1 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
3 PR rtl-optimization/57193
4 * ira-color.c (struct allocno_color_data): Add member
5 conflict_allocno_hard_prefs.
6 (update_conflict_allocno_hard_prefs): New.
7 (bucket_allocno_compare_func): Add a preference based on
8 conflict_allocno_hard_prefs.
9 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
10 (color_allocnos): Remove a dead code. Initiate
11 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
13 2018-02-09 Jakub Jelinek <jakub@redhat.com>
16 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
17 constraint from =wa to wa. Avoid a subreg on the output operand,
18 instead use a pseudo and subreg it in a move.
19 (p9_xxbrd_<mode>): Changed to ...
20 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
21 (p9_xxbrd_v2df): New expander.
22 (p9_xxbrw_<mode>): Changed to ...
23 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
24 (p9_xxbrw_v4sf): New expander.
26 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
28 * config/rx.md: updated "movsicc" expand to be matched by GCC
29 * testsuite/gcc.target/rx/movsicc.c: new test case
31 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
34 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
35 multiply in 32-bit mode.
36 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
37 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
40 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
42 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
43 to allow or block "symbol_ref" depending on value of TARGET_JSR
44 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
45 call_value_internal insns
47 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
50 * dwarf2out.c (is_trivial_indirect_ref): New function.
51 (dwarf2out_late_global_decl): Do not generate a location
52 attribute for variables that have a non-trivial DECL_VALUE_EXPR
53 and that are not defined in the current unit.
55 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
57 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
58 instead of a libcall for UNORDERED.
60 2018-02-09 Tamar Christina <tamar.christina@arm.com>
63 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
64 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
66 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
69 * config/s390/s390.c (s390_set_current_function): Invoke
70 s390_indirect_branch_settings also if fndecl didn't change.
72 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
74 * config/rs6000/rs6000.md (blockage): Set length to zero.
76 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
78 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
80 2018-02-09 Jakub Jelinek <jakub@redhat.com>
83 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
84 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
88 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
89 PARALLEL incoming that failed vt_get_decl_and_offset check.
92 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
93 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
95 (get_variable_section): For decls in named .bss* sections pass true as
96 second argument to bss_initializer_p.
98 2018-02-09 Marek Polacek <polacek@redhat.com>
99 Jakub Jelinek <jakub@redhat.com>
102 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
103 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
104 Sync some changes from cxx_fold_indirect_ref.
106 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
108 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
110 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
111 (BLOCK_INLINE_ENTRY_LABEL): New.
112 (dwarf2out_var_location): Disregard inline entry markers.
113 (inline_entry_data): New struct.
114 (inline_entry_data_hasher): New hashtable type.
115 (inline_entry_data_hasher::hash): New.
116 (inline_entry_data_hasher::equal): New.
117 (inline_entry_data_table): New variable.
118 (add_high_low_attributes): Add DW_AT_entry_pc and
119 DW_AT_GNU_entry_view attributes if a pending entry is found
120 in inline_entry_data_table. Add old entry_pc attribute only
121 if debug nonbinding markers are disabled.
122 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
124 (block_within_block_p, dwarf2out_inline_entry): New.
125 (dwarf2out_finish): Check that no entries remained in
126 inline_entry_data_table.
127 * final.c (reemit_insn_block_notes): Handle inline entry notes.
128 (final_scan_insn, notice_source_line): Likewise.
129 (rest_of_clean_state): Skip inline entry markers.
130 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
132 * gimple.c (gimple_build_debug_inline_entry): New.
133 * gimple.h (enum gimple_debug_subcode): Add
134 GIMPLE_DEBUG_INLINE_ENTRY.
135 (gimple_build_debug_inline_entry): Declare.
136 (gimple_debug_inline_entry_p): New.
137 (gimple_debug_nonbind_marker_p): Adjust.
138 * insn-notes.def (INLINE_ENTRY): New.
139 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
140 inline entry marker notes.
141 (print_insn): Likewise.
142 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
143 (INSN_DEBUG_MARKER_KIND): Likewise.
144 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
145 * tree-inline.c (expand_call_inline): Build and insert
146 debug_inline_entry stmt.
147 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
148 inline entry blocks early, if nonbind markers are enabled.
149 (dump_scope_block): Dump fragment info.
150 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
151 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
152 (gimple_build_debug_inline_entry): New.
153 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
154 Enable/disable inline entry points too.
155 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
156 (DEBUG_INSN): Describe inline entry markers.
158 * common.opt (gvariable-location-views): New.
159 (gvariable-location-views=incompat5): New.
160 * config.in: Rebuilt.
161 * configure: Rebuilt.
162 * configure.ac: Test assembler for view support.
163 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
164 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
165 * dwarf2out.c (var_loc_view): New typedef.
166 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
167 (dwarf2out_locviews_in_attribute): New.
168 (dwarf2out_locviews_in_loclist): New.
169 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
170 (enum dw_line_info_opcode): Add LI_adv_address.
171 (struct dw_line_info_table): Add view.
172 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
173 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
174 (zero_view_p): New variable.
175 (ZERO_VIEW_P): New macro.
176 (output_asm_line_debug_info): New.
177 (struct var_loc_node): Add view.
178 (add_AT_view_list, AT_loc_list): New.
179 (add_var_loc_to_decl): Add view param. Test it against last.
180 (new_loc_list): Add view params. Record them.
181 (AT_loc_list_ptr): Handle loc and view lists.
182 (view_list_to_loc_list_val_node): New.
183 (print_dw_val): Handle dw_val_class_view_list.
184 (size_of_die): Likewise.
185 (value_format): Likewise.
186 (loc_list_has_views): New.
187 (gen_llsym): Set vl_symbol too.
188 (maybe_gen_llsym, skip_loc_list_entry): New.
189 (dwarf2out_maybe_output_loclist_view_pair): New.
190 (output_loc_list): Output view list or entries too.
191 (output_view_list_offset): New.
192 (output_die): Handle dw_val_class_view_list.
193 (output_dwarf_version): New.
194 (output_compilation_unit_header): Use it.
195 (output_skeleton_debug_sections): Likewise.
196 (output_rnglists, output_line_info): Likewise.
197 (output_pubnames, output_aranges): Update version comments.
198 (output_one_line_info_table): Output view numbers in asm comments.
199 (dw_loc_list): Determine current endview, pass it to new_loc_list.
200 Call maybe_gen_llsym.
201 (loc_list_from_tree_1): Adjust.
202 (add_AT_location_description): Create view list attribute if
203 needed, check it's absent otherwise.
204 (convert_cfa_to_fb_loc_list): Adjust.
205 (maybe_emit_file): Call output_asm_line_debug_info for test.
206 (dwarf2out_var_location): Reset views as needed. Precompute
207 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
209 (new_line_info_table): Reset next view.
210 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
211 (dwarf2out_source_line): Likewise. Output view resets and labels to
212 the assembler, or select appropriate line info opcodes.
213 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
214 (optimize_string_length): Catch it. Adjust.
215 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
216 dw_val_class_view_list, and remove it if no longer needed.
217 (hash_loc_list): Hash view numbers.
218 (loc_list_hasher::equal): Compare them.
219 (optimize_location_lists): Check whether a view list symbol is
220 needed, and whether the locview attribute is present, and
221 whether they match. Remove the locview attribute if no longer
223 (index_location_lists): Call skip_loc_list_entry for test.
224 (dwarf2out_finish): Call output_asm_line_debug_info for test.
225 Use output_dwarf_version.
226 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
227 (struct dw_val_node): Add val_view_list.
228 * final.c (SEEN_NEXT_VIEW): New.
229 (set_next_view_needed): New.
230 (clear_next_view_needed): New.
231 (maybe_output_next_view): New.
232 (final_start_function): Rename to...
233 (final_start_function_1): ... this. Take pointer to FIRST,
234 add SEEN parameter. Emit param bindings in the initial view.
235 (final_start_function): Reintroduce SEEN-less interface.
236 (final): Rename to...
237 (final_1): ... this. Take SEEN parameter. Output final pending
238 next view at the end.
239 (final): Reintroduce seen-less interface.
240 (final_scan_insn): Output pending next view before switching
241 sections or ending a block. Mark the next view as needed when
242 outputting variable locations. Notify debug backend of section
243 changes, and of location view changes.
244 (rest_of_handle_final): Adjust.
245 * toplev.c (process_options): Autodetect value for debug variable
246 location views option. Warn on incompat5 without -gdwarf-5.
247 * doc/invoke.texi (gvariable-location-views): New.
248 (gvariable-location-views=incompat5): New.
249 (gno-variable-location-views): New.
251 2018-02-08 David Malcolm <dmalcolm@redhat.com>
253 PR tree-optimization/84136
254 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
255 that the result of find_edge is non-NULL.
257 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
260 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
261 storing integer register in SImode. Fix cost of 256 and 512
262 byte aligned SSE register store.
264 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
266 * config/i386/i386.c (ix86_multiplication_cost): Fix
267 multiplication cost for TARGET_AVX512DQ.
269 2018-02-08 Marek Polacek <polacek@redhat.com>
271 PR tree-optimization/84238
272 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
275 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
277 PR tree-optimization/84265
278 * tree-vect-stmts.c (vectorizable_store): Don't treat
279 VMAT_CONTIGUOUS accesses as grouped.
280 (vectorizable_load): Likewise.
282 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
284 PR tree-optimization/81635
285 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
286 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
287 (test_round_for_mask): New functions.
288 (wide_int_cc_tests): Call test_round_for_mask.
289 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
290 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
291 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
292 range returned by get_range_info.
294 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
297 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
298 * symtab.c: Include builtins.h
299 (symtab_node::output_to_lto_symbol_table_p): Move here
300 from lto-streamer-out.c:output_symbol_p.
301 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
302 (output_symbol_p): Move all logic to symtab.c
303 (produce_symtab): Update.
305 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
307 * config/s390/s390-opts.h (enum indirect_branch): Define.
308 * config/s390/s390-protos.h (s390_return_addr_from_memory)
309 (s390_indirect_branch_via_thunk)
310 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
311 (enum s390_indirect_branch_type): Define.
312 * config/s390/s390.c (struct s390_frame_layout, struct
313 machine_function): Remove.
314 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
315 (indirect_branch_table_label_no, indirect_branch_table_name):
317 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
318 (enum s390_indirect_branch_option): Define.
319 (s390_return_addr_from_memory): New function.
320 (s390_handle_string_attribute): New function.
321 (s390_attribute_table): Add new attribute handler.
322 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
323 (s390_indirect_branch_via_thunk): New function.
324 (s390_indirect_branch_via_inline_thunk): New function.
325 (s390_function_ok_for_sibcall): When jumping via thunk disallow
326 sibling call optimization for non z10 compiles.
327 (s390_emit_call): Force indirect branch target to be a single
328 register. Add r1 clobber for non-z10 compiles.
329 (s390_emit_epilogue): Emit return jump via return_use expander.
330 (s390_reorg): Handle JUMP_INSNs as execute targets.
331 (s390_option_override_internal): Perform validity checks for the
332 new command line options.
333 (s390_indirect_branch_attrvalue): New function.
334 (s390_indirect_branch_settings): New function.
335 (s390_set_current_function): Invoke s390_indirect_branch_settings.
336 (s390_output_indirect_thunk_function): New function.
337 (s390_code_end): Implement target hook.
338 (s390_case_values_threshold): Implement target hook.
339 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
341 * config/s390/s390.h (struct s390_frame_layout)
342 (struct machine_function): Move here from s390.c.
343 (TARGET_INDIRECT_BRANCH_NOBP_RET)
344 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
345 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
346 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
347 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
348 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
349 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
350 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
351 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
352 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
353 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
354 (mnemonic attribute): Add values which aren't recognized
356 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
357 pattern for branch conversion. Fix mnemonic attribute.
358 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
359 indirect branch via thunk if requested.
360 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
361 ("*indirect_jump"): Disable for branch conversion using out of
363 ("indirect_jump_via_thunk<mode>_z10")
364 ("indirect_jump_via_thunk<mode>")
365 ("indirect_jump_via_inlinethunk<mode>_z10")
366 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
367 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
368 ("casesi_jump_via_inlinethunk<mode>_z10")
369 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
370 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
371 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
372 ("*indirect2_jump"): Disable for branch conversion.
373 ("casesi_jump"): Turn into expander and expand patterns for branch
375 ("return_use"): New expander.
376 ("*return"): Emit return via thunk and rename it to ...
377 ("*return<mode>"): ... this one.
378 * config/s390/s390.opt: Add new options and and enum for the
381 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
383 * lra-constraints.c (match_reload): Unconditionally use
384 gen_lowpart_SUBREG, rather than selecting between that
385 and equivalent gen_rtx_SUBREG code.
387 2018-02-08 Richard Biener <rguenther@suse.de>
389 PR tree-optimization/84233
390 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
391 changed flag instead of boguously re-using phi_inserted.
393 2018-02-08 Martin Jambor <mjambor@suse.cz>
395 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
396 static local variables.
398 2018-02-08 Richard Biener <rguenther@suse.de>
400 PR tree-optimization/84278
401 * tree-vect-stmts.c (vectorizable_store): When looking for
402 smaller vector types to perform grouped strided loads/stores
403 make sure the mode is supported by the target.
404 (vectorizable_load): Likewise.
406 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
408 * config/aarch64/aarch64.c (aarch64_components_for_bb):
409 Increase LDP/STP opportunities by adding adjacent callee-saves.
411 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
413 PR rtl-optimization/84068
414 PR rtl-optimization/83459
415 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
417 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
419 PR tree-optimization/84224
420 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
421 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
424 2018-02-07 Iain Sandoe <iain@codesourcery.com>
427 * config/rs6000/altivec.md (*restore_world): Remove LR use.
428 * config/rs6000/predicates.md (restore_world_operation): Adjust op
429 count, remove one USE.
431 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
433 * doc/install.texi (Configuration): Document the
434 --with-long-double-format={ibm,ieee} PowerPC configuration
438 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
439 Convert from define_expand to be define_insn_and_split. Rework
440 float/double/_Float128 conversions to QI/HI/SImode to work with
441 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
442 conversions to QI/HImode types did a store and then a load to
443 truncate the value. For conversions to VSX registers, don't split
444 the insn, instead emit the code directly. Use the code iterator
445 any_fix to combine signed and unsigned conversions.
446 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
447 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
448 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
449 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
450 (fix_<mode>di2_hw): Likewise.
451 (fixuns_<mode>di2_hw): Likewise.
452 (fix_<mode>si2_hw): Likewise.
453 (fixuns_<mode>si2_hw): Likewise.
454 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
455 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
456 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
457 fix<uns>_trunc<SFDF:mode>si2_p8.
458 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
460 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
461 (fix<uns>_<mode>_mem): Likewise.
462 (fctiw<u>z_<mode>_mem): Likewise.
463 (fix<uns>_<mode>_mem): Likewise.
464 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
465 the register allocator from doing a direct move to the GPRs to do
466 a store, and instead use the ISA 3.0 store byte/half-word from
467 vector register instruction. For IEEE 128-bit floating point,
468 also optimize stores of 32-bit ints.
469 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
471 2018-02-07 Alan Hayward <alan.hayward@arm.com>
473 * genextract.c (push_pathstr_operand): New function to support
475 (walk_rtx): Call push_pathstr_operand.
476 (print_path): Support [a-zA-Z].
478 2018-02-07 Richard Biener <rguenther@suse.de>
480 PR tree-optimization/84037
481 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
482 (cse_and_gimplify_to_preheader): Declare.
483 (vect_get_place_in_interleaving_chain): Likewise.
484 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
486 (_loop_vec_info::~_loop_vec_info): Delete it.
487 (cse_and_gimplify_to_preheader): New function.
488 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
489 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
490 (vectorizable_load): Likewise. For grouped stores always base
491 the IV on the first element.
492 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
493 condition before gimplifying.
495 2018-02-07 Jakub Jelinek <jakub@redhat.com>
497 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
498 *DIV_EXPR and *MOD_EXPR.
500 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
503 * config/i386/i386.c (ix86_option_override_internal): Mask out
504 the CF_SET bit when checking -fcf-protection.
506 2018-02-07 Tom de Vries <tom@codesourcery.com>
509 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
512 2018-02-07 Richard Biener <rguenther@suse.de>
514 PR tree-optimization/84204
515 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
518 PR tree-optimization/84205
519 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
520 special-case isl_ast_op_zdiv_r.
522 PR tree-optimization/84223
523 * graphite-scop-detection.c (gather_bbs::before_dom_children):
524 Only add conditions from within the region.
525 (gather_bbs::after_dom_children): Adjust.
527 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
530 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
531 * config/avr/avr.md: Only post-reload split REG-REG moves if
532 either register is REGERAL_REG_P.
534 2018-02-07 Jakub Jelinek <jakub@redhat.com>
536 PR tree-optimization/84235
537 * tree-ssa-scopedtables.c
538 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
539 if the subtraction is performed in floating point type where NaNs are
540 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
541 build 1. Formatting fix.
543 2018-02-06 Jakub Jelinek <jakub@redhat.com>
546 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
547 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
548 and skip it regardless of bb boundaries. Use CALL_P macro,
549 don't test INSN_P (insn) together with CALL_P or JUMP_P check
550 unnecessarily, formatting fix.
552 2018-02-06 Michael Collison <michael.collison@arm.com>
554 * config/arm/thumb2.md:
555 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
556 (*thumb_mov_notscc): Ditto.
558 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
561 * config/rs6000/rs6000.md (su code attribute): Use "u" for
562 unsigned_fix, not "s".
564 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
566 * configure.ac (gcc_fn_eh_frame_ro): New function.
567 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
568 correct .eh_frame permissions.
569 * configure: Regenerate.
571 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
573 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
576 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
578 * config/rs6000/rs6000.c (rs6000_option_override_internal):
579 Display warning message for -mno-speculate-indirect-jumps.
581 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
583 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
585 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
587 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
589 PR tree-optimization/84225
590 * tree-eh.c (find_trapping_overflow): Only call
591 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
593 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
596 * config/i386/i386.c: Reimplement the check of possible options
597 -mibt/-mshstk conbination. Change error messages.
598 * doc/invoke.texi: Fix a typo: remove extra '='.
600 2018-02-06 Marek Polacek <polacek@redhat.com>
602 PR tree-optimization/84228
603 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
605 2018-02-06 Tamar Christina <tamar.christina@arm.com>
608 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
609 emitted arch directives.
610 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
611 __ARM_FEATURE_COPROC before changing architectures.
613 2018-02-06 Richard Biener <rguenther@suse.de>
615 * config/i386/i386.c (print_reg): Fix typo.
616 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
618 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
620 * configure: Regenerate.
622 2018-02-05 Martin Sebor <msebor@redhat.com>
624 PR tree-optimization/83369
625 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
628 2018-02-05 Martin Liska <mliska@suse.cz>
630 * doc/invoke.texi: Cherry-pick upstream r323995.
632 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
634 * ira.c (ira_init_register_move_cost): Adjust comment.
636 2018-02-05 Martin Liska <mliska@suse.cz>
638 PR gcov-profile/84137
639 * doc/gcov.texi: Fix typo in documentation.
641 2018-02-05 Martin Liska <mliska@suse.cz>
643 PR gcov-profile/83879
644 * doc/gcov.texi: Document necessity of --dynamic-list-data when
645 using dlopen functionality.
647 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
649 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
650 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
651 _mm_maskz_range_ss, _mm_mask_range_round_ss,
652 _mm_maskz_range_round_ss): New intrinsics.
653 (__builtin_ia32_rangesd128_round)
654 (__builtin_ia32_rangess128_round): Remove.
655 (__builtin_ia32_rangesd128_mask_round,
656 __builtin_ia32_rangess128_mask_round): New builtins.
657 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
658 __builtin_ia32_rangess128_round): Remove.
659 (__builtin_ia32_rangesd128_mask_round,
660 __builtin_ia32_rangess128_mask_round): New builtins.
661 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
662 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
663 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
664 "<round_saeonly_constraint>")): Changed to ...
665 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
666 "<round_saeonly_scalar_constraint>")): ... this.
667 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
668 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
669 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
670 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
671 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
673 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
675 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
677 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
678 Remove all values except native, 8540 and 8548.
680 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
682 * config/i386/i386.c (ix86_output_function_return): Pass
683 INVALID_REGNUM, instead of -1, as invalid register number to
684 indirect_thunk_name and output_indirect_thunk.
686 2018-02-02 Julia Koval <julia.koval@intel.com>
688 * config.gcc: Add -march=icelake.
689 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
690 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
691 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
692 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
693 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
694 (processor_target_table): Add icelake.
695 (ix86_option_override_internal): Handle new PTAs.
696 (get_builtin_code_for_version): Handle icelake.
697 (M_INTEL_COREI7_ICELAKE): New.
698 (fold_builtin_cpu): Handle icelake.
699 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
700 * doc/invoke.texi: Add -march=icelake.
702 2018-02-02 Julia Koval <julia.koval@intel.com>
704 * config/i386/i386.c (ix86_option_override_internal): Change flags type
706 * wide-int-bitmask.h: New.
708 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
711 * config/i386/i386.md: Replace Pmode with word_mode in
712 builtin_setjmp_setup and builtin_longjmp to support x32.
714 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
718 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
720 (rs6000_supported_cpu_names): New static variable.
721 (linux_cpu_translation_table): Likewise.
722 (elf_platform) <cpu>: Define new static variable and use it.
723 Translate kernel AT_PLATFORM name to canonical name if needed.
724 Error if platform name is unknown.
726 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
729 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
731 2018-02-01 Jeff Law <law@redhat.com>
734 * config/i386/i386.c (release_scratch_register_on_entry): Add new
735 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
736 the scratch if RELEASE_VIA_POP is false.
737 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
738 If we have to save a temporary register, decrement SIZE appropriately.
739 Pass new arguments to release_scratch_register_on_entry.
740 (ix86_adjust_stack_and_probe): Likewise.
741 (ix86_emit_probe_stack_range): Pass new arguments to
742 release_scratch_register_on_entry.
744 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
746 PR rtl-optimization/84157
747 * combine.c (change_zero_ext): Use REG_P predicate in
748 front of HARD_REGISTER_P predicate.
750 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
752 * config/avr/avr.c (avr_option_override): Move disabling of
753 -fdelete-null-pointer-checks to...
754 * common/config/avr/avr-common.c (avr_option_optimization_table):
757 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
759 PR tree-optimization/81635
760 * tree-data-ref.c (split_constant_offset_1): For types that
761 wrap on overflow, try to use range info to prove that wrapping
764 2018-02-01 Renlin Li <renlin.li@arm.com>
767 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
769 (aarch64_register_move_cost): Likewise.
770 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
772 (REG_CLASS_NAMES): Likewise.
773 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
774 TAILCALL_ADDR_REGS. Remove IP registers.
775 * config/aarch64/aarch64.md (Ucs): Update register constraint.
777 2018-02-01 Richard Biener <rguenther@suse.de>
779 * domwalk.h (dom_walker::dom_walker): Add additional constructor
780 for specifying RPO order and allow NULL for that.
781 * domwalk.c (dom_walker::dom_walker): Likewise.
782 (dom_walker::walk): Handle NULL RPO order.
783 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
785 (rewrite_update_dom_walker): Likewise.
786 (mark_def_dom_walker): Likewise.
788 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
790 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
791 (aarch64_maybe_expand_sve_subreg_move): Declare.
792 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
793 * config/aarch64/predicates.md (aarch64_any_register_operand): New
795 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
796 that are semantically a reverse operation.
797 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
798 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
799 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
801 (aarch64_can_change_mode_class): For big-endian, forbid changes
802 between two SVE modes if they have different element sizes.
804 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
806 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
807 the TImode handling for big-endian targets.
