1 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
4 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
5 multiply in 32-bit mode.
6 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
7 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
10 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
12 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
13 to allow or block "symbol_ref" depending on value of TARGET_JSR
14 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
15 call_value_internal insns
17 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
20 * dwarf2out.c (is_trivial_indirect_ref): New function.
21 (dwarf2out_late_global_decl): Do not generate a location
22 attribute for variables that have a non-trivial DECL_VALUE_EXPR
23 and that are not defined in the current unit.
25 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
27 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
28 instead of a libcall for UNORDERED.
30 2018-02-09 Tamar Christina <tamar.christina@arm.com>
33 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
34 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
36 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
39 * config/s390/s390.c (s390_set_current_function): Invoke
40 s390_indirect_branch_settings also if fndecl didn't change.
42 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
44 * config/rs6000/rs6000.md (blockage): Set length to zero.
46 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
48 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
50 2018-02-09 Jakub Jelinek <jakub@redhat.com>
53 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
54 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
58 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
59 PARALLEL incoming that failed vt_get_decl_and_offset check.
62 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
63 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
65 (get_variable_section): For decls in named .bss* sections pass true as
66 second argument to bss_initializer_p.
68 2018-02-09 Marek Polacek <polacek@redhat.com>
69 Jakub Jelinek <jakub@redhat.com>
72 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
73 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
74 Sync some changes from cxx_fold_indirect_ref.
76 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
78 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
80 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
81 (BLOCK_INLINE_ENTRY_LABEL): New.
82 (dwarf2out_var_location): Disregard inline entry markers.
83 (inline_entry_data): New struct.
84 (inline_entry_data_hasher): New hashtable type.
85 (inline_entry_data_hasher::hash): New.
86 (inline_entry_data_hasher::equal): New.
87 (inline_entry_data_table): New variable.
88 (add_high_low_attributes): Add DW_AT_entry_pc and
89 DW_AT_GNU_entry_view attributes if a pending entry is found
90 in inline_entry_data_table. Add old entry_pc attribute only
91 if debug nonbinding markers are disabled.
92 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
94 (block_within_block_p, dwarf2out_inline_entry): New.
95 (dwarf2out_finish): Check that no entries remained in
96 inline_entry_data_table.
97 * final.c (reemit_insn_block_notes): Handle inline entry notes.
98 (final_scan_insn, notice_source_line): Likewise.
99 (rest_of_clean_state): Skip inline entry markers.
100 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
102 * gimple.c (gimple_build_debug_inline_entry): New.
103 * gimple.h (enum gimple_debug_subcode): Add
104 GIMPLE_DEBUG_INLINE_ENTRY.
105 (gimple_build_debug_inline_entry): Declare.
106 (gimple_debug_inline_entry_p): New.
107 (gimple_debug_nonbind_marker_p): Adjust.
108 * insn-notes.def (INLINE_ENTRY): New.
109 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
110 inline entry marker notes.
111 (print_insn): Likewise.
112 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
113 (INSN_DEBUG_MARKER_KIND): Likewise.
114 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
115 * tree-inline.c (expand_call_inline): Build and insert
116 debug_inline_entry stmt.
117 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
118 inline entry blocks early, if nonbind markers are enabled.
119 (dump_scope_block): Dump fragment info.
120 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
121 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
122 (gimple_build_debug_inline_entry): New.
123 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
124 Enable/disable inline entry points too.
125 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
126 (DEBUG_INSN): Describe inline entry markers.
128 * common.opt (gvariable-location-views): New.
129 (gvariable-location-views=incompat5): New.
130 * config.in: Rebuilt.
131 * configure: Rebuilt.
132 * configure.ac: Test assembler for view support.
133 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
134 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
135 * dwarf2out.c (var_loc_view): New typedef.
136 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
137 (dwarf2out_locviews_in_attribute): New.
138 (dwarf2out_locviews_in_loclist): New.
139 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
140 (enum dw_line_info_opcode): Add LI_adv_address.
141 (struct dw_line_info_table): Add view.
142 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
143 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
144 (zero_view_p): New variable.
145 (ZERO_VIEW_P): New macro.
146 (output_asm_line_debug_info): New.
147 (struct var_loc_node): Add view.
148 (add_AT_view_list, AT_loc_list): New.
149 (add_var_loc_to_decl): Add view param. Test it against last.
150 (new_loc_list): Add view params. Record them.
151 (AT_loc_list_ptr): Handle loc and view lists.
152 (view_list_to_loc_list_val_node): New.
153 (print_dw_val): Handle dw_val_class_view_list.
154 (size_of_die): Likewise.
155 (value_format): Likewise.
156 (loc_list_has_views): New.
157 (gen_llsym): Set vl_symbol too.
158 (maybe_gen_llsym, skip_loc_list_entry): New.
159 (dwarf2out_maybe_output_loclist_view_pair): New.
160 (output_loc_list): Output view list or entries too.
161 (output_view_list_offset): New.
162 (output_die): Handle dw_val_class_view_list.
163 (output_dwarf_version): New.
164 (output_compilation_unit_header): Use it.
165 (output_skeleton_debug_sections): Likewise.
166 (output_rnglists, output_line_info): Likewise.
167 (output_pubnames, output_aranges): Update version comments.
168 (output_one_line_info_table): Output view numbers in asm comments.
169 (dw_loc_list): Determine current endview, pass it to new_loc_list.
170 Call maybe_gen_llsym.
171 (loc_list_from_tree_1): Adjust.
172 (add_AT_location_description): Create view list attribute if
173 needed, check it's absent otherwise.
174 (convert_cfa_to_fb_loc_list): Adjust.
175 (maybe_emit_file): Call output_asm_line_debug_info for test.
176 (dwarf2out_var_location): Reset views as needed. Precompute
177 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
179 (new_line_info_table): Reset next view.
180 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
181 (dwarf2out_source_line): Likewise. Output view resets and labels to
182 the assembler, or select appropriate line info opcodes.
183 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
184 (optimize_string_length): Catch it. Adjust.
185 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
186 dw_val_class_view_list, and remove it if no longer needed.
187 (hash_loc_list): Hash view numbers.
188 (loc_list_hasher::equal): Compare them.
189 (optimize_location_lists): Check whether a view list symbol is
190 needed, and whether the locview attribute is present, and
191 whether they match. Remove the locview attribute if no longer
193 (index_location_lists): Call skip_loc_list_entry for test.
194 (dwarf2out_finish): Call output_asm_line_debug_info for test.
195 Use output_dwarf_version.
196 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
197 (struct dw_val_node): Add val_view_list.
198 * final.c (SEEN_NEXT_VIEW): New.
199 (set_next_view_needed): New.
200 (clear_next_view_needed): New.
201 (maybe_output_next_view): New.
202 (final_start_function): Rename to...
203 (final_start_function_1): ... this. Take pointer to FIRST,
204 add SEEN parameter. Emit param bindings in the initial view.
205 (final_start_function): Reintroduce SEEN-less interface.
206 (final): Rename to...
207 (final_1): ... this. Take SEEN parameter. Output final pending
208 next view at the end.
209 (final): Reintroduce seen-less interface.
210 (final_scan_insn): Output pending next view before switching
211 sections or ending a block. Mark the next view as needed when
212 outputting variable locations. Notify debug backend of section
213 changes, and of location view changes.
214 (rest_of_handle_final): Adjust.
215 * toplev.c (process_options): Autodetect value for debug variable
216 location views option. Warn on incompat5 without -gdwarf-5.
217 * doc/invoke.texi (gvariable-location-views): New.
218 (gvariable-location-views=incompat5): New.
219 (gno-variable-location-views): New.
221 2018-02-08 David Malcolm <dmalcolm@redhat.com>
223 PR tree-optimization/84136
224 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
225 that the result of find_edge is non-NULL.
227 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
230 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
231 storing integer register in SImode. Fix cost of 256 and 512
232 byte aligned SSE register store.
234 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
236 * config/i386/i386.c (ix86_multiplication_cost): Fix
237 multiplication cost for TARGET_AVX512DQ.
239 2018-02-08 Marek Polacek <polacek@redhat.com>
241 PR tree-optimization/84238
242 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
245 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
247 PR tree-optimization/84265
248 * tree-vect-stmts.c (vectorizable_store): Don't treat
249 VMAT_CONTIGUOUS accesses as grouped.
250 (vectorizable_load): Likewise.
252 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
254 PR tree-optimization/81635
255 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
256 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
257 (test_round_for_mask): New functions.
258 (wide_int_cc_tests): Call test_round_for_mask.
259 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
260 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
261 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
262 range returned by get_range_info.
264 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
267 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
268 * symtab.c: Include builtins.h
269 (symtab_node::output_to_lto_symbol_table_p): Move here
270 from lto-streamer-out.c:output_symbol_p.
271 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
272 (output_symbol_p): Move all logic to symtab.c
273 (produce_symtab): Update.
275 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
277 * config/s390/s390-opts.h (enum indirect_branch): Define.
278 * config/s390/s390-protos.h (s390_return_addr_from_memory)
279 (s390_indirect_branch_via_thunk)
280 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
281 (enum s390_indirect_branch_type): Define.
282 * config/s390/s390.c (struct s390_frame_layout, struct
283 machine_function): Remove.
284 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
285 (indirect_branch_table_label_no, indirect_branch_table_name):
287 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
288 (enum s390_indirect_branch_option): Define.
289 (s390_return_addr_from_memory): New function.
290 (s390_handle_string_attribute): New function.
291 (s390_attribute_table): Add new attribute handler.
292 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
293 (s390_indirect_branch_via_thunk): New function.
294 (s390_indirect_branch_via_inline_thunk): New function.
295 (s390_function_ok_for_sibcall): When jumping via thunk disallow
296 sibling call optimization for non z10 compiles.
297 (s390_emit_call): Force indirect branch target to be a single
298 register. Add r1 clobber for non-z10 compiles.
299 (s390_emit_epilogue): Emit return jump via return_use expander.
300 (s390_reorg): Handle JUMP_INSNs as execute targets.
301 (s390_option_override_internal): Perform validity checks for the
302 new command line options.
303 (s390_indirect_branch_attrvalue): New function.
304 (s390_indirect_branch_settings): New function.
305 (s390_set_current_function): Invoke s390_indirect_branch_settings.
306 (s390_output_indirect_thunk_function): New function.
307 (s390_code_end): Implement target hook.
308 (s390_case_values_threshold): Implement target hook.
309 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
311 * config/s390/s390.h (struct s390_frame_layout)
312 (struct machine_function): Move here from s390.c.
313 (TARGET_INDIRECT_BRANCH_NOBP_RET)
314 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
315 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
316 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
317 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
318 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
319 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
320 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
321 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
322 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
323 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
324 (mnemonic attribute): Add values which aren't recognized
326 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
327 pattern for branch conversion. Fix mnemonic attribute.
328 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
329 indirect branch via thunk if requested.
330 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
331 ("*indirect_jump"): Disable for branch conversion using out of
333 ("indirect_jump_via_thunk<mode>_z10")
334 ("indirect_jump_via_thunk<mode>")
335 ("indirect_jump_via_inlinethunk<mode>_z10")
336 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
337 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
338 ("casesi_jump_via_inlinethunk<mode>_z10")
339 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
340 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
341 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
342 ("*indirect2_jump"): Disable for branch conversion.
343 ("casesi_jump"): Turn into expander and expand patterns for branch
345 ("return_use"): New expander.
346 ("*return"): Emit return via thunk and rename it to ...
347 ("*return<mode>"): ... this one.
348 * config/s390/s390.opt: Add new options and and enum for the
351 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
353 * lra-constraints.c (match_reload): Unconditionally use
354 gen_lowpart_SUBREG, rather than selecting between that
355 and equivalent gen_rtx_SUBREG code.
357 2018-02-08 Richard Biener <rguenther@suse.de>
359 PR tree-optimization/84233
360 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
361 changed flag instead of boguously re-using phi_inserted.
363 2018-02-08 Martin Jambor <mjambor@suse.cz>
365 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
366 static local variables.
368 2018-02-08 Richard Biener <rguenther@suse.de>
370 PR tree-optimization/84278
371 * tree-vect-stmts.c (vectorizable_store): When looking for
372 smaller vector types to perform grouped strided loads/stores
373 make sure the mode is supported by the target.
374 (vectorizable_load): Likewise.
376 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
378 * config/aarch64/aarch64.c (aarch64_components_for_bb):
379 Increase LDP/STP opportunities by adding adjacent callee-saves.
381 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
383 PR rtl-optimization/84068
384 PR rtl-optimization/83459
385 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
387 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
389 PR tree-optimization/84224
390 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
391 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
394 2018-02-07 Iain Sandoe <iain@codesourcery.com>
397 * config/rs6000/altivec.md (*restore_world): Remove LR use.
398 * config/rs6000/predicates.md (restore_world_operation): Adjust op
399 count, remove one USE.
401 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
403 * doc/install.texi (Configuration): Document the
404 --with-long-double-format={ibm,ieee} PowerPC configuration
408 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
409 Convert from define_expand to be define_insn_and_split. Rework
410 float/double/_Float128 conversions to QI/HI/SImode to work with
411 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
412 conversions to QI/HImode types did a store and then a load to
413 truncate the value. For conversions to VSX registers, don't split
414 the insn, instead emit the code directly. Use the code iterator
415 any_fix to combine signed and unsigned conversions.
416 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
417 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
418 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
419 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
420 (fix_<mode>di2_hw): Likewise.
421 (fixuns_<mode>di2_hw): Likewise.
422 (fix_<mode>si2_hw): Likewise.
