't tell at compile time
5964 which vector contains the final result.
5966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5967 Alan Hayward <alan.hayward@arm.com>
5968 David Sherwood <david.sherwood@arm.com>
5970 * tree-vect-loop.c (vectorizable_induction): Treat the number
5971 of units as polynomial. Punt on SLP inductions. Use an integer
5972 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5973 cast of such a series for variable-length floating-point
5976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5977 Alan Hayward <alan.hayward@arm.com>
5978 David Sherwood <david.sherwood@arm.com>
5980 * tree.h (build_index_vector): Declare.
5981 * tree.c (build_index_vector): New function.
5982 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5983 of units as polynomial, forcibly converting it to a constant if
5984 vectorizable_reduction has already enforced the condition.
5985 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5986 to create a {1,2,3,...} vector.
5987 (vectorizable_reduction): Treat the number of units as polynomial.
5988 Choose vectype_in based on the largest scalar element size rather
5989 than the smallest number of units. Enforce the restrictions
5992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5993 Alan Hayward <alan.hayward@arm.com>
5994 David Sherwood <david.sherwood@arm.com>
5996 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5997 number of units as polynomial.
5999 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6000 Alan Hayward <alan.hayward@arm.com>
6001 David Sherwood <david.sherwood@arm.com>
6003 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6004 * target.def (autovectorize_vector_sizes): Return the vector sizes
6005 by pointer, using vector_sizes rather than a bitmask.
6006 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6007 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6008 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6010 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6011 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6012 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6013 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6014 * omp-general.c (omp_max_vf): Likewise.
6015 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6016 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6017 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6018 * tree-vect-slp.c (vect_slp_bb): Likewise.
6019 * doc/tm.texi: Regenerate.
6020 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6022 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6023 the vector size as a poly_uint64 rather than an unsigned int.
6024 (current_vector_size): Change from an unsigned int to a poly_uint64.
6025 (get_vectype_for_scalar_type): Update accordingly.
6026 * tree.h (build_truth_vector_type): Take the size and number of
6027 units as a poly_uint64 rather than an unsigned int.
6028 (build_vector_type): Add a temporary overload that takes
6029 the number of units as a poly_uint64 rather than an unsigned int.
6030 * tree.c (make_vector_type): Likewise.
6031 (build_truth_vector_type): Take the number of units as a poly_uint64
6032 rather than an unsigned int.
6034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6035 Alan Hayward <alan.hayward@arm.com>
6036 David Sherwood <david.sherwood@arm.com>
6038 * target.def (get_mask_mode): Take the number of units and length
6039 as poly_uint64s rather than unsigned ints.
6040 * targhooks.h (default_get_mask_mode): Update accordingly.
6041 * targhooks.c (default_get_mask_mode): Likewise.
6042 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6043 * doc/tm.texi: Regenerate.
6045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6046 Alan Hayward <alan.hayward@arm.com>
6047 David Sherwood <david.sherwood@arm.com>
6049 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6050 * omp-general.c (omp_max_vf): Likewise.
6051 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6052 (expand_omp_simd): Handle polynomial safelen.
6053 * omp-low.c (omplow_simd_context): Add a default constructor.
6054 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6055 (lower_rec_simd_input_clauses): Update accordingly.
6056 (lower_rec_input_clauses): Likewise.
6058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6059 Alan Hayward <alan.hayward@arm.com>
6060 David Sherwood <david.sherwood@arm.com>
6062 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6063 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6064 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6065 (vect_analyze_slp_cost): Likewise.
6066 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6067 (vect_model_load_cost): Likewise.
6069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6070 Alan Hayward <alan.hayward@arm.com>
6071 David Sherwood <david.sherwood@arm.com>
6073 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6074 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6075 from an unsigned int * to a poly_uint64_pod *.
6076 (calculate_unrolling_factor): New function.
6077 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6080 Alan Hayward <alan.hayward@arm.com>
6081 David Sherwood <david.sherwood@arm.com>
6083 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6084 from an unsigned int to a poly_uint64.
6085 (_loop_vec_info::slp_unrolling_factor): Likewise.
6086 (_loop_vec_info::vectorization_factor): Change from an int
6088 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6089 (vect_get_num_vectors): New function.
6090 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6091 (vect_get_num_copies): Use vect_get_num_vectors.
6092 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6093 to an unsigned int *.