809 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
811 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
812 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
814 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
815 Remove BSWAP handing for big-endian targets and use the form of
816 LD1RQ appropariate for the mode.
818 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
820 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
821 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
824 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
827 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
828 check for operands that need to go through aarch64_sve_reload_be.
830 2018-02-01 Jakub Jelinek <jakub@redhat.com>
832 PR tree-optimization/81661
833 PR tree-optimization/84117
834 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
835 * tree-eh.c: Include gimplify.h.
836 (find_trapping_overflow, replace_trapping_overflow,
837 rewrite_to_non_trapping_overflow): New functions.
838 * tree-vect-loop.c: Include tree-eh.h.
839 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
840 * tree-data-ref.c: Include tree-eh.h.
841 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
843 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
845 PR rtl-optimization/84123
846 * combine.c (change_zero_ext): Check if hard register satisfies
847 can_change_dest_mode before calling gen_lowpart_SUBREG.
849 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
852 * ira.c (ira_init_register_move_cost): Remove assert.
854 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
856 PR rtl-optimization/84071
857 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
858 * doc/tm.texi: Regenerate.
860 2018-01-31 Richard Biener <rguenther@suse.de>
862 PR tree-optimization/84132
863 * tree-data-ref.c (analyze_miv_subscript): Properly
864 check whether evolution_function_is_affine_multivariate_p
865 before calling gcd_of_steps_may_divide_p.
867 2018-01-31 Julia Koval <julia.koval@intel.com>
870 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
871 * config/i386/i386.md (rdpid_rex64) New.
872 (rdpid): Make 32bit only.
874 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
877 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
878 an IDENTIFIER_NODE for FUNCTION_TYPE's.
880 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
883 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
885 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
887 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
889 PR rtl-optimization/84071
890 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
891 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
893 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
895 * config/arc/arc.c (arc_handle_aux_attribute): New function.
896 (arc_attribute_table): Add 'aux' attribute.
897 (arc_in_small_data_p): Consider aux like variables.
898 (arc_is_aux_reg_p): New function.
899 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
900 (arc_get_aux_arg): New function.
901 (prepare_move_operands): Handle aux-register access.
902 (arc_handle_aux_attribute): New function.
903 * doc/extend.texi (ARC Variable attributes): Add subsection.
905 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
907 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
908 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
909 (arc_attribute_table): Add 'uncached' attribute.
910 (arc_print_operand): Print '.di' flag for uncached memory
912 (arc_in_small_data_p): Do not consider for small data the uncached
914 (arc_is_uncached_mem_p): New function.
915 * config/arc/predicates.md (compact_store_memory_operand): Check
916 for uncached memory accesses.
917 (nonvol_nonimm_operand): Likewise.
918 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
920 2018-01-31 Jakub Jelinek <jakub@redhat.com>
923 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
924 falign-loops=): Add Optimization flag.
926 2018-01-30 Jeff Law <law@redhat.com>
929 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
930 INT_REGISTERS_SAVED. Check it prior to calling
931 get_scratch_register_on_entry.
932 (ix86_adjust_stack_and_probe): Similarly.
933 (ix86_emit_probe_stack_range): Similarly.
934 (ix86_expand_prologue): Corresponding changes.
936 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
939 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
940 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
942 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
945 * lra-constraints.c (curr_insn_transform): Process AND in the
948 2018-01-30 Jakub Jelinek <jakub@redhat.com>
950 PR rtl-optimization/83986
951 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
952 dependence against last_pending_memory_flush in addition to
955 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
957 PR tree-optimization/81611
958 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
961 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
963 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
966 2018-01-30 Richard Biener <rguenther@suse.de>
967 Jakub Jelinek <jakub@redhat.com>
969 PR tree-optimization/84111
970 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
971 inner loops added during recursion, as they don't have up-to-date
974 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
977 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
978 (can_inline_edge_by_limits_p): ... here.
979 (can_early_inline_edge_p, check_callers,
980 update_caller_keys, update_callee_keys, recursive_inlining,
981 add_new_edges_to_heap, speculation_useful_p,
982 inline_small_functions,
983 inline_small_functions, flatten_function,
984 inline_to_all_callers_1): Update.
986 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
988 * profile-count.c (profile_count::combine_with_ipa_count): Handle
991 2018-01-30 Richard Biener <rguenther@suse.de>
993 PR tree-optimization/83008
994 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
995 invariant and constant vector uses in stmts when they need
998 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1001 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1002 * configure: Regenerate.
1004 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1006 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1008 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1009 Use gen_rtx_REG rather than gen_lowpart.
1011 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1013 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1014 rather than 0 when creating partial subregs.
1016 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1018 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1021 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1024 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1025 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1026 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1027 flags. This restores the settings used before the 2017-07-24.
1028 Turning off pre increment/decrement/modify allows IVOPTS to
1029 optimize DF/SF loops where the index is an int.
1031 2018-01-29 Richard Biener <rguenther@suse.de>
1032 Kelvin Nilsen <kelvin@gcc.gnu.org>
1035 * tree-vect-stmts.c (vectorizable_call): Don't call
1036 targetm.vectorize_builtin_md_vectorized_function if callee is
1039 2018-01-22 Carl Love <cel@us.ibm.com>
1041 * doc/extend.tex: Fix typo in second arg in
1042 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1044 2018-01-29 Richard Biener <rguenther@suse.de>
1046 PR tree-optimization/84086
1047 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1048 (flush_ssaname_freelist): When SSA names were released reset
1049 the SCEV hash table.
1051 2018-01-29 Richard Biener <rguenther@suse.de>
1053 PR tree-optimization/84057
1054 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1055 removed paths when removing edges.
1057 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1059 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1060 -mfunction-return=@var{choice}.
1062 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1065 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1067 (layout::print_source_line): Likewise.
1068 (test_get_line_width_without_trailing_whitespace): Add test cases.
1070 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1073 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1076 2018-01-26 Jim Wilson <jimw@sifive.com>
1078 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1080 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1083 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1085 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1086 and CMP + SUB-immediate -> SUBS.
1088 2018-01-26 Martin Sebor <msebor@redhat.com>
1090 PR tree-optimization/83896
1091 * tree-ssa-strlen.c (get_string_len): Rename...
1092 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1093 Avoid assuming length is constant.
1094 (handle_char_store): Use HOST_WIDE_INT for string length.
1096 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1099 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1100 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1102 2018-01-26 Richard Biener <rguenther@suse.de>
1104 PR rtl-optimization/84003
1105 * dse.c (record_store): Only record redundant stores when
1106 the earlier store aliases at least all accesses the later one does.
1108 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1110 PR rtl-optimization/83985
1111 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1112 REG_CFA_RESTORE insns.
1113 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1114 REG_CFA_RESTORE insns here.
1117 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1118 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1120 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1122 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1123 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1124 (arc_init): Likewise.
1125 (arc_override_options): Likewise.
1126 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1128 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1130 * config/arc/arc.h (TARGET_DBNZ): Define.
1131 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1132 properly set the tune attribute.
1133 (dbnz): Use TARGET_DBNZ guard.
1134 * config/arc/arc.opt (mtune): Add core3 option.
1136 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1138 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1139 recognize new pic like addresses.
1140 (arc_delegitimize_address): Clean up.
1142 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1144 * config/arc/arc-arches.def: Option mrf16 valid for all
1146 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1147 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1148 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1149 * config/arc/arc-tables.opt: Regenerate.
1150 * config/arc/arc.c (arc_conditional_register_usage): Handle
1151 reduced register file case.
1152 (arc_file_start): Set must have build attributes.
1153 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1155 * config/arc/arc.opt (mrf16): Add new option.
1156 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1157 * config/arc/genmultilib.awk: Handle new mrf16 option.
1158 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1159 * config/arc/t-multilib: Regenerate.
1160 * doc/invoke.texi (ARC Options): Document mrf16 option.
1162 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1164 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1165 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1166 (arc_attribute_table): Add 'secure_call' attribute.
1167 (arc_print_operand): Print secure call operand.
1168 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1170 (arc_is_secure_call_p): New function. * config/arc/arc.md
1171 (call_i): Add support for sjli instruction.
1172 (call_value_i): Likewise.
1173 * config/arc/constraints.md (Csc): New constraint.
1175 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1176 John Eric Martin <John.Martin@emmicro-us.com>
1178 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1179 * config/arc/arc.c (_arc_jli_section): New struct.
1180 (arc_jli_section): New type.
1181 (rc_jli_sections): New static variable.
1182 (arc_handle_jli_attribute): New function.
1183 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1184 (arc_file_end): New function.
1185 (TARGET_ASM_FILE_END): Define.
1186 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1187 (arc_add_jli_section): New function.
1188 (jli_call_scan): Likewise.
1189 (arc_reorg): Call jli_call_scan.
1190 (arc_output_addsi): Remove 'S' from printing asm operand.
1191 (arc_is_jli_call_p): New function.
1192 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1194 (movhi_insn): Likewise.
1195 (movsi_insn): Likewise.
1196 (movsi_set_cc_insn): Likewise.
1197 (loadqi_update): Likewise.
1198 (load_zeroextendqisi_update): Likewise.
1199 (load_signextendqisi_update): Likewise.
1200 (loadhi_update): Likewise.
1201 (load_zeroextendhisi_update): Likewise.
1202 (load_signextendhisi_update): Likewise.
1203 (loadsi_update): Likewise.
1204 (loadsf_update): Likewise.
1205 (movsicc_insn): Likewise.
1206 (bset_insn): Likewise.
1207 (bxor_insn): Likewise.
1208 (bclr_insn): Likewise.
1209 (bmsk_insn): Likewise.
1210 (bicsi3_insn): Likewise.
1211 (cmpsi_cc_c_insn): Likewise.
1212 (movsi_ne): Likewise.
1213 (movsi_cond_exec): Likewise.
1214 (clrsbsi2): Likewise.
1225 (call_i): Remove 'S' asm letter, add jli instruction.
1226 (call_value_i): Likewise.
1227 * config/arc/arc.op (mjli-always): New option.
1228 * config/arc/constraints.md (Cji): New constraint.
1229 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1231 (subsf3_fpx): Likewise.
1232 (mulsf3_fpx): Likewise.
1233 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1235 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1237 * doc/invoke.texi (ARC): Document mjli-always option.
1239 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1241 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1242 and use incw and decw where possible
1243 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1245 2018-01-26 Richard Biener <rguenther@suse.de>
1247 PR tree-optimization/81082
1248 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1249 association if it requires casting to unsigned.
1250 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1251 from fold_plusminus_mult_expr to catch important cases late when
1252 range info is available.
1254 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1256 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1257 * configure.ac (hidden_linkonce): New test.
1258 * configure: Regenerate.
1259 * config.in: Regenerate.
1261 2018-01-26 Julia Koval <julia.koval@intel.com>
1263 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1264 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1265 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1266 _mm_mask_bitshuffle_epi64_mask): Fix type.
1267 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1268 USI_FTYPE_V4DI_V4DI_USI): Remove.
1269 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1270 __builtin_ia32_vpshufbitqmb256_mask,
1271 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1272 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1273 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1275 2018-01-26 Alan Modra <amodra@gmail.com>
1278 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1279 UNSPEC_VBPERMQ. Sort other unspecs.
1281 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1283 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1285 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1288 * predict.c (drop_profile): Do not push/pop cfun; update also
1290 (handle_missing_profiles): Fix logic looking for zero profiles.
1292 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1295 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1296 on functions with #pragma omp declare simd or functions with simd
1298 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1299 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1300 Remove trailing \n from warning_at calls.
1302 2018-01-25 Tom de Vries <tom@codesourcery.com>
1305 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1306 for neutered workers.
1308 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1311 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1312 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1314 2018-01-24 Jeff Law <law@redhat.com>
1317 * i386.c (get_probe_interval): Move to earlier point.
1318 (ix86_compute_frame_layout): If -fstack-clash-protection and
1319 the frame is larger than the probe interval, then use pushes
1320 to save registers rather than reg->mem moves.
1321 (ix86_expand_prologue): Remove conditional for int_registers_saved
1324 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1327 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1328 min/max for never referenced object.
1330 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1333 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1335 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1336 attributes from DECL_ATTRIBUTES (decl) without affecting
1337 DECL_ATTRIBUTES (current_function_decl).
1338 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1339 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1341 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1343 PR tree-optimization/83979
1344 * fold-const.c (fold_comparison): Use constant_boolean_node
1345 instead of boolean_{true,false}_node.
1347 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1349 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1352 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1354 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1355 Simplify the clause that sets the length attribute.
1356 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1357 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1358 clause that sets the length attribute.
1359 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1361 2018-01-24 Tom de Vries <tom@codesourcery.com>
1364 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1365 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1366 Add strict parameter.
1367 (prevent_branch_around_nothing): Insert dummy insn between branch to
1368 label and label with no ptx insn inbetween.
1369 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1371 2018-01-24 Tom de Vries <tom@codesourcery.com>
1374 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1375 for neutered threads in warp.
1376 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1378 2018-01-24 Richard Biener <rguenther@suse.de>
1380 PR tree-optimization/83176
1381 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1384 2018-01-24 Richard Biener <rguenther@suse.de>
1386 PR tree-optimization/82819
1387 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1388 code generating pluses that are no-ops in the target precision.
1390 2018-01-24 Richard Biener <rguenther@suse.de>
1393 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1395 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1397 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1398 to merge probabilities.
1399 * predict.c (probably_never_executed): Also mark as cold functions
1400 with global 0 profile and guessed local profile.
1401 * profile-count.c (profile_probability::combine_with_count): New
1403 * profile-count.h (profile_probability::operator*,
1404 profile_probability::operator*=, profile_probability::operator/,
1405 profile_probability::operator/=): Reduce precision to adjusted
1406 and set value to guessed on contradictory divisions.
1407 (profile_probability::combine_with_freq): Remove.
1408 (profile_probability::combine_wiht_count): Declare.
1409 (profile_count::force_nonzero):: Set to adjusted.
1410 (profile_count::probability_in):: Set quality to adjusted.
1411 * tree-ssa-tail-merge.c (replace_block_by): Use
1414 2018-01-23 Andrew Waterman <andrew@sifive.com>
1415 Jim Wilson <jimw@sifive.com>
1417 * config/riscv/riscv.c (riscv_stack_boundary): New.
1418 (riscv_option_override): Set riscv_stack_boundary. Handle
1419 riscv_preferred_stack_boundary_arg.
1420 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1421 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1422 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1423 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1424 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1425 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1427 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1430 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1431 of struct ix86_frame.
1432 (ix86_expand_epilogue): Likewise. Add a local variable for
1433 the reg_save_offset field in struct ix86_frame.
1435 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1437 PR tree-optimization/82604
1438 * tree-loop-distribution.c (enum partition_kind): New enum item
1439 PKIND_PARTIAL_MEMSET.
1440 (partition_builtin_p): Support above new enum item.
1441 (generate_code_for_partition): Ditto.
1442 (compute_access_range): Differentiate cases that equality can be
1443 proven at all loops, the innermost loops or no loops.
1444 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1445 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1446 (finalize_partitions, distribute_loop): Don't fuse partition of
1447 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1448 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1451 2018-01-23 Martin Liska <mliska@suse.cz>
1453 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1454 order to ignore the predictor.
1455 (PRED_POLYMORPHIC_CALL): Likewise.
1456 (PRED_RECURSIVE_CALL): Likewise.
1458 2018-01-23 Martin Liska <mliska@suse.cz>
1460 * tree-profile.c (tree_profiling): Print function header to
1461 aware reader which function we are working on.
1462 * value-prof.c (gimple_find_values_to_profile): Do not print
1463 not interesting value histograms.
1465 2018-01-23 Martin Liska <mliska@suse.cz>
1467 * profile-count.h (enum profile_quality): Add
1468 profile_uninitialized as the first value. Do not number values
1469 as they are zero based.
1470 (profile_count::verify): Update sanity check.
1471 (profile_probability::verify): Likewise.
1473 2018-01-23 Nathan Sidwell <nathan@acm.org>
1475 * doc/invoke.texi (ffor-scope): Deprecate.
1477 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1479 PR tree-optimization/83510
1480 * domwalk.c (set_all_edges_as_executable): New function.
1481 (dom_walker::dom_walker): Convert bool param
1482 "skip_unreachable_blocks" to enum reachability. Move setup of
1483 edge flags to set_all_edges_as_executable and only do it when
1484 reachability is REACHABLE_BLOCKS.
1485 * domwalk.h (enum dom_walker::reachability): New enum.
1486 (dom_walker::dom_walker): Convert bool param
1487 "skip_unreachable_blocks" to enum reachability.
1488 (set_all_edges_as_executable): New decl.
1489 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1490 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1492 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1493 but converting true to REACHABLE_BLOCKS.
1494 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1496 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1497 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1498 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1500 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1501 if check_all_array_refs will be called.
1503 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1505 * tree.c (selftest::test_location_wrappers): Add more test
1508 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1510 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1511 (selftest::test_bit_in_range): Likewise.
1513 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1516 * doc/sourcebuild.texi (vect_float): Say that the selector
1517 only describes the situation when -funsafe-math-optimizations is on.
1518 (vect_float_strict): Document.
1520 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1522 PR tree-optimization/83965
1523 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1524 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1525 instead of checking only for a reduction.
1526 (vect_recog_widen_sum_pattern): Likewise.
1528 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1530 * predict.c (probably_never_executed): Only use precise profile info.
1531 (compute_function_frequency): Skip after inlining hack since we now
1532 have quality checking.
1534 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1536 * profile-count.h (profile_probability::very_unlikely,
1537 profile_probability::unlikely, profile_probability::even): Set
1538 precision to guessed.
1540 2018-01-23 Richard Biener <rguenther@suse.de>
1542 PR tree-optimization/83963
1543 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1544 Properly terminate dominator walk when crossing the exit edge not
1545 when visiting its source block.
1547 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1550 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1551 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1553 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1555 PR tree-optimization/83957
1556 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1557 semicolon after for body surrounded by braces.
1559 PR tree-optimization/83081
1560 * profile-count.h (profile_probability::split): New method.
1561 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1562 Use profile_probability::split.
1563 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1564 when splitting a single conditional jump into 2.
1566 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1568 PR tree-optimization/69452
1569 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1572 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1574 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1575 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1576 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1578 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1580 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1581 * config/rl78/rl78.md: New define_expand "movdi"
1582 * config/rl78/rl78.c: New function definition rl78_split_movdi
1584 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1587 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1589 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1590 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1591 128-bit to produce an UNSPEC move to get the double word with the
1592 signbit and then a shift directly to do signbit.
1593 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1594 implementation with a new version that just does either a direct
1595 move or a regular move. Move memory interface to separate insns.
1596 Move insns so they are next to the expander.
1597 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1598 with signbit move. Split big and little endian case.
1599 (signbit<mode>2_dm_mem_le): Likewise.
1600 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1601 (signbit<mode>2_dm2): Likewise.
1603 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1605 * config/rl78/rl78.md: New define_expand "anddi3".
1607 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1609 * config/rl78/rl78.md: New define_expand "umindi3".
1611 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1613 * config/rl78/rl78.md: New define_expand "smindi3".
1615 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1617 * config/rl78/rl78.md: New define_expand "smaxdi3".
1619 2018-01-22 Carl Love <cel@us.ibm.com>
1621 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1622 LVX_V1TI): Add macro expansion.
1623 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1624 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1625 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1626 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1627 Change check to determine if the instruction is a byte reversing
1628 entry. Fix typo in comment.
1629 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1630 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1631 Add def_builtin calls for new builtins.
1632 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1633 Add define_insn expansion.
1635 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1637 * config/rl78/rl78.md: New define_expand "umaxdi3".
1639 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1641 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1642 for non-QImode registers
1644 2018-01-22 Richard Biener <rguenther@suse.de>
1646 PR tree-optimization/83963
1647 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1648 including the loop exit block.
1649 (scop_detection::merge_sese): Likewise.
1650 (scop_detection::add_scop): Do it here instead.
1652 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1654 * doc/sourcebuild.texi (arm_softfloat): Document.
1656 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1659 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1660 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1661 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1663 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1664 David Edelsohn <dje.gcc@gmail.com>
1667 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1668 Change "crset eq" to "crset 2".
1669 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1670 (*call_indirect_aix<mode>_nospec): Likewise.
1671 (*call_value_indirect_aix<mode>_nospec): Likewise.
1672 (*call_indirect_elfv2<mode>_nospec): Likewise.
1673 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1674 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1675 change assembly output from . to $.
1676 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1677 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1678 (*tablejump<mode>_internal1_nospec): Likewise.
1680 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1683 * config/sh/sh_optimize_sett_clrt.cc:
1684 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1686 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1688 PR tree-optimization/83940
1689 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1690 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1691 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1692 use it to pass back the definition type.
1693 (vect_check_store_rhs): Likewise rhs_dt_out.
1694 (vect_build_gather_load_calls): Add a mask_dt argument and use
1695 it instead of a call to vect_is_simple_use.
1696 (vectorizable_store): Update calls to vect_check_load_store_mask
1697 and vect_check_store_rhs. Use the dt returned by the latter instead
1698 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1699 instead of calls to vect_is_simple_use. Pass the scalar rather
1700 than the vector operand to vect_is_simple_use when handling
1701 second and subsequent copies of an rhs value.
1702 (vectorizable_load): Update calls to vect_check_load_store_mask
1703 and vect_build_gather_load_calls. Use the cached mask_dt and
1704 gs_info.offset_dt instead of calls to vect_is_simple_use.
1706 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1709 * tree-emutls.c: Include gimplify.h.
1710 (lower_emutls_2): New function.
1711 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1712 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1713 it before further processing.
1716 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1717 UINTVAL (trueop1) instead of INTVAL (op1).
1719 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1723 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1724 INCOMING_FRAME_SP_OFFSET if not defined.
1725 (scan_trace): Add ENTRY argument. If true and
1726 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1727 emit a note to adjust the CFA offset.
1728 (create_cfi_notes): Adjust scan_trace callers.
1729 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1730 INCOMING_FRAME_SP_OFFSET in the CIE.
1731 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1732 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1734 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1735 * doc/tm.texi: Regenerated.
1737 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1739 PR rtl-optimization/83147
1740 * lra-constraints.c (remove_inheritance_pseudos): Use
1741 lra_substitute_pseudo_within_insn.
1743 2018-01-19 Tom de Vries <tom@codesourcery.com>
1744 Cesar Philippidis <cesar@codesourcery.com>
1747 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1749 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1752 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1753 spaces for function labels.
1755 2018-01-19 Martin Liska <mliska@suse.cz>
1757 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1758 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1759 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1760 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1761 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1762 (PRED_CONST_RETURN): Change from 69 to 65.
1763 (PRED_NULL_RETURN): Change from 91 to 71.
1764 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1765 (PRED_LOOP_GUARD): Change from 66 to 73.
1767 2018-01-19 Martin Liska <mliska@suse.cz>
1769 * predict.c (predict_insn_def): Add new assert.
1770 (struct branch_predictor): Change type to signed integer.
1771 (test_prediction_value_range): Amend test to cover
1773 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1774 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1775 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1776 (PRED_LOOP_IV_COMPARE): Likewise.
1777 * predict.h (PROB_UNINITIALIZED): Define new constant.
1779 2018-01-19 Martin Liska <mliska@suse.cz>
1781 * predict.c (dump_prediction): Add new format for
1782 analyze_brprob.py script which is enabled with -details
1784 * profile-count.h (precise_p): New function.
1786 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1788 PR tree-optimization/83922
1789 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1790 there are no statements that need masking.
1791 (vect_active_double_reduction_p): New function.
1792 (vect_analyze_loop_operations): Use it when handling phis that
1793 are not in the loop header.
1795 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1797 PR tree-optimization/83914
1798 * tree-vect-loop.c (vectorizable_induction): Don't convert
1799 init_expr or apply the peeling adjustment for inductions
1800 that are nested within the vectorized loop.
1802 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1804 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1807 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1811 * function.h (gimplify_parameters): Add gimple_seq * argument.