423 (fixuns_<mode>si2_hw): Likewise.
424 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
425 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
426 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
427 fix<uns>_trunc<SFDF:mode>si2_p8.
428 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
430 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
431 (fix<uns>_<mode>_mem): Likewise.
432 (fctiw<u>z_<mode>_mem): Likewise.
433 (fix<uns>_<mode>_mem): Likewise.
434 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
435 the register allocator from doing a direct move to the GPRs to do
436 a store, and instead use the ISA 3.0 store byte/half-word from
437 vector register instruction. For IEEE 128-bit floating point,
438 also optimize stores of 32-bit ints.
439 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
441 2018-02-07 Alan Hayward <alan.hayward@arm.com>
443 * genextract.c (push_pathstr_operand): New function to support
445 (walk_rtx): Call push_pathstr_operand.
446 (print_path): Support [a-zA-Z].
448 2018-02-07 Richard Biener <rguenther@suse.de>
450 PR tree-optimization/84037
451 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
452 (cse_and_gimplify_to_preheader): Declare.
453 (vect_get_place_in_interleaving_chain): Likewise.
454 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
456 (_loop_vec_info::~_loop_vec_info): Delete it.
457 (cse_and_gimplify_to_preheader): New function.
458 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
459 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
460 (vectorizable_load): Likewise. For grouped stores always base
461 the IV on the first element.
462 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
463 condition before gimplifying.
465 2018-02-07 Jakub Jelinek <jakub@redhat.com>
467 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
468 *DIV_EXPR and *MOD_EXPR.
470 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
473 * config/i386/i386.c (ix86_option_override_internal): Mask out
474 the CF_SET bit when checking -fcf-protection.
476 2018-02-07 Tom de Vries <tom@codesourcery.com>
479 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
482 2018-02-07 Richard Biener <rguenther@suse.de>
484 PR tree-optimization/84204
485 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
488 PR tree-optimization/84205
489 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
490 special-case isl_ast_op_zdiv_r.
492 PR tree-optimization/84223
493 * graphite-scop-detection.c (gather_bbs::before_dom_children):
494 Only add conditions from within the region.
495 (gather_bbs::after_dom_children): Adjust.
497 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
500 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
501 * config/avr/avr.md: Only post-reload split REG-REG moves if
502 either register is REGERAL_REG_P.
504 2018-02-07 Jakub Jelinek <jakub@redhat.com>
506 PR tree-optimization/84235
507 * tree-ssa-scopedtables.c
508 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
509 if the subtraction is performed in floating point type where NaNs are
510 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
511 build 1. Formatting fix.
513 2018-02-06 Jakub Jelinek <jakub@redhat.com>
516 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
517 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
518 and skip it regardless of bb boundaries. Use CALL_P macro,
519 don't test INSN_P (insn) together with CALL_P or JUMP_P check
520 unnecessarily, formatting fix.
522 2018-02-06 Michael Collison <michael.collison@arm.com>
524 * config/arm/thumb2.md:
525 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
526 (*thumb_mov_notscc): Ditto.
528 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
531 * config/rs6000/rs6000.md (su code attribute): Use "u" for
532 unsigned_fix, not "s".
534 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
536 * configure.ac (gcc_fn_eh_frame_ro): New function.
537 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
538 correct .eh_frame permissions.
539 * configure: Regenerate.
541 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
543 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
546 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
548 * config/rs6000/rs6000.c (rs6000_option_override_internal):
549 Display warning message for -mno-speculate-indirect-jumps.
551 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
553 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
555 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
557 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
559 PR tree-optimization/84225
560 * tree-eh.c (find_trapping_overflow): Only call
561 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
563 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
566 * config/i386/i386.c: Reimplement the check of possible options
567 -mibt/-mshstk conbination. Change error messages.
568 * doc/invoke.texi: Fix a typo: remove extra '='.
570 2018-02-06 Marek Polacek <polacek@redhat.com>
572 PR tree-optimization/84228
573 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
575 2018-02-06 Tamar Christina <tamar.christina@arm.com>
578 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
579 emitted arch directives.
580 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
581 __ARM_FEATURE_COPROC before changing architectures.
583 2018-02-06 Richard Biener <rguenther@suse.de>
585 * config/i386/i386.c (print_reg): Fix typo.
586 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
588 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
590 * configure: Regenerate.
592 2018-02-05 Martin Sebor <msebor@redhat.com>
594 PR tree-optimization/83369
595 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
598 2018-02-05 Martin Liska <mliska@suse.cz>
600 * doc/invoke.texi: Cherry-pick upstream r323995.
602 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
604 * ira.c (ira_init_register_move_cost): Adjust comment.
606 2018-02-05 Martin Liska <mliska@suse.cz>
608 PR gcov-profile/84137
609 * doc/gcov.texi: Fix typo in documentation.
611 2018-02-05 Martin Liska <mliska@suse.cz>
613 PR gcov-profile/83879
614 * doc/gcov.texi: Document necessity of --dynamic-list-data when
615 using dlopen functionality.
617 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
619 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
620 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
621 _mm_maskz_range_ss, _mm_mask_range_round_ss,
622 _mm_maskz_range_round_ss): New intrinsics.
623 (__builtin_ia32_rangesd128_round)
624 (__builtin_ia32_rangess128_round): Remove.
625 (__builtin_ia32_rangesd128_mask_round,
626 __builtin_ia32_rangess128_mask_round): New builtins.
627 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
628 __builtin_ia32_rangess128_round): Remove.
629 (__builtin_ia32_rangesd128_mask_round,
630 __builtin_ia32_rangess128_mask_round): New builtins.
631 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
632 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
633 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
634 "<round_saeonly_constraint>")): Changed to ...
635 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
636 "<round_saeonly_scalar_constraint>")): ... this.
637 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
638 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
639 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
640 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
641 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
643 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
645 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
647 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
648 Remove all values except native, 8540 and 8548.
650 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
652 * config/i386/i386.c (ix86_output_function_return): Pass
653 INVALID_REGNUM, instead of -1, as invalid register number to
654 indirect_thunk_name and output_indirect_thunk.
656 2018-02-02 Julia Koval <julia.koval@intel.com>
658 * config.gcc: Add -march=icelake.
659 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
660 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
661 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
662 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
663 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
664 (processor_target_table): Add icelake.
665 (ix86_option_override_internal): Handle new PTAs.
666 (get_builtin_code_for_version): Handle icelake.
667 (M_INTEL_COREI7_ICELAKE): New.
668 (fold_builtin_cpu): Handle icelake.
669 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
670 * doc/invoke.texi: Add -march=icelake.
672 2018-02-02 Julia Koval <julia.koval@intel.com>
674 * config/i386/i386.c (ix86_option_override_internal): Change flags type
676 * wide-int-bitmask.h: New.
678 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
681 * config/i386/i386.md: Replace Pmode with word_mode in
682 builtin_setjmp_setup and builtin_longjmp to support x32.
684 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
688 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
690 (rs6000_supported_cpu_names): New static variable.
691 (linux_cpu_translation_table): Likewise.
692 (elf_platform) <cpu>: Define new static variable and use it.
693 Translate kernel AT_PLATFORM name to canonical name if needed.
694 Error if platform name is unknown.
696 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
699 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
701 2018-02-01 Jeff Law <law@redhat.com>
704 * config/i386/i386.c (release_scratch_register_on_entry): Add new
705 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
706 the scratch if RELEASE_VIA_POP is false.
707 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
708 If we have to save a temporary register, decrement SIZE appropriately.
709 Pass new arguments to release_scratch_register_on_entry.
710 (ix86_adjust_stack_and_probe): Likewise.
711 (ix86_emit_probe_stack_range): Pass new arguments to
712 release_scratch_register_on_entry.
714 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
716 PR rtl-optimization/84157
717 * combine.c (change_zero_ext): Use REG_P predicate in
718 front of HARD_REGISTER_P predicate.
720 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
722 * config/avr/avr.c (avr_option_override): Move disabling of
723 -fdelete-null-pointer-checks to...
724 * common/config/avr/avr-common.c (avr_option_optimization_table):
727 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
729 PR tree-optimization/81635
730 * tree-data-ref.c (split_constant_offset_1): For types that
731 wrap on overflow, try to use range info to prove that wrapping
734 2018-02-01 Renlin Li <renlin.li@arm.com>
737 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
739 (aarch64_register_move_cost): Likewise.
740 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
742 (REG_CLASS_NAMES): Likewise.
743 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
744 TAILCALL_ADDR_REGS. Remove IP registers.
745 * config/aarch64/aarch64.md (Ucs): Update register constraint.
747 2018-02-01 Richard Biener <rguenther@suse.de>
749 * domwalk.h (dom_walker::dom_walker): Add additional constructor
750 for specifying RPO order and allow NULL for that.
751 * domwalk.c (dom_walker::dom_walker): Likewise.
752 (dom_walker::walk): Handle NULL RPO order.
753 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
755 (rewrite_update_dom_walker): Likewise.
756 (mark_def_dom_walker): Likewise.
758 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
760 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
761 (aarch64_maybe_expand_sve_subreg_move): Declare.
762 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
763 * config/aarch64/predicates.md (aarch64_any_register_operand): New
765 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
766 that are semantically a reverse operation.
767 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
768 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
769 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
771 (aarch64_can_change_mode_class): For big-endian, forbid changes
772 between two SVE modes if they have different element sizes.
774 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
776 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
777 the TImode handling for big-endian targets.
779 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
781 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
782 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
784 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
785 Remove BSWAP handing for big-endian targets and use the form of
786 LD1RQ appropariate for the mode.
788 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
790 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
791 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
794 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
797 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
798 check for operands that need to go through aarch64_sve_reload_be.
800 2018-02-01 Jakub Jelinek <jakub@redhat.com>
802 PR tree-optimization/81661
803 PR tree-optimization/84117
804 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
805 * tree-eh.c: Include gimplify.h.
806 (find_trapping_overflow, replace_trapping_overflow,
807 rewrite_to_non_trapping_overflow): New functions.
808 * tree-vect-loop.c: Include tree-eh.h.
809 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
810 * tree-data-ref.c: Include tree-eh.h.
811 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
813 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
815 PR rtl-optimization/84123
816 * combine.c (change_zero_ext): Check if hard register satisfies
817 can_change_dest_mode before calling gen_lowpart_SUBREG.
819 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
822 * ira.c (ira_init_register_move_cost): Remove assert.
824 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
826 PR rtl-optimization/84071
827 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
828 * doc/tm.texi: Regenerate.
830 2018-01-31 Richard Biener <rguenther@suse.de>
832 PR tree-optimization/84132
833 * tree-data-ref.c (analyze_miv_subscript): Properly
834 check whether evolution_function_is_affine_multivariate_p
835 before calling gcd_of_steps_may_divide_p.
837 2018-01-31 Julia Koval <julia.koval@intel.com>
840 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
841 * config/i386/i386.md (rdpid_rex64) New.
842 (rdpid): Make 32bit only.
844 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
847 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
848 an IDENTIFIER_NODE for FUNCTION_TYPE's.
850 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
853 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
855 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
857 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
859 PR rtl-optimization/84071
860 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
861 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
863 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
865 * config/arc/arc.c (arc_handle_aux_attribute): New function.
866 (arc_attribute_table): Add 'aux' attribute.
867 (arc_in_small_data_p): Consider aux like variables.
868 (arc_is_aux_reg_p): New function.
869 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
870 (arc_get_aux_arg): New function.
871 (prepare_move_operands): Handle aux-register access.
872 (arc_handle_aux_attribute): New function.
873 * doc/extend.texi (ARC Variable attributes): Add subsection.
875 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
877 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
878 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
879 (arc_attribute_table): Add 'uncached' attribute.
880 (arc_print_operand): Print '.di' flag for uncached memory
882 (arc_in_small_data_p): Do not consider for small data the uncached
884 (arc_is_uncached_mem_p): New function.
885 * config/arc/predicates.md (compact_store_memory_operand): Check
886 for uncached memory accesses.
887 (nonvol_nonimm_operand): Likewise.
888 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
890 2018-01-31 Jakub Jelinek <jakub@redhat.com>
893 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
894 falign-loops=): Add Optimization flag.
896 2018-01-30 Jeff Law <law@redhat.com>
899 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
900 INT_REGISTERS_SAVED. Check it prior to calling
901 get_scratch_register_on_entry.
902 (ix86_adjust_stack_and_probe): Similarly.
903 (ix86_emit_probe_stack_range): Similarly.
904 (ix86_expand_prologue): Corresponding changes.
906 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
909 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
910 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
912 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
915 * lra-constraints.c (curr_insn_transform): Process AND in the
918 2018-01-30 Jakub Jelinek <jakub@redhat.com>
920 PR rtl-optimization/83986
921 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
922 dependence against last_pending_memory_flush in addition to
925 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
927 PR tree-optimization/81611
928 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
931 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
933 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
936 2018-01-30 Richard Biener <rguenther@suse.de>
937 Jakub Jelinek <jakub@redhat.com>
939 PR tree-optimization/84111
940 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
941 inner loops added during recursion, as they don't have up-to-date
944 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
947 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
948 (can_inline_edge_by_limits_p): ... here.
949 (can_early_inline_edge_p, check_callers,
950 update_caller_keys, update_callee_keys, recursive_inlining,
951 add_new_edges_to_heap, speculation_useful_p,
952 inline_small_functions,
953 inline_small_functions, flatten_function,
954 inline_to_all_callers_1): Update.