6094 (vect_analyze_data_refs): Change min_vf from an int * to a
6096 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6097 than an unsigned HOST_WIDE_INT.
6098 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6099 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6100 to an unsigned int *.
6101 (vect_analyze_data_ref_dependences): Likewise.
6102 (vect_compute_data_ref_alignment): Handle polynomial vf.
6103 (vect_enhance_data_refs_alignment): Likewise.
6104 (vect_prune_runtime_alias_test_list): Likewise.
6105 (vect_shift_permute_load_chain): Likewise.
6106 (vect_supportable_dr_alignment): Likewise.
6107 (dependence_distance_ge_vf): Take the vectorization factor as a
6108 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6109 (vect_analyze_data_refs): Change min_vf from an int * to a
6111 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6112 vfm1 as a poly_uint64 rather than an int. Make the same change
6113 for the returned bound_scalar.
6114 (vect_gen_vector_loop_niters): Handle polynomial vf.
6115 (vect_do_peeling): Likewise. Update call to
6116 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6117 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6119 * tree-vect-loop.c (vect_determine_vectorization_factor)
6120 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6121 (vect_get_known_peeling_cost): Likewise.
6122 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6123 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6124 (vect_transform_loop): Likewise. Use the lowest possible VF when
6125 updating the upper bounds of the loop.
6126 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6128 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6129 polynomial unroll factors.
6130 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6131 (vect_make_slp_decision): Likewise.
6132 (vect_supported_load_permutation_p): Likewise, and polynomial
6134 (vect_analyze_slp_cost): Handle polynomial vf.
6135 (vect_slp_analyze_node_operations): Likewise.
6136 (vect_slp_analyze_bb_1): Likewise.
6137 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6138 than an unsigned HOST_WIDE_INT.
6139 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6140 (vectorizable_load): Handle polynomial vf.
6141 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6143 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6146 Alan Hayward <alan.hayward@arm.com>
6147 David Sherwood <david.sherwood@arm.com>
6149 * match.pd: Handle bit operations involving three constants
6150 and try to fold one pair.
6152 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6154 * tree-vect-loop-manip.c: Include gimple-fold.h.
6155 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6156 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6157 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6158 Add a path that uses a step of VF instead of 1, but disable it
6160 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6161 and niters_no_overflow parameters. Update calls to
6162 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6163 Create a new SSA name if the latter choses to use a ste other
6164 than zero, and return it via niters_vector_mult_vf_var.
6165 * tree-vect-loop.c (vect_transform_loop): Update calls to
6166 vect_do_peeling, vect_gen_vector_loop_niters and
6167 slpeel_make_loop_iterate_ntimes.
6168 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6169 (vect_gen_vector_loop_niters): Update declarations after above changes.
6171 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6173 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6174 128-bit round to integer instructions.
6175 (ceil<mode>2): Likewise.
6176 (btrunc<mode>2): Likewise.
6177 (round<mode>2): Likewise.
6179 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6181 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6182 unaligned VSX load/store on P8/P9.
6183 (expand_block_clear): Allow the use of unaligned VSX
6184 load/store on P8/P9.
6186 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6188 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6190 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6191 swap associated with both a load and a store.
6193 2018-01-02 Andrew Waterman <andrew@sifive.com>
6195 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6196 * config/riscv/riscv.md (clear_cache): Use it.
6198 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6200 * web.c: Remove out-of-date comment.
6202 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6204 * expr.c (fixup_args_size_notes): Check that any existing
6205 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6206 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6207 (emit_single_push_insn): ...here.
6209 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6211 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6212 (const_vector_encoded_nelts): New function.
6213 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6214 (const_vector_int_elt, const_vector_elt): Declare.
6215 * emit-rtl.c (const_vector_int_elt_1): New function.
6216 (const_vector_elt): Likewise.
6217 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6218 of CONST_VECTOR_ELT.
6220 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6222 * expr.c: Include rtx-vector-builder.h.
6223 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6224 directly on the tree encoding.
6225 (const_vector_from_tree): Likewise.
6226 * optabs.c: Include rtx-vector-builder.h.
6227 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6228 sequence of "u" values.
6229 * vec-perm-indices.c: Include rtx-vector-builder.h.
6230 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6231 directly on the vec_perm_indices encoding.
6233 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6235 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6236 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6237 * rtx-vector-builder.h: New file.