1812 * function.c: Include gimple.h and options.h.
1813 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1814 for the added local temporaries if needed.
1815 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1816 if there are any parameter cleanups, wrap whole body into a
1817 try/finally with the cleanups.
1819 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1822 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1823 Use GET_MODE_CLASS for scalar floating point.
1825 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1829 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1830 Fix call of call_cgraph_insertion_hooks.
1832 2018-01-18 Martin Sebor <msebor@redhat.com>
1834 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1836 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1839 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1842 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1845 * common.opt: (-ffile-prefix-map): New option.
1846 * opts.c (common_handle_option): Defer it.
1847 * opts-global.c (handle_common_deferred_options): Handle it.
1848 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1849 * file-prefix-map.h: New file.
1850 (remap_debug_filename, add_debug_prefix_map): ...here.
1851 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1852 * final.c (debug_prefix_map, add_debug_prefix_map
1853 remap_debug_filename): Move to...
1854 * file-prefix-map.c: New file.
1855 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1856 generalize, get rid of alloca(), use strrchr() instead of strchr().
1857 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1858 Implement in terms of add_prefix_map().
1859 (remap_macro_filename, remap_debug_filename): Implement in term of
1861 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1862 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1863 * dbxout.c: Include file-prefix-map.h.
1864 * varasm.c: Likewise.
1865 * vmsdbgout.c: Likewise.
1866 * xcoffout.c: Likewise.
1867 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1868 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1869 * doc/invoke.texi (-ffile-prefix-map): Document.
1870 (-fdebug-prefix-map): Update description.
1872 2018-01-18 Martin Liska <mliska@suse.cz>
1874 * config/i386/i386.c (indirect_thunk_name): Document that also
1876 (output_indirect_thunk): Document why both instructions
1877 (pause and lfence) are generated.
1879 2018-01-18 Richard Biener <rguenther@suse.de>
1881 PR tree-optimization/83887
1882 * graphite-scop-detection.c
1883 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1884 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1885 (scop_detection::merge_sese): Re-implement with a flood-fill
1886 algorithm that properly finds a SESE region if it exists.
1888 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1891 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1892 pointer_diff optimizations use view_convert instead of convert.
1894 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1896 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1897 Generate different code for -mno-speculate-indirect-jumps.
1898 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1899 (*call_indirect_aix<mode>): Disable for
1900 -mno-speculate-indirect-jumps.
1901 (*call_indirect_aix<mode>_nospec): New define_insn.
1902 (*call_value_indirect_aix<mode>): Disable for
1903 -mno-speculate-indirect-jumps.
1904 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1905 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1906 -mno-speculate-indirect-jumps.
1907 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1909 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1911 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1912 long double type, set the flags for noting the default long double
1913 type, even if we don't pass or return a long double type.
1915 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1918 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1921 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1923 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1924 support for merge[hl].
1925 (fold_mergehl_helper): New helper function.
1926 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1927 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1928 (altivec_vmrglw_direct): Add xxmrglw insn.
1930 2018-01-17 Andrew Waterman <andrew@sifive.com>
1932 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1933 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1935 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1938 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1939 call the lto_location_cache before reading the
1940 DECL_SOURCE_LOCATION of the types.
1942 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1943 Richard Sandiford <richard.sandiford@linaro.org>
1945 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1946 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1947 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1948 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1949 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1951 * config/aarch64/constraints.md (aarch64_movti_operand):
1953 * config/aarch64/predicates.md (Uti): Add new constraint.
1955 2018-01-17 Carl Love <cel@us.ibm.com>
1956 * config/rs6000/vsx.md (define_expand xl_len_r,
1957 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1958 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1960 (define_expand, define_insn): Move the shift left from the
1961 define_insn to the define_expand for lxvl and stxvl instructions.
1962 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1963 and XL_LEN_R definitions to PURE.
1965 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1967 * config/i386/i386.c (indirect_thunk_name): Declare regno
1968 as unsigned int. Compare regno with INVALID_REGNUM.
1969 (output_indirect_thunk): Ditto.
1970 (output_indirect_thunk_function): Ditto.
1971 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1972 in the call to output_indirect_thunk_function.
1974 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1977 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1978 rather than the size of inner_type to determine the stack slot size
1979 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1981 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1984 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1987 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1989 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1990 endian Linux systems to optionally enable multilibs for selecting
1991 the long double type if the user configured an explicit type.
1992 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1993 have no long double multilibs if not defined.
1994 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1995 warn if the user used -mabi={ieee,ibm}longdouble and we built
1996 multilibs for long double.
1997 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1998 appropriate multilib option.
1999 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2001 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2002 for building long double multilibs.
2003 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2005 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2007 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2010 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2012 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2015 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2018 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2021 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2023 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2024 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2026 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2028 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2029 different rtl trees depending on TARGET_64BIT.
2030 (rs6000_gen_lvx): Likewise.
2032 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2034 * config/visium/visium.md (nop): Tweak comment.
2035 (hazard_nop): Likewise.
2037 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2039 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2040 -mspeculate-indirect-jumps.
2041 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2042 for -mno-speculate-indirect-jumps.
2043 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2044 (*call_value_indirect_elfv2<mode>): Disable for
2045 -mno-speculate-indirect-jumps.
2046 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2047 (indirect_jump): Emit different RTL for
2048 -mno-speculate-indirect-jumps.
2049 (*indirect_jump<mode>): Disable for
2050 -mno-speculate-indirect-jumps.
2051 (*indirect_jump<mode>_nospec): New define_insn.
2052 (tablejump): Emit different RTL for
2053 -mno-speculate-indirect-jumps.
2054 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2055 (tablejumpsi_nospec): New define_expand.
2056 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2057 (tablejumpdi_nospec): New define_expand.
2058 (*tablejump<mode>_internal1): Disable for
2059 -mno-speculate-indirect-jumps.
2060 (*tablejump<mode>_internal1_nospec): New define_insn.
2061 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2064 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2066 * caller-save.c (insert_save): Drop unnecessary parameter. All
2069 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2070 Richard Biener <rguenth@suse.de>
2073 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2074 return early, inline manually is_gimple_sizepos. Make sure if we
2075 call gimplify_expr we don't end up with a gimple constant.
2076 * tree.c (variably_modified_type_p): Don't return true for
2077 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2078 * gimplify.h (is_gimple_sizepos): Remove.
2080 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2082 PR tree-optimization/83857
2083 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2084 vectorizable_live_operation for pure SLP statements.
2085 (vectorizable_live_operation): Handle PHIs.
2087 2018-01-16 Richard Biener <rguenther@suse.de>
2089 PR tree-optimization/83867
2090 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2091 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2093 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2096 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2097 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2098 If off is not INTEGER_CST, issue a may not be aligned warning
2099 rather than isn't aligned. Use isn%'t rather than isn't.
2100 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2102 <case MULT_EXPR>: Improve the case when bottom and one of the
2103 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2104 operand, in that case check if the other operand is multiple of
2105 bottom divided by the INTEGER_CST operand.
2107 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2110 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2111 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2112 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2113 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2114 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2115 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2116 (pa_function_arg_size): New function.
2118 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2120 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2121 in a separate statement.
2123 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2125 PR tree-optimization/83847
2126 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2127 group gathers and scatters.
2129 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2131 PR rtl-optimization/86620
2132 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2134 PR rtl-optimization/83213
2135 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2136 to last if both are JUMP_INSNs.
2138 PR tree-optimization/83843
2139 * gimple-ssa-store-merging.c
2140 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2141 store_immediate_info for bswap/nop orig_stores.
2143 2018-01-15 Andrew Waterman <andrew@sifive.com>
2145 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2147 <UDIV>: Increase cost if !TARGET_DIV.
2149 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2151 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2152 (define_attr "cr_logical_3op"): New.
2153 (cceq_ior_compare): Adjust.
2154 (cceq_ior_compare_complement): Adjust.
2155 (*cceq_rev_compare): Adjust.
2156 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2157 (is_cracked_insn): Adjust.
2158 (insn_must_be_first_in_group): Adjust.
2159 * config/rs6000/40x.md: Adjust.
2160 * config/rs6000/440.md: Adjust.
2161 * config/rs6000/476.md: Adjust.
2162 * config/rs6000/601.md: Adjust.
2163 * config/rs6000/603.md: Adjust.
2164 * config/rs6000/6xx.md: Adjust.
2165 * config/rs6000/7450.md: Adjust.
2166 * config/rs6000/7xx.md: Adjust.
2167 * config/rs6000/8540.md: Adjust.
2168 * config/rs6000/cell.md: Adjust.
2169 * config/rs6000/e300c2c3.md: Adjust.
2170 * config/rs6000/e500mc.md: Adjust.
2171 * config/rs6000/e500mc64.md: Adjust.
2172 * config/rs6000/e5500.md: Adjust.
2173 * config/rs6000/e6500.md: Adjust.
2174 * config/rs6000/mpc.md: Adjust.
2175 * config/rs6000/power4.md: Adjust.
2176 * config/rs6000/power5.md: Adjust.
2177 * config/rs6000/power6.md: Adjust.
2178 * config/rs6000/power7.md: Adjust.
2179 * config/rs6000/power8.md: Adjust.
2180 * config/rs6000/power9.md: Adjust.
2181 * config/rs6000/rs64.md: Adjust.
2182 * config/rs6000/titan.md: Adjust.
2184 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2186 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2187 ix86_indirect_branch_register logic.
2189 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2191 * config/i386/constraints.md (Bs): Update
2192 ix86_indirect_branch_register check. Don't check
2193 ix86_indirect_branch_register with GOT_memory_operand.
2195 * config/i386/predicates.md (GOT_memory_operand): Don't check
2196 ix86_indirect_branch_register here.
2197 (GOT32_symbol_operand): Likewise.
2199 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2201 * config/i386/predicates.md (constant_call_address_operand):
2202 Rewrite ix86_indirect_branch_register logic.
2203 (sibcall_insn_operand): Likewise.
2205 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2207 * config/i386/constraints.md (Bs): Replace
2208 ix86_indirect_branch_thunk_register with
2209 ix86_indirect_branch_register.
2211 * config/i386/i386.md (indirect_jump): Likewise.
2212 (tablejump): Likewise.
2213 (*sibcall_memory): Likewise.
2214 (*sibcall_value_memory): Likewise.
2215 Peepholes of indirect call and jump via memory: Likewise.
2216 * config/i386/i386.opt: Likewise.
2217 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2218 (GOT_memory_operand): Likewise.
2219 (call_insn_operand): Likewise.
2220 (sibcall_insn_operand): Likewise.
2221 (GOT32_symbol_operand): Likewise.
2223 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2226 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2227 type rather than type addr's type points to.
2228 (expand_omp_atomic_mutex): Likewise.
2229 (expand_omp_atomic): Likewise.
2231 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2234 * config/i386/i386.c (output_indirect_thunk_function): Use
2235 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2236 for __x86_return_thunk.
2238 2018-01-15 Richard Biener <rguenther@suse.de>
2241 * expmed.c (extract_bit_field_1): Fix typo.
2243 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2246 * config/arm/iterators.md (VF): New mode iterator.
2247 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2248 Remove integer-related logic from pattern.
2249 (neon_vabd<mode>_3): Likewise.
2251 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2254 * common.opt (fstrict-overflow): No longer an alias.
2255 (fwrapv-pointer): New option.
2256 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2257 also for pointer types based on flag_wrapv_pointer.
2258 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2259 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2260 opts->x_flag_wrapv got set.
2261 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2262 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2263 POINTER_TYPE_OVERFLOW_UNDEFINED.
2264 * match.pd: Likewise in address comparison pattern.
2265 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2267 2018-01-15 Richard Biener <rguenther@suse.de>
2270 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2271 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2272 Reset type names to their identifier if their TYPE_DECL doesn't
2273 have linkage (and thus is used for ODR and devirt).
2274 (save_debug_info_for_decl): Remove.
2275 (save_debug_info_for_type): Likewise.
2276 (add_tree_to_fld_list): Adjust.
2277 * tree-pretty-print.c (dump_generic_node): Make dumping of
2278 type names more robust.
2280 2018-01-15 Richard Biener <rguenther@suse.de>
2282 * BASE-VER: Bump to 8.0.1.
2284 2018-01-14 Martin Sebor <msebor@redhat.com>
2287 * builtins.c (check_access): Avoid warning when the no-warning bit
2290 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2292 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2293 * ira-color (allocno_hard_regs_compare): Likewise.
2295 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2298 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2299 Use .pushsection/.popsection.
2301 2018-01-14 Martin Sebor <msebor@redhat.com>
2304 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2306 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2308 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2309 entry from extra_headers.
2310 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2311 extra_headers, make the list bitwise identical to the i?86-*-* one.
2313 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2315 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2316 -mcmodel=large with -mindirect-branch=thunk,
2317 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2318 -mfunction-return=thunk-extern.
2319 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2320 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2321 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2323 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2325 * config/i386/i386.c (print_reg): Print the name of the full
2326 integer register without '%'.
2327 (ix86_print_operand): Handle 'V'.
2328 * doc/extend.texi: Document 'V' modifier.
2330 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2332 * config/i386/constraints.md (Bs): Disallow memory operand for
2333 -mindirect-branch-register.
2335 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2336 (GOT_memory_operand): Likewise.
2337 (call_insn_operand): Likewise.
2338 (sibcall_insn_operand): Likewise.
2339 (GOT32_symbol_operand): Likewise.
2340 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2341 for -mindirect-branch-register.
2342 (tablejump): Likewise.
2343 (*sibcall_memory): Likewise.
2344 (*sibcall_value_memory): Likewise.
2345 Disallow peepholes of indirect call and jump via memory for
2346 -mindirect-branch-register.
2347 (*call_pop): Replace m with Bw.
2348 (*call_value_pop): Likewise.
2349 (*sibcall_pop_memory): Replace m with Bs.
2350 * config/i386/i386.opt (mindirect-branch-register): New option.
2351 * doc/invoke.texi: Document -mindirect-branch-register option.
2353 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2355 * config/i386/i386-protos.h (ix86_output_function_return): New.
2356 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2357 set function_return_type.
2358 (indirect_thunk_name): Add ret_p to indicate thunk for function
2360 (output_indirect_thunk_function): Pass false to
2361 indirect_thunk_name.
2362 (ix86_output_indirect_branch_via_reg): Likewise.
2363 (ix86_output_indirect_branch_via_push): Likewise.
2364 (output_indirect_thunk_function): Create alias for function
2365 return thunk if regno < 0.
2366 (ix86_output_function_return): New function.
2367 (ix86_handle_fndecl_attribute): Handle function_return.
2368 (ix86_attribute_table): Add function_return.
2369 * config/i386/i386.h (machine_function): Add
2370 function_return_type.
2371 * config/i386/i386.md (simple_return_internal): Use
2372 ix86_output_function_return.
2373 (simple_return_internal_long): Likewise.
2374 * config/i386/i386.opt (mfunction-return=): New option.
2375 (indirect_branch): Mention -mfunction-return=.
2376 * doc/extend.texi: Document function_return function attribute.
2377 * doc/invoke.texi: Document -mfunction-return= option.
2379 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2381 * config/i386/i386-opts.h (indirect_branch): New.
2382 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2383 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2384 with local indirect jump when converting indirect call and jump.
2385 (ix86_set_indirect_branch_type): New.
2386 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2387 (indirectlabelno): New.
2388 (indirect_thunk_needed): Likewise.
2389 (indirect_thunk_bnd_needed): Likewise.
2390 (indirect_thunks_used): Likewise.
2391 (indirect_thunks_bnd_used): Likewise.
2392 (INDIRECT_LABEL): Likewise.
2393 (indirect_thunk_name): Likewise.
2394 (output_indirect_thunk): Likewise.
2395 (output_indirect_thunk_function): Likewise.
2396 (ix86_output_indirect_branch_via_reg): Likewise.
2397 (ix86_output_indirect_branch_via_push): Likewise.
2398 (ix86_output_indirect_branch): Likewise.
2399 (ix86_output_indirect_jmp): Likewise.
2400 (ix86_code_end): Call output_indirect_thunk_function if needed.
2401 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2403 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2404 (ix86_attribute_table): Add indirect_branch.
2405 * config/i386/i386.h (machine_function): Add indirect_branch_type
2406 and has_local_indirect_jump.
2407 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2409 (tablejump): Likewise.
2410 (*indirect_jump): Use ix86_output_indirect_jmp.
2411 (*tablejump_1): Likewise.
2412 (simple_return_indirect_internal): Likewise.
2413 * config/i386/i386.opt (mindirect-branch=): New option.
2414 (indirect_branch): New.
2417 (thunk-inline): Likewise.
2418 (thunk-extern): Likewise.
2419 * doc/extend.texi: Document indirect_branch function attribute.
2420 * doc/invoke.texi: Document -mindirect-branch= option.
2422 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2425 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2427 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2429 * ipa-inline.c (want_inline_small_function_p): Return false if
2430 inlining has already failed with CIF_FINAL_ERROR.
2431 (update_caller_keys): Call want_inline_small_function_p before
2433 (update_callee_keys): Likewise.
2435 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2437 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2439 (rs6000_quadword_masked_address_p): Likewise.
2440 (quad_aligned_load_p): Likewise.
2441 (quad_aligned_store_p): Likewise.
2442 (const_load_sequence_p): Add comment to describe the outer-most loop.
2443 (mimic_memory_attributes_and_flags): New function.
2444 (rs6000_gen_stvx): Likewise.
2445 (replace_swapped_aligned_store): Likewise.
2446 (rs6000_gen_lvx): Likewise.
2447 (replace_swapped_aligned_load): Likewise.
2448 (replace_swapped_load_constant): Capitalize argument name in
2449 comment describing this function.
2450 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2451 and stores that access quad-word aligned addresses and replace
2452 with stvx or lvx instructions when appropriate.
2453 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2454 New function prototype.
2455 (rs6000_quadword_masked_address_p): Likewise.
2456 (rs6000_gen_lvx): Likewise.
2457 (rs6000_gen_stvx): Likewise.
2458 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2459 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2460 when memory address is aligned.
2461 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2462 this split to select lvx instruction when memory address is aligned.
2463 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2464 instruction when memory address is aligned.
2465 (*vsx_le_perm_load_v16qi): Likewise.
2466 (four unnamed splitters): Modify to select the stvx instruction
2467 when memory is aligned.
2469 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2471 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2472 which appears in the queue multiple times.
2474 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2475 Alan Hayward <alan.hayward@arm.com>
2476 David Sherwood <david.sherwood@arm.com>
2478 * tree-vectorizer.h (vec_lower_bound): New structure.
2479 (_loop_vec_info): Add check_nonzero and lower_bounds.
2480 (LOOP_VINFO_CHECK_NONZERO): New macro.
2481 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2482 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2483 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2484 fields. Make seg_len the distance travelled, not including the
2486 (dr_direction_indicator): Declare.
2487 (dr_zero_step_indicator): Likewise.
2488 (dr_known_forward_stride_p): Likewise.
2489 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2491 (runtime_alias_check_p): Allow runtime alias checks with
2493 (operator ==): Compare access_size and align.
2494 (prune_runtime_alias_test_list): Rework for new distinction between
2495 the access_size and seg_len.
2496 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2498 (get_segment_min_max): New function.
2499 (create_intersect_range_checks): Use it.
2500 (dr_step_indicator): New function.
2501 (dr_direction_indicator): Likewise.
2502 (dr_zero_step_indicator): Likewise.
2503 (dr_known_forward_stride_p): Likewise.
2504 * tree-loop-distribution.c (data_ref_segment_size): Return
2505 DR_STEP * (niters - 1).
2506 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2508 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2509 (vect_preserves_scalar_order_p): New function, split out from...
2510 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2511 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2512 (vect_vfa_access_size): New function.
2513 (vect_vfa_align): Likewise.
2514 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2515 (dump_lower_bound): New function.
2516 (vect_check_lower_bound): Likewise.
2517 (vect_small_gap_p): Likewise.
2518 (vectorizable_with_step_bound_p): Likewise.
2519 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2520 depencies if the vectorization factor is 1. Convert the checks
2521 for nonzero steps into checks on the bounds of DR_STEP. Try using
2522 a bunds check for variable steps if the minimum required step is
2523 relatively small. Update calls to the dr_with_seg_len
2524 constructor and to vect_compile_time_alias.
2525 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2527 (vect_loop_versioning): Call it.
2528 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2530 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2532 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2533 Alan Hayward <alan.hayward@arm.com>
2534 David Sherwood <david.sherwood@arm.com>
2536 * doc/sourcebuild.texi (vect_scatter_store): Document.
2537 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2539 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2541 * genopinit.c (main): Add supports_vec_scatter_store and
2542 supports_vec_scatter_store_cached to target_optabs.
2543 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2544 IFN_MASK_SCATTER_STORE.
2545 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2547 * internal-fn.h (internal_store_fn_p): Declare.
2548 (internal_fn_stored_value_index): Likewise.
2549 * internal-fn.c (scatter_store_direct): New macro.
2550 (expand_scatter_store_optab_fn): New function.
2551 (direct_scatter_store_optab_supported_p): New macro.
2552 (internal_store_fn_p): New function.
2553 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2554 IFN_MASK_SCATTER_STORE.
2555 (internal_fn_mask_index): Likewise.
2556 (internal_fn_stored_value_index): New function.
2557 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2559 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2560 * optabs-query.c (supports_vec_scatter_store_p): New function.
2561 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2562 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2563 true for scatter stores.
2564 (vect_gather_scatter_fn_p): Handle scatter stores too.
2565 (vect_check_gather_scatter): Consider using scatter stores if
2566 supports_vec_scatter_store_p.
2567 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2569 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2570 internal_fn_stored_value_index.
2571 (check_load_store_masking): Handle scatter stores too.
2572 (vect_get_store_rhs): Make public.
2573 (vectorizable_call): Use internal_store_fn_p.
2574 (vectorizable_store): Handle scatter store internal functions.
2575 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2576 when deciding whether the end of the group has been reached.
2577 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2578 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2579 (mask_scatter_store<mode>): New insns.
2581 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2582 Alan Hayward <alan.hayward@arm.com>
2583 David Sherwood <david.sherwood@arm.com>
2585 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2586 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2587 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2589 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2590 Use vect_truncate_gather_scatter_offset if we can't treat the
2591 operation as a normal gather load or scatter store.
2592 (get_group_load_store_type): Take the gather_scatter_info
2593 as argument. Try using a gather load or scatter store for
2594 single-element groups.
2595 (get_load_store_type): Update calls to get_group_load_store_type
2596 and vect_use_strided_gather_scatters_p.
2598 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2599 Alan Hayward <alan.hayward@arm.com>
2600 David Sherwood <david.sherwood@arm.com>
2602 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2603 optional tree argument.
2604 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2606 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2607 but continue to use the current value as a fallback.
2608 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2609 to compare the updates.
2610 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2611 (get_load_store_type): Use it when handling a strided access.
2612 (vect_get_strided_load_store_ops): New function.
2613 (vect_get_data_ptr_increment): Likewise.
2614 (vectorizable_load): Handle strided gather loads. Always pass
2615 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2617 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2618 Alan Hayward <alan.hayward@arm.com>
2619 David Sherwood <david.sherwood@arm.com>
2621 * doc/md.texi (gather_load@var{m}): Document.
2622 (mask_gather_load@var{m}): Likewise.
2623 * genopinit.c (main): Add supports_vec_gather_load and
2624 supports_vec_gather_load_cached to target_optabs.
2625 * optabs-tree.c (init_tree_optimization_optabs): Use
2626 ggc_cleared_alloc to allocate target_optabs.
2627 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2628 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2630 * internal-fn.h (internal_load_fn_p): Declare.
2631 (internal_gather_scatter_fn_p): Likewise.
2632 (internal_fn_mask_index): Likewise.