956 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
958 * profile-count.c (profile_count::combine_with_ipa_count): Handle
961 2018-01-30 Richard Biener <rguenther@suse.de>
963 PR tree-optimization/83008
964 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
965 invariant and constant vector uses in stmts when they need
968 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
971 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
972 * configure: Regenerate.
974 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
976 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
978 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
979 Use gen_rtx_REG rather than gen_lowpart.
981 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
983 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
984 rather than 0 when creating partial subregs.
986 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
988 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
991 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
994 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
995 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
996 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
997 flags. This restores the settings used before the 2017-07-24.
998 Turning off pre increment/decrement/modify allows IVOPTS to
999 optimize DF/SF loops where the index is an int.
1001 2018-01-29 Richard Biener <rguenther@suse.de>
1002 Kelvin Nilsen <kelvin@gcc.gnu.org>
1005 * tree-vect-stmts.c (vectorizable_call): Don't call
1006 targetm.vectorize_builtin_md_vectorized_function if callee is
1009 2018-01-22 Carl Love <cel@us.ibm.com>
1011 * doc/extend.tex: Fix typo in second arg in
1012 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1014 2018-01-29 Richard Biener <rguenther@suse.de>
1016 PR tree-optimization/84086
1017 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1018 (flush_ssaname_freelist): When SSA names were released reset
1019 the SCEV hash table.
1021 2018-01-29 Richard Biener <rguenther@suse.de>
1023 PR tree-optimization/84057
1024 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1025 removed paths when removing edges.
1027 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1029 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1030 -mfunction-return=@var{choice}.
1032 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1035 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1037 (layout::print_source_line): Likewise.
1038 (test_get_line_width_without_trailing_whitespace): Add test cases.
1040 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1043 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1046 2018-01-26 Jim Wilson <jimw@sifive.com>
1048 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1050 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1053 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1055 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1056 and CMP + SUB-immediate -> SUBS.
1058 2018-01-26 Martin Sebor <msebor@redhat.com>
1060 PR tree-optimization/83896
1061 * tree-ssa-strlen.c (get_string_len): Rename...
1062 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1063 Avoid assuming length is constant.
1064 (handle_char_store): Use HOST_WIDE_INT for string length.
1066 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1069 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1070 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1072 2018-01-26 Richard Biener <rguenther@suse.de>
1074 PR rtl-optimization/84003
1075 * dse.c (record_store): Only record redundant stores when
1076 the earlier store aliases at least all accesses the later one does.
1078 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1080 PR rtl-optimization/83985
1081 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1082 REG_CFA_RESTORE insns.
1083 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1084 REG_CFA_RESTORE insns here.
1087 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1088 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1090 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1092 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1093 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1094 (arc_init): Likewise.
1095 (arc_override_options): Likewise.
1096 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1098 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1100 * config/arc/arc.h (TARGET_DBNZ): Define.
1101 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1102 properly set the tune attribute.
1103 (dbnz): Use TARGET_DBNZ guard.
1104 * config/arc/arc.opt (mtune): Add core3 option.
1106 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1108 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1109 recognize new pic like addresses.
1110 (arc_delegitimize_address): Clean up.
1112 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1114 * config/arc/arc-arches.def: Option mrf16 valid for all
1116 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1117 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1118 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1119 * config/arc/arc-tables.opt: Regenerate.
1120 * config/arc/arc.c (arc_conditional_register_usage): Handle
1121 reduced register file case.
1122 (arc_file_start): Set must have build attributes.
1123 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1125 * config/arc/arc.opt (mrf16): Add new option.
1126 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1127 * config/arc/genmultilib.awk: Handle new mrf16 option.
1128 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1129 * config/arc/t-multilib: Regenerate.
1130 * doc/invoke.texi (ARC Options): Document mrf16 option.
1132 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1134 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1135 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1136 (arc_attribute_table): Add 'secure_call' attribute.
1137 (arc_print_operand): Print secure call operand.
1138 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1140 (arc_is_secure_call_p): New function. * config/arc/arc.md
1141 (call_i): Add support for sjli instruction.
1142 (call_value_i): Likewise.
1143 * config/arc/constraints.md (Csc): New constraint.
1145 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1146 John Eric Martin <John.Martin@emmicro-us.com>
1148 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1149 * config/arc/arc.c (_arc_jli_section): New struct.
1150 (arc_jli_section): New type.
1151 (rc_jli_sections): New static variable.
1152 (arc_handle_jli_attribute): New function.
1153 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1154 (arc_file_end): New function.
1155 (TARGET_ASM_FILE_END): Define.
1156 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1157 (arc_add_jli_section): New function.
1158 (jli_call_scan): Likewise.
1159 (arc_reorg): Call jli_call_scan.
1160 (arc_output_addsi): Remove 'S' from printing asm operand.
1161 (arc_is_jli_call_p): New function.
1162 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1164 (movhi_insn): Likewise.
1165 (movsi_insn): Likewise.
1166 (movsi_set_cc_insn): Likewise.
1167 (loadqi_update): Likewise.
1168 (load_zeroextendqisi_update): Likewise.
1169 (load_signextendqisi_update): Likewise.
1170 (loadhi_update): Likewise.
1171 (load_zeroextendhisi_update): Likewise.
1172 (load_signextendhisi_update): Likewise.
1173 (loadsi_update): Likewise.
1174 (loadsf_update): Likewise.
1175 (movsicc_insn): Likewise.
1176 (bset_insn): Likewise.
1177 (bxor_insn): Likewise.
1178 (bclr_insn): Likewise.
1179 (bmsk_insn): Likewise.
1180 (bicsi3_insn): Likewise.
1181 (cmpsi_cc_c_insn): Likewise.
1182 (movsi_ne): Likewise.
1183 (movsi_cond_exec): Likewise.
1184 (clrsbsi2): Likewise.
1195 (call_i): Remove 'S' asm letter, add jli instruction.
1196 (call_value_i): Likewise.
1197 * config/arc/arc.op (mjli-always): New option.
1198 * config/arc/constraints.md (Cji): New constraint.
1199 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1201 (subsf3_fpx): Likewise.
1202 (mulsf3_fpx): Likewise.
1203 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1205 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1207 * doc/invoke.texi (ARC): Document mjli-always option.
1209 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1211 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1212 and use incw and decw where possible
1213 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1215 2018-01-26 Richard Biener <rguenther@suse.de>
1217 PR tree-optimization/81082
1218 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1219 association if it requires casting to unsigned.
1220 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1221 from fold_plusminus_mult_expr to catch important cases late when
1222 range info is available.
1224 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1226 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1227 * configure.ac (hidden_linkonce): New test.
1228 * configure: Regenerate.
1229 * config.in: Regenerate.
1231 2018-01-26 Julia Koval <julia.koval@intel.com>
1233 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1234 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1235 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1236 _mm_mask_bitshuffle_epi64_mask): Fix type.
1237 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1238 USI_FTYPE_V4DI_V4DI_USI): Remove.
1239 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1240 __builtin_ia32_vpshufbitqmb256_mask,
1241 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1242 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1243 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1245 2018-01-26 Alan Modra <amodra@gmail.com>
1248 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1249 UNSPEC_VBPERMQ. Sort other unspecs.
1251 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1253 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1255 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1258 * predict.c (drop_profile): Do not push/pop cfun; update also
1260 (handle_missing_profiles): Fix logic looking for zero profiles.
1262 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1265 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1266 on functions with #pragma omp declare simd or functions with simd
1268 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1269 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1270 Remove trailing \n from warning_at calls.
1272 2018-01-25 Tom de Vries <tom@codesourcery.com>
1275 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1276 for neutered workers.
1278 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1281 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1282 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1284 2018-01-24 Jeff Law <law@redhat.com>
1287 * i386.c (get_probe_interval): Move to earlier point.
1288 (ix86_compute_frame_layout): If -fstack-clash-protection and
1289 the frame is larger than the probe interval, then use pushes
1290 to save registers rather than reg->mem moves.
1291 (ix86_expand_prologue): Remove conditional for int_registers_saved
1294 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1297 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1298 min/max for never referenced object.
1300 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1303 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1305 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1306 attributes from DECL_ATTRIBUTES (decl) without affecting
1307 DECL_ATTRIBUTES (current_function_decl).
1308 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1309 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1311 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1313 PR tree-optimization/83979
1314 * fold-const.c (fold_comparison): Use constant_boolean_node
1315 instead of boolean_{true,false}_node.
1317 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1319 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1322 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1324 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1325 Simplify the clause that sets the length attribute.
1326 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1327 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1328 clause that sets the length attribute.
1329 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1331 2018-01-24 Tom de Vries <tom@codesourcery.com>
1334 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1335 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1336 Add strict parameter.
1337 (prevent_branch_around_nothing): Insert dummy insn between branch to
1338 label and label with no ptx insn inbetween.
1339 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1341 2018-01-24 Tom de Vries <tom@codesourcery.com>
1344 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1345 for neutered threads in warp.
1346 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1348 2018-01-24 Richard Biener <rguenther@suse.de>
1350 PR tree-optimization/83176
1351 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1354 2018-01-24 Richard Biener <rguenther@suse.de>
1356 PR tree-optimization/82819
1357 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1358 code generating pluses that are no-ops in the target precision.
1360 2018-01-24 Richard Biener <rguenther@suse.de>
1363 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1365 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1367 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1368 to merge probabilities.
1369 * predict.c (probably_never_executed): Also mark as cold functions
1370 with global 0 profile and guessed local profile.
1371 * profile-count.c (profile_probability::combine_with_count): New
1373 * profile-count.h (profile_probability::operator*,
1374 profile_probability::operator*=, profile_probability::operator/,
1375 profile_probability::operator/=): Reduce precision to adjusted
1376 and set value to guessed on contradictory divisions.
1377 (profile_probability::combine_with_freq): Remove.
1378 (profile_probability::combine_wiht_count): Declare.
1379 (profile_count::force_nonzero):: Set to adjusted.
1380 (profile_count::probability_in):: Set quality to adjusted.
1381 * tree-ssa-tail-merge.c (replace_block_by): Use
1384 2018-01-23 Andrew Waterman <andrew@sifive.com>
1385 Jim Wilson <jimw@sifive.com>
1387 * config/riscv/riscv.c (riscv_stack_boundary): New.
1388 (riscv_option_override): Set riscv_stack_boundary. Handle
1389 riscv_preferred_stack_boundary_arg.
1390 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1391 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1392 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1393 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1394 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1395 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1397 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1400 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1401 of struct ix86_frame.
1402 (ix86_expand_epilogue): Likewise. Add a local variable for
1403 the reg_save_offset field in struct ix86_frame.
1405 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1407 PR tree-optimization/82604
1408 * tree-loop-distribution.c (enum partition_kind): New enum item
1409 PKIND_PARTIAL_MEMSET.
1410 (partition_builtin_p): Support above new enum item.
1411 (generate_code_for_partition): Ditto.
1412 (compute_access_range): Differentiate cases that equality can be
1413 proven at all loops, the innermost loops or no loops.
1414 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1415 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1416 (finalize_partitions, distribute_loop): Don't fuse partition of
1417 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1418 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1421 2018-01-23 Martin Liska <mliska@suse.cz>
1423 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1424 order to ignore the predictor.
1425 (PRED_POLYMORPHIC_CALL): Likewise.
1426 (PRED_RECURSIVE_CALL): Likewise.
1428 2018-01-23 Martin Liska <mliska@suse.cz>
1430 * tree-profile.c (tree_profiling): Print function header to
1431 aware reader which function we are working on.
1432 * value-prof.c (gimple_find_values_to_profile): Do not print
1433 not interesting value histograms.
1435 2018-01-23 Martin Liska <mliska@suse.cz>
1437 * profile-count.h (enum profile_quality): Add
1438 profile_uninitialized as the first value. Do not number values
1439 as they are zero based.
1440 (profile_count::verify): Update sanity check.
1441 (profile_probability::verify): Likewise.
1443 2018-01-23 Nathan Sidwell <nathan@acm.org>
1445 * doc/invoke.texi (ffor-scope): Deprecate.
1447 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1449 PR tree-optimization/83510
1450 * domwalk.c (set_all_edges_as_executable): New function.
1451 (dom_walker::dom_walker): Convert bool param
1452 "skip_unreachable_blocks" to enum reachability. Move setup of
1453 edge flags to set_all_edges_as_executable and only do it when
1454 reachability is REACHABLE_BLOCKS.
1455 * domwalk.h (enum dom_walker::reachability): New enum.
1456 (dom_walker::dom_walker): Convert bool param
1457 "skip_unreachable_blocks" to enum reachability.
1458 (set_all_edges_as_executable): New decl.
1459 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1460 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1462 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1463 but converting true to REACHABLE_BLOCKS.
1464 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1466 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1467 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1468 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1470 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1471 if check_all_array_refs will be called.
1473 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1475 * tree.c (selftest::test_location_wrappers): Add more test
1478 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1480 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1481 (selftest::test_bit_in_range): Likewise.
1483 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1486 * doc/sourcebuild.texi (vect_float): Say that the selector
1487 only describes the situation when -funsafe-math-optimizations is on.
1488 (vect_float_strict): Document.
1490 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1492 PR tree-optimization/83965
1493 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1494 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1495 instead of checking only for a reduction.
1496 (vect_recog_widen_sum_pattern): Likewise.
1498 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1500 * predict.c (probably_never_executed): Only use precise profile info.
1501 (compute_function_frequency): Skip after inlining hack since we now
1502 have quality checking.
1504 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1506 * profile-count.h (profile_probability::very_unlikely,
1507 profile_probability::unlikely, profile_probability::even): Set
1508 precision to guessed.