6238 * rtx-vector-builder.c: Likewise.
6239 * rtl.h (rtx_def::u2): Add a const_vector field.
6240 (CONST_VECTOR_NPATTERNS): New macro.
6241 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6242 (CONST_VECTOR_DUPLICATE_P): Likewise.
6243 (CONST_VECTOR_STEPPED_P): Likewise.
6244 (CONST_VECTOR_ENCODED_ELT): Likewise.
6245 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6246 (unwrap_const_vec_duplicate): Likewise.
6247 (const_vec_series_p): Check for a non-duplicated vector encoding.
6248 Say that the function only returns true for integer vectors.
6249 * emit-rtl.c: Include rtx-vector-builder.h.
6250 (gen_const_vec_duplicate_1): Delete.
6251 (gen_const_vector): Call gen_const_vec_duplicate instead of
6252 gen_const_vec_duplicate_1.
6253 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6254 (gen_const_vec_duplicate): Use rtx_vector_builder.
6255 (gen_const_vec_series): Likewise.
6256 (gen_rtx_CONST_VECTOR): Likewise.
6257 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6258 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6259 Build a new vector rather than modifying a CONST_VECTOR in-place.
6260 (handle_special_swappables): Update call accordingly.
6261 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6262 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6263 Build a new vector rather than modifying a CONST_VECTOR in-place.
6264 (handle_special_swappables): Update call accordingly.
6266 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6268 * simplify-rtx.c (simplify_const_binary_operation): Use
6269 CONST_VECTOR_ELT instead of XVECEXP.
6271 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6273 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6274 the selector elements to be different from the data elements
6275 if the selector is a VECTOR_CST.
6276 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6277 ssizetype for the selector.
6279 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6281 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6282 before testing each element individually.
6283 * tree-vect-generic.c (lower_vec_perm): Likewise.
6285 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6287 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6288 * selftest-run-tests.c (selftest::run_tests): Call it.
6289 * vector-builder.h (vector_builder::operator ==): New function.
6290 (vector_builder::operator !=): Likewise.
6291 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6292 (vec_perm_indices::all_from_input_p): New function.
6293 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6294 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6295 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6296 instead of reading the VECTOR_CST directly. Detect whether both
6297 vector inputs are the same before constructing the vec_perm_indices,
6298 and update the number of inputs argument accordingly. Use the
6299 utility functions added above. Only construct sel2 if we need to.
6301 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6303 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6304 the broadcast of the low byte.
6305 (expand_mult_highpart): Use an explicit encoding for the permutes.
6306 * optabs-query.c (can_mult_highpart_p): Likewise.
6307 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6308 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6309 (vectorizable_bswap): Likewise.
6310 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6311 explicit encoding for the power-of-2 permutes.
6312 (vect_permute_store_chain): Likewise.
6313 (vect_grouped_load_supported): Likewise.
6314 (vect_permute_load_chain): Likewise.
6316 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6318 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6319 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6320 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6321 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6322 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6323 (vect_gen_perm_mask_any): Likewise.
6325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6327 * int-vector-builder.h: New file.
6328 * vec-perm-indices.h: Include int-vector-builder.h.
6329 (vec_perm_indices): Redefine as an int_vector_builder.
6330 (auto_vec_perm_indices): Delete.
6331 (vec_perm_builder): Redefine as a stand-alone class.
6332 (vec_perm_indices::vec_perm_indices): New function.
6333 (vec_perm_indices::clamp): Likewise.
6334 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6335 (vec_perm_indices::new_vector): New function.
6336 (vec_perm_indices::new_expanded_vector): Update for new
6337 vec_perm_indices class.
6338 (vec_perm_indices::rotate_inputs): New function.
6339 (vec_perm_indices::all_in_range_p): Operate directly on the
6340 encoded form, without computing elided elements.
6341 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6342 encoding. Update for new vec_perm_indices class.
6343 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6344 the given vec_perm_builder.
6345 (expand_vec_perm_var): Update vec_perm_builder constructor.
6346 (expand_mult_highpart): Use vec_perm_builder instead of
6347 auto_vec_perm_indices.
6348 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6349 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6350 or double series encoding as appropriate.
6351 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6352 vec_perm_indices instead of auto_vec_perm_indices.
6353 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6354 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6355 (vect_permute_store_chain): Likewise.