2633 (internal_gather_scatter_fn_supported_p): Likewise.
2634 * internal-fn.c (gather_load_direct): New macro.
2635 (expand_gather_load_optab_fn): New function.
2636 (direct_gather_load_optab_supported_p): New macro.
2637 (direct_internal_fn_optab): New function.
2638 (internal_load_fn_p): Likewise.
2639 (internal_gather_scatter_fn_p): Likewise.
2640 (internal_fn_mask_index): Likewise.
2641 (internal_gather_scatter_fn_supported_p): Likewise.
2642 * optabs-query.c (supports_at_least_one_mode_p): New function.
2643 (supports_vec_gather_load_p): Likewise.
2644 * optabs-query.h (supports_vec_gather_load_p): Declare.
2645 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2646 and memory_type field.
2647 (NUM_PATTERNS): Bump to 15.
2648 * tree-vect-data-refs.c: Include internal-fn.h.
2649 (vect_gather_scatter_fn_p): New function.
2650 (vect_describe_gather_scatter_call): Likewise.
2651 (vect_check_gather_scatter): Try using internal functions for
2652 gather loads. Recognize existing calls to a gather load function.
2653 (vect_analyze_data_refs): Consider using gather loads if
2654 supports_vec_gather_load_p.
2655 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2656 (vect_get_gather_scatter_offset_type): Likewise.
2657 (vect_convert_mask_for_vectype): Likewise.
2658 (vect_add_conversion_to_patterm): Likewise.
2659 (vect_try_gather_scatter_pattern): Likewise.
2660 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2661 (vect_vect_recog_func_ptrs): Add it.
2662 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2663 internal_fn_mask_index and internal_gather_scatter_fn_p.
2664 (check_load_store_masking): Take the gather_scatter_info as an
2665 argument and handle gather loads.
2666 (vect_get_gather_scatter_ops): New function.
2667 (vectorizable_call): Check internal_load_fn_p.
2668 (vectorizable_load): Likewise. Handle gather load internal
2670 (vectorizable_store): Update call to check_load_store_masking.
2671 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2672 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2673 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2674 (aarch64_gather_scale_operand_d): New predicates.
2675 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2676 (mask_gather_load<mode>): New insns.
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 * optabs.def (fold_left_plus_optab): New optab.
2683 * doc/md.texi (fold_left_plus_@var{m}): Document.
2684 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2685 * internal-fn.c (fold_left_direct): Define.
2686 (expand_fold_left_optab_fn): Likewise.
2687 (direct_fold_left_optab_supported_p): Likewise.
2688 * fold-const-call.c (fold_const_fold_left): New function.
2689 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2690 * tree-parloops.c (valid_reduction_p): New function.
2691 (gather_scalar_reductions): Use it.
2692 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2693 (vect_finish_replace_stmt): Declare.
2694 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2695 (needs_fold_left_reduction_p): New function, split out from...
2696 (vect_is_simple_reduction): ...here. Accept reductions that
2697 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2698 (vect_force_simple_reduction): Also store the reduction type in
2699 the assignment's STMT_VINFO_REDUC_TYPE.
2700 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2701 (merge_with_identity): New function.
2702 (vect_expand_fold_left): Likewise.
2703 (vectorize_fold_left_reduction): Likewise.
2704 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2705 scalar phi in place for it. Check for target support and reject
2706 cases that would reassociate the operation. Defer the transform
2707 phase to vectorize_fold_left_reduction.
2708 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2709 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2710 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2714 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2715 call to ifc_temp_var.
2717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2718 Alan Hayward <alan.hayward@arm.com>
2719 David Sherwood <david.sherwood@arm.com>
2721 * target.def (legitimize_address_displacement): Take the original
2722 offset as a poly_int.
2723 * targhooks.h (default_legitimize_address_displacement): Update
2725 * targhooks.c (default_legitimize_address_displacement): Likewise.
2726 * doc/tm.texi: Regenerate.
2727 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2728 as an argument, moving assert of ad->disp == ad->disp_term to...
2729 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2730 Try calling targetm.legitimize_address_displacement before expanding
2731 the address rather than afterwards, and adjust for the new interface.
2732 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2733 Match the new hook interface. Handle SVE addresses.
2734 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2737 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2739 * Makefile.in (OBJS): Add early-remat.o.
2740 * target.def (select_early_remat_modes): New hook.
2741 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2742 * doc/tm.texi: Regenerate.
2743 * targhooks.h (default_select_early_remat_modes): Declare.
2744 * targhooks.c (default_select_early_remat_modes): New function.
2745 * timevar.def (TV_EARLY_REMAT): New timevar.
2746 * passes.def (pass_early_remat): New pass.
2747 * tree-pass.h (make_pass_early_remat): Declare.
2748 * early-remat.c: New file.
2749 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2751 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2753 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2754 Alan Hayward <alan.hayward@arm.com>
2755 David Sherwood <david.sherwood@arm.com>
2757 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2758 vfm1 with a bound_epilog parameter.
2759 (vect_do_peeling): Update calls accordingly, and move the prologue
2760 call earlier in the function. Treat the base bound_epilog as 0 for
2761 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2762 this base when peeling for gaps.
2763 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2764 with fully-masked loops.
2765 (vect_estimate_min_profitable_iters): Handle the single peeled
2766 iteration in that case.
2768 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2769 Alan Hayward <alan.hayward@arm.com>
2770 David Sherwood <david.sherwood@arm.com>
2772 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2773 single-element interleaving even if the size is not a power of 2.
2774 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2775 accesses for single-element interleaving if the group size is
2778 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2779 Alan Hayward <alan.hayward@arm.com>
2780 David Sherwood <david.sherwood@arm.com>
2782 * doc/md.texi (fold_extract_last_@var{m}): Document.
2783 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2784 * optabs.def (fold_extract_last_optab): New optab.
2785 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2786 * internal-fn.c (fold_extract_direct): New macro.
2787 (expand_fold_extract_optab_fn): Likewise.
2788 (direct_fold_extract_optab_supported_p): Likewise.
2789 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2790 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2791 EXTRACT_LAST_REDUCTION.
2792 (get_initial_def_for_reduction): Do not create an initial vector
2793 for EXTRACT_LAST_REDUCTION reductions.
2794 (vectorizable_reduction): Leave the scalar phi in place for
2795 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2796 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2797 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2798 transform phase to vectorizable_condition.
2799 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2801 (vect_finish_stmt_generation): ...here.
2802 (vect_finish_replace_stmt): New function.
2803 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2804 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2806 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2808 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2809 Alan Hayward <alan.hayward@arm.com>
2810 David Sherwood <david.sherwood@arm.com>
2812 * doc/md.texi (extract_last_@var{m}): Document.
2813 * optabs.def (extract_last_optab): New optab.
2814 * internal-fn.def (EXTRACT_LAST): New internal function.
2815 * internal-fn.c (cond_unary_direct): New macro.
2816 (expand_cond_unary_optab_fn): Likewise.
2817 (direct_cond_unary_optab_supported_p): Likewise.
2818 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2819 loops using EXTRACT_LAST.
2820 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2821 (extract_last_<mode>): ...this optab.
2822 (vec_extract<mode><Vel>): Update accordingly.
2824 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2825 Alan Hayward <alan.hayward@arm.com>
2826 David Sherwood <david.sherwood@arm.com>
2828 * target.def (empty_mask_is_expensive): New hook.
2829 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2830 * doc/tm.texi: Regenerate.
2831 * targhooks.h (default_empty_mask_is_expensive): Declare.
2832 * targhooks.c (default_empty_mask_is_expensive): New function.
2833 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2834 if the target says that empty masks are expensive.
2835 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2837 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2839 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2840 Alan Hayward <alan.hayward@arm.com>
2841 David Sherwood <david.sherwood@arm.com>
2843 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2844 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2845 (vect_use_loop_mask_for_alignment_p): New function.
2846 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2847 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2848 niters_skip argument. Make sure that the first niters_skip elements
2849 of the first iteration are inactive.
2850 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2851 Update call to vect_set_loop_masks_directly.
2852 (get_misalign_in_elems): New function, split out from...
2853 (vect_gen_prolog_loop_niters): ...here.
2854 (vect_update_init_of_dr): Take a code argument that specifies whether
2855 the adjustment should be added or subtracted.
2856 (vect_update_init_of_drs): Likewise.
2857 (vect_prepare_for_masked_peels): New function.
2858 (vect_do_peeling): Skip prologue peeling if we're using a mask
2859 instead. Update call to vect_update_inits_of_drs.
2860 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2862 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2863 alignment. Do not include the number of peeled iterations in
2864 the minimum threshold in that case.
2865 (vectorizable_induction): Adjust the start value down by
2866 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2867 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2868 Take the number of skipped iterations into account when calculating
2870 * tree-vect-stmts.c (vect_gen_while_not): New function.
2872 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2873 Alan Hayward <alan.hayward@arm.com>
2874 David Sherwood <david.sherwood@arm.com>
2876 * doc/sourcebuild.texi (vect_fully_masked): Document.
2877 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2879 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2881 (vect_analyze_loop_2): ...here. Don't check the vectorization
2882 factor against the number of loop iterations if the loop is
2885 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2886 Alan Hayward <alan.hayward@arm.com>
2887 David Sherwood <david.sherwood@arm.com>
2889 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2890 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2891 (dump_groups): Update accordingly.
2892 (iv_use::mem_type): New member variable.
2893 (address_p): New function.
2894 (record_use): Add a mem_type argument and initialize the new
2896 (record_group_use): Add a mem_type argument. Use address_p.
2897 Remove obsolete null checks of base_object. Update call to record_use.
2898 (find_interesting_uses_op): Update call to record_group_use.
2899 (find_interesting_uses_cond): Likewise.
2900 (find_interesting_uses_address): Likewise.
2901 (get_mem_type_for_internal_fn): New function.
2902 (find_address_like_use): Likewise.
2903 (find_interesting_uses_stmt): Try find_address_like_use before
2904 calling find_interesting_uses_op.
2905 (addr_offset_valid_p): Use the iv mem_type field as the type
2906 of the addressed memory.
2907 (add_autoinc_candidates): Likewise.
2908 (get_address_cost): Likewise.
2909 (split_small_address_groups_p): Use address_p.
2910 (split_address_groups): Likewise.
2911 (add_iv_candidate_for_use): Likewise.
2912 (autoinc_possible_for_pair): Likewise.
2913 (rewrite_groups): Likewise.
2914 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2915 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2916 (get_alias_ptr_type_for_ptr_address): New function.
2917 (rewrite_use_address): Rewrite address uses in calls that were
2918 identified by find_address_like_use.
2920 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2921 Alan Hayward <alan.hayward@arm.com>
2922 David Sherwood <david.sherwood@arm.com>
2924 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2926 * gimple-expr.h (is_gimple_addressable: Likewise.
2927 * gimple-expr.c (is_gimple_address): Likewise.
2928 * internal-fn.c (expand_call_mem_ref): New function.
2929 (expand_mask_load_optab_fn): Use it.
2930 (expand_mask_store_optab_fn): Likewise.
2932 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2933 Alan Hayward <alan.hayward@arm.com>
2934 David Sherwood <david.sherwood@arm.com>
2936 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2937 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2938 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2939 (cond_umax@var{mode}): Document.
2940 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2941 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2942 (cond_umin_optab, cond_umax_optab): New optabs.
2943 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2944 (COND_IOR, COND_XOR): New internal functions.
2945 * internal-fn.h (get_conditional_internal_fn): Declare.
2946 * internal-fn.c (cond_binary_direct): New macro.
2947 (expand_cond_binary_optab_fn): Likewise.
2948 (direct_cond_binary_optab_supported_p): Likewise.
2949 (get_conditional_internal_fn): New function.
2950 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2951 Cope with reduction statements that are vectorized as calls rather
2953 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2954 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2955 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2956 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2957 (UNSPEC_COND_EOR): New unspecs.
2958 (optab): Add mappings for them.
2959 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2960 (sve_int_op, sve_fp_op): New int attributes.
2962 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2963 Alan Hayward <alan.hayward@arm.com>
2964 David Sherwood <david.sherwood@arm.com>
2966 * optabs.def (while_ult_optab): New optab.
2967 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2968 * internal-fn.def (WHILE_ULT): New internal function.
2969 * internal-fn.h (direct_internal_fn_supported_p): New override
2970 that takes two types as argument.
2971 * internal-fn.c (while_direct): New macro.
2972 (expand_while_optab_fn): New function.
2973 (convert_optab_supported_p): Likewise.
2974 (direct_while_optab_supported_p): New macro.
2975 * wide-int.h (wi::udiv_ceil): New function.
2976 * tree-vectorizer.h (rgroup_masks): New structure.
2977 (vec_loop_masks): New typedef.
2978 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2980 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2981 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2982 (vect_max_vf): New function.
2983 (slpeel_make_loop_iterate_ntimes): Delete.
2984 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2985 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2986 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2987 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2988 internal-fn.h, stor-layout.h and optabs-query.h.
2989 (vect_set_loop_mask): New function.
2990 (add_preheader_seq): Likewise.
2991 (add_header_seq): Likewise.
2992 (interleave_supported_p): Likewise.
2993 (vect_maybe_permute_loop_masks): Likewise.
2994 (vect_set_loop_masks_directly): Likewise.
2995 (vect_set_loop_condition_masked): Likewise.
2996 (vect_set_loop_condition_unmasked): New function, split out from
2997 slpeel_make_loop_iterate_ntimes.
2998 (slpeel_make_loop_iterate_ntimes): Rename to..
2999 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3000 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3001 (vect_do_peeling): Update call accordingly.
3002 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3004 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3005 mask_compare_type, can_fully_mask_p and fully_masked_p.
3006 (release_vec_loop_masks): New function.
3007 (_loop_vec_info): Use it to free the loop masks.
3008 (can_produce_all_loop_masks_p): New function.
3009 (vect_get_max_nscalars_per_iter): Likewise.
3010 (vect_verify_full_masking): Likewise.
3011 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3012 retries, and free the mask rgroups before retrying. Check loop-wide
3013 reasons for disallowing fully-masked loops. Make the final decision
3014 about whether use a fully-masked loop or not.
3015 (vect_estimate_min_profitable_iters): Do not assume that peeling
3016 for the number of iterations will be needed for fully-masked loops.
3017 (vectorizable_reduction): Disable fully-masked loops.
3018 (vectorizable_live_operation): Likewise.
3019 (vect_halve_mask_nunits): New function.
3020 (vect_double_mask_nunits): Likewise.
3021 (vect_record_loop_mask): Likewise.
3022 (vect_get_loop_mask): Likewise.
3023 (vect_transform_loop): Handle the case in which the final loop
3024 iteration might handle a partial vector. Call vect_set_loop_condition
3025 instead of slpeel_make_loop_iterate_ntimes.
3026 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3027 (check_load_store_masking): New function.
3028 (prepare_load_store_mask): Likewise.
3029 (vectorizable_store): Handle fully-masked loops.
3030 (vectorizable_load): Likewise.
3031 (supportable_widening_operation): Use vect_halve_mask_nunits for
3033 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3034 (vect_gen_while): New function.
3035 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3036 (aarch64_uqdec<mode>): New insn.
3038 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3039 Alan Hayward <alan.hayward@arm.com>
3040 David Sherwood <david.sherwood@arm.com>
3042 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3043 (reduc_xor_scal_optab): New optabs.
3044 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3045 (reduc_xor_scal_@var{m}): Document.
3046 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3047 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3049 * fold-const-call.c (fold_const_call): Handle them.
3050 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3051 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3052 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3053 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3054 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3055 (UNSPEC_XORV): New unspecs.
3056 (optab): Add entries for them.
3057 (BITWISEV): New int iterator.
3058 (bit_reduc_op): New int attributes.
3060 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3061 Alan Hayward <alan.hayward@arm.com>
3062 David Sherwood <david.sherwood@arm.com>
3064 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3065 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3066 * optabs.def (vec_shl_insert_optab): New optab.
3067 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3068 (duplicate_and_interleave): Likewise.
3069 * tree-vect-loop.c: Include internal-fn.h.
3070 (neutral_op_for_slp_reduction): New function, split out from
3071 get_initial_defs_for_reduction.
3072 (get_initial_def_for_reduction): Handle option 2 for variable-length
3073 vectors by loading the neutral value into a vector and then shifting
3074 the initial value into element 0.
3075 (get_initial_defs_for_reduction): Replace the code argument with
3076 the neutral value calculated by neutral_op_for_slp_reduction.
3077 Use gimple_build_vector for constant-length vectors.
3078 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3079 but the first group_size elements have a neutral value.
3080 Use duplicate_and_interleave otherwise.
3081 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3082 Update call to get_initial_defs_for_reduction. Handle SLP
3083 reductions for variable-length vectors by creating one vector
3084 result for each scalar result, with the elements associated
3085 with other scalar results stubbed out with the neutral value.
3086 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3087 Require IFN_VEC_SHL_INSERT for double reductions on
3088 variable-length vectors, or SLP reductions that have
3089 a neutral value. Require can_duplicate_and_interleave_p
3090 support for variable-length unchained SLP reductions if there
3091 is no neutral value, such as for MIN/MAX reductions. Also require
3092 the number of vector elements to be a multiple of the number of
3093 SLP statements when doing variable-length unchained SLP reductions.
3094 Update call to vect_create_epilog_for_reduction.
3095 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3096 and remove initial values.
3097 (duplicate_and_interleave): Make public.
3098 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3099 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3101 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3102 Alan Hayward <alan.hayward@arm.com>
3103 David Sherwood <david.sherwood@arm.com>
3105 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3106 (can_duplicate_and_interleave_p): New function.
3107 (vect_get_and_check_slp_defs): Take the vector of statements
3108 rather than just the current one. Remove excess parentheses.
3109 Restriction rejectinon of vect_constant_def and vect_external_def
3110 for variable-length vectors to boolean types, or types for which
3111 can_duplicate_and_interleave_p is false.
3112 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3113 (duplicate_and_interleave): New function.
3114 (vect_get_constant_vectors): Use gimple_build_vector for
3115 constant-length vectors and suitable variable-length constant
3116 vectors. Use duplicate_and_interleave for other variable-length
3117 vectors. Don't defer the update when inserting new statements.
3119 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3120 Alan Hayward <alan.hayward@arm.com>
3121 David Sherwood <david.sherwood@arm.com>
3123 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3124 min_profitable_iters doesn't go negative.
3126 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3127 Alan Hayward <alan.hayward@arm.com>
3128 David Sherwood <david.sherwood@arm.com>
3130 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3131 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3132 * optabs.def (vec_mask_load_lanes_optab): New optab.
3133 (vec_mask_store_lanes_optab): Likewise.
3134 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3135 (MASK_STORE_LANES): Likewise.
3136 * internal-fn.c (mask_load_lanes_direct): New macro.
3137 (mask_store_lanes_direct): Likewise.
3138 (expand_mask_load_optab_fn): Handle masked operations.
3139 (expand_mask_load_lanes_optab_fn): New macro.
3140 (expand_mask_store_optab_fn): Handle masked operations.
3141 (expand_mask_store_lanes_optab_fn): New macro.
3142 (direct_mask_load_lanes_optab_supported_p): Likewise.
3143 (direct_mask_store_lanes_optab_supported_p): Likewise.
3144 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3146 (vect_load_lanes_supported): Likewise.
3147 * tree-vect-data-refs.c (strip_conversion): New function.
3148 (can_group_stmts_p): Likewise.
3149 (vect_analyze_data_ref_accesses): Use it instead of checking
3150 for a pair of assignments.
3151 (vect_store_lanes_supported): Take a masked_p parameter.
3152 (vect_load_lanes_supported): Likewise.
3153 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3154 vect_store_lanes_supported and vect_load_lanes_supported.
3155 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3156 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3157 parameter. Don't allow gaps for masked accesses.
3158 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3159 and vect_load_lanes_supported.
3160 (get_load_store_type): Take a masked_p parameter and update
3161 call to get_group_load_store_type.
3162 (vectorizable_store): Update call to get_load_store_type.
3163 Handle IFN_MASK_STORE_LANES.
3164 (vectorizable_load): Update call to get_load_store_type.
3165 Handle IFN_MASK_LOAD_LANES.
3167 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3168 Alan Hayward <alan.hayward@arm.com>
3169 David Sherwood <david.sherwood@arm.com>
3171 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3173 * config/aarch64/aarch64-protos.h
3174 (aarch64_sve_struct_memory_operand_p): Declare.
3175 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3176 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3177 (VPRED, vpred): Handle SVE structure modes.
3178 * config/aarch64/constraints.md (Utx): New constraint.
3179 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3180 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3181 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3182 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3183 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3184 structure modes. Split into pieces after RA.
3185 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3186 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3188 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3189 SVE structure modes.
3190 (aarch64_classify_address): Likewise.
3191 (sizetochar): Move earlier in file.
3192 (aarch64_print_operand): Handle SVE register lists.
3193 (aarch64_array_mode): New function.
3194 (aarch64_sve_struct_memory_operand_p): Likewise.
3195 (TARGET_ARRAY_MODE): Redefine.
3197 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3198 Alan Hayward <alan.hayward@arm.com>
3199 David Sherwood <david.sherwood@arm.com>
3201 * target.def (array_mode): New target hook.
3202 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3203 * doc/tm.texi: Regenerate.
3204 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3205 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3206 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3208 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3211 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3212 Alan Hayward <alan.hayward@arm.com>
3213 David Sherwood <david.sherwood@arm.com>
3215 * fold-const.c (fold_binary_loc): Check the argument types
3216 rather than the result type when testing for a vector operation.
3218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3220 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3221 * doc/tm.texi: Regenerate.
3223 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3224 Alan Hayward <alan.hayward@arm.com>
3225 David Sherwood <david.sherwood@arm.com>
3227 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3228 (sve): Document new AArch64 extension.
3229 * doc/md.texi (w): Extend the description of the AArch64
3230 constraint to include SVE vectors.
3231 (Upl, Upa): Document new AArch64 predicate constraints.
3232 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3234 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3235 (msve-vector-bits=): New option.
3236 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3237 SVE when these are disabled.
3238 (sve): New extension.
3239 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3240 modes. Adjust their number of units based on aarch64_sve_vg.
3241 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3242 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3243 aarch64_addr_query_type.
3244 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3245 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3246 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3247 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3248 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3249 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3250 (aarch64_simd_imm_zero_p): Delete.
3251 (aarch64_check_zero_based_sve_index_immediate): Declare.
3252 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3253 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3254 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3255 (aarch64_sve_float_mul_immediate_p): Likewise.
3256 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3258 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3259 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3260 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3261 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3262 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3263 (aarch64_regmode_natural_size): Likewise.
3264 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3265 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3267 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3268 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3269 for VG and the SVE predicate registers.
3270 (V_ALIASES): Add a "z"-prefixed alias.
3271 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3272 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3273 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3274 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3275 (REG_CLASS_NAMES): Add entries for them.
3276 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3277 and the predicate registers.
3278 (aarch64_sve_vg): Declare.
3279 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3280 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3281 (REGMODE_NATURAL_SIZE): Define.
3282 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3284 * config/aarch64/aarch64.c: Include cfgrtl.h.
3285 (simd_immediate_info): Add a constructor for series vectors,
3286 and an associated step field.
3287 (aarch64_sve_vg): New variable.
3288 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3289 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3290 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3291 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3292 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3293 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3294 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3295 (aarch64_get_mask_mode): New functions.
3296 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3297 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3298 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3299 predicate modes and predicate registers. Explicitly restrict
3300 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3301 to store a vector mode if it is recognized by
3302 aarch64_classify_vector_mode.
3303 (aarch64_regmode_natural_size): New function.
3304 (aarch64_hard_regno_caller_save_mode): Return the original mode
3306 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3307 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3308 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3309 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3311 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3312 does not overlap dest if the function is frame-related. Handle
3314 (aarch64_split_add_offset): New function.