1510 2018-01-23 Richard Biener <rguenther@suse.de>
1512 PR tree-optimization/83963
1513 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1514 Properly terminate dominator walk when crossing the exit edge not
1515 when visiting its source block.
1517 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1520 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1521 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1523 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1525 PR tree-optimization/83957
1526 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1527 semicolon after for body surrounded by braces.
1529 PR tree-optimization/83081
1530 * profile-count.h (profile_probability::split): New method.
1531 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1532 Use profile_probability::split.
1533 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1534 when splitting a single conditional jump into 2.
1536 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1538 PR tree-optimization/69452
1539 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1542 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1544 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1545 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1546 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1548 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1550 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1551 * config/rl78/rl78.md: New define_expand "movdi"
1552 * config/rl78/rl78.c: New function definition rl78_split_movdi
1554 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1557 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1559 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1560 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1561 128-bit to produce an UNSPEC move to get the double word with the
1562 signbit and then a shift directly to do signbit.
1563 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1564 implementation with a new version that just does either a direct
1565 move or a regular move. Move memory interface to separate insns.
1566 Move insns so they are next to the expander.
1567 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1568 with signbit move. Split big and little endian case.
1569 (signbit<mode>2_dm_mem_le): Likewise.
1570 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1571 (signbit<mode>2_dm2): Likewise.
1573 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1575 * config/rl78/rl78.md: New define_expand "anddi3".
1577 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1579 * config/rl78/rl78.md: New define_expand "umindi3".
1581 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1583 * config/rl78/rl78.md: New define_expand "smindi3".
1585 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1587 * config/rl78/rl78.md: New define_expand "smaxdi3".
1589 2018-01-22 Carl Love <cel@us.ibm.com>
1591 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1592 LVX_V1TI): Add macro expansion.
1593 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1594 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1595 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1596 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1597 Change check to determine if the instruction is a byte reversing
1598 entry. Fix typo in comment.
1599 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1600 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1601 Add def_builtin calls for new builtins.
1602 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1603 Add define_insn expansion.
1605 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1607 * config/rl78/rl78.md: New define_expand "umaxdi3".
1609 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1611 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1612 for non-QImode registers
1614 2018-01-22 Richard Biener <rguenther@suse.de>
1616 PR tree-optimization/83963
1617 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1618 including the loop exit block.
1619 (scop_detection::merge_sese): Likewise.
1620 (scop_detection::add_scop): Do it here instead.
1622 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1624 * doc/sourcebuild.texi (arm_softfloat): Document.
1626 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1629 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1630 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1631 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1633 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1634 David Edelsohn <dje.gcc@gmail.com>
1637 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1638 Change "crset eq" to "crset 2".
1639 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1640 (*call_indirect_aix<mode>_nospec): Likewise.
1641 (*call_value_indirect_aix<mode>_nospec): Likewise.
1642 (*call_indirect_elfv2<mode>_nospec): Likewise.
1643 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1644 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1645 change assembly output from . to $.
1646 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1647 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1648 (*tablejump<mode>_internal1_nospec): Likewise.
1650 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1653 * config/sh/sh_optimize_sett_clrt.cc:
1654 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1656 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1658 PR tree-optimization/83940
1659 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1660 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1661 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1662 use it to pass back the definition type.
1663 (vect_check_store_rhs): Likewise rhs_dt_out.
1664 (vect_build_gather_load_calls): Add a mask_dt argument and use
1665 it instead of a call to vect_is_simple_use.
1666 (vectorizable_store): Update calls to vect_check_load_store_mask
1667 and vect_check_store_rhs. Use the dt returned by the latter instead
1668 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1669 instead of calls to vect_is_simple_use. Pass the scalar rather
1670 than the vector operand to vect_is_simple_use when handling
1671 second and subsequent copies of an rhs value.
1672 (vectorizable_load): Update calls to vect_check_load_store_mask
1673 and vect_build_gather_load_calls. Use the cached mask_dt and
1674 gs_info.offset_dt instead of calls to vect_is_simple_use.
1676 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1679 * tree-emutls.c: Include gimplify.h.
1680 (lower_emutls_2): New function.
1681 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1682 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1683 it before further processing.
1686 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1687 UINTVAL (trueop1) instead of INTVAL (op1).
1689 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1693 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1694 INCOMING_FRAME_SP_OFFSET if not defined.
1695 (scan_trace): Add ENTRY argument. If true and
1696 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1697 emit a note to adjust the CFA offset.
1698 (create_cfi_notes): Adjust scan_trace callers.
1699 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1700 INCOMING_FRAME_SP_OFFSET in the CIE.
1701 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1702 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1704 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1705 * doc/tm.texi: Regenerated.
1707 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1709 PR rtl-optimization/83147
1710 * lra-constraints.c (remove_inheritance_pseudos): Use
1711 lra_substitute_pseudo_within_insn.
1713 2018-01-19 Tom de Vries <tom@codesourcery.com>
1714 Cesar Philippidis <cesar@codesourcery.com>
1717 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1719 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1722 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1723 spaces for function labels.
1725 2018-01-19 Martin Liska <mliska@suse.cz>
1727 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1728 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1729 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1730 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1731 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1732 (PRED_CONST_RETURN): Change from 69 to 65.
1733 (PRED_NULL_RETURN): Change from 91 to 71.
1734 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1735 (PRED_LOOP_GUARD): Change from 66 to 73.
1737 2018-01-19 Martin Liska <mliska@suse.cz>
1739 * predict.c (predict_insn_def): Add new assert.
1740 (struct branch_predictor): Change type to signed integer.
1741 (test_prediction_value_range): Amend test to cover
1743 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1744 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1745 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1746 (PRED_LOOP_IV_COMPARE): Likewise.
1747 * predict.h (PROB_UNINITIALIZED): Define new constant.
1749 2018-01-19 Martin Liska <mliska@suse.cz>
1751 * predict.c (dump_prediction): Add new format for
1752 analyze_brprob.py script which is enabled with -details
1754 * profile-count.h (precise_p): New function.
1756 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1758 PR tree-optimization/83922
1759 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1760 there are no statements that need masking.
1761 (vect_active_double_reduction_p): New function.
1762 (vect_analyze_loop_operations): Use it when handling phis that
1763 are not in the loop header.
1765 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1767 PR tree-optimization/83914
1768 * tree-vect-loop.c (vectorizable_induction): Don't convert
1769 init_expr or apply the peeling adjustment for inductions
1770 that are nested within the vectorized loop.
1772 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1774 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1777 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1781 * function.h (gimplify_parameters): Add gimple_seq * argument.
1782 * function.c: Include gimple.h and options.h.
1783 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1784 for the added local temporaries if needed.
1785 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1786 if there are any parameter cleanups, wrap whole body into a
1787 try/finally with the cleanups.
1789 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1792 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1793 Use GET_MODE_CLASS for scalar floating point.
1795 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1799 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1800 Fix call of call_cgraph_insertion_hooks.
1802 2018-01-18 Martin Sebor <msebor@redhat.com>
1804 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1806 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1809 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1812 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1815 * common.opt: (-ffile-prefix-map): New option.
1816 * opts.c (common_handle_option): Defer it.
1817 * opts-global.c (handle_common_deferred_options): Handle it.
1818 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1819 * file-prefix-map.h: New file.
1820 (remap_debug_filename, add_debug_prefix_map): ...here.
1821 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1822 * final.c (debug_prefix_map, add_debug_prefix_map
1823 remap_debug_filename): Move to...
1824 * file-prefix-map.c: New file.
1825 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1826 generalize, get rid of alloca(), use strrchr() instead of strchr().
1827 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1828 Implement in terms of add_prefix_map().
1829 (remap_macro_filename, remap_debug_filename): Implement in term of
1831 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1832 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1833 * dbxout.c: Include file-prefix-map.h.
1834 * varasm.c: Likewise.
1835 * vmsdbgout.c: Likewise.
1836 * xcoffout.c: Likewise.
1837 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1838 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1839 * doc/invoke.texi (-ffile-prefix-map): Document.
1840 (-fdebug-prefix-map): Update description.
1842 2018-01-18 Martin Liska <mliska@suse.cz>
1844 * config/i386/i386.c (indirect_thunk_name): Document that also
1846 (output_indirect_thunk): Document why both instructions
1847 (pause and lfence) are generated.
1849 2018-01-18 Richard Biener <rguenther@suse.de>
1851 PR tree-optimization/83887
1852 * graphite-scop-detection.c
1853 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1854 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1855 (scop_detection::merge_sese): Re-implement with a flood-fill
1856 algorithm that properly finds a SESE region if it exists.
1858 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1861 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1862 pointer_diff optimizations use view_convert instead of convert.
1864 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1866 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1867 Generate different code for -mno-speculate-indirect-jumps.
1868 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1869 (*call_indirect_aix<mode>): Disable for
1870 -mno-speculate-indirect-jumps.
1871 (*call_indirect_aix<mode>_nospec): New define_insn.
1872 (*call_value_indirect_aix<mode>): Disable for
1873 -mno-speculate-indirect-jumps.
1874 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1875 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1876 -mno-speculate-indirect-jumps.
1877 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1879 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1881 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1882 long double type, set the flags for noting the default long double
1883 type, even if we don't pass or return a long double type.
1885 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1888 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1891 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1893 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1894 support for merge[hl].
1895 (fold_mergehl_helper): New helper function.
1896 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1897 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1898 (altivec_vmrglw_direct): Add xxmrglw insn.
1900 2018-01-17 Andrew Waterman <andrew@sifive.com>
1902 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1903 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1905 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1908 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1909 call the lto_location_cache before reading the
1910 DECL_SOURCE_LOCATION of the types.
1912 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1913 Richard Sandiford <richard.sandiford@linaro.org>
1915 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1916 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1917 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1918 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1919 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1921 * config/aarch64/constraints.md (aarch64_movti_operand):
1923 * config/aarch64/predicates.md (Uti): Add new constraint.
1925 2018-01-17 Carl Love <cel@us.ibm.com>
1926 * config/rs6000/vsx.md (define_expand xl_len_r,
1927 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1928 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1930 (define_expand, define_insn): Move the shift left from the
1931 define_insn to the define_expand for lxvl and stxvl instructions.
1932 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1933 and XL_LEN_R definitions to PURE.
1935 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1937 * config/i386/i386.c (indirect_thunk_name): Declare regno
1938 as unsigned int. Compare regno with INVALID_REGNUM.
1939 (output_indirect_thunk): Ditto.
1940 (output_indirect_thunk_function): Ditto.
1941 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1942 in the call to output_indirect_thunk_function.
1944 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1947 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1948 rather than the size of inner_type to determine the stack slot size
1949 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1951 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1954 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1957 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1959 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1960 endian Linux systems to optionally enable multilibs for selecting
1961 the long double type if the user configured an explicit type.
1962 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1963 have no long double multilibs if not defined.
1964 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1965 warn if the user used -mabi={ieee,ibm}longdouble and we built
1966 multilibs for long double.
1967 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1968 appropriate multilib option.
1969 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1971 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1972 for building long double multilibs.
1973 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1975 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1977 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1980 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1982 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1985 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1988 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1991 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1993 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1994 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1996 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1998 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1999 different rtl trees depending on TARGET_64BIT.
2000 (rs6000_gen_lvx): Likewise.
2002 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2004 * config/visium/visium.md (nop): Tweak comment.
2005 (hazard_nop): Likewise.
2007 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2009 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2010 -mspeculate-indirect-jumps.
2011 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2012 for -mno-speculate-indirect-jumps.
2013 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2014 (*call_value_indirect_elfv2<mode>): Disable for
2015 -mno-speculate-indirect-jumps.
2016 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2017 (indirect_jump): Emit different RTL for
2018 -mno-speculate-indirect-jumps.
2019 (*indirect_jump<mode>): Disable for
2020 -mno-speculate-indirect-jumps.
2021 (*indirect_jump<mode>_nospec): New define_insn.
2022 (tablejump): Emit different RTL for
2023 -mno-speculate-indirect-jumps.
2024 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2025 (tablejumpsi_nospec): New define_expand.
2026 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2027 (tablejumpdi_nospec): New define_expand.
2028 (*tablejump<mode>_internal1): Disable for
2029 -mno-speculate-indirect-jumps.
2030 (*tablejump<mode>_internal1_nospec): New define_insn.
2031 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2034 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2036 * caller-save.c (insert_save): Drop unnecessary parameter. All
2039 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2040 Richard Biener <rguenth@suse.de>
2043 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2044 return early, inline manually is_gimple_sizepos. Make sure if we
2045 call gimplify_expr we don't end up with a gimple constant.
2046 * tree.c (variably_modified_type_p): Don't return true for
2047 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2048 * gimplify.h (is_gimple_sizepos): Remove.
2050 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2052 PR tree-optimization/83857
2053 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2054 vectorizable_live_operation for pure SLP statements.
2055 (vectorizable_live_operation): Handle PHIs.
2057 2018-01-16 Richard Biener <rguenther@suse.de>
2059 PR tree-optimization/83867
2060 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2061 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2063 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2066 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2067 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2068 If off is not INTEGER_CST, issue a may not be aligned warning
2069 rather than isn't aligned. Use isn%'t rather than isn't.
2070 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2072 <case MULT_EXPR>: Improve the case when bottom and one of the
2073 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2074 operand, in that case check if the other operand is multiple of
2075 bottom divided by the INTEGER_CST operand.
2077 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2080 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2081 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2082 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2083 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2084 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2085 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2086 (pa_function_arg_size): New function.
2088 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2090 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2091 in a separate statement.
2093 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2095 PR tree-optimization/83847
2096 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2097 group gathers and scatters.