6356 (vect_grouped_load_supported): Likewise.
6357 (vect_permute_load_chain): Likewise.
6358 (vect_shift_permute_load_chain): Likewise.
6359 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6360 (vect_transform_slp_perm_load): Likewise.
6361 (vect_schedule_slp_instance): Likewise.
6362 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6363 (vectorizable_mask_load_store): Likewise.
6364 (vectorizable_bswap): Likewise.
6365 (vectorizable_store): Likewise.
6366 (vectorizable_load): Likewise.
6367 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6368 vec_perm_indices instead of auto_vec_perm_indices. Use
6369 tree_to_vec_perm_builder to read the vector from a tree.
6370 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6371 vec_perm_builder instead of a vec_perm_indices.
6372 (have_whole_vector_shift): Use vec_perm_builder and
6373 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6374 truncation to calc_vec_perm_mask_for_shift.
6375 (vect_create_epilog_for_reduction): Likewise.
6376 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6377 from auto_vec_perm_indices to vec_perm_indices.
6378 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6379 instead of changing individual elements.
6380 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6381 the vector in d.perm.
6382 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6383 from auto_vec_perm_indices to vec_perm_indices.
6384 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6385 instead of changing individual elements.
6386 (arm_vectorize_vec_perm_const): Use new_vector to install
6387 the vector in d.perm.
6388 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6389 Update vec_perm_builder constructor.
6390 (rs6000_expand_interleave): Likewise.
6391 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6392 (rs6000_expand_interleave): Likewise.
6394 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6396 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6397 to qimode could truncate the indices.
6398 * optabs.c (expand_vec_perm_var): Likewise.
6400 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6402 * Makefile.in (OBJS): Add vec-perm-indices.o.
6403 * vec-perm-indices.h: New file.
6404 * vec-perm-indices.c: Likewise.
6405 * target.h (vec_perm_indices): Replace with a forward class
6407 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6408 * optabs.h: Include vec-perm-indices.h.
6409 (expand_vec_perm): Delete.
6410 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6411 (expand_vec_perm_const): Declare.
6412 * target.def (vec_perm_const_ok): Replace with...
6413 (vec_perm_const): ...this new hook.
6414 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6415 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6416 * doc/tm.texi: Regenerate.
6417 * optabs.def (vec_perm_const): Delete.
6418 * doc/md.texi (vec_perm_const): Likewise.
6419 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6420 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6421 expand_vec_perm for constant permutation vectors. Assert that
6422 the mode of variable permutation vectors is the integer equivalent
6423 of the mode that is being permuted.
6424 * optabs-query.h (selector_fits_mode_p): Declare.
6425 * optabs-query.c: Include vec-perm-indices.h.
6426 (selector_fits_mode_p): New function.
6427 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6428 is defined, instead of checking whether the vec_perm_const_optab
6429 exists. Use targetm.vectorize.vec_perm_const instead of
6430 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6431 fit in the vector mode before using a variable permute.
6432 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6433 vec_perm_indices instead of an rtx.
6434 (expand_vec_perm): Replace with...
6435 (expand_vec_perm_const): ...this new function. Take the selector
6436 as a vec_perm_indices rather than an rtx. Also take the mode of
6437 the selector. Update call to shift_amt_for_vec_perm_mask.
6438 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6439 Use vec_perm_indices::new_expanded_vector to expand the original
6440 selector into bytes. Check whether the indices fit in the vector
6441 mode before using a variable permute.
6442 (expand_vec_perm_var): Make global.
6443 (expand_mult_highpart): Use expand_vec_perm_const.
6444 * fold-const.c: Includes vec-perm-indices.h.
6445 * tree-ssa-forwprop.c: Likewise.
6446 * tree-vect-data-refs.c: Likewise.
6447 * tree-vect-generic.c: Likewise.
6448 * tree-vect-loop.c: Likewise.
6449 * tree-vect-slp.c: Likewise.
6450 * tree-vect-stmts.c: Likewise.
6451 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6453 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6454 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6455 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6456 (aarch64_vectorize_vec_perm_const): ...this new function.
6457 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6458 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6459 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6460 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6461 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6462 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6463 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6465 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6466 check for NEON modes.
6467 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6468 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6469 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6470 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6472 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6473 the old VEC_PERM_CONST conditions.