3315 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3316 them aarch64_add_offset.
3317 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3318 and update call to aarch64_sub_sp.
3319 (aarch64_add_cfa_expression): New function.
3320 (aarch64_expand_prologue): Pass extra temporary registers to the
3321 functions above. Handle the case in which we need to emit new
3322 DW_CFA_expressions for registers that were originally saved
3323 relative to the stack pointer, but now have to be expressed
3324 relative to the frame pointer.
3325 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3327 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3328 IP0 and IP1 values for SVE frames.
3329 (aarch64_expand_vec_series): New function.
3330 (aarch64_expand_sve_widened_duplicate): Likewise.
3331 (aarch64_expand_sve_const_vector): Likewise.
3332 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3333 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3334 into the register, rather than emitting a SET directly.
3335 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3336 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3337 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3338 (offset_9bit_signed_scaled_p): New functions.
3339 (aarch64_replicate_bitmask_imm): New function.
3340 (aarch64_bitmask_imm): Use it.
3341 (aarch64_cannot_force_const_mem): Reject expressions involving
3342 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3343 (aarch64_classify_index): Handle SVE indices, by requiring
3344 a plain register index with a scale that matches the element size.
3345 (aarch64_classify_address): Handle SVE addresses. Assert that
3346 the mode of the address is VOIDmode or an integer mode.
3347 Update call to aarch64_classify_symbol.
3348 (aarch64_classify_symbolic_expression): Update call to
3349 aarch64_classify_symbol.
3350 (aarch64_const_vec_all_in_range_p): New function.
3351 (aarch64_print_vector_float_operand): Likewise.
3352 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3353 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3354 and the FP immediates 1.0 and 0.5.
3355 (aarch64_print_address_internal): Handle SVE addresses.
3356 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3357 (aarch64_regno_regclass): Handle predicate registers.
3358 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3360 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3361 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3362 (aarch64_convert_sve_vector_bits): New function.
3363 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3364 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3366 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3367 Handle SVE vector and predicate modes. Accept VL-based constants
3368 that need only one temporary register, and VL offsets that require
3369 no temporary registers.
3370 (aarch64_conditional_register_usage): Mark the predicate registers
3371 as fixed if SVE isn't available.
3372 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3373 Return true for SVE vector and predicate modes.
3374 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3375 rather than an unsigned int. Handle SVE modes.
3376 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3378 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3380 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3381 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3382 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3383 (aarch64_sve_float_mul_immediate_p): New functions.
3384 (aarch64_sve_valid_immediate): New function.
3385 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3386 Explicitly reject structure modes. Check for INDEX constants.
3387 Handle PTRUE and PFALSE constants.
3388 (aarch64_check_zero_based_sve_index_immediate): New function.
3389 (aarch64_simd_imm_zero_p): Delete.
3390 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3391 vector modes. Accept constants in the range of CNT[BHWD].
3392 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3393 ask for an Advanced SIMD mode.
3394 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3395 (aarch64_simd_vector_alignment): Handle SVE predicates.
3396 (aarch64_vectorize_preferred_vector_alignment): New function.
3397 (aarch64_simd_vector_alignment_reachable): Use it instead of
3399 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3400 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3402 (MAX_VECT_LEN): Delete.
3403 (expand_vec_perm_d): Add a vec_flags field.
3404 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3405 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3406 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3408 (aarch64_evpc_rev): Rename to...
3409 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3410 (aarch64_evpc_rev_global): New function.
3411 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3412 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3414 (aarch64_evpc_sve_tbl): New function.
3415 (aarch64_expand_vec_perm_const_1): Update after rename of
3416 aarch64_evpc_rev. Handle SVE permutes too, trying
3417 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3418 than aarch64_evpc_tbl.
3419 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3420 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3421 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3422 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3423 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3424 (aarch64_expand_sve_vcond): New functions.
3425 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3426 of aarch64_vector_mode_p.
3427 (aarch64_dwarf_poly_indeterminate_value): New function.
3428 (aarch64_compute_pressure_classes): Likewise.
3429 (aarch64_can_change_mode_class): Likewise.
3430 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3431 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3432 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3433 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3434 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3435 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3436 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3437 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3439 (Dn, Dl, Dr): Accept const as well as const_vector.
3440 (Dz): Likewise. Compare against CONST0_RTX.
3441 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3442 of "vector" where appropriate.
3443 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3444 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3445 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3446 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3447 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3448 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3449 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3450 (v_int_equiv): Extend to SVE modes.
3451 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3453 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3454 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3455 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3456 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3457 (SVE_COND_FP_CMP): New int iterators.
3458 (perm_hilo): Handle the new unpack unspecs.
3459 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3461 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3462 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3463 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3464 (aarch64_equality_operator, aarch64_constant_vector_operand)
3465 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3466 (aarch64_sve_nonimmediate_operand): Likewise.
3467 (aarch64_sve_general_operand): Likewise.
3468 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3469 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3470 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3471 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3472 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3473 (aarch64_sve_float_arith_immediate): Likewise.
3474 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3475 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3476 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3477 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3478 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3479 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3480 (aarch64_sve_float_arith_operand): Likewise.
3481 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3482 (aarch64_sve_float_mul_operand): Likewise.
3483 (aarch64_sve_vec_perm_operand): Likewise.
3484 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3485 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3486 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3487 as well as const_vector.
3488 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3489 in file. Use CONST0_RTX and CONSTM1_RTX.
3490 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3491 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3492 Use aarch64_simd_imm_zero.
3493 * config/aarch64/aarch64-sve.md: New file.
3494 * config/aarch64/aarch64.md: Include it.
3495 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3496 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3497 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3498 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3499 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3500 (sve): New attribute.
3501 (enabled): Disable instructions with the sve attribute unless
3503 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3504 aarch64_expand_mov_immediate.
3505 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3506 CNT[BHSD] immediates.
3507 (movti): Split CONST_POLY_INT moves into two halves.
3508 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3509 Split additions that need a temporary here if the destination
3510 is the stack pointer.
3511 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3512 (*add<mode>3_poly_1): New instruction.
3513 (set_clobber_cc): New expander.
3515 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3517 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3518 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3519 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3520 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3521 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3522 Change innermode from fixed_mode_size to machine_mode.
3523 (simplify_subreg): Update call accordingly. Handle a constant-sized
3524 subreg of a variable-length CONST_VECTOR.
3526 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3527 Alan Hayward <alan.hayward@arm.com>
3528 David Sherwood <david.sherwood@arm.com>
3530 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3531 (add_offset_to_base): New function, split out from...
3532 (create_mem_ref): ...here. When handling a scale other than 1,
3533 check first whether the address is valid without the offset.
3534 Add it into the base if so, leaving the index and scale as-is.
3536 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3539 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3540 fold_for_warn before checking if arg2 is INTEGER_CST.
3542 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3544 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3545 (store_multiple_operation): Delete.
3546 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3547 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3548 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3549 guarded by TARGET_STRING.
3550 (rs6000_output_load_multiple): Delete.
3551 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3552 OPTION_MASK_STRING / TARGET_STRING handling.
3553 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3554 (const rs6000_opt_masks) <"string">: Change mask to 0.
3555 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3556 (MASK_STRING): Delete.
3557 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3559 (load_multiple): Delete.
3566 (store_multiple): Delete.
3573 (movmemsi_8reg): Delete.
3574 (corresponding unnamed define_insn): Delete.
3575 (movmemsi_6reg): Delete.
3576 (corresponding unnamed define_insn): Delete.
3577 (movmemsi_4reg): Delete.
3578 (corresponding unnamed define_insn): Delete.
3579 (movmemsi_2reg): Delete.
3580 (corresponding unnamed define_insn): Delete.
3581 (movmemsi_1reg): Delete.
3582 (corresponding unnamed define_insn): Delete.
3583 * config/rs6000/rs6000.opt (mno-string): New.
3584 (mstring): Replace by deprecation warning stub.
3585 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3587 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3589 * regrename.c (regrename_do_replace): If replacing the same
3590 reg multiple times, try to reuse last created gen_raw_REG.
3593 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3594 main to workaround a bug in GDB.
3596 2018-01-12 Tom de Vries <tom@codesourcery.com>
3599 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3601 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3603 PR rtl-optimization/80481
3604 * ira-color.c (get_cap_member): New function.
3605 (allocnos_conflict_by_live_ranges_p): Use it.
3606 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3607 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3609 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3612 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3613 (*saddl_se_1): Ditto.
3615 (*ssubl_se_1): Ditto.
3617 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3619 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3620 rather than wi::to_widest for DR_INITs.
3621 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3622 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3623 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3625 (vect_analyze_group_access_1): Note that here.
3627 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3629 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3630 polynomial type sizes.
3632 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3634 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3635 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3636 (gimple_add_tmp_var): Likewise.
3638 2018-01-12 Martin Liska <mliska@suse.cz>
3640 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3641 (gimple_alloc_sizes): Likewise.
3642 (dump_gimple_statistics): Use PRIu64 in printf format.
3643 * gimple.h: Change uint64_t to int.
3645 2018-01-12 Martin Liska <mliska@suse.cz>
3647 * tree-core.h: Use uint64_t instead of int.
3648 * tree.c (tree_node_counts): Likewise.
3649 (tree_node_sizes): Likewise.
3650 (dump_tree_statistics): Use PRIu64 in printf format.
3652 2018-01-12 Martin Liska <mliska@suse.cz>
3654 * Makefile.in: As qsort_chk is implemented in vec.c, add
3655 vec.o to linkage of gencfn-macros.
3656 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3657 passing the info to record_node_allocation_statistics.
3658 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3660 * ggc-common.c (struct ggc_usage): Add operator== and use
3661 it in operator< and compare function.
3662 * mem-stats.h (struct mem_usage): Likewise.
3663 * vec.c (struct vec_usage): Remove operator< and compare
3664 function. Can be simply inherited.
3666 2018-01-12 Martin Jambor <mjambor@suse.cz>
3669 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3670 * tree-ssa-math-opts.c: Include domwalk.h.
3671 (convert_mult_to_fma_1): New function.
3672 (fma_transformation_info): New type.
3673 (fma_deferring_state): Likewise.
3674 (cancel_fma_deferring): New function.
3675 (result_of_phi): Likewise.
3676 (last_fma_candidate_feeds_initial_phi): Likewise.
3677 (convert_mult_to_fma): Added deferring logic, split actual
3678 transformation to convert_mult_to_fma_1.
3679 (math_opts_dom_walker): New type.
3680 (math_opts_dom_walker::after_dom_children): New method, body moved
3681 here from pass_optimize_widening_mul::execute, added deferring logic
3683 (pass_optimize_widening_mul::execute): Moved most of code to
3684 math_opts_dom_walker::after_dom_children.
3685 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3686 * config/i386/i386.c (ix86_option_override_internal): Added
3687 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3689 2018-01-12 Richard Biener <rguenther@suse.de>
3692 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3693 inline instance vars.
3695 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3698 * config/rx/rx.c (rx_is_restricted_memory_address):
3701 2018-01-12 Richard Biener <rguenther@suse.de>
3703 PR tree-optimization/80846
3704 * target.def (split_reduction): New target hook.
3705 * targhooks.c (default_split_reduction): New function.
3706 * targhooks.h (default_split_reduction): Declare.
3707 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3708 target requests first reduce vectors by combining low and high
3710 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3711 (get_vectype_for_scalar_type_and_size): Export.
3712 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3713 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3714 * doc/tm.texi: Regenerate.
3715 * config/i386/i386.c (ix86_split_reduction): Implement
3716 TARGET_VECTORIZE_SPLIT_REDUCTION.
3718 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3721 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3722 in PIC mode except for TARGET_VXWORKS_RTP.
3723 * config/sparc/sparc.c: Include cfgrtl.h.
3724 (TARGET_INIT_PIC_REG): Define.
3725 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3726 (sparc_pic_register_p): New predicate.
3727 (sparc_legitimate_address_p): Use it.
3728 (sparc_legitimize_pic_address): Likewise.
3729 (sparc_delegitimize_address): Likewise.
3730 (sparc_mode_dependent_address_p): Likewise.
3731 (gen_load_pcrel_sym): Remove 4th parameter.
3732 (load_got_register): Adjust call to above. Remove obsolete stuff.
3733 (sparc_expand_prologue): Do not call load_got_register here.
3734 (sparc_flat_expand_prologue): Likewise.
3735 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3736 (sparc_use_pseudo_pic_reg): New function.
3737 (sparc_init_pic_reg): Likewise.
3738 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3739 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3741 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3743 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3744 Add item for branch_cost.
3746 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3748 PR rtl-optimization/83565
3749 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3750 not extend the result to a larger mode for rotate operations.
3751 (num_sign_bit_copies1): Likewise.
3753 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3756 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3758 Use values-Xc.o for -pedantic.
3759 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3761 2018-01-12 Martin Liska <mliska@suse.cz>
3764 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3766 (possible_polymorphic_call_targets): Use it.
3767 (ipa_devirt): Likewise.
3769 2018-01-12 Martin Liska <mliska@suse.cz>
3771 * profile-count.h (enum profile_quality): Use 0 as invalid
3772 enum value of profile_quality.
3774 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3776 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3777 -mext-string options.
3779 2018-01-12 Richard Biener <rguenther@suse.de>
3781 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3782 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3783 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3785 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3787 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3789 * configure.ac (--with-long-double-format): Add support for the
3790 configuration option to change the default long double format on
3792 * config.gcc (powerpc*-linux*-*): Likewise.
3793 * configure: Regenerate.
3794 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3795 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3796 used without modification.
3798 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3800 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3801 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3802 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3803 MISC_BUILTIN_SPEC_BARRIER.
3804 (rs6000_init_builtins): Likewise.
3805 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3807 (speculation_barrier): New define_insn.
3808 * doc/extend.texi: Document __builtin_speculation_barrier.
3810 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3813 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3814 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3815 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3817 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3818 integral modes instead of "ss" and "sd".
3819 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3820 vectors with 32-bit and 64-bit elements.
3821 (vecdupssescalarmodesuffix): New mode attribute.
3822 (vec_dup<mode>): Use it.
3824 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3827 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3828 frame if argument is passed on stack.
3830 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3833 * ree.c (combine_reaching_defs): Optimize also
3834 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3835 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3837 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3840 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3842 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3845 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3846 after they are computed.
3848 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3850 PR tree-optimization/83695
3851 * gimple-loop-linterchange.cc
3852 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3853 reset cached scev information after interchange.
3854 (pass_linterchange::execute): Remove call to scev_reset_htab.
3856 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3858 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3859 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3860 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3861 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3862 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3863 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3864 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3865 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3866 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3867 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3868 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3869 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3870 (V_lane_reg): Likewise.
3871 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3873 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3874 (vfmal_lane_low<mode>_intrinsic,
3875 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3876 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3877 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3878 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3879 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3880 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3882 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3884 * config/arm/arm-cpus.in (fp16fml): New feature.
3885 (ALL_SIMD): Add fp16fml.
3886 (armv8.2-a): Add fp16fml as an option.
3887 (armv8.3-a): Likewise.
3888 (armv8.4-a): Add fp16fml as part of fp16.
3889 * config/arm/arm.h (TARGET_FP16FML): Define.
3890 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3892 * config/arm/arm-modes.def (V2HF): Define.
3893 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3894 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3895 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3896 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3897 vfmsl_low, vfmsl_high): New set of builtins.
3898 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3899 (vfml_op): New code attribute.
3900 (VFMLHALVES): New int iterator.
3901 (VFML, VFMLSEL): New mode attributes.
3902 (V_reg): Define mapping for V2HF.
3903 (V_hi, V_lo): New mode attributes.
3904 (VF_constraint): Likewise.
3905 (vfml_half, vfml_half_selector): New int attributes.
3906 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3908 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3909 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3911 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3912 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3913 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3914 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3916 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3917 Document new effective target and option set.
3919 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3921 * config/arm/arm-cpus.in (armv8_4): New feature.
3922 (ARMv8_4a): New fgroup.
3923 (armv8.4-a): New arch.
3924 * config/arm/arm-tables.opt: Regenerate.
3925 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3926 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3927 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3928 Add matching rules for -march=armv8.4-a and extensions.
3929 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3931 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3934 * config/rx/rx.md (BW): New mode attribute.
3935 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3937 2018-01-11 Richard Biener <rguenther@suse.de>
3939 PR tree-optimization/83435
3940 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3941 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3942 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3944 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3945 Alan Hayward <alan.hayward@arm.com>
3946 David Sherwood <david.sherwood@arm.com>
3948 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3950 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3951 (aarch64_print_address_internal): Use it to check for a zero offset.
3953 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3954 Alan Hayward <alan.hayward@arm.com>
3955 David Sherwood <david.sherwood@arm.com>
3957 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3958 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3959 Return a poly_int64 rather than a HOST_WIDE_INT.
3960 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3961 rather than a HOST_WIDE_INT.
3962 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3963 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3964 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3965 final_offset from HOST_WIDE_INT to poly_int64.
3966 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3967 to_constant when getting the number of units in an Advanced SIMD
3969 (aarch64_builtin_vectorized_function): Check for a constant number
3971 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3973 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3974 attribute instead of GET_MODE_NUNITS.
3975 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3976 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3977 GET_MODE_SIZE for fixed-size registers.
3978 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3979 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3980 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3981 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3982 (aarch64_print_operand, aarch64_print_address_internal)
3983 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3984 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3985 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3986 Handle polynomial GET_MODE_SIZE.
3987 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3988 wider than SImode without modification.
3989 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3990 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3991 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3992 passing and returning SVE modes.
3993 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3994 rather than GEN_INT.
3995 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3996 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3997 (aarch64_allocate_and_probe_stack_space): Likewise.
3998 (aarch64_layout_frame): Cope with polynomial offsets.
3999 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4000 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4002 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4003 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4004 poly_int64 rather than a HOST_WIDE_INT.
4005 (aarch64_get_separate_components, aarch64_process_components)
4006 (aarch64_expand_prologue, aarch64_expand_epilogue)
4007 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4008 (aarch64_anchor_offset): New function, split out from...
4009 (aarch64_legitimize_address): ...here.
4010 (aarch64_builtin_vectorization_cost): Handle polynomial
4011 TYPE_VECTOR_SUBPARTS.
4012 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4014 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4015 number of elements from the PARALLEL rather than the mode.
4016 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4017 rather than GET_MODE_BITSIZE.
4018 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4019 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4020 (aarch64_expand_vec_perm_const_1): Handle polynomial
4021 d->perm.length () and d->perm elements.
4022 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4023 Apply to_constant to d->perm elements.
4024 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4025 polynomial CONST_VECTOR_NUNITS.
4026 (aarch64_move_pointer): Take amount as a poly_int64 rather
4028 (aarch64_progress_pointer): Avoid temporary variable.
4029 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4030 the mode attribute instead of GET_MODE.
4032 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4033 Alan Hayward <alan.hayward@arm.com>
4034 David Sherwood <david.sherwood@arm.com>
4036 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4037 x exists before using it.
4038 (aarch64_add_constant_internal): Rename to...
4039 (aarch64_add_offset_1): ...this. Replace regnum with separate
4040 src and dest rtxes. Handle the case in which they're different,
4041 including when the offset is zero. Replace scratchreg with an rtx.
4042 Use 2 additions if there is no spare register into which we can
4043 move a 16-bit constant.
4044 (aarch64_add_constant): Delete.
4045 (aarch64_add_offset): Replace reg with separate src and dest
4046 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4047 Use aarch64_add_offset_1.
4048 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4049 an rtx rather than an int. Take the delta as a poly_int64
4050 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4051 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4052 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4053 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4054 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4056 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4057 aarch64_add_constant.
4059 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4061 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4062 Use scalar_float_mode.
4064 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4066 * config/aarch64/aarch64-simd.md
4067 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4068 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4069 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4070 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4071 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4072 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4073 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4074 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4075 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4076 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4078 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4081 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4082 targ_options->x_arm_arch_string is non NULL.
4084 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4086 * config/aarch64/aarch64.h
4087 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4089 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4092 * expmed.c (emit_store_flag_force): Swap if const op0
4093 and change VOIDmode to mode of op0.
4095 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4097 PR rtl-optimization/83761
4098 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4099 than bytes to mode_for_size.
4101 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4104 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4105 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4108 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4111 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4112 when in layout mode.
4113 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4114 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4117 2018-01-10 Michael Collison <michael.collison@arm.com>
4119 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4120 * config/aarch64/aarch64-option-extension.def: Add
4121 AARCH64_OPT_EXTENSION of 'fp16fml'.
4122 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4123 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4124 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4125 * config/aarch64/constraints.md (Ui7): New constraint.
4126 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4127 (VFMLA_SEL_W): Ditto.
4130 (VFMLA16_LOW): New int iterator.
4131 (VFMLA16_HIGH): Ditto.
4132 (UNSPEC_FMLAL): New unspec.
4133 (UNSPEC_FMLSL): Ditto.
4134 (UNSPEC_FMLAL2): Ditto.
4135 (UNSPEC_FMLSL2): Ditto.
4136 (f16mac): New code attribute.
4137 * config/aarch64/aarch64-simd-builtins.def
4138 (aarch64_fmlal_lowv2sf): Ditto.
4139 (aarch64_fmlsl_lowv2sf): Ditto.
4140 (aarch64_fmlalq_lowv4sf): Ditto.
4141 (aarch64_fmlslq_lowv4sf): Ditto.
4142 (aarch64_fmlal_highv2sf): Ditto.
4143 (aarch64_fmlsl_highv2sf): Ditto.
4144 (aarch64_fmlalq_highv4sf): Ditto.
4145 (aarch64_fmlslq_highv4sf): Ditto.
4146 (aarch64_fmlal_lane_lowv2sf): Ditto.
4147 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4148 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4149 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4150 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4151 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4152 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4153 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4154 (aarch64_fmlal_lane_highv2sf): Ditto.
4155 (aarch64_fmlsl_lane_highv2sf): Ditto.
4156 (aarch64_fmlal_laneq_highv2sf): Ditto.
4157 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4158 (aarch64_fmlalq_lane_highv4sf): Ditto.
4159 (aarch64_fmlsl_lane_highv4sf): Ditto.
4160 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4161 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4162 * config/aarch64/aarch64-simd.md:
4163 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4164 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4165 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4166 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4167 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4168 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4169 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4170 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4171 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4172 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4173 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4174 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4175 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4176 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4177 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4178 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4179 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4180 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4181 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4182 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4183 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4184 (vfmlsl_low_u32): Ditto.
4185 (vfmlalq_low_u32): Ditto.
4186 (vfmlslq_low_u32): Ditto.
4187 (vfmlal_high_u32): Ditto.
4188 (vfmlsl_high_u32): Ditto.
4189 (vfmlalq_high_u32): Ditto.
4190 (vfmlslq_high_u32): Ditto.
4191 (vfmlal_lane_low_u32): Ditto.
4192 (vfmlsl_lane_low_u32): Ditto.
4193 (vfmlal_laneq_low_u32): Ditto.
4194 (vfmlsl_laneq_low_u32): Ditto.
4195 (vfmlalq_lane_low_u32): Ditto.
4196 (vfmlslq_lane_low_u32): Ditto.
4197 (vfmlalq_laneq_low_u32): Ditto.
4198 (vfmlslq_laneq_low_u32): Ditto.
4199 (vfmlal_lane_high_u32): Ditto.
4200 (vfmlsl_lane_high_u32): Ditto.
4201 (vfmlal_laneq_high_u32): Ditto.
4202 (vfmlsl_laneq_high_u32): Ditto.
4203 (vfmlalq_lane_high_u32): Ditto.
4204 (vfmlslq_lane_high_u32): Ditto.
4205 (vfmlalq_laneq_high_u32): Ditto.
4206 (vfmlslq_laneq_high_u32): Ditto.
4207 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4208 (AARCH64_FL_FOR_ARCH8_4): New.