2099 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2101 PR rtl-optimization/86620
2102 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2104 PR rtl-optimization/83213
2105 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2106 to last if both are JUMP_INSNs.
2108 PR tree-optimization/83843
2109 * gimple-ssa-store-merging.c
2110 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2111 store_immediate_info for bswap/nop orig_stores.
2113 2018-01-15 Andrew Waterman <andrew@sifive.com>
2115 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2117 <UDIV>: Increase cost if !TARGET_DIV.
2119 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2121 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2122 (define_attr "cr_logical_3op"): New.
2123 (cceq_ior_compare): Adjust.
2124 (cceq_ior_compare_complement): Adjust.
2125 (*cceq_rev_compare): Adjust.
2126 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2127 (is_cracked_insn): Adjust.
2128 (insn_must_be_first_in_group): Adjust.
2129 * config/rs6000/40x.md: Adjust.
2130 * config/rs6000/440.md: Adjust.
2131 * config/rs6000/476.md: Adjust.
2132 * config/rs6000/601.md: Adjust.
2133 * config/rs6000/603.md: Adjust.
2134 * config/rs6000/6xx.md: Adjust.
2135 * config/rs6000/7450.md: Adjust.
2136 * config/rs6000/7xx.md: Adjust.
2137 * config/rs6000/8540.md: Adjust.
2138 * config/rs6000/cell.md: Adjust.
2139 * config/rs6000/e300c2c3.md: Adjust.
2140 * config/rs6000/e500mc.md: Adjust.
2141 * config/rs6000/e500mc64.md: Adjust.
2142 * config/rs6000/e5500.md: Adjust.
2143 * config/rs6000/e6500.md: Adjust.
2144 * config/rs6000/mpc.md: Adjust.
2145 * config/rs6000/power4.md: Adjust.
2146 * config/rs6000/power5.md: Adjust.
2147 * config/rs6000/power6.md: Adjust.
2148 * config/rs6000/power7.md: Adjust.
2149 * config/rs6000/power8.md: Adjust.
2150 * config/rs6000/power9.md: Adjust.
2151 * config/rs6000/rs64.md: Adjust.
2152 * config/rs6000/titan.md: Adjust.
2154 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2156 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2157 ix86_indirect_branch_register logic.
2159 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2161 * config/i386/constraints.md (Bs): Update
2162 ix86_indirect_branch_register check. Don't check
2163 ix86_indirect_branch_register with GOT_memory_operand.
2165 * config/i386/predicates.md (GOT_memory_operand): Don't check
2166 ix86_indirect_branch_register here.
2167 (GOT32_symbol_operand): Likewise.
2169 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2171 * config/i386/predicates.md (constant_call_address_operand):
2172 Rewrite ix86_indirect_branch_register logic.
2173 (sibcall_insn_operand): Likewise.
2175 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2177 * config/i386/constraints.md (Bs): Replace
2178 ix86_indirect_branch_thunk_register with
2179 ix86_indirect_branch_register.
2181 * config/i386/i386.md (indirect_jump): Likewise.
2182 (tablejump): Likewise.
2183 (*sibcall_memory): Likewise.
2184 (*sibcall_value_memory): Likewise.
2185 Peepholes of indirect call and jump via memory: Likewise.
2186 * config/i386/i386.opt: Likewise.
2187 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2188 (GOT_memory_operand): Likewise.
2189 (call_insn_operand): Likewise.
2190 (sibcall_insn_operand): Likewise.
2191 (GOT32_symbol_operand): Likewise.
2193 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2196 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2197 type rather than type addr's type points to.
2198 (expand_omp_atomic_mutex): Likewise.
2199 (expand_omp_atomic): Likewise.
2201 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2204 * config/i386/i386.c (output_indirect_thunk_function): Use
2205 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2206 for __x86_return_thunk.
2208 2018-01-15 Richard Biener <rguenther@suse.de>
2211 * expmed.c (extract_bit_field_1): Fix typo.
2213 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2216 * config/arm/iterators.md (VF): New mode iterator.
2217 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2218 Remove integer-related logic from pattern.
2219 (neon_vabd<mode>_3): Likewise.
2221 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2224 * common.opt (fstrict-overflow): No longer an alias.
2225 (fwrapv-pointer): New option.
2226 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2227 also for pointer types based on flag_wrapv_pointer.
2228 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2229 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2230 opts->x_flag_wrapv got set.
2231 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2232 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2233 POINTER_TYPE_OVERFLOW_UNDEFINED.
2234 * match.pd: Likewise in address comparison pattern.
2235 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2237 2018-01-15 Richard Biener <rguenther@suse.de>
2240 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2241 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2242 Reset type names to their identifier if their TYPE_DECL doesn't
2243 have linkage (and thus is used for ODR and devirt).
2244 (save_debug_info_for_decl): Remove.
2245 (save_debug_info_for_type): Likewise.
2246 (add_tree_to_fld_list): Adjust.
2247 * tree-pretty-print.c (dump_generic_node): Make dumping of
2248 type names more robust.
2250 2018-01-15 Richard Biener <rguenther@suse.de>
2252 * BASE-VER: Bump to 8.0.1.
2254 2018-01-14 Martin Sebor <msebor@redhat.com>
2257 * builtins.c (check_access): Avoid warning when the no-warning bit
2260 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2262 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2263 * ira-color (allocno_hard_regs_compare): Likewise.
2265 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2268 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2269 Use .pushsection/.popsection.
2271 2018-01-14 Martin Sebor <msebor@redhat.com>
2274 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2276 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2278 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2279 entry from extra_headers.
2280 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2281 extra_headers, make the list bitwise identical to the i?86-*-* one.
2283 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2285 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2286 -mcmodel=large with -mindirect-branch=thunk,
2287 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2288 -mfunction-return=thunk-extern.
2289 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2290 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2291 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2293 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2295 * config/i386/i386.c (print_reg): Print the name of the full
2296 integer register without '%'.
2297 (ix86_print_operand): Handle 'V'.
2298 * doc/extend.texi: Document 'V' modifier.
2300 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2302 * config/i386/constraints.md (Bs): Disallow memory operand for
2303 -mindirect-branch-register.
2305 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2306 (GOT_memory_operand): Likewise.
2307 (call_insn_operand): Likewise.
2308 (sibcall_insn_operand): Likewise.
2309 (GOT32_symbol_operand): Likewise.
2310 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2311 for -mindirect-branch-register.
2312 (tablejump): Likewise.
2313 (*sibcall_memory): Likewise.
2314 (*sibcall_value_memory): Likewise.
2315 Disallow peepholes of indirect call and jump via memory for
2316 -mindirect-branch-register.
2317 (*call_pop): Replace m with Bw.
2318 (*call_value_pop): Likewise.
2319 (*sibcall_pop_memory): Replace m with Bs.
2320 * config/i386/i386.opt (mindirect-branch-register): New option.
2321 * doc/invoke.texi: Document -mindirect-branch-register option.
2323 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2325 * config/i386/i386-protos.h (ix86_output_function_return): New.
2326 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2327 set function_return_type.
2328 (indirect_thunk_name): Add ret_p to indicate thunk for function
2330 (output_indirect_thunk_function): Pass false to
2331 indirect_thunk_name.
2332 (ix86_output_indirect_branch_via_reg): Likewise.
2333 (ix86_output_indirect_branch_via_push): Likewise.
2334 (output_indirect_thunk_function): Create alias for function
2335 return thunk if regno < 0.
2336 (ix86_output_function_return): New function.
2337 (ix86_handle_fndecl_attribute): Handle function_return.
2338 (ix86_attribute_table): Add function_return.
2339 * config/i386/i386.h (machine_function): Add
2340 function_return_type.
2341 * config/i386/i386.md (simple_return_internal): Use
2342 ix86_output_function_return.
2343 (simple_return_internal_long): Likewise.
2344 * config/i386/i386.opt (mfunction-return=): New option.
2345 (indirect_branch): Mention -mfunction-return=.
2346 * doc/extend.texi: Document function_return function attribute.
2347 * doc/invoke.texi: Document -mfunction-return= option.
2349 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2351 * config/i386/i386-opts.h (indirect_branch): New.
2352 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2353 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2354 with local indirect jump when converting indirect call and jump.
2355 (ix86_set_indirect_branch_type): New.
2356 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2357 (indirectlabelno): New.
2358 (indirect_thunk_needed): Likewise.
2359 (indirect_thunk_bnd_needed): Likewise.
2360 (indirect_thunks_used): Likewise.
2361 (indirect_thunks_bnd_used): Likewise.
2362 (INDIRECT_LABEL): Likewise.
2363 (indirect_thunk_name): Likewise.
2364 (output_indirect_thunk): Likewise.
2365 (output_indirect_thunk_function): Likewise.
2366 (ix86_output_indirect_branch_via_reg): Likewise.
2367 (ix86_output_indirect_branch_via_push): Likewise.
2368 (ix86_output_indirect_branch): Likewise.
2369 (ix86_output_indirect_jmp): Likewise.
2370 (ix86_code_end): Call output_indirect_thunk_function if needed.
2371 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2373 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2374 (ix86_attribute_table): Add indirect_branch.
2375 * config/i386/i386.h (machine_function): Add indirect_branch_type
2376 and has_local_indirect_jump.
2377 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2379 (tablejump): Likewise.
2380 (*indirect_jump): Use ix86_output_indirect_jmp.
2381 (*tablejump_1): Likewise.
2382 (simple_return_indirect_internal): Likewise.
2383 * config/i386/i386.opt (mindirect-branch=): New option.
2384 (indirect_branch): New.
2387 (thunk-inline): Likewise.
2388 (thunk-extern): Likewise.
2389 * doc/extend.texi: Document indirect_branch function attribute.
2390 * doc/invoke.texi: Document -mindirect-branch= option.
2392 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2395 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2397 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2399 * ipa-inline.c (want_inline_small_function_p): Return false if
2400 inlining has already failed with CIF_FINAL_ERROR.
2401 (update_caller_keys): Call want_inline_small_function_p before
2403 (update_callee_keys): Likewise.
2405 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2407 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2409 (rs6000_quadword_masked_address_p): Likewise.
2410 (quad_aligned_load_p): Likewise.
2411 (quad_aligned_store_p): Likewise.
2412 (const_load_sequence_p): Add comment to describe the outer-most loop.
2413 (mimic_memory_attributes_and_flags): New function.
2414 (rs6000_gen_stvx): Likewise.
2415 (replace_swapped_aligned_store): Likewise.
2416 (rs6000_gen_lvx): Likewise.
2417 (replace_swapped_aligned_load): Likewise.
2418 (replace_swapped_load_constant): Capitalize argument name in
2419 comment describing this function.
2420 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2421 and stores that access quad-word aligned addresses and replace
2422 with stvx or lvx instructions when appropriate.
2423 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2424 New function prototype.
2425 (rs6000_quadword_masked_address_p): Likewise.
2426 (rs6000_gen_lvx): Likewise.
2427 (rs6000_gen_stvx): Likewise.
2428 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2429 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2430 when memory address is aligned.
2431 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2432 this split to select lvx instruction when memory address is aligned.
2433 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2434 instruction when memory address is aligned.
2435 (*vsx_le_perm_load_v16qi): Likewise.
2436 (four unnamed splitters): Modify to select the stvx instruction
2437 when memory is aligned.
2439 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2441 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2442 which appears in the queue multiple times.
2444 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2445 Alan Hayward <alan.hayward@arm.com>
2446 David Sherwood <david.sherwood@arm.com>
2448 * tree-vectorizer.h (vec_lower_bound): New structure.
2449 (_loop_vec_info): Add check_nonzero and lower_bounds.
2450 (LOOP_VINFO_CHECK_NONZERO): New macro.
2451 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2452 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2453 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2454 fields. Make seg_len the distance travelled, not including the
2456 (dr_direction_indicator): Declare.
2457 (dr_zero_step_indicator): Likewise.
2458 (dr_known_forward_stride_p): Likewise.
2459 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2461 (runtime_alias_check_p): Allow runtime alias checks with
2463 (operator ==): Compare access_size and align.
2464 (prune_runtime_alias_test_list): Rework for new distinction between
2465 the access_size and seg_len.
2466 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2468 (get_segment_min_max): New function.
2469 (create_intersect_range_checks): Use it.
2470 (dr_step_indicator): New function.
2471 (dr_direction_indicator): Likewise.
2472 (dr_zero_step_indicator): Likewise.
2473 (dr_known_forward_stride_p): Likewise.
2474 * tree-loop-distribution.c (data_ref_segment_size): Return
2475 DR_STEP * (niters - 1).
2476 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2478 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2479 (vect_preserves_scalar_order_p): New function, split out from...
2480 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2481 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2482 (vect_vfa_access_size): New function.
2483 (vect_vfa_align): Likewise.
2484 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2485 (dump_lower_bound): New function.
2486 (vect_check_lower_bound): Likewise.
2487 (vect_small_gap_p): Likewise.
2488 (vectorizable_with_step_bound_p): Likewise.
2489 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2490 depencies if the vectorization factor is 1. Convert the checks
2491 for nonzero steps into checks on the bounds of DR_STEP. Try using
2492 a bunds check for variable steps if the minimum required step is
2493 relatively small. Update calls to the dr_with_seg_len
2494 constructor and to vect_compile_time_alias.
2495 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2497 (vect_loop_versioning): Call it.
2498 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2500 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2503 Alan Hayward <alan.hayward@arm.com>
2504 David Sherwood <david.sherwood@arm.com>
2506 * doc/sourcebuild.texi (vect_scatter_store): Document.