6474 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6475 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6476 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6477 (ia64_vectorize_vec_perm_const_ok): Merge into...
6478 (ia64_vectorize_vec_perm_const): ...this new function.
6479 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6480 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6481 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6482 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6483 * config/mips/mips.c (mips_expand_vec_perm_const)
6484 (mips_vectorize_vec_perm_const_ok): Merge into...
6485 (mips_vectorize_vec_perm_const): ...this new function.
6486 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6487 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6488 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6489 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6490 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6491 (rs6000_expand_vec_perm_const): Delete.
6492 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6494 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6495 (altivec_expand_vec_perm_const_le): Take each operand individually.
6496 Operate on constant selectors rather than rtxes.
6497 (altivec_expand_vec_perm_const): Likewise. Update call to
6498 altivec_expand_vec_perm_const_le.
6499 (rs6000_expand_vec_perm_const): Delete.
6500 (rs6000_vectorize_vec_perm_const_ok): Delete.
6501 (rs6000_vectorize_vec_perm_const): New function.
6502 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6503 an element count and rtx array.
6504 (rs6000_expand_extract_even): Update call accordingly.
6505 (rs6000_expand_interleave): Likewise.
6506 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6507 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6508 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6509 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6510 (rs6000_expand_vec_perm_const): Delete.
6511 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6512 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6513 (altivec_expand_vec_perm_const_le): Take each operand individually.
6514 Operate on constant selectors rather than rtxes.
6515 (altivec_expand_vec_perm_const): Likewise. Update call to
6516 altivec_expand_vec_perm_const_le.
6517 (rs6000_expand_vec_perm_const): Delete.
6518 (rs6000_vectorize_vec_perm_const_ok): Delete.
6519 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6520 reference to the SPE evmerge intructions.
6521 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6522 an element count and rtx array.
6523 (rs6000_expand_extract_even): Update call accordingly.
6524 (rs6000_expand_interleave): Likewise.
6525 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6526 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6528 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6530 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6532 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6533 vector mode and that that mode matches the mode of the data
6535 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6536 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6537 directly using expand_vec_perm_1 when forcing selectors into
6539 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6541 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6543 * optabs-query.h (can_vec_perm_p): Delete.
6544 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6545 * optabs-query.c (can_vec_perm_p): Split into...
6546 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6547 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6548 particular selector is valid.
6549 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6550 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6551 (vect_grouped_load_supported): Likewise.
6552 (vect_shift_permute_load_chain): Likewise.
6553 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6554 (vect_transform_slp_perm_load): Likewise.
6555 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6556 (vectorizable_bswap): Likewise.
6557 (vect_gen_perm_mask_checked): Likewise.
6558 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6559 implementations of variable permutation vectors into account
6560 when deciding which selector to use.
6561 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6562 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6563 with a false third argument.
6564 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6565 to test whether the constant selector is valid and can_vec_perm_var_p
6566 to test whether a variable selector is valid.
6568 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6570 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6571 * optabs-query.c (can_vec_perm_p): Likewise.
6572 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6573 instead of vec_perm_indices.
6574 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6575 (vect_gen_perm_mask_checked): Likewise,
6576 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6577 (vect_gen_perm_mask_checked): Likewise,
6579 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6581 * optabs-query.h (qimode_for_vec_perm): Declare.
6582 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6583 (qimode_for_vec_perm): ...this new function.
6584 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6586 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6588 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6589 does not have a conditional at the top.
6591 2018-01-02 Richard Biener <rguenther@suse.de>
6593 * ipa-inline.c (big_speedup_p): Fix expression.
6595 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6598 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6601 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6605 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6606 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6607 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6608 cond_taken_branch_cost 3->4.
6610 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6612 PR tree-optimization/83581
6613 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6614 TODO_cleanup_cfg if any changes have been made.
6617 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6618 convert_modes if target mode has the right side, but different mode
6622 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6623 last argument when extracting from CONCAT. If either from_real or
6624 from_imag is NULL, use expansion through memory. If result is not
6625 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6626 the parts directly to inner mode, if even that fails, use expansion
6630 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6631 check for bswap in mode rather than HImode and use that in expand_unop
6634 Copyright (C) 2018 Free Software Foundation, Inc.
6636 Copying and distribution of this file, with or without modification,
6637 are permitted in any medium without royalty provided the copyright
6638 notice and this notice are preserved.