4209 (AARCH64_ISA_F16FML): New ISA flag.
4210 (TARGET_F16FML): New feature flag for fp16fml.
4211 (doc/invoke.texi): Document new fp16fml option.
4213 2018-01-10 Michael Collison <michael.collison@arm.com>
4215 * config/aarch64/aarch64-builtins.c:
4216 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4217 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4218 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4219 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4220 (AARCH64_ISA_SHA3): New ISA flag.
4221 (TARGET_SHA3): New feature flag for sha3.
4222 * config/aarch64/iterators.md (sha512_op): New int attribute.
4223 (CRYPTO_SHA512): New int iterator.
4224 (UNSPEC_SHA512H): New unspec.
4225 (UNSPEC_SHA512H2): Ditto.
4226 (UNSPEC_SHA512SU0): Ditto.
4227 (UNSPEC_SHA512SU1): Ditto.
4228 * config/aarch64/aarch64-simd-builtins.def
4229 (aarch64_crypto_sha512hqv2di): New builtin.
4230 (aarch64_crypto_sha512h2qv2di): Ditto.
4231 (aarch64_crypto_sha512su0qv2di): Ditto.
4232 (aarch64_crypto_sha512su1qv2di): Ditto.
4233 (aarch64_eor3qv8hi): Ditto.
4234 (aarch64_rax1qv2di): Ditto.
4235 (aarch64_xarqv2di): Ditto.
4236 (aarch64_bcaxqv8hi): Ditto.
4237 * config/aarch64/aarch64-simd.md:
4238 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4239 (aarch64_crypto_sha512su0qv2di): Ditto.
4240 (aarch64_crypto_sha512su1qv2di): Ditto.
4241 (aarch64_eor3qv8hi): Ditto.
4242 (aarch64_rax1qv2di): Ditto.
4243 (aarch64_xarqv2di): Ditto.
4244 (aarch64_bcaxqv8hi): Ditto.
4245 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4246 (vsha512h2q_u64): Ditto.
4247 (vsha512su0q_u64): Ditto.
4248 (vsha512su1q_u64): Ditto.
4249 (veor3q_u16): Ditto.
4250 (vrax1q_u64): Ditto.
4252 (vbcaxq_u16): Ditto.
4253 * config/arm/types.md (crypto_sha512): New type attribute.
4254 (crypto_sha3): Ditto.
4255 (doc/invoke.texi): Document new sha3 option.
4257 2018-01-10 Michael Collison <michael.collison@arm.com>
4259 * config/aarch64/aarch64-builtins.c:
4260 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4261 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4262 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4263 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4264 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4265 (AARCH64_ISA_SM4): New ISA flag.
4266 (TARGET_SM4): New feature flag for sm4.
4267 * config/aarch64/aarch64-simd-builtins.def
4268 (aarch64_sm3ss1qv4si): Ditto.
4269 (aarch64_sm3tt1aq4si): Ditto.
4270 (aarch64_sm3tt1bq4si): Ditto.
4271 (aarch64_sm3tt2aq4si): Ditto.
4272 (aarch64_sm3tt2bq4si): Ditto.
4273 (aarch64_sm3partw1qv4si): Ditto.
4274 (aarch64_sm3partw2qv4si): Ditto.
4275 (aarch64_sm4eqv4si): Ditto.
4276 (aarch64_sm4ekeyqv4si): Ditto.
4277 * config/aarch64/aarch64-simd.md:
4278 (aarch64_sm3ss1qv4si): Ditto.
4279 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4280 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4281 (aarch64_sm4eqv4si): Ditto.
4282 (aarch64_sm4ekeyqv4si): Ditto.
4283 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4284 (sm3part_op): Ditto.
4285 (CRYPTO_SM3TT): Ditto.
4286 (CRYPTO_SM3PART): Ditto.
4287 (UNSPEC_SM3SS1): New unspec.
4288 (UNSPEC_SM3TT1A): Ditto.
4289 (UNSPEC_SM3TT1B): Ditto.
4290 (UNSPEC_SM3TT2A): Ditto.
4291 (UNSPEC_SM3TT2B): Ditto.
4292 (UNSPEC_SM3PARTW1): Ditto.
4293 (UNSPEC_SM3PARTW2): Ditto.
4294 (UNSPEC_SM4E): Ditto.
4295 (UNSPEC_SM4EKEY): Ditto.
4296 * config/aarch64/constraints.md (Ui2): New constraint.
4297 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4298 * config/arm/types.md (crypto_sm3): New type attribute.
4299 (crypto_sm4): Ditto.
4300 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4301 (vsm3tt1aq_u32): Ditto.
4302 (vsm3tt1bq_u32): Ditto.
4303 (vsm3tt2aq_u32): Ditto.
4304 (vsm3tt2bq_u32): Ditto.
4305 (vsm3partw1q_u32): Ditto.
4306 (vsm3partw2q_u32): Ditto.
4307 (vsm4eq_u32): Ditto.
4308 (vsm4ekeyq_u32): Ditto.
4309 (doc/invoke.texi): Document new sm4 option.
4311 2018-01-10 Michael Collison <michael.collison@arm.com>
4313 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4314 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4315 (AARCH64_FL_FOR_ARCH8_4): New.
4316 (AARCH64_FL_V8_4): New flag.
4317 (doc/invoke.texi): Document new armv8.4-a option.
4319 2018-01-10 Michael Collison <michael.collison@arm.com>
4321 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4322 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4323 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4324 * config/aarch64/aarch64-option-extension.def: Add
4325 AARCH64_OPT_EXTENSION of 'sha2'.
4326 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4327 (crypto): Disable sha2 and aes if crypto disabled.
4328 (crypto): Enable aes and sha2 if enabled.
4329 (simd): Disable sha2 and aes if simd disabled.
4330 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4332 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4333 (TARGET_SHA2): New feature flag for sha2.
4334 (TARGET_AES): New feature flag for aes.
4335 * config/aarch64/aarch64-simd.md:
4336 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4337 conditional on TARGET_AES.
4338 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4339 (aarch64_crypto_sha1hsi): Make pattern conditional
4341 (aarch64_crypto_sha1hv4si): Ditto.
4342 (aarch64_be_crypto_sha1hv4si): Ditto.
4343 (aarch64_crypto_sha1su1v4si): Ditto.
4344 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4345 (aarch64_crypto_sha1su0v4si): Ditto.
4346 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4347 (aarch64_crypto_sha256su0v4si): Ditto.
4348 (aarch64_crypto_sha256su1v4si): Ditto.
4349 (doc/invoke.texi): Document new aes and sha2 options.
4351 2018-01-10 Martin Sebor <msebor@redhat.com>
4353 PR tree-optimization/83781
4354 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4357 2018-01-11 Martin Sebor <msebor@gmail.com>
4358 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4360 PR tree-optimization/83501
4361 PR tree-optimization/81703
4363 * tree-ssa-strlen.c (get_string_cst): Rename...
4364 (get_string_len): ...to this. Handle global constants.
4365 (handle_char_store): Adjust.
4367 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4368 Jim Wilson <jimw@sifive.com>
4370 * config/riscv/riscv-protos.h (riscv_output_return): New.
4371 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4372 (riscv_attribute_table, riscv_output_return),
4373 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4374 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4375 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4377 (riscv_expand_prologue): Add early return for naked function.
4378 (riscv_expand_epilogue): Likewise.
4379 (riscv_function_ok_for_sibcall): Return false for naked function.
4380 (riscv_set_current_function): New.
4381 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4382 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4383 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4384 * doc/extend.texi (RISC-V Function Attributes): New.
4386 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4388 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4389 check for 128-bit long double before checking TCmode.
4390 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4391 128-bit long doubles before checking TFmode or TCmode.
4392 (FLOAT128_IBM_P): Likewise.
4394 2018-01-10 Martin Sebor <msebor@redhat.com>
4396 PR tree-optimization/83671
4397 * builtins.c (c_strlen): Unconditionally return zero for the empty
4399 Use -Warray-bounds for warnings.
4400 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4401 for non-constant array indices with COMPONENT_REF, arrays of
4402 arrays, and pointers to arrays.
4403 (gimple_fold_builtin_strlen): Determine and set length range for
4404 non-constant character arrays.
4406 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4409 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4412 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4414 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4416 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4419 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4420 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4421 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4422 indexed_or_indirect_operand predicate.
4423 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4424 (*vsx_le_perm_load_v8hi): Likewise.
4425 (*vsx_le_perm_load_v16qi): Likewise.
4426 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4427 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4428 (*vsx_le_perm_store_v8hi): Likewise.
4429 (*vsx_le_perm_store_v16qi): Likewise.
4430 (eight unnamed splitters): Likewise.
4432 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4434 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4435 * config/rs6000/emmintrin.h: Likewise.
4436 * config/rs6000/mmintrin.h: Likewise.
4437 * config/rs6000/xmmintrin.h: Likewise.
4439 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4442 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4444 * tree.c (tree_nop_conversion): Return true for location wrapper
4446 (maybe_wrap_with_location): New function.
4447 (selftest::check_strip_nops): New function.
4448 (selftest::test_location_wrappers): New function.
4449 (selftest::tree_c_tests): Call it.
4450 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4451 (maybe_wrap_with_location): New decl.
4452 (EXPR_LOCATION_WRAPPER_P): New macro.
4453 (location_wrapper_p): New inline function.
4454 (tree_strip_any_location_wrapper): New inline function.
4456 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4459 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4460 stack_realign_offset for the largest alignment of stack slot
4462 (ix86_find_max_used_stack_alignment): New function.
4463 (ix86_finalize_stack_frame_flags): Use it. Set
4464 max_used_stack_alignment if we don't realign stack.
4465 * config/i386/i386.h (machine_function): Add
4466 max_used_stack_alignment.
4468 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4470 * config/arm/arm.opt (-mbranch-cost): New option.
4471 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4474 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4477 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4478 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4480 2018-01-10 Richard Biener <rguenther@suse.de>
4483 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4484 early out so it also covers the case where we have a non-NULL
4487 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4489 PR tree-optimization/83753
4490 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4491 for non-strided grouped accesses if the number of elements is 1.
4493 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4496 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4497 * i386.h (TARGET_USE_GATHER): Define.
4498 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4500 2018-01-10 Martin Liska <mliska@suse.cz>
4503 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4504 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4506 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4507 CLEANUP_NO_PARTITIONING is not set.
4509 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4511 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4512 for vectors, as a partial revert of r254296.
4513 * rtl.h (const_vec_p): Delete.
4514 (const_vec_duplicate_p): Don't test for vector CONSTs.
4515 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4516 * expmed.c (make_tree): Likewise.
4519 * common.md (E, F): Use CONSTANT_P instead of checking for
4521 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4522 checking for CONST_VECTOR.
4524 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4527 * predict.c (force_edge_cold): Handle in more sane way edges
4530 2018-01-09 Carl Love <cel@us.ibm.com>
4532 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4534 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4535 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4536 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4537 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4538 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4539 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4540 * config/rs6000/rs6000-protos.h: Add extern defition for
4541 rs6000_generate_float2_double_code.
4542 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4544 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4545 (float2_v2df): Add define_expand.
4547 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4550 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4551 op_mode in the force_to_mode call.
4553 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4555 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4556 instead of checking each element individually.
4557 (aarch64_evpc_uzp): Likewise.
4558 (aarch64_evpc_zip): Likewise.
4559 (aarch64_evpc_ext): Likewise.
4560 (aarch64_evpc_rev): Likewise.
4561 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4562 instead of checking each element individually. Return true without
4564 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4565 whether all selected elements come from the same input, instead of
4566 checking each element individually. Remove calls to gen_rtx_REG,
4567 start_sequence and end_sequence and instead assert that no rtl is
4570 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4572 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4573 order of HIGH and CONST checks.
4575 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4577 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4578 if the destination isn't an SSA_NAME.
4580 2018-01-09 Richard Biener <rguenther@suse.de>
4582 PR tree-optimization/83668
4583 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4585 (canonicalize_loop_form): ... here, renamed from ...
4586 (canonicalize_loop_closed_ssa_form): ... this and amended to
4587 swap successor edges for loop exit blocks to make us use
4588 the RPO order we need for initial schedule generation.
4590 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4592 PR tree-optimization/64811
4593 * match.pd: When optimizing comparisons with Inf, avoid
4594 introducing or losing exceptions from comparisons with NaN.
4596 2018-01-09 Martin Liska <mliska@suse.cz>
4599 * asan.c (shadow_mem_size): Add gcc_assert.
4601 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4603 Don't save registers in main().
4606 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4607 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4608 * config/avr/avr.c (avr_set_current_function): Don't error if
4609 naked, OS_task or OS_main are specified at the same time.
4610 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4612 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4614 * common/config/avr/avr-common.c (avr_option_optimization_table):
4615 Switch on -mmain-is-OS_task for optimizing compilations.
4617 2018-01-09 Richard Biener <rguenther@suse.de>
4619 PR tree-optimization/83572
4620 * graphite.c: Include cfganal.h.
4621 (graphite_transform_loops): Connect infinite loops to exit
4622 and remove fake edges at the end.
4624 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4626 * ipa-inline.c (edge_badness): Revert accidental checkin.
4628 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4631 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4632 symbols; not inline clones.
4634 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4637 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4638 hard registers. Formatting fixes.
4640 PR preprocessor/83722
4641 * gcc.c (try_generate_repro): Pass
4642 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4643 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4646 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4647 Kito Cheng <kito.cheng@gmail.com>
4649 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4650 (riscv_leaf_function_p): Delete.
4651 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4653 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4655 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4657 (do_ifelse): New function.
4658 (do_isel): New function.
4659 (do_sub3): New function.
4660 (do_add3): New function.
4661 (do_load_mask_compare): New function.
4662 (do_overlap_load_compare): New function.
4663 (expand_compare_loop): New function.
4664 (expand_block_compare): Call expand_compare_loop() when appropriate.
4665 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4667 (-mblock-compare-inline-loop-limit): New option.
4669 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4672 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4673 Reverse order of second and third operands in first alternative.
4674 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4675 of first and second elements in UNSPEC_VPERMR vector.
4676 (altivec_expand_vec_perm_le): Likewise.
4678 2018-01-08 Jeff Law <law@redhat.com>
4680 PR rtl-optimizatin/81308
4681 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4682 (process_switch): If group_case_labels makes a change, then set
4684 (pass_convert_switch::execute): If a switch is converted, then
4685 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4687 PR rtl-optimization/81308
4688 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4691 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4693 PR target/83663 - Revert r255946
4694 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4695 generation for cases where splatting a value is not useful.
4696 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4697 across a vec_duplicate and a paradoxical subreg forming a vector
4698 mode to a vec_concat.
4700 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4702 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4703 -march=armv8.3-a variants.
4704 * config/arm/t-multilib: Likewise.
4705 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4707 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4709 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4711 (cceq_ior_compare_complement): Give it a name so I can use it, and
4712 change boolean_or_operator predicate to boolean_operator so it can
4713 be used to generate a crand.
4714 (eqne): New code iterator.
4715 (bd/bd_neg): New code_attrs.
4716 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4717 a single define_insn.
4718 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4719 decrement (bdnzt/bdnzf/bdzt/bdzf).
4720 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4721 with the new names of the branch decrement patterns, and added the
4722 names of the branch decrement conditional patterns.
4724 2018-01-08 Richard Biener <rguenther@suse.de>
4726 PR tree-optimization/83563
4727 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4730 2018-01-08 Richard Biener <rguenther@suse.de>
4733 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4735 2018-01-08 Richard Biener <rguenther@suse.de>
4737 PR tree-optimization/83685
4738 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4739 references to abnormals.
4741 2018-01-08 Richard Biener <rguenther@suse.de>
4744 * dwarf2out.c (output_indirect_strings): Handle empty
4745 skeleton_debug_str_hash.
4746 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4748 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4750 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4751 (emit_store_direct): Likewise.
4752 (arc_trampoline_adjust_address): Likewise.
4753 (arc_asm_trampoline_template): New function.
4754 (arc_initialize_trampoline): Use asm_trampoline_template.
4755 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4756 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4757 * config/arc/arc.md (flush_icache): Delete pattern.
4759 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4761 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4762 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4765 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4768 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4769 by not USED_FOR_TARGET.
4770 (make_pass_resolve_sw_modes): Likewise.
4772 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4774 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4777 2018-01-08 Richard Biener <rguenther@suse.de>
4780 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4782 2018-01-08 Richard Biener <rguenther@suse.de>
4785 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4787 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4790 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4791 basic blocks with a small number of successors.
4792 (convert_control_dep_chain_into_preds): Improve handling of
4794 (dump_predicates): Split apart into...
4795 (dump_pred_chain): ...here...
4796 (dump_pred_info): ...and here.
4797 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4798 (can_chain_union_be_invalidated_p): Improve check for invalidation
4800 (uninit_uses_cannot_happen): Avoid unnecessary if
4801 convert_control_dep_chain_into_preds yielded nothing.
4803 2018-01-06 Martin Sebor <msebor@redhat.com>
4805 PR tree-optimization/83640
4806 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4807 subtracting negative offset from size.
4808 (builtin_access::overlap): Adjust offset bounds of the access to fall
4809 within the size of the object if possible.
4811 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4813 PR rtl-optimization/83699
4814 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4815 extract_bit_field_as_subreg to cases in which the extracted
4816 value is also a vector.
4818 * lra-constraints.c (process_alt_operands): Test for the equivalence
4819 substitutions when detecting a possible reload cycle.
4821 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4824 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4825 by default if flag_selective_schedling{,2}. Formatting fixes.
4827 PR rtl-optimization/83682
4828 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4829 if it has non-VECTOR_MODE element mode.
4830 (vec_duplicate_p): Likewise.
4833 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4834 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4836 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4839 * config/i386/i386-builtin.def
4840 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4841 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4842 Require also OPTION_MASK_ISA_AVX512F in addition to
4843 OPTION_MASK_ISA_GFNI.
4844 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4845 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4846 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4847 to OPTION_MASK_ISA_GFNI.
4848 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4849 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4850 OPTION_MASK_ISA_AVX512BW.
4851 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4852 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4853 addition to OPTION_MASK_ISA_GFNI.
4854 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4855 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4856 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4857 to OPTION_MASK_ISA_GFNI.
4858 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4859 a requirement for all ISAs rather than any of them with a few
4861 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4863 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4864 bitmasks to be enabled with 3 exceptions, instead of requiring any
4865 enabled ISA with lots of exceptions.
4866 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4867 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4868 Change avx512bw in isa attribute to avx512f.
4869 * config/i386/sgxintrin.h: Add license boilerplate.
4870 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4871 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4872 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4873 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4875 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4876 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4877 temporarily sse2 rather than sse if not enabled already.
4880 * config/i386/sse.md (VI248_VLBW): Rename to ...
4881 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4882 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4883 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4884 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4885 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4886 mode iterator instead of VI248_VLBW.
4888 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4890 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4891 (record_modified): Skip clobbers; add debug output.
4892 (param_change_prob): Use sreal frequencies.
4894 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4896 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4897 punt for user-aligned variables.
4899 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4901 * tree-chrec.c (chrec_contains_symbols): Return true for
4904 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4907 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4908 of (x|y) == x for BICS pattern.
4910 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4912 PR tree-optimization/83605
4913 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4914 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4917 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4919 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4920 * config/epiphany/rtems.h: New file.
4922 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4923 Uros Bizjak <ubizjak@gmail.com>
4926 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4927 QIreg_operand instead of register_operand predicate.
4928 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4929 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4930 comments instead of -fmitigate[-_]rop.
4932 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4935 * cgraphunit.c (symbol_table::compile): Switch to text_section
4936 before calling assembly_start debug hook.
4937 * run-rtl-passes.c (run_rtl_passes): Likewise.
4940 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4942 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4943 range_int_cst_p rather than !symbolic_range_p before calling
4944 extract_range_from_multiplicative_op_1.
4946 2018-01-04 Jeff Law <law@redhat.com>
4948 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4949 redundant test in assertion.
4951 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4953 * doc/rtl.texi: Document machine_mode wrapper classes.
4955 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4957 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4960 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4962 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4963 the VEC_PERM_EXPR fold to fail.
4965 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4968 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4969 to switched_sections.
4971 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4974 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4977 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4980 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4981 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4983 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4986 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4987 is BLKmode and bitpos not zero or mode change is needed.
4989 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4992 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4995 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4998 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4999 instead of MULT rtx. Update all corresponding splitters.
5001 (*ssub<modesuffix>): Ditto.
5003 (*cmp_sadd_di): Update split patterns.
5004 (*cmp_sadd_si): Ditto.
5005 (*cmp_sadd_sidi): Ditto.
5006 (*cmp_ssub_di): Ditto.
5007 (*cmp_ssub_si): Ditto.
5008 (*cmp_ssub_sidi): Ditto.
5009 * config/alpha/predicates.md (const23_operand): New predicate.
5010 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5011 Look for ASHIFT, not MULT inner operand.
5012 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5014 2018-01-04 Martin Liska <mliska@suse.cz>
5016 PR gcov-profile/83669
5017 * gcov.c (output_intermediate_file): Add version to intermediate
5019 * doc/gcov.texi: Document new field 'version' in intermediate
5020 file format. Fix location of '-k' option of gcov command.
5022 2018-01-04 Martin Liska <mliska@suse.cz>
5025 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5027 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5029 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5031 2018-01-03 Martin Sebor <msebor@redhat.com>
5033 PR tree-optimization/83655
5034 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5035 checking calls with invalid arguments.
5037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5039 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5040 (vectorizable_mask_load_store): Delete.
5041 (vectorizable_call): Return false for masked loads and stores.
5042 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5043 instead of gimple_assign_rhs1.
5044 (vectorizable_load): Handle IFN_MASK_LOAD.
5045 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5049 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5051 (vectorizable_mask_load_store): ...here.
5052 (vectorizable_load): ...and here.
5054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5056 * tree-vect-stmts.c (vect_build_all_ones_mask)
5057 (vect_build_zero_merge_argument): New functions, split out from...
5058 (vectorizable_load): ...here.
5060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5062 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5064 (vectorizable_mask_load_store): ...here.
5065 (vectorizable_store): ...and here.
5067 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5069 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5071 (vectorizable_mask_load_store): ...here.
5073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5075 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5076 (vect_model_store_cost): Take a vec_load_store_type instead of a
5078 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5079 (vect_model_store_cost): Take a vec_load_store_type instead of a
5081 (vectorizable_mask_load_store): Update accordingly.
5082 (vectorizable_store): Likewise.
5083 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5087 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5088 IFN_MASK_LOAD calls here rather than...
5089 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5092 Alan Hayward <alan.hayward@arm.com>
5093 David Sherwood <david.sherwood@arm.com>
5095 * expmed.c (extract_bit_field_1): For vector extracts,
5096 fall back to extract_bit_field_as_subreg if vec_extract
5099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5100 Alan Hayward <alan.hayward@arm.com>
5101 David Sherwood <david.sherwood@arm.com>
5103 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5104 they are variable or constant sized.
5105 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5106 slots for constant-sized data.
5108 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5109 Alan Hayward <alan.hayward@arm.com>
5110 David Sherwood <david.sherwood@arm.com>
5112 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5113 handling COND_EXPRs with boolean comparisons, try to find a better
5114 basis for the mask type than the boolean itself.
5116 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5118 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5119 is calculated and how it can be overridden.
5120 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5121 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5123 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5127 Alan Hayward <alan.hayward@arm.com>
5128 David Sherwood <david.sherwood@arm.com>
5130 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5131 Remove the mode argument.
5132 (aarch64_simd_valid_immediate): Remove the mode and inverse
5134 * config/aarch64/iterators.md (bitsize): New iterator.
5135 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5136 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5137 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5138 aarch64_simd_valid_immediate.
5139 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5140 (aarch64_reg_or_bic_imm): Likewise.