2507 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2509 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2511 * genopinit.c (main): Add supports_vec_scatter_store and
2512 supports_vec_scatter_store_cached to target_optabs.
2513 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2514 IFN_MASK_SCATTER_STORE.
2515 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2517 * internal-fn.h (internal_store_fn_p): Declare.
2518 (internal_fn_stored_value_index): Likewise.
2519 * internal-fn.c (scatter_store_direct): New macro.
2520 (expand_scatter_store_optab_fn): New function.
2521 (direct_scatter_store_optab_supported_p): New macro.
2522 (internal_store_fn_p): New function.
2523 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2524 IFN_MASK_SCATTER_STORE.
2525 (internal_fn_mask_index): Likewise.
2526 (internal_fn_stored_value_index): New function.
2527 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2529 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2530 * optabs-query.c (supports_vec_scatter_store_p): New function.
2531 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2532 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2533 true for scatter stores.
2534 (vect_gather_scatter_fn_p): Handle scatter stores too.
2535 (vect_check_gather_scatter): Consider using scatter stores if
2536 supports_vec_scatter_store_p.
2537 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2539 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2540 internal_fn_stored_value_index.
2541 (check_load_store_masking): Handle scatter stores too.
2542 (vect_get_store_rhs): Make public.
2543 (vectorizable_call): Use internal_store_fn_p.
2544 (vectorizable_store): Handle scatter store internal functions.
2545 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2546 when deciding whether the end of the group has been reached.
2547 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2548 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2549 (mask_scatter_store<mode>): New insns.
2551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2552 Alan Hayward <alan.hayward@arm.com>
2553 David Sherwood <david.sherwood@arm.com>
2555 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2556 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2557 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2559 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2560 Use vect_truncate_gather_scatter_offset if we can't treat the
2561 operation as a normal gather load or scatter store.
2562 (get_group_load_store_type): Take the gather_scatter_info
2563 as argument. Try using a gather load or scatter store for
2564 single-element groups.
2565 (get_load_store_type): Update calls to get_group_load_store_type
2566 and vect_use_strided_gather_scatters_p.
2568 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2569 Alan Hayward <alan.hayward@arm.com>
2570 David Sherwood <david.sherwood@arm.com>
2572 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2573 optional tree argument.
2574 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2576 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2577 but continue to use the current value as a fallback.
2578 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2579 to compare the updates.
2580 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2581 (get_load_store_type): Use it when handling a strided access.
2582 (vect_get_strided_load_store_ops): New function.
2583 (vect_get_data_ptr_increment): Likewise.
2584 (vectorizable_load): Handle strided gather loads. Always pass
2585 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2587 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2588 Alan Hayward <alan.hayward@arm.com>
2589 David Sherwood <david.sherwood@arm.com>
2591 * doc/md.texi (gather_load@var{m}): Document.
2592 (mask_gather_load@var{m}): Likewise.
2593 * genopinit.c (main): Add supports_vec_gather_load and
2594 supports_vec_gather_load_cached to target_optabs.
2595 * optabs-tree.c (init_tree_optimization_optabs): Use
2596 ggc_cleared_alloc to allocate target_optabs.
2597 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2598 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2600 * internal-fn.h (internal_load_fn_p): Declare.
2601 (internal_gather_scatter_fn_p): Likewise.
2602 (internal_fn_mask_index): Likewise.
2603 (internal_gather_scatter_fn_supported_p): Likewise.
2604 * internal-fn.c (gather_load_direct): New macro.
2605 (expand_gather_load_optab_fn): New function.
2606 (direct_gather_load_optab_supported_p): New macro.
2607 (direct_internal_fn_optab): New function.
2608 (internal_load_fn_p): Likewise.
2609 (internal_gather_scatter_fn_p): Likewise.
2610 (internal_fn_mask_index): Likewise.
2611 (internal_gather_scatter_fn_supported_p): Likewise.
2612 * optabs-query.c (supports_at_least_one_mode_p): New function.
2613 (supports_vec_gather_load_p): Likewise.
2614 * optabs-query.h (supports_vec_gather_load_p): Declare.
2615 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2616 and memory_type field.
2617 (NUM_PATTERNS): Bump to 15.
2618 * tree-vect-data-refs.c: Include internal-fn.h.
2619 (vect_gather_scatter_fn_p): New function.
2620 (vect_describe_gather_scatter_call): Likewise.
2621 (vect_check_gather_scatter): Try using internal functions for
2622 gather loads. Recognize existing calls to a gather load function.
2623 (vect_analyze_data_refs): Consider using gather loads if
2624 supports_vec_gather_load_p.
2625 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2626 (vect_get_gather_scatter_offset_type): Likewise.
2627 (vect_convert_mask_for_vectype): Likewise.
2628 (vect_add_conversion_to_patterm): Likewise.
2629 (vect_try_gather_scatter_pattern): Likewise.
2630 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2631 (vect_vect_recog_func_ptrs): Add it.
2632 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2633 internal_fn_mask_index and internal_gather_scatter_fn_p.
2634 (check_load_store_masking): Take the gather_scatter_info as an
2635 argument and handle gather loads.
2636 (vect_get_gather_scatter_ops): New function.
2637 (vectorizable_call): Check internal_load_fn_p.
2638 (vectorizable_load): Likewise. Handle gather load internal
2640 (vectorizable_store): Update call to check_load_store_masking.
2641 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2642 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2643 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2644 (aarch64_gather_scale_operand_d): New predicates.
2645 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2646 (mask_gather_load<mode>): New insns.
2648 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2649 Alan Hayward <alan.hayward@arm.com>
2650 David Sherwood <david.sherwood@arm.com>
2652 * optabs.def (fold_left_plus_optab): New optab.
2653 * doc/md.texi (fold_left_plus_@var{m}): Document.
2654 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2655 * internal-fn.c (fold_left_direct): Define.
2656 (expand_fold_left_optab_fn): Likewise.
2657 (direct_fold_left_optab_supported_p): Likewise.
2658 * fold-const-call.c (fold_const_fold_left): New function.
2659 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2660 * tree-parloops.c (valid_reduction_p): New function.
2661 (gather_scalar_reductions): Use it.
2662 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2663 (vect_finish_replace_stmt): Declare.
2664 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2665 (needs_fold_left_reduction_p): New function, split out from...
2666 (vect_is_simple_reduction): ...here. Accept reductions that
2667 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2668 (vect_force_simple_reduction): Also store the reduction type in
2669 the assignment's STMT_VINFO_REDUC_TYPE.
2670 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2671 (merge_with_identity): New function.
2672 (vect_expand_fold_left): Likewise.
2673 (vectorize_fold_left_reduction): Likewise.
2674 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2675 scalar phi in place for it. Check for target support and reject
2676 cases that would reassociate the operation. Defer the transform
2677 phase to vectorize_fold_left_reduction.
2678 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2679 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2680 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2682 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2684 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2685 call to ifc_temp_var.
2687 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2688 Alan Hayward <alan.hayward@arm.com>
2689 David Sherwood <david.sherwood@arm.com>
2691 * target.def (legitimize_address_displacement): Take the original
2692 offset as a poly_int.
2693 * targhooks.h (default_legitimize_address_displacement): Update
2695 * targhooks.c (default_legitimize_address_displacement): Likewise.
2696 * doc/tm.texi: Regenerate.
2697 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2698 as an argument, moving assert of ad->disp == ad->disp_term to...
2699 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2700 Try calling targetm.legitimize_address_displacement before expanding
2701 the address rather than afterwards, and adjust for the new interface.
2702 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2703 Match the new hook interface. Handle SVE addresses.
2704 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2707 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2709 * Makefile.in (OBJS): Add early-remat.o.
2710 * target.def (select_early_remat_modes): New hook.
2711 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2712 * doc/tm.texi: Regenerate.
2713 * targhooks.h (default_select_early_remat_modes): Declare.
2714 * targhooks.c (default_select_early_remat_modes): New function.
2715 * timevar.def (TV_EARLY_REMAT): New timevar.
2716 * passes.def (pass_early_remat): New pass.
2717 * tree-pass.h (make_pass_early_remat): Declare.
2718 * early-remat.c: New file.
2719 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2721 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2723 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2724 Alan Hayward <alan.hayward@arm.com>
2725 David Sherwood <david.sherwood@arm.com>
2727 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2728 vfm1 with a bound_epilog parameter.
2729 (vect_do_peeling): Update calls accordingly, and move the prologue
2730 call earlier in the function. Treat the base bound_epilog as 0 for
2731 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2732 this base when peeling for gaps.
2733 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2734 with fully-masked loops.
2735 (vect_estimate_min_profitable_iters): Handle the single peeled
2736 iteration in that case.
2738 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2739 Alan Hayward <alan.hayward@arm.com>
2740 David Sherwood <david.sherwood@arm.com>
2742 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2743 single-element interleaving even if the size is not a power of 2.
2744 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2745 accesses for single-element interleaving if the group size is
2748 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2749 Alan Hayward <alan.hayward@arm.com>
2750 David Sherwood <david.sherwood@arm.com>
2752 * doc/md.texi (fold_extract_last_@var{m}): Document.
2753 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2754 * optabs.def (fold_extract_last_optab): New optab.
2755 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2756 * internal-fn.c (fold_extract_direct): New macro.
2757 (expand_fold_extract_optab_fn): Likewise.
2758 (direct_fold_extract_optab_supported_p): Likewise.
2759 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2760 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2761 EXTRACT_LAST_REDUCTION.
2762 (get_initial_def_for_reduction): Do not create an initial vector
2763 for EXTRACT_LAST_REDUCTION reductions.
2764 (vectorizable_reduction): Leave the scalar phi in place for
2765 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2766 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2767 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2768 transform phase to vectorizable_condition.
2769 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2771 (vect_finish_stmt_generation): ...here.
2772 (vect_finish_replace_stmt): New function.
2773 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2774 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2776 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
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 (extract_last_@var{m}): Document.
2783 * optabs.def (extract_last_optab): New optab.
2784 * internal-fn.def (EXTRACT_LAST): New internal function.
2785 * internal-fn.c (cond_unary_direct): New macro.
2786 (expand_cond_unary_optab_fn): Likewise.
2787 (direct_cond_unary_optab_supported_p): Likewise.
2788 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2789 loops using EXTRACT_LAST.
2790 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2791 (extract_last_<mode>): ...this optab.
2792 (vec_extract<mode><Vel>): Update accordingly.
2794 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2795 Alan Hayward <alan.hayward@arm.com>
2796 David Sherwood <david.sherwood@arm.com>
2798 * target.def (empty_mask_is_expensive): New hook.
2799 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2800 * doc/tm.texi: Regenerate.
2801 * targhooks.h (default_empty_mask_is_expensive): Declare.
2802 * targhooks.c (default_empty_mask_is_expensive): New function.
2803 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2804 if the target says that empty masks are expensive.
2805 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2807 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2809 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2810 Alan Hayward <alan.hayward@arm.com>
2811 David Sherwood <david.sherwood@arm.com>
2813 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2814 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2815 (vect_use_loop_mask_for_alignment_p): New function.
2816 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2817 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2818 niters_skip argument. Make sure that the first niters_skip elements
2819 of the first iteration are inactive.
2820 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2821 Update call to vect_set_loop_masks_directly.
2822 (get_misalign_in_elems): New function, split out from...
2823 (vect_gen_prolog_loop_niters): ...here.
2824 (vect_update_init_of_dr): Take a code argument that specifies whether
2825 the adjustment should be added or subtracted.
2826 (vect_update_init_of_drs): Likewise.
2827 (vect_prepare_for_masked_peels): New function.
2828 (vect_do_peeling): Skip prologue peeling if we're using a mask
2829 instead. Update call to vect_update_inits_of_drs.
2830 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2832 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2833 alignment. Do not include the number of peeled iterations in
2834 the minimum threshold in that case.
2835 (vectorizable_induction): Adjust the start value down by
2836 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2837 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2838 Take the number of skipped iterations into account when calculating
2840 * tree-vect-stmts.c (vect_gen_while_not): New function.
2842 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2843 Alan Hayward <alan.hayward@arm.com>
2844 David Sherwood <david.sherwood@arm.com>
2846 * doc/sourcebuild.texi (vect_fully_masked): Document.
2847 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2849 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2851 (vect_analyze_loop_2): ...here. Don't check the vectorization
2852 factor against the number of loop iterations if the loop is
2855 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2856 Alan Hayward <alan.hayward@arm.com>
2857 David Sherwood <david.sherwood@arm.com>
2859 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2860 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2861 (dump_groups): Update accordingly.
2862 (iv_use::mem_type): New member variable.
2863 (address_p): New function.
2864 (record_use): Add a mem_type argument and initialize the new
2866 (record_group_use): Add a mem_type argument. Use address_p.
2867 Remove obsolete null checks of base_object. Update call to record_use.
2868 (find_interesting_uses_op): Update call to record_group_use.
2869 (find_interesting_uses_cond): Likewise.
2870 (find_interesting_uses_address): Likewise.
2871 (get_mem_type_for_internal_fn): New function.
2872 (find_address_like_use): Likewise.
2873 (find_interesting_uses_stmt): Try find_address_like_use before
2874 calling find_interesting_uses_op.
2875 (addr_offset_valid_p): Use the iv mem_type field as the type
2876 of the addressed memory.
2877 (add_autoinc_candidates): Likewise.
2878 (get_address_cost): Likewise.
2879 (split_small_address_groups_p): Use address_p.
2880 (split_address_groups): Likewise.
2881 (add_iv_candidate_for_use): Likewise.
2882 (autoinc_possible_for_pair): Likewise.