5141 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5142 with an insn_type enum and msl with a modifier_type enum.
5143 Replace element_width with a scalar_mode. Change the shift
5144 to unsigned int. Add constructors for scalar_float_mode and
5145 scalar_int_mode elements.
5146 (aarch64_vect_float_const_representable_p): Delete.
5147 (aarch64_can_const_movi_rtx_p)
5148 (aarch64_simd_scalar_immediate_valid_for_move)
5149 (aarch64_simd_make_constant): Update call to
5150 aarch64_simd_valid_immediate.
5151 (aarch64_advsimd_valid_immediate_hs): New function.
5152 (aarch64_advsimd_valid_immediate): Likewise.
5153 (aarch64_simd_valid_immediate): Remove mode and inverse
5154 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5155 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5156 and aarch64_float_const_representable_p on the result.
5157 (aarch64_output_simd_mov_immediate): Remove mode argument.
5158 Update call to aarch64_simd_valid_immediate and use of
5159 simd_immediate_info.
5160 (aarch64_output_scalar_simd_mov_immediate): Update call
5163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5164 Alan Hayward <alan.hayward@arm.com>
5165 David Sherwood <david.sherwood@arm.com>
5167 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5168 (mode_nunits): Likewise CONST_MODE_NUNITS.
5169 * machmode.def (ADJUST_NUNITS): Document.
5170 * genmodes.c (mode_data::need_nunits_adj): New field.
5171 (blank_mode): Update accordingly.
5172 (adj_nunits): New variable.
5173 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5175 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5176 listed in adj_nunits.
5177 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5178 listed in adj_nunits. Don't emit case statements for such modes.
5179 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5180 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5181 nothing if adj_nunits is nonnull.
5182 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5183 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5184 (emit_mode_fbit): Update use of print_maybe_const_decl.
5185 (emit_move_size): Likewise. Treat the array as non-const
5187 (emit_mode_adjustments): Handle adj_nunits.
5189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5191 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5192 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5193 (VECTOR_MODES): Use it.
5194 (make_vector_modes): Take the prefix as an argument.
5196 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5197 Alan Hayward <alan.hayward@arm.com>
5198 David Sherwood <david.sherwood@arm.com>
5200 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5201 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5202 for MODE_VECTOR_BOOL.
5203 * machmode.def (VECTOR_BOOL_MODE): Document.
5204 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5205 (make_vector_bool_mode): New function.
5206 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5208 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5209 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5211 * stor-layout.c (int_mode_for_mode): Likewise.
5212 * tree.c (build_vector_type_for_mode): Likewise.
5213 * varasm.c (output_constant_pool_2): Likewise.
5214 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5215 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5216 for MODE_VECTOR_BOOL.
5217 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5218 of mode class checks.
5219 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5220 instead of a list of mode class checks.
5221 (expand_vector_scalar_condition): Likewise.
5222 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5225 Alan Hayward <alan.hayward@arm.com>
5226 David Sherwood <david.sherwood@arm.com>
5228 * machmode.h (mode_size): Change from unsigned short to
5230 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5231 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5232 or if measurement_type is not polynomial.
5233 (fixed_size_mode::includes_p): Check for constant-sized modes.
5234 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5235 return a poly_uint16 rather than an unsigned short.
5236 (emit_mode_size): Change the type of mode_size from unsigned short
5237 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5238 (emit_mode_adjustments): Cope with polynomial vector sizes.
5239 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5241 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5243 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5244 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5245 * caller-save.c (setup_save_areas): Likewise.
5246 (replace_reg_with_saved_mem): Likewise.
5247 * calls.c (emit_library_call_value_1): Likewise.
5248 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5249 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5250 (gen_lowpart_for_combine): Likewise.
5251 * convert.c (convert_to_integer_1): Likewise.
5252 * cse.c (equiv_constant, cse_insn): Likewise.
5253 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5254 (cselib_subst_to_values): Likewise.
5255 * dce.c (word_dce_process_block): Likewise.
5256 * df-problems.c (df_word_lr_mark_ref): Likewise.
5257 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5258 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5259 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5260 (rtl_for_decl_location): Likewise.
5261 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5262 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5263 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5264 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5265 (expand_expr_real_1): Likewise.
5266 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5267 (pad_below): Likewise.
5268 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5269 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5270 * ira.c (get_subreg_tracking_sizes): Likewise.
5271 * ira-build.c (ira_create_allocno_objects): Likewise.
5272 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5273 (ira_sort_regnos_for_alter_reg): Likewise.
5274 * ira-costs.c (record_operand_costs): Likewise.
5275 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5276 (resolve_simple_move): Likewise.
5277 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5278 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5279 (lra_constraints): Likewise.
5280 (CONST_POOL_OK_P): Reject variable-sized modes.
5281 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5282 (add_pseudo_to_slot, lra_spill): Likewise.
5283 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5284 * optabs-query.c (get_best_extraction_insn): Likewise.
5285 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5286 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5287 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5288 * recog.c (offsettable_address_addr_space_p): Likewise.
5289 * regcprop.c (maybe_mode_change): Likewise.
5290 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5291 * regrename.c (build_def_use): Likewise.
5292 * regstat.c (dump_reg_info): Likewise.
5293 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5294 (find_reloads, find_reloads_subreg_address): Likewise.
5295 * reload1.c (eliminate_regs_1): Likewise.
5296 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5297 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5298 (simplify_binary_operation_1, simplify_subreg): Likewise.
5299 * targhooks.c (default_function_arg_padding): Likewise.
5300 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5301 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5302 (verify_gimple_assign_ternary): Likewise.
5303 * tree-inline.c (estimate_move_cost): Likewise.
5304 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5305 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5306 (get_address_cost_ainc): Likewise.
5307 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5308 (vect_supportable_dr_alignment): Likewise.
5309 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5310 (vectorizable_reduction): Likewise.
5311 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5312 (vectorizable_operation, vectorizable_load): Likewise.
5313 * tree.c (build_same_sized_truth_vector_type): Likewise.
5314 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5315 * var-tracking.c (emit_note_insn_var_location): Likewise.
5316 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5317 (ADDR_VEC_ALIGN): Likewise.
5319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5320 Alan Hayward <alan.hayward@arm.com>
5321 David Sherwood <david.sherwood@arm.com>
5323 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5325 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5326 or if measurement_type is polynomial.
5327 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5328 * combine.c (make_extraction): Likewise.
5329 * dse.c (find_shift_sequence): Likewise.
5330 * dwarf2out.c (mem_loc_descriptor): Likewise.
5331 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5332 (extract_bit_field, extract_low_bits): Likewise.
5333 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5334 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5335 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5336 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5337 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5338 * reload.c (find_reloads): Likewise.
5339 * reload1.c (alter_reg): Likewise.
5340 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5341 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5342 * tree-if-conv.c (predicate_mem_writes): Likewise.
5343 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5344 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5345 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5346 * valtrack.c (dead_debug_insert_temp): Likewise.
5347 * varasm.c (mergeable_constant_section): Likewise.
5348 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5350 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5351 Alan Hayward <alan.hayward@arm.com>
5352 David Sherwood <david.sherwood@arm.com>
5354 * expr.c (expand_assignment): Cope with polynomial mode sizes
5355 when assigning to a CONCAT.
5357 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5358 Alan Hayward <alan.hayward@arm.com>
5359 David Sherwood <david.sherwood@arm.com>
5361 * machmode.h (mode_precision): Change from unsigned short to
5363 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5365 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5366 or if measurement_type is not polynomial.
5367 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5368 in which the mode is already known to be a scalar_int_mode.
5369 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5370 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5372 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5373 for GET_MODE_PRECISION.
5374 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5375 for GET_MODE_PRECISION.
5376 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5378 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5379 (expand_field_assignment, make_extraction): Likewise.
5380 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5381 (get_last_value): Likewise.
5382 * convert.c (convert_to_integer_1): Likewise.
5383 * cse.c (cse_insn): Likewise.
5384 * expr.c (expand_expr_real_1): Likewise.
5385 * lra-constraints.c (simplify_operand_subreg): Likewise.
5386 * optabs-query.c (can_atomic_load_p): Likewise.
5387 * optabs.c (expand_atomic_load): Likewise.
5388 (expand_atomic_store): Likewise.
5389 * ree.c (combine_reaching_defs): Likewise.
5390 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5391 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5392 * tree.h (type_has_mode_precision_p): Likewise.
5393 * ubsan.c (instrument_si_overflow): Likewise.
5395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5396 Alan Hayward <alan.hayward@arm.com>
5397 David Sherwood <david.sherwood@arm.com>
5399 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5400 polynomial numbers of units.
5401 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5402 (valid_vector_subparts_p): New function.
5403 (build_vector_type): Remove temporary shim and take the number
5404 of units as a poly_uint64 rather than an int.
5405 (build_opaque_vector_type): Take the number of units as a
5406 poly_uint64 rather than an int.
5407 * tree.c (build_vector_from_ctor): Handle polynomial
5408 TYPE_VECTOR_SUBPARTS.
5409 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5410 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5411 (build_vector_from_val): If the number of units is variable,
5412 use build_vec_duplicate_cst for constant operands and
5413 VEC_DUPLICATE_EXPR otherwise.
5414 (make_vector_type): Remove temporary is_constant ().
5415 (build_vector_type, build_opaque_vector_type): Take the number of
5416 units as a poly_uint64 rather than an int.
5417 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5419 * cfgexpand.c (expand_debug_expr): Likewise.
5420 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5421 (store_constructor, expand_expr_real_1): Likewise.
5422 (const_scalar_mask_from_tree): Likewise.
5423 * fold-const-call.c (fold_const_reduction): Likewise.
5424 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5425 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5426 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5427 (fold_relational_const): Likewise.
5428 (native_interpret_vector): Likewise. Change the size from an
5429 int to an unsigned int.
5430 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5431 TYPE_VECTOR_SUBPARTS.
5432 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5433 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5434 duplicating a non-constant operand into a variable-length vector.
5435 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5436 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5437 * ipa-icf.c (sem_variable::equals): Likewise.
5438 * match.pd: Likewise.
5439 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5440 * print-tree.c (print_node): Likewise.
5441 * stor-layout.c (layout_type): Likewise.
5442 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5443 * tree-cfg.c (verify_gimple_comparison): Likewise.
5444 (verify_gimple_assign_binary): Likewise.
5445 (verify_gimple_assign_ternary): Likewise.
5446 (verify_gimple_assign_single): Likewise.
5447 * tree-pretty-print.c (dump_generic_node): Likewise.
5448 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5449 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5450 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5451 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5452 (vect_shift_permute_load_chain): Likewise.
5453 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5454 (expand_vector_condition, optimize_vector_constructor): Likewise.
5455 (lower_vec_perm, get_compute_type): Likewise.
5456 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5457 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5458 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5459 (vect_recog_mask_conversion_pattern): Likewise.
5460 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5461 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5462 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5463 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5464 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5465 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5466 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5467 (supportable_widening_operation): Likewise.
5468 (supportable_narrowing_operation): Likewise.
5469 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5471 * varasm.c (output_constant): Likewise.
5473 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5474 Alan Hayward <alan.hayward@arm.com>
5475 David Sherwood <david.sherwood@arm.com>
5477 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5478 so that both the length == 3 and length != 3 cases set up their
5479 own permute vectors. Add comments explaining why we know the
5480 number of elements is constant.
5481 (vect_permute_load_chain): Likewise.
5483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5484 Alan Hayward <alan.hayward@arm.com>
5485 David Sherwood <david.sherwood@arm.com>
5487 * machmode.h (mode_nunits): Change from unsigned char to
5489 (ONLY_FIXED_SIZE_MODES): New macro.
5490 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5491 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5492 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5494 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5495 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5496 or if measurement_type is not polynomial.
5497 * genmodes.c (ZERO_COEFFS): New macro.
5498 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5500 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5501 Use ZERO_COEFFS when emitting initializers.
5502 * data-streamer.h (bp_pack_poly_value): New function.
5503 (bp_unpack_poly_value): Likewise.
5504 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5505 for GET_MODE_NUNITS.
5506 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5507 for GET_MODE_NUNITS.
5508 * tree.c (make_vector_type): Remove temporary shim and make
5509 the real function take the number of units as a poly_uint64
5511 (build_vector_type_for_mode): Handle polynomial nunits.
5512 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5513 * emit-rtl.c (const_vec_series_p_1): Likewise.
5514 (gen_rtx_CONST_VECTOR): Likewise.
5515 * fold-const.c (test_vec_duplicate_folding): Likewise.
5516 * genrecog.c (validate_pattern): Likewise.
5517 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5518 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5519 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5520 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5521 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5522 * rtlanal.c (subreg_get_info): Likewise.
5523 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5524 (vect_grouped_load_supported): Likewise.
5525 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5526 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5527 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5528 (simplify_const_unary_operation, simplify_binary_operation_1)
5529 (simplify_const_binary_operation, simplify_ternary_operation)
5530 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5531 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5532 instead of CONST_VECTOR_NUNITS.
5533 * varasm.c (output_constant_pool_2): Likewise.
5534 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5535 explicit-encoded elements in the XVEC for variable-length vectors.
5537 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5539 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5541 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5542 Alan Hayward <alan.hayward@arm.com>
5543 David Sherwood <david.sherwood@arm.com>
5545 * coretypes.h (fixed_size_mode): Declare.
5546 (fixed_size_mode_pod): New typedef.
5547 * builtins.h (target_builtins::x_apply_args_mode)
5548 (target_builtins::x_apply_result_mode): Change type to
5549 fixed_size_mode_pod.
5550 * builtins.c (apply_args_size, apply_result_size, result_vector)
5551 (expand_builtin_apply_args_1, expand_builtin_apply)
5552 (expand_builtin_return): Update accordingly.
5554 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5556 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5557 * cselib.c (cselib_hash_rtx): Likewise.
5558 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5559 CONST_VECTOR encoding.
5561 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5562 Jeff Law <law@redhat.com>
5565 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5566 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5567 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5568 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5571 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5572 explicitly probe *sp in a noreturn function if there were any callee
5573 register saves or frame pointer is needed.
5575 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5578 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5579 BLKmode for ternary, binary or unary expressions.
5582 * var-tracking.c (delete_vta_debug_insn): New inline function.
5583 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5584 insns from get_insns () to NULL instead of each bb separately.
5585 Use delete_vta_debug_insn. No longer static.
5586 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5587 delete_vta_debug_insns callers.
5588 * rtl.h (delete_vta_debug_insns): Declare.
5589 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5590 instead of variable_tracking_main.
5592 2018-01-03 Martin Sebor <msebor@redhat.com>
5594 PR tree-optimization/83603
5595 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5596 arguments past the endof the argument list in functions declared
5597 without a prototype.
5598 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5599 Avoid checking when arguments are null.
5601 2018-01-03 Martin Sebor <msebor@redhat.com>
5604 * doc/extend.texi (attribute const): Fix a typo.
5605 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5606 issuing -Wsuggest-attribute for void functions.
5608 2018-01-03 Martin Sebor <msebor@redhat.com>
5610 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5611 offset_int::from instead of wide_int::to_shwi.
5612 (maybe_diag_overlap): Remove assertion.
5613 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5614 * gimple-ssa-sprintf.c (format_directive): Same.
5615 (parse_directive): Same.
5616 (sprintf_dom_walker::compute_format_length): Same.
5617 (try_substitute_return_value): Same.
5619 2018-01-03 Jeff Law <law@redhat.com>
5622 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5623 non-constant residual for zero at runtime and avoid probing in
5624 that case. Reorganize code for trailing problem to mirror handling
5627 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5629 PR tree-optimization/83501
5630 * tree-ssa-strlen.c (get_string_cst): New.
5631 (handle_char_store): Call get_string_cst.
5633 2018-01-03 Martin Liska <mliska@suse.cz>
5635 PR tree-optimization/83593
5636 * tree-ssa-strlen.c: Include tree-cfg.h.
5637 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5638 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5639 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5641 (strlen_dom_walker::before_dom_children): Call
5642 gimple_purge_dead_eh_edges. Dump tranformation with details
5644 (strlen_dom_walker::before_dom_children): Update call by adding
5645 new argument cleanup_eh.
5646 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5648 2018-01-03 Martin Liska <mliska@suse.cz>
5651 * cif-code.def (VARIADIC_THUNK): New enum value.
5652 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5655 2018-01-03 Jan Beulich <jbeulich@suse.com>
5657 * sse.md (mov<mode>_internal): Tighten condition for when to use
5658 vmovdqu<ssescalarsize> for TI and OI modes.
5660 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5662 Update copyright years.
5664 2018-01-03 Martin Liska <mliska@suse.cz>
5667 * ipa-visibility.c (function_and_variable_visibility): Skip
5668 functions with noipa attribure.
5670 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5672 * gcc.c (process_command): Update copyright notice dates.
5673 * gcov-dump.c (print_version): Ditto.
5674 * gcov.c (print_version): Ditto.
5675 * gcov-tool.c (print_version): Ditto.
5676 * gengtype.c (create_file): Ditto.
5677 * doc/cpp.texi: Bump @copying's copyright year.
5678 * doc/cppinternals.texi: Ditto.
5679 * doc/gcc.texi: Ditto.
5680 * doc/gccint.texi: Ditto.
5681 * doc/gcov.texi: Ditto.
5682 * doc/install.texi: Ditto.
5683 * doc/invoke.texi: Ditto.
5685 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5687 * vector-builder.h (vector_builder::m_full_nelts): Change from
5688 unsigned int to poly_uint64.
5689 (vector_builder::full_nelts): Update prototype accordingly.
5690 (vector_builder::new_vector): Likewise.
5691 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5692 (vector_builder::operator ==): Likewise.
5693 (vector_builder::finalize): Likewise.
5694 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5695 Take the number of elements as a poly_uint64 rather than an
5697 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5698 from unsigned int to poly_uint64.
5699 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5700 (vec_perm_indices::new_vector): Likewise.
5701 (vec_perm_indices::length): Likewise.
5702 (vec_perm_indices::nelts_per_input): Likewise.
5703 (vec_perm_indices::input_nelts): Likewise.
5704 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5705 number of elements per input as a poly_uint64 rather than an
5706 unsigned int. Use the original encoding for variable-length
5707 vectors, rather than clamping each individual element.
5708 For the second and subsequent elements in each pattern,
5709 clamp the step and base before clamping their sum.
5710 (vec_perm_indices::series_p): Handle polynomial element counts.
5711 (vec_perm_indices::all_in_range_p): Likewise.
5712 (vec_perm_indices_to_tree): Likewise.
5713 (vec_perm_indices_to_rtx): Likewise.
5714 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5715 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5716 (tree_vector_builder::new_binary_operation): Handle polynomial
5717 element counts. Return false if we need to know the number
5718 of elements at compile time.
5719 * fold-const.c (fold_vec_perm): Punt if the number of elements
5720 isn't known at compile time.
5722 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5724 * vec-perm-indices.h (vec_perm_builder): Change element type
5725 from HOST_WIDE_INT to poly_int64.
5726 (vec_perm_indices::element_type): Update accordingly.
5727 (vec_perm_indices::clamp): Handle polynomial element_types.
5728 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5729 (vec_perm_indices::all_in_range_p): Likewise.
5730 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5732 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5733 polynomial vec_perm_indices element types.
5734 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5735 * fold-const.c (fold_vec_perm): Likewise.
5736 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5737 * tree-vect-generic.c (lower_vec_perm): Likewise.
5738 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5739 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5740 element type to HOST_WIDE_INT.
5742 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5743 Alan Hayward <alan.hayward@arm.com>
5744 David Sherwood <david.sherwood@arm.com>
5746 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5747 rather than an int. Use plus_constant.
5748 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5749 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5752 Alan Hayward <alan.hayward@arm.com>
5753 David Sherwood <david.sherwood@arm.com>
5755 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5756 a HOST_WIDE_INT to a poly_int64.
5758 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5759 Alan Hayward <alan.hayward@arm.com>
5760 David Sherwood <david.sherwood@arm.com>
5762 * calls.c (load_register_parameters): Cope with polynomial
5763 mode sizes. Require a constant size for BLKmode parameters
5764 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5765 forces a parameter to be padded at the lsb end in order to
5766 fill a complete number of words, require the parameter size
5767 to be ordered wrt UNITS_PER_WORD.
5769 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5770 Alan Hayward <alan.hayward@arm.com>
5771 David Sherwood <david.sherwood@arm.com>
5773 * reload1.c (spill_stack_slot_width): Change element type
5774 from unsigned int to poly_uint64_pod.
5775 (alter_reg): Treat mode sizes as polynomial.
5777 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5778 Alan Hayward <alan.hayward@arm.com>
5779 David Sherwood <david.sherwood@arm.com>
5781 * reload.c (complex_word_subreg_p): New function.
5782 (reload_inner_reg_of_subreg, push_reload): Use it.
5784 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5785 Alan Hayward <alan.hayward@arm.com>
5786 David Sherwood <david.sherwood@arm.com>
5788 * lra-constraints.c (process_alt_operands): Reject matched
5789 operands whose sizes aren't ordered.
5790 (match_reload): Refer to this check here.
5792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5793 Alan Hayward <alan.hayward@arm.com>
5794 David Sherwood <david.sherwood@arm.com>
5796 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5797 that the mode size is in the set {1, 2, 4, 8, 16}.
5799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5800 Alan Hayward <alan.hayward@arm.com>
5801 David Sherwood <david.sherwood@arm.com>
5803 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5804 Use plus_constant instead of gen_rtx_PLUS.
5806 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5807 Alan Hayward <alan.hayward@arm.com>
5808 David Sherwood <david.sherwood@arm.com>
5810 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5811 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5812 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5813 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5814 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5815 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5816 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5817 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5818 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5819 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5821 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5822 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5823 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5824 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5825 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5826 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5827 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5828 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5829 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5830 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5832 * expr.c (emit_move_resolve_push): Treat the input and result
5833 of PUSH_ROUNDING as a poly_int64.
5834 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5835 (emit_push_insn): Likewise.
5836 * lra-eliminations.c (mark_not_eliminable): Likewise.
5837 * recog.c (push_operand): Likewise.
5838 * reload1.c (elimination_effects): Likewise.
5839 * rtlanal.c (nonzero_bits1): Likewise.
5840 * calls.c (store_one_arg): Likewise. Require the padding to be
5841 known at compile time.
5843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5844 Alan Hayward <alan.hayward@arm.com>
5845 David Sherwood <david.sherwood@arm.com>
5847 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5848 Use plus_constant instead of gen_rtx_PLUS.
5850 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5851 Alan Hayward <alan.hayward@arm.com>
5852 David Sherwood <david.sherwood@arm.com>
5854 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5857 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5858 Alan Hayward <alan.hayward@arm.com>
5859 David Sherwood <david.sherwood@arm.com>
5861 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5862 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5863 via stack temporaries. Treat the mode size as polynomial too.
5865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5866 Alan Hayward <alan.hayward@arm.com>
5867 David Sherwood <david.sherwood@arm.com>
5869 * expr.c (expand_expr_real_2): When handling conversions involving
5870 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5871 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5872 as a poly_uint64 too.
5874 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5875 Alan Hayward <alan.hayward@arm.com>
5876 David Sherwood <david.sherwood@arm.com>
5878 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5880 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5881 Alan Hayward <alan.hayward@arm.com>
5882 David Sherwood <david.sherwood@arm.com>
5884 * combine.c (can_change_dest_mode): Handle polynomial
5885 REGMODE_NATURAL_SIZE.
5886 * expmed.c (store_bit_field_1): Likewise.
5887 * expr.c (store_constructor): Likewise.
5888 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5889 and polynomial REGMODE_NATURAL_SIZE.
5890 (gen_lowpart_common): Likewise.
5891 * reginfo.c (record_subregs_of_mode): Likewise.
5892 * rtlanal.c (read_modify_subreg_p): Likewise.
5894 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5895 Alan Hayward <alan.hayward@arm.com>
5896 David Sherwood <david.sherwood@arm.com>
5898 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5899 numbers of elements.
5901 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5902 Alan Hayward <alan.hayward@arm.com>
5903 David Sherwood <david.sherwood@arm.com>
5905 * match.pd: Cope with polynomial numbers of vector elements.
5907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5908 Alan Hayward <alan.hayward@arm.com>
5909 David Sherwood <david.sherwood@arm.com>
5911 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5912 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
5919 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5920 (ipa_simd_modify_function_body): Likewise.
5922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5923 Alan Hayward <alan.hayward@arm.com>
5924 David Sherwood <david.sherwood@arm.com>
5926 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5927 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5928 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5929 (expand_vector_condition, vector_element): Likewise.
5930 (subparts_gt): New function.
5931 (get_compute_type): Use subparts_gt.
5932 (count_type_subparts): Delete.
5933 (expand_vector_operations_1): Use subparts_gt instead of
5934 count_type_subparts.
5936 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5937 Alan Hayward <alan.hayward@arm.com>
5938 David Sherwood <david.sherwood@arm.com>
5940 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5941 (vect_compile_time_alias): ...this new function. Do the calculation
5942 on poly_ints rather than trees.
5943 (vect_prune_runtime_alias_test_list): Update call accordingly.
5945 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5946 Alan Hayward <alan.hayward@arm.com>
5947 David Sherwood <david.sherwood@arm.com>
5949 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5951 (vect_schedule_slp_instance): Likewise.
5953 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5954 Alan Hayward <alan.hayward@arm.com>
5955 David Sherwood <david.sherwood@arm.com>
5957 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5958 constant and extern definitions for variable-length vectors.
5959 (vect_get_constant_vectors): Note that the number of units
5960 is known to be constant.
5962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5963 Alan Hayward <alan.hayward@arm.com>
5964 David Sherwood <david.sherwood@arm.com>
5966 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5967 of units as polynomial. Choose between WIDE and NARROW based
5970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5971 Alan Hayward <alan.hayward@arm.com>
5972 David Sherwood <david.sherwood@arm.com>
5974 * tree-vect-stmts.c (simd_clone_subparts): New function.
5975 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5978 Alan Hayward <alan.hayward@arm.com>
5979 David Sherwood <david.sherwood@arm.com>
5981 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5982 vectors as polynomial. Use build_index_vector for
5985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5986 Alan Hayward <alan.hayward@arm.com>
5987 David Sherwood <david.sherwood@arm.com>
5989 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5990 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5991 for variable-length vectors.
5992 (vectorizable_mask_load_store): Treat the number of units as
5993 polynomial, asserting that it is constant if the condition has
5994 already been enforced.
5995 (vectorizable_store, vectorizable_load): Likewise.
5997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5998 Alan Hayward <alan.hayward@arm.com>
5999 David Sherwood <david.sherwood@arm.com>
6001 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6002 of units as polynomial. Punt if we can't tell at compile time
6003 which vector contains the final result.
6005 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6006 Alan Hayward <alan.hayward@arm.com>
6007 David Sherwood <david.sherwood@arm.com>
6009 * tree-vect-loop.c (vectorizable_induction): Treat the number
6010 of units as polynomial. Punt on SLP inductions. Use an integer
6011 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6012 cast of such a series for variable-length floating-point
6015 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6016 Alan Hayward <alan.hayward@arm.com>
6017 David Sherwood <david.sherwood@arm.com>
6019 * tree.h (build_index_vector): Declare.
6020 * tree.c (build_index_vector): New function.
6021 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6022 of units as polynomial, forcibly converting it to a constant if
6023 vectorizable_reduction has already enforced the condition.
6024 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6025 to create a {1,2,3,...} vector.
6026 (vectorizable_reduction): Treat the number of units as polynomial.
6027 Choose vectype_in based on the largest scalar element size rather
6028 than the smallest number of units. Enforce the restrictions
6031 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6032 Alan Hayward <alan.hayward@arm.com>
6033 David Sherwood <david.sherwood@arm.com>
6035 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6036 number of units as polynomial.
6038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6039 Alan Hayward <alan.hayward@arm.com>
6040 David Sherwood <david.sherwood@arm.com>
6042 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6043 * target.def (autovectorize_vector_sizes): Return the vector sizes
6044 by pointer, using vector_sizes rather than a bitmask.
6045 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6046 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6047 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6049 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6050 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6051 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6052 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6053 * omp-general.c (omp_max_vf): Likewise.
6054 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6055 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6056 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6057 * tree-vect-slp.c (vect_slp_bb): Likewise.
6058 * doc/tm.texi: Regenerate.
6059 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6061 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6062 the vector size as a poly_uint64 rather than an unsigned int.
6063 (current_vector_size): Change from an unsigned int to a poly_uint64.
6064 (get_vectype_for_scalar_type): Update accordingly.
6065 * tree.h (build_truth_vector_type): Take the size and number of
6066 units as a poly_uint64 rather than an unsigned int.
6067 (build_vector_type): Add a temporary overload that takes
6068 the number of units as a poly_uint64 rather than an unsigned int.
6069 * tree.c (make_vector_type): Likewise.
6070 (build_truth_vector_type): Take the number of units as a poly_uint64
6071 rather than an unsigned int.
6073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6074 Alan Hayward <alan.hayward@arm.com>
6075 David Sherwood <david.sherwood@arm.com>
6077 * target.def (get_mask_mode): Take the number of units and length
6078 as poly_uint64s rather than unsigned ints.
6079 * targhooks.h (default_get_mask_mode): Update accordingly.
6080 * targhooks.c (default_get_mask_mode): Likewise.
6081 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6082 * doc/tm.texi: Regenerate.
6084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6085 Alan Hayward <alan.hayward@arm.com>
6086 David Sherwood <david.sherwood@arm.com>
6088 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6089 * omp-general.c (omp_max_vf): Likewise.
6090 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6091 (expand_omp_simd): Handle polynomial safelen.
6092 * omp-low.c (omplow_simd_context): Add a default constructor.
6093 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6094 (lower_rec_simd_input_clauses): Update accordingly.
6095 (lower_rec_input_clauses): Likewise.
6097 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6098 Alan Hayward <alan.hayward@arm.com>
6099 David Sherwood <david.sherwood@arm.com>
6101 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6102 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6103 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6104 (vect_analyze_slp_cost): Likewise.
6105 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6106 (vect_model_load_cost): Likewise.
6108 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6109 Alan Hayward <alan.hayward@arm.com>
6110 David Sherwood <david.sherwood@arm.com>
6112 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6113 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6114 from an unsigned int * to a poly_uint64_pod *.
6115 (calculate_unrolling_factor): New function.
6116 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6118 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6119 Alan Hayward <alan.hayward@arm.com>
6120 David Sherwood <david.sherwood@arm.com>
6122 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6123 from an unsigned int to a poly_uint64.
6124 (_loop_vec_info::slp_unrolling_factor): Likewise.
6125 (_loop_vec_info::vectorization_factor): Change from an int
6127 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6128 (vect_get_num_vectors): New function.
6129 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6130 (vect_get_num_copies): Use vect_get_num_vectors.
6131 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6132 to an unsigned int *.
6133 (vect_analyze_data_refs): Change min_vf from an int * to a
6135 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6136 than an unsigned HOST_WIDE_INT.
6137 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6138 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6139 to an unsigned int *.
6140 (vect_analyze_data_ref_dependences): Likewise.
6141 (vect_compute_data_ref_alignment): Handle polynomial vf.
6142 (vect_enhance_data_refs_alignment): Likewise.
6143 (vect_prune_runtime_alias_test_list): Likewise.
6144 (vect_shift_permute_load_chain): Likewise.
6145 (vect_supportable_dr_alignment): Likewise.
6146 (dependence_distance_ge_vf): Take the vectorization factor as a
6147 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6148 (vect_analyze_data_refs): Change min_vf from an int * to a
6150 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6151 vfm1 as a poly_uint64 rather than an int. Make the same change
6152 for the returned bound_scalar.
6153 (vect_gen_vector_loop_niters): Handle polynomial vf.
6154 (vect_do_peeling): Likewise. Update call to
6155 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6156 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6158 * tree-vect-loop.c (vect_determine_vectorization_factor)
6159 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6160 (vect_get_known_peeling_cost): Likewise.
6161 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6162 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6163 (vect_transform_loop): Likewise. Use the lowest possible VF when
6164 updating the upper bounds of the loop.
6165 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6167 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6168 polynomial unroll factors.
6169 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6170 (vect_make_slp_decision): Likewise.
6171 (vect_supported_load_permutation_p): Likewise, and polynomial
6173 (vect_analyze_slp_cost): Handle polynomial vf.
6174 (vect_slp_analyze_node_operations): Likewise.
6175 (vect_slp_analyze_bb_1): Likewise.
6176 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6177 than an unsigned HOST_WIDE_INT.
6178 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6179 (vectorizable_load): Handle polynomial vf.
6180 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6182 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6184 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6185 Alan Hayward <alan.hayward@arm.com>
6186 David Sherwood <david.sherwood@arm.com>
6188 * match.pd: Handle bit operations involving three constants
6189 and try to fold one pair.
6191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6193 * tree-vect-loop-manip.c: Include gimple-fold.h.
6194 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6195 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6196 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6197 Add a path that uses a step of VF instead of 1, but disable it
6199 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6200 and niters_no_overflow parameters. Update calls to
6201 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6202 Create a new SSA name if the latter choses to use a ste other
6203 than zero, and return it via niters_vector_mult_vf_var.
6204 * tree-vect-loop.c (vect_transform_loop): Update calls to
6205 vect_do_peeling, vect_gen_vector_loop_niters and
6206 slpeel_make_loop_iterate_ntimes.
6207 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6208 (vect_gen_vector_loop_niters): Update declarations after above changes.
6210 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6212 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6213 128-bit round to integer instructions.
6214 (ceil<mode>2): Likewise.
6215 (btrunc<mode>2): Likewise.
6216 (round<mode>2): Likewise.
6218 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6220 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6221 unaligned VSX load/store on P8/P9.
6222 (expand_block_clear): Allow the use of unaligned VSX
6223 load/store on P8/P9.
6225 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6227 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6229 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6230 swap associated with both a load and a store.
6232 2018-01-02 Andrew Waterman <andrew@sifive.com>
6234 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6235 * config/riscv/riscv.md (clear_cache): Use it.
6237 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6239 * web.c: Remove out-of-date comment.
6241 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6243 * expr.c (fixup_args_size_notes): Check that any existing
6244 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6245 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6246 (emit_single_push_insn): ...here.
6248 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6250 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6251 (const_vector_encoded_nelts): New function.
6252 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6253 (const_vector_int_elt, const_vector_elt): Declare.
6254 * emit-rtl.c (const_vector_int_elt_1): New function.
6255 (const_vector_elt): Likewise.
6256 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6257 of CONST_VECTOR_ELT.
6259 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6261 * expr.c: Include rtx-vector-builder.h.
6262 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6263 directly on the tree encoding.
6264 (const_vector_from_tree): Likewise.
6265 * optabs.c: Include rtx-vector-builder.h.
6266 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6267 sequence of "u" values.
6268 * vec-perm-indices.c: Include rtx-vector-builder.h.
6269 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6270 directly on the vec_perm_indices encoding.
6272 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6274 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6275 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6276 * rtx-vector-builder.h: New file.
6277 * rtx-vector-builder.c: Likewise.
6278 * rtl.h (rtx_def::u2): Add a const_vector field.
6279 (CONST_VECTOR_NPATTERNS): New macro.
6280 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6281 (CONST_VECTOR_DUPLICATE_P): Likewise.
6282 (CONST_VECTOR_STEPPED_P): Likewise.
6283 (CONST_VECTOR_ENCODED_ELT): Likewise.
6284 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6285 (unwrap_const_vec_duplicate): Likewise.
6286 (const_vec_series_p): Check for a non-duplicated vector encoding.
6287 Say that the function only returns true for integer vectors.
6288 * emit-rtl.c: Include rtx-vector-builder.h.
6289 (gen_const_vec_duplicate_1): Delete.
6290 (gen_const_vector): Call gen_const_vec_duplicate instead of
6291 gen_const_vec_duplicate_1.
6292 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6293 (gen_const_vec_duplicate): Use rtx_vector_builder.
6294 (gen_const_vec_series): Likewise.
6295 (gen_rtx_CONST_VECTOR): Likewise.
6296 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6297 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6298 Build a new vector rather than modifying a CONST_VECTOR in-place.
6299 (handle_special_swappables): Update call accordingly.
6300 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6301 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6302 Build a new vector rather than modifying a CONST_VECTOR in-place.
6303 (handle_special_swappables): Update call accordingly.
6305 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6307 * simplify-rtx.c (simplify_const_binary_operation): Use
6308 CONST_VECTOR_ELT instead of XVECEXP.
6310 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6312 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6313 the selector elements to be different from the data elements
6314 if the selector is a VECTOR_CST.
6315 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6316 ssizetype for the selector.
6318 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6320 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6321 before testing each element individually.
6322 * tree-vect-generic.c (lower_vec_perm): Likewise.
6324 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6326 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6327 * selftest-run-tests.c (selftest::run_tests): Call it.
6328 * vector-builder.h (vector_builder::operator ==): New function.
6329 (vector_builder::operator !=): Likewise.
6330 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6331 (vec_perm_indices::all_from_input_p): New function.
6332 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6333 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6334 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6335 instead of reading the VECTOR_CST directly. Detect whether both
6336 vector inputs are the same before constructing the vec_perm_indices,
6337 and update the number of inputs argument accordingly. Use the
6338 utility functions added above. Only construct sel2 if we need to.
6340 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6342 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6343 the broadcast of the low byte.
6344 (expand_mult_highpart): Use an explicit encoding for the permutes.
6345 * optabs-query.c (can_mult_highpart_p): Likewise.
6346 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6347 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6348 (vectorizable_bswap): Likewise.
6349 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6350 explicit encoding for the power-of-2 permutes.
6351 (vect_permute_store_chain): Likewise.
6352 (vect_grouped_load_supported): Likewise.
6353 (vect_permute_load_chain): Likewise.
6355 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6357 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6358 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6359 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6360 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6361 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6362 (vect_gen_perm_mask_any): Likewise.
6364 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6366 * int-vector-builder.h: New file.
6367 * vec-perm-indices.h: Include int-vector-builder.h.
6368 (vec_perm_indices): Redefine as an int_vector_builder.
6369 (auto_vec_perm_indices): Delete.
6370 (vec_perm_builder): Redefine as a stand-alone class.
6371 (vec_perm_indices::vec_perm_indices): New function.
6372 (vec_perm_indices::clamp): Likewise.
6373 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6374 (vec_perm_indices::new_vector): New function.
6375 (vec_perm_indices::new_expanded_vector): Update for new
6376 vec_perm_indices class.
6377 (vec_perm_indices::rotate_inputs): New function.
6378 (vec_perm_indices::all_in_range_p): Operate directly on the
6379 encoded form, without computing elided elements.
6380 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6381 encoding. Update for new vec_perm_indices class.
6382 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6383 the given vec_perm_builder.
6384 (expand_vec_perm_var): Update vec_perm_builder constructor.
6385 (expand_mult_highpart): Use vec_perm_builder instead of
6386 auto_vec_perm_indices.
6387 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6388 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6389 or double series encoding as appropriate.
6390 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6391 vec_perm_indices instead of auto_vec_perm_indices.
6392 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6393 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6394 (vect_permute_store_chain): Likewise.
6395 (vect_grouped_load_supported): Likewise.
6396 (vect_permute_load_chain): Likewise.
6397 (vect_shift_permute_load_chain): Likewise.
6398 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6399 (vect_transform_slp_perm_load): Likewise.
6400 (vect_schedule_slp_instance): Likewise.
6401 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6402 (vectorizable_mask_load_store): Likewise.
6403 (vectorizable_bswap): Likewise.
6404 (vectorizable_store): Likewise.
6405 (vectorizable_load): Likewise.
6406 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6407 vec_perm_indices instead of auto_vec_perm_indices. Use
6408 tree_to_vec_perm_builder to read the vector from a tree.
6409 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6410 vec_perm_builder instead of a vec_perm_indices.
6411 (have_whole_vector_shift): Use vec_perm_builder and
6412 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6413 truncation to calc_vec_perm_mask_for_shift.
6414 (vect_create_epilog_for_reduction): Likewise.
6415 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6416 from auto_vec_perm_indices to vec_perm_indices.
6417 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6418 instead of changing individual elements.
6419 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6420 the vector in d.perm.
6421 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6422 from auto_vec_perm_indices to vec_perm_indices.
6423 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6424 instead of changing individual elements.
6425 (arm_vectorize_vec_perm_const): Use new_vector to install
6426 the vector in d.perm.
6427 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6428 Update vec_perm_builder constructor.
6429 (rs6000_expand_interleave): Likewise.
6430 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6431 (rs6000_expand_interleave): Likewise.
6433 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6435 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6436 to qimode could truncate the indices.
6437 * optabs.c (expand_vec_perm_var): Likewise.
6439 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6441 * Makefile.in (OBJS): Add vec-perm-indices.o.
6442 * vec-perm-indices.h: New file.
6443 * vec-perm-indices.c: Likewise.
6444 * target.h (vec_perm_indices): Replace with a forward class
6446 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6447 * optabs.h: Include vec-perm-indices.h.
6448 (expand_vec_perm): Delete.
6449 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6450 (expand_vec_perm_const): Declare.
6451 * target.def (vec_perm_const_ok): Replace with...
6452 (vec_perm_const): ...this new hook.
6453 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6454 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6455 * doc/tm.texi: Regenerate.
6456 * optabs.def (vec_perm_const): Delete.
6457 * doc/md.texi (vec_perm_const): Likewise.
6458 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6459 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6460 expand_vec_perm for constant permutation vectors. Assert that
6461 the mode of variable permutation vectors is the integer equivalent
6462 of the mode that is being permuted.
6463 * optabs-query.h (selector_fits_mode_p): Declare.
6464 * optabs-query.c: Include vec-perm-indices.h.
6465 (selector_fits_mode_p): New function.
6466 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6467 is defined, instead of checking whether the vec_perm_const_optab
6468 exists. Use targetm.vectorize.vec_perm_const instead of
6469 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6470 fit in the vector mode before using a variable permute.
6471 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6472 vec_perm_indices instead of an rtx.
6473 (expand_vec_perm): Replace with...
6474 (expand_vec_perm_const): ...this new function. Take the selector
6475 as a vec_perm_indices rather than an rtx. Also take the mode of
6476 the selector. Update call to shift_amt_for_vec_perm_mask.
6477 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6478 Use vec_perm_indices::new_expanded_vector to expand the original
6479 selector into bytes. Check whether the indices fit in the vector
6480 mode before using a variable permute.
6481 (expand_vec_perm_var): Make global.
6482 (expand_mult_highpart): Use expand_vec_perm_const.
6483 * fold-const.c: Includes vec-perm-indices.h.
6484 * tree-ssa-forwprop.c: Likewise.
6485 * tree-vect-data-refs.c: Likewise.
6486 * tree-vect-generic.c: Likewise.
6487 * tree-vect-loop.c: Likewise.
6488 * tree-vect-slp.c: Likewise.
6489 * tree-vect-stmts.c: Likewise.
6490 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6492 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6493 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6494 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6495 (aarch64_vectorize_vec_perm_const): ...this new function.
6496 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6497 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6498 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6499 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6500 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6501 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6502 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6504 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6505 check for NEON modes.
6506 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6507 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6508 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6509 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6511 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6512 the old VEC_PERM_CONST conditions.
6513 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6514 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6515 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6516 (ia64_vectorize_vec_perm_const_ok): Merge into...
6517 (ia64_vectorize_vec_perm_const): ...this new function.
6518 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6519 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6520 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6521 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6522 * config/mips/mips.c (mips_expand_vec_perm_const)
6523 (mips_vectorize_vec_perm_const_ok): Merge into...
6524 (mips_vectorize_vec_perm_const): ...this new function.
6525 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6526 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6527 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6528 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6529 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6530 (rs6000_expand_vec_perm_const): Delete.
6531 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6533 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6534 (altivec_expand_vec_perm_const_le): Take each operand individually.
6535 Operate on constant selectors rather than rtxes.
6536 (altivec_expand_vec_perm_const): Likewise. Update call to
6537 altivec_expand_vec_perm_const_le.
6538 (rs6000_expand_vec_perm_const): Delete.
6539 (rs6000_vectorize_vec_perm_const_ok): Delete.
6540 (rs6000_vectorize_vec_perm_const): New function.
6541 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6542 an element count and rtx array.
6543 (rs6000_expand_extract_even): Update call accordingly.
6544 (rs6000_expand_interleave): Likewise.
6545 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6546 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6547 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6548 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6549 (rs6000_expand_vec_perm_const): Delete.
6550 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6551 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6552 (altivec_expand_vec_perm_const_le): Take each operand individually.
6553 Operate on constant selectors rather than rtxes.
6554 (altivec_expand_vec_perm_const): Likewise. Update call to
6555 altivec_expand_vec_perm_const_le.
6556 (rs6000_expand_vec_perm_const): Delete.
6557 (rs6000_vectorize_vec_perm_const_ok): Delete.
6558 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6559 reference to the SPE evmerge intructions.
6560 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6561 an element count and rtx array.
6562 (rs6000_expand_extract_even): Update call accordingly.
6563 (rs6000_expand_interleave): Likewise.
6564 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6565 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6567 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6569 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6571 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6572 vector mode and that that mode matches the mode of the data
6574 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6575 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6576 directly using expand_vec_perm_1 when forcing selectors into
6578 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6580 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6582 * optabs-query.h (can_vec_perm_p): Delete.
6583 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6584 * optabs-query.c (can_vec_perm_p): Split into...
6585 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6586 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6587 particular selector is valid.
6588 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6589 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6590 (vect_grouped_load_supported): Likewise.
6591 (vect_shift_permute_load_chain): Likewise.
6592 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6593 (vect_transform_slp_perm_load): Likewise.
6594 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6595 (vectorizable_bswap): Likewise.
6596 (vect_gen_perm_mask_checked): Likewise.
6597 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6598 implementations of variable permutation vectors into account
6599 when deciding which selector to use.
6600 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6601 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6602 with a false third argument.
6603 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6604 to test whether the constant selector is valid and can_vec_perm_var_p
6605 to test whether a variable selector is valid.
6607 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6609 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6610 * optabs-query.c (can_vec_perm_p): Likewise.
6611 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6612 instead of vec_perm_indices.
6613 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6614 (vect_gen_perm_mask_checked): Likewise,
6615 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6616 (vect_gen_perm_mask_checked): Likewise,
6618 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6620 * optabs-query.h (qimode_for_vec_perm): Declare.
6621 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6622 (qimode_for_vec_perm): ...this new function.
6623 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6625 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6627 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6628 does not have a conditional at the top.
6630 2018-01-02 Richard Biener <rguenther@suse.de>
6632 * ipa-inline.c (big_speedup_p): Fix expression.
6634 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6637 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6640 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6644 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6645 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6646 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6647 cond_taken_branch_cost 3->4.
6649 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6651 PR tree-optimization/83581
6652 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6653 TODO_cleanup_cfg if any changes have been made.
6656 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6657 convert_modes if target mode has the right side, but different mode
6661 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6662 last argument when extracting from CONCAT. If either from_real or
6663 from_imag is NULL, use expansion through memory. If result is not
6664 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6665 the parts directly to inner mode, if even that fails, use expansion
6669 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6670 check for bswap in mode rather than HImode and use that in expand_unop
6673 Copyright (C) 2018 Free Software Foundation, Inc.
6675 Copying and distribution of this file, with or without modification,
6676 are permitted in any medium without royalty provided the copyright
6677 notice and this notice are preserved.