2883 (rewrite_groups): Likewise.
2884 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2885 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2886 (get_alias_ptr_type_for_ptr_address): New function.
2887 (rewrite_use_address): Rewrite address uses in calls that were
2888 identified by find_address_like_use.
2890 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2891 Alan Hayward <alan.hayward@arm.com>
2892 David Sherwood <david.sherwood@arm.com>
2894 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2896 * gimple-expr.h (is_gimple_addressable: Likewise.
2897 * gimple-expr.c (is_gimple_address): Likewise.
2898 * internal-fn.c (expand_call_mem_ref): New function.
2899 (expand_mask_load_optab_fn): Use it.
2900 (expand_mask_store_optab_fn): Likewise.
2902 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2903 Alan Hayward <alan.hayward@arm.com>
2904 David Sherwood <david.sherwood@arm.com>
2906 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2907 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2908 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2909 (cond_umax@var{mode}): Document.
2910 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2911 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2912 (cond_umin_optab, cond_umax_optab): New optabs.
2913 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2914 (COND_IOR, COND_XOR): New internal functions.
2915 * internal-fn.h (get_conditional_internal_fn): Declare.
2916 * internal-fn.c (cond_binary_direct): New macro.
2917 (expand_cond_binary_optab_fn): Likewise.
2918 (direct_cond_binary_optab_supported_p): Likewise.
2919 (get_conditional_internal_fn): New function.
2920 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2921 Cope with reduction statements that are vectorized as calls rather
2923 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2924 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2925 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2926 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2927 (UNSPEC_COND_EOR): New unspecs.
2928 (optab): Add mappings for them.
2929 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2930 (sve_int_op, sve_fp_op): New int attributes.
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 * optabs.def (while_ult_optab): New optab.
2937 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2938 * internal-fn.def (WHILE_ULT): New internal function.
2939 * internal-fn.h (direct_internal_fn_supported_p): New override
2940 that takes two types as argument.
2941 * internal-fn.c (while_direct): New macro.
2942 (expand_while_optab_fn): New function.
2943 (convert_optab_supported_p): Likewise.
2944 (direct_while_optab_supported_p): New macro.
2945 * wide-int.h (wi::udiv_ceil): New function.
2946 * tree-vectorizer.h (rgroup_masks): New structure.
2947 (vec_loop_masks): New typedef.
2948 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2950 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2951 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2952 (vect_max_vf): New function.
2953 (slpeel_make_loop_iterate_ntimes): Delete.
2954 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2955 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2956 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2957 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2958 internal-fn.h, stor-layout.h and optabs-query.h.
2959 (vect_set_loop_mask): New function.
2960 (add_preheader_seq): Likewise.
2961 (add_header_seq): Likewise.
2962 (interleave_supported_p): Likewise.
2963 (vect_maybe_permute_loop_masks): Likewise.
2964 (vect_set_loop_masks_directly): Likewise.
2965 (vect_set_loop_condition_masked): Likewise.
2966 (vect_set_loop_condition_unmasked): New function, split out from
2967 slpeel_make_loop_iterate_ntimes.
2968 (slpeel_make_loop_iterate_ntimes): Rename to..
2969 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2970 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2971 (vect_do_peeling): Update call accordingly.
2972 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2974 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2975 mask_compare_type, can_fully_mask_p and fully_masked_p.
2976 (release_vec_loop_masks): New function.
2977 (_loop_vec_info): Use it to free the loop masks.
2978 (can_produce_all_loop_masks_p): New function.
2979 (vect_get_max_nscalars_per_iter): Likewise.
2980 (vect_verify_full_masking): Likewise.
2981 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2982 retries, and free the mask rgroups before retrying. Check loop-wide
2983 reasons for disallowing fully-masked loops. Make the final decision
2984 about whether use a fully-masked loop or not.
2985 (vect_estimate_min_profitable_iters): Do not assume that peeling
2986 for the number of iterations will be needed for fully-masked loops.
2987 (vectorizable_reduction): Disable fully-masked loops.
2988 (vectorizable_live_operation): Likewise.
2989 (vect_halve_mask_nunits): New function.
2990 (vect_double_mask_nunits): Likewise.
2991 (vect_record_loop_mask): Likewise.
2992 (vect_get_loop_mask): Likewise.
2993 (vect_transform_loop): Handle the case in which the final loop
2994 iteration might handle a partial vector. Call vect_set_loop_condition
2995 instead of slpeel_make_loop_iterate_ntimes.
2996 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2997 (check_load_store_masking): New function.
2998 (prepare_load_store_mask): Likewise.
2999 (vectorizable_store): Handle fully-masked loops.
3000 (vectorizable_load): Likewise.
3001 (supportable_widening_operation): Use vect_halve_mask_nunits for
3003 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3004 (vect_gen_while): New function.
3005 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3006 (aarch64_uqdec<mode>): New insn.
3008 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3009 Alan Hayward <alan.hayward@arm.com>
3010 David Sherwood <david.sherwood@arm.com>
3012 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3013 (reduc_xor_scal_optab): New optabs.
3014 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3015 (reduc_xor_scal_@var{m}): Document.
3016 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3017 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3019 * fold-const-call.c (fold_const_call): Handle them.
3020 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3021 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3022 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3023 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3024 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3025 (UNSPEC_XORV): New unspecs.
3026 (optab): Add entries for them.
3027 (BITWISEV): New int iterator.
3028 (bit_reduc_op): New int attributes.
3030 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3031 Alan Hayward <alan.hayward@arm.com>
3032 David Sherwood <david.sherwood@arm.com>
3034 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3035 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3036 * optabs.def (vec_shl_insert_optab): New optab.
3037 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3038 (duplicate_and_interleave): Likewise.
3039 * tree-vect-loop.c: Include internal-fn.h.
3040 (neutral_op_for_slp_reduction): New function, split out from
3041 get_initial_defs_for_reduction.
3042 (get_initial_def_for_reduction): Handle option 2 for variable-length
3043 vectors by loading the neutral value into a vector and then shifting
3044 the initial value into element 0.
3045 (get_initial_defs_for_reduction): Replace the code argument with
3046 the neutral value calculated by neutral_op_for_slp_reduction.
3047 Use gimple_build_vector for constant-length vectors.
3048 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3049 but the first group_size elements have a neutral value.
3050 Use duplicate_and_interleave otherwise.
3051 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3052 Update call to get_initial_defs_for_reduction. Handle SLP
3053 reductions for variable-length vectors by creating one vector
3054 result for each scalar result, with the elements associated
3055 with other scalar results stubbed out with the neutral value.
3056 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3057 Require IFN_VEC_SHL_INSERT for double reductions on
3058 variable-length vectors, or SLP reductions that have
3059 a neutral value. Require can_duplicate_and_interleave_p
3060 support for variable-length unchained SLP reductions if there
3061 is no neutral value, such as for MIN/MAX reductions. Also require
3062 the number of vector elements to be a multiple of the number of
3063 SLP statements when doing variable-length unchained SLP reductions.
3064 Update call to vect_create_epilog_for_reduction.
3065 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3066 and remove initial values.
3067 (duplicate_and_interleave): Make public.
3068 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3069 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3071 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3072 Alan Hayward <alan.hayward@arm.com>
3073 David Sherwood <david.sherwood@arm.com>
3075 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3076 (can_duplicate_and_interleave_p): New function.
3077 (vect_get_and_check_slp_defs): Take the vector of statements
3078 rather than just the current one. Remove excess parentheses.
3079 Restriction rejectinon of vect_constant_def and vect_external_def
3080 for variable-length vectors to boolean types, or types for which
3081 can_duplicate_and_interleave_p is false.
3082 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3083 (duplicate_and_interleave): New function.
3084 (vect_get_constant_vectors): Use gimple_build_vector for
3085 constant-length vectors and suitable variable-length constant
3086 vectors. Use duplicate_and_interleave for other variable-length
3087 vectors. Don't defer the update when inserting new statements.
3089 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3090 Alan Hayward <alan.hayward@arm.com>
3091 David Sherwood <david.sherwood@arm.com>
3093 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3094 min_profitable_iters doesn't go negative.
3096 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3097 Alan Hayward <alan.hayward@arm.com>
3098 David Sherwood <david.sherwood@arm.com>
3100 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3101 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3102 * optabs.def (vec_mask_load_lanes_optab): New optab.
3103 (vec_mask_store_lanes_optab): Likewise.
3104 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3105 (MASK_STORE_LANES): Likewise.
3106 * internal-fn.c (mask_load_lanes_direct): New macro.
3107 (mask_store_lanes_direct): Likewise.
3108 (expand_mask_load_optab_fn): Handle masked operations.
3109 (expand_mask_load_lanes_optab_fn): New macro.
3110 (expand_mask_store_optab_fn): Handle masked operations.
3111 (expand_mask_store_lanes_optab_fn): New macro.
3112 (direct_mask_load_lanes_optab_supported_p): Likewise.
3113 (direct_mask_store_lanes_optab_supported_p): Likewise.
3114 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3116 (vect_load_lanes_supported): Likewise.
3117 * tree-vect-data-refs.c (strip_conversion): New function.
3118 (can_group_stmts_p): Likewise.
3119 (vect_analyze_data_ref_accesses): Use it instead of checking
3120 for a pair of assignments.
3121 (vect_store_lanes_supported): Take a masked_p parameter.
3122 (vect_load_lanes_supported): Likewise.
3123 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3124 vect_store_lanes_supported and vect_load_lanes_supported.
3125 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3126 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3127 parameter. Don't allow gaps for masked accesses.
3128 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3129 and vect_load_lanes_supported.
3130 (get_load_store_type): Take a masked_p parameter and update
3131 call to get_group_load_store_type.
3132 (vectorizable_store): Update call to get_load_store_type.
3133 Handle IFN_MASK_STORE_LANES.
3134 (vectorizable_load): Update call to get_load_store_type.
3135 Handle IFN_MASK_LOAD_LANES.
3137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3138 Alan Hayward <alan.hayward@arm.com>
3139 David Sherwood <david.sherwood@arm.com>
3141 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3143 * config/aarch64/aarch64-protos.h
3144 (aarch64_sve_struct_memory_operand_p): Declare.
3145 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3146 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3147 (VPRED, vpred): Handle SVE structure modes.
3148 * config/aarch64/constraints.md (Utx): New constraint.
3149 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3150 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3151 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3152 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3153 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3154 structure modes. Split into pieces after RA.
3155 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3156 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3158 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3159 SVE structure modes.
3160 (aarch64_classify_address): Likewise.
3161 (sizetochar): Move earlier in file.
3162 (aarch64_print_operand): Handle SVE register lists.
3163 (aarch64_array_mode): New function.
3164 (aarch64_sve_struct_memory_operand_p): Likewise.
3165 (TARGET_ARRAY_MODE): Redefine.
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 * target.def (array_mode): New target hook.
3172 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3173 * doc/tm.texi: Regenerate.
3174 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3175 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3176 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3178 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3181 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3182 Alan Hayward <alan.hayward@arm.com>
3183 David Sherwood <david.sherwood@arm.com>
3185 * fold-const.c (fold_binary_loc): Check the argument types
3186 rather than the result type when testing for a vector operation.
3188 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3190 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3191 * doc/tm.texi: Regenerate.
3193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3194 Alan Hayward <alan.hayward@arm.com>
3195 David Sherwood <david.sherwood@arm.com>
3197 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3198 (sve): Document new AArch64 extension.
3199 * doc/md.texi (w): Extend the description of the AArch64
3200 constraint to include SVE vectors.
3201 (Upl, Upa): Document new AArch64 predicate constraints.
3202 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3204 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3205 (msve-vector-bits=): New option.
3206 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3207 SVE when these are disabled.
3208 (sve): New extension.
3209 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3210 modes. Adjust their number of units based on aarch64_sve_vg.
3211 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3212 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3213 aarch64_addr_query_type.
3214 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3215 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3216 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3217 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3218 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3219 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3220 (aarch64_simd_imm_zero_p): Delete.
3221 (aarch64_check_zero_based_sve_index_immediate): Declare.
3222 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3223 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3224 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3225 (aarch64_sve_float_mul_immediate_p): Likewise.
3226 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3228 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3229 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3230 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3231 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3232 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3233 (aarch64_regmode_natural_size): Likewise.
3234 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3235 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3237 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3238 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3239 for VG and the SVE predicate registers.
3240 (V_ALIASES): Add a "z"-prefixed alias.
3241 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3242 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3243 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3244 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3245 (REG_CLASS_NAMES): Add entries for them.
3246 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3247 and the predicate registers.
3248 (aarch64_sve_vg): Declare.
3249 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3250 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3251 (REGMODE_NATURAL_SIZE): Define.
3252 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3254 * config/aarch64/aarch64.c: Include cfgrtl.h.
3255 (simd_immediate_info): Add a constructor for series vectors,
3256 and an associated step field.
3257 (aarch64_sve_vg): New variable.
3258 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3259 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3260 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3261 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3262 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3263 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3264 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3265 (aarch64_get_mask_mode): New functions.
3266 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3267 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3268 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3269 predicate modes and predicate registers. Explicitly restrict
3270 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3271 to store a vector mode if it is recognized by
3272 aarch64_classify_vector_mode.
3273 (aarch64_regmode_natural_size): New function.
3274 (aarch64_hard_regno_caller_save_mode): Return the original mode
3276 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3277 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3278 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3279 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3281 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3282 does not overlap dest if the function is frame-related. Handle
3284 (aarch64_split_add_offset): New function.
3285 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3286 them aarch64_add_offset.
3287 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3288 and update call to aarch64_sub_sp.
3289 (aarch64_add_cfa_expression): New function.
3290 (aarch64_expand_prologue): Pass extra temporary registers to the
3291 functions above. Handle the case in which we need to emit new
3292 DW_CFA_expressions for registers that were originally saved
3293 relative to the stack pointer, but now have to be expressed
3294 relative to the frame pointer.
3295 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3297 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3298 IP0 and IP1 values for SVE frames.
3299 (aarch64_expand_vec_series): New function.
3300 (aarch64_expand_sve_widened_duplicate): Likewise.
3301 (aarch64_expand_sve_const_vector): Likewise.
3302 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3303 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3304 into the register, rather than emitting a SET directly.
3305 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3306 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3307 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3308 (offset_9bit_signed_scaled_p): New functions.
3309 (aarch64_replicate_bitmask_imm): New function.
3310 (aarch64_bitmask_imm): Use it.
3311 (aarch64_cannot_force_const_mem): Reject expressions involving
3312 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3313 (aarch64_classify_index): Handle SVE indices, by requiring
3314 a plain register index with a scale that matches the element size.
3315 (aarch64_classify_address): Handle SVE addresses. Assert that
3316 the mode of the address is VOIDmode or an integer mode.
3317 Update call to aarch64_classify_symbol.
3318 (aarch64_classify_symbolic_expression): Update call to
3319 aarch64_classify_symbol.
3320 (aarch64_const_vec_all_in_range_p): New function.
3321 (aarch64_print_vector_float_operand): Likewise.
3322 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3323 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3324 and the FP immediates 1.0 and 0.5.
3325 (aarch64_print_address_internal): Handle SVE addresses.
3326 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3327 (aarch64_regno_regclass): Handle predicate registers.
3328 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3330 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3331 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3332 (aarch64_convert_sve_vector_bits): New function.
3333 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3334 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3336 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3337 Handle SVE vector and predicate modes. Accept VL-based constants
3338 that need only one temporary register, and VL offsets that require
3339 no temporary registers.
3340 (aarch64_conditional_register_usage): Mark the predicate registers
3341 as fixed if SVE isn't available.
3342 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3343 Return true for SVE vector and predicate modes.
3344 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3345 rather than an unsigned int. Handle SVE modes.
3346 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3348 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3350 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3351 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3352 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3353 (aarch64_sve_float_mul_immediate_p): New functions.
3354 (aarch64_sve_valid_immediate): New function.
3355 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3356 Explicitly reject structure modes. Check for INDEX constants.
3357 Handle PTRUE and PFALSE constants.
3358 (aarch64_check_zero_based_sve_index_immediate): New function.
3359 (aarch64_simd_imm_zero_p): Delete.
3360 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3361 vector modes. Accept constants in the range of CNT[BHWD].
3362 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3363 ask for an Advanced SIMD mode.
3364 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3365 (aarch64_simd_vector_alignment): Handle SVE predicates.
3366 (aarch64_vectorize_preferred_vector_alignment): New function.
3367 (aarch64_simd_vector_alignment_reachable): Use it instead of
3369 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3370 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3372 (MAX_VECT_LEN): Delete.
3373 (expand_vec_perm_d): Add a vec_flags field.
3374 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3375 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3376 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3378 (aarch64_evpc_rev): Rename to...
3379 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3380 (aarch64_evpc_rev_global): New function.
3381 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3382 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3384 (aarch64_evpc_sve_tbl): New function.
3385 (aarch64_expand_vec_perm_const_1): Update after rename of
3386 aarch64_evpc_rev. Handle SVE permutes too, trying
3387 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3388 than aarch64_evpc_tbl.
3389 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3390 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3391 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3392 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3393 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3394 (aarch64_expand_sve_vcond): New functions.
3395 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3396 of aarch64_vector_mode_p.
3397 (aarch64_dwarf_poly_indeterminate_value): New function.
3398 (aarch64_compute_pressure_classes): Likewise.
3399 (aarch64_can_change_mode_class): Likewise.
3400 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3401 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3402 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3403 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3404 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3405 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3406 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3407 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3409 (Dn, Dl, Dr): Accept const as well as const_vector.
3410 (Dz): Likewise. Compare against CONST0_RTX.
3411 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3412 of "vector" where appropriate.
3413 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3414 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3415 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3416 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3417 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3418 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3419 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3420 (v_int_equiv): Extend to SVE modes.
3421 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3423 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3424 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3425 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3426 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3427 (SVE_COND_FP_CMP): New int iterators.
3428 (perm_hilo): Handle the new unpack unspecs.
3429 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3431 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3432 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3433 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3434 (aarch64_equality_operator, aarch64_constant_vector_operand)
3435 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3436 (aarch64_sve_nonimmediate_operand): Likewise.
3437 (aarch64_sve_general_operand): Likewise.
3438 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3439 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3440 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3441 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3442 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3443 (aarch64_sve_float_arith_immediate): Likewise.
3444 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3445 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3446 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3447 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3448 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3449 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3450 (aarch64_sve_float_arith_operand): Likewise.
3451 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3452 (aarch64_sve_float_mul_operand): Likewise.
3453 (aarch64_sve_vec_perm_operand): Likewise.
3454 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3455 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3456 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3457 as well as const_vector.
3458 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3459 in file. Use CONST0_RTX and CONSTM1_RTX.
3460 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3461 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3462 Use aarch64_simd_imm_zero.
3463 * config/aarch64/aarch64-sve.md: New file.
3464 * config/aarch64/aarch64.md: Include it.
3465 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3466 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3467 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3468 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3469 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3470 (sve): New attribute.
3471 (enabled): Disable instructions with the sve attribute unless
3473 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3474 aarch64_expand_mov_immediate.
3475 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3476 CNT[BHSD] immediates.
3477 (movti): Split CONST_POLY_INT moves into two halves.
3478 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3479 Split additions that need a temporary here if the destination
3480 is the stack pointer.
3481 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3482 (*add<mode>3_poly_1): New instruction.
3483 (set_clobber_cc): New expander.
3485 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3487 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3488 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3489 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3490 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3491 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3492 Change innermode from fixed_mode_size to machine_mode.
3493 (simplify_subreg): Update call accordingly. Handle a constant-sized
3494 subreg of a variable-length CONST_VECTOR.
3496 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3497 Alan Hayward <alan.hayward@arm.com>
3498 David Sherwood <david.sherwood@arm.com>
3500 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3501 (add_offset_to_base): New function, split out from...
3502 (create_mem_ref): ...here. When handling a scale other than 1,
3503 check first whether the address is valid without the offset.
3504 Add it into the base if so, leaving the index and scale as-is.
3506 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3509 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3510 fold_for_warn before checking if arg2 is INTEGER_CST.
3512 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3514 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3515 (store_multiple_operation): Delete.
3516 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3517 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3518 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3519 guarded by TARGET_STRING.
3520 (rs6000_output_load_multiple): Delete.
3521 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3522 OPTION_MASK_STRING / TARGET_STRING handling.
3523 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3524 (const rs6000_opt_masks) <"string">: Change mask to 0.
3525 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3526 (MASK_STRING): Delete.
3527 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3529 (load_multiple): Delete.
3536 (store_multiple): Delete.
3543 (movmemsi_8reg): Delete.
3544 (corresponding unnamed define_insn): Delete.
3545 (movmemsi_6reg): Delete.
3546 (corresponding unnamed define_insn): Delete.
3547 (movmemsi_4reg): Delete.
3548 (corresponding unnamed define_insn): Delete.
3549 (movmemsi_2reg): Delete.
3550 (corresponding unnamed define_insn): Delete.
3551 (movmemsi_1reg): Delete.
3552 (corresponding unnamed define_insn): Delete.
3553 * config/rs6000/rs6000.opt (mno-string): New.
3554 (mstring): Replace by deprecation warning stub.
3555 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3557 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3559 * regrename.c (regrename_do_replace): If replacing the same
3560 reg multiple times, try to reuse last created gen_raw_REG.
3563 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3564 main to workaround a bug in GDB.
3566 2018-01-12 Tom de Vries <tom@codesourcery.com>
3569 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3571 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3573 PR rtl-optimization/80481
3574 * ira-color.c (get_cap_member): New function.
3575 (allocnos_conflict_by_live_ranges_p): Use it.
3576 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3577 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3579 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3582 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3583 (*saddl_se_1): Ditto.
3585 (*ssubl_se_1): Ditto.
3587 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3589 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3590 rather than wi::to_widest for DR_INITs.
3591 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3592 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3593 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3595 (vect_analyze_group_access_1): Note that here.
3597 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3599 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3600 polynomial type sizes.
3602 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3604 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3605 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3606 (gimple_add_tmp_var): Likewise.
3608 2018-01-12 Martin Liska <mliska@suse.cz>
3610 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3611 (gimple_alloc_sizes): Likewise.
3612 (dump_gimple_statistics): Use PRIu64 in printf format.
3613 * gimple.h: Change uint64_t to int.
3615 2018-01-12 Martin Liska <mliska@suse.cz>
3617 * tree-core.h: Use uint64_t instead of int.
3618 * tree.c (tree_node_counts): Likewise.
3619 (tree_node_sizes): Likewise.
3620 (dump_tree_statistics): Use PRIu64 in printf format.
3622 2018-01-12 Martin Liska <mliska@suse.cz>
3624 * Makefile.in: As qsort_chk is implemented in vec.c, add
3625 vec.o to linkage of gencfn-macros.
3626 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3627 passing the info to record_node_allocation_statistics.
3628 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3630 * ggc-common.c (struct ggc_usage): Add operator== and use
3631 it in operator< and compare function.
3632 * mem-stats.h (struct mem_usage): Likewise.
3633 * vec.c (struct vec_usage): Remove operator< and compare
3634 function. Can be simply inherited.
3636 2018-01-12 Martin Jambor <mjambor@suse.cz>
3639 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3640 * tree-ssa-math-opts.c: Include domwalk.h.
3641 (convert_mult_to_fma_1): New function.
3642 (fma_transformation_info): New type.
3643 (fma_deferring_state): Likewise.
3644 (cancel_fma_deferring): New function.
3645 (result_of_phi): Likewise.
3646 (last_fma_candidate_feeds_initial_phi): Likewise.
3647 (convert_mult_to_fma): Added deferring logic, split actual
3648 transformation to convert_mult_to_fma_1.
3649 (math_opts_dom_walker): New type.
3650 (math_opts_dom_walker::after_dom_children): New method, body moved
3651 here from pass_optimize_widening_mul::execute, added deferring logic
3653 (pass_optimize_widening_mul::execute): Moved most of code to
3654 math_opts_dom_walker::after_dom_children.
3655 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3656 * config/i386/i386.c (ix86_option_override_internal): Added
3657 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3659 2018-01-12 Richard Biener <rguenther@suse.de>
3662 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3663 inline instance vars.
3665 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3668 * config/rx/rx.c (rx_is_restricted_memory_address):
3671 2018-01-12 Richard Biener <rguenther@suse.de>
3673 PR tree-optimization/80846
3674 * target.def (split_reduction): New target hook.
3675 * targhooks.c (default_split_reduction): New function.
3676 * targhooks.h (default_split_reduction): Declare.
3677 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3678 target requests first reduce vectors by combining low and high
3680 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3681 (get_vectype_for_scalar_type_and_size): Export.
3682 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3683 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3684 * doc/tm.texi: Regenerate.
3685 * config/i386/i386.c (ix86_split_reduction): Implement
3686 TARGET_VECTORIZE_SPLIT_REDUCTION.
3688 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3691 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3692 in PIC mode except for TARGET_VXWORKS_RTP.
3693 * config/sparc/sparc.c: Include cfgrtl.h.
3694 (TARGET_INIT_PIC_REG): Define.
3695 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3696 (sparc_pic_register_p): New predicate.
3697 (sparc_legitimate_address_p): Use it.
3698 (sparc_legitimize_pic_address): Likewise.
3699 (sparc_delegitimize_address): Likewise.
3700 (sparc_mode_dependent_address_p): Likewise.
3701 (gen_load_pcrel_sym): Remove 4th parameter.
3702 (load_got_register): Adjust call to above. Remove obsolete stuff.
3703 (sparc_expand_prologue): Do not call load_got_register here.
3704 (sparc_flat_expand_prologue): Likewise.
3705 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3706 (sparc_use_pseudo_pic_reg): New function.
3707 (sparc_init_pic_reg): Likewise.
3708 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3709 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3711 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3713 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3714 Add item for branch_cost.
3716 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3718 PR rtl-optimization/83565
3719 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3720 not extend the result to a larger mode for rotate operations.
3721 (num_sign_bit_copies1): Likewise.
3723 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3726 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3728 Use values-Xc.o for -pedantic.
3729 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3731 2018-01-12 Martin Liska <mliska@suse.cz>
3734 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3736 (possible_polymorphic_call_targets): Use it.
3737 (ipa_devirt): Likewise.
3739 2018-01-12 Martin Liska <mliska@suse.cz>
3741 * profile-count.h (enum profile_quality): Use 0 as invalid
3742 enum value of profile_quality.
3744 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3746 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3747 -mext-string options.
3749 2018-01-12 Richard Biener <rguenther@suse.de>
3751 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3752 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3753 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3755 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3757 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3759 * configure.ac (--with-long-double-format): Add support for the
3760 configuration option to change the default long double format on
3762 * config.gcc (powerpc*-linux*-*): Likewise.
3763 * configure: Regenerate.
3764 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3765 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3766 used without modification.
3768 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3770 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3771 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3772 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3773 MISC_BUILTIN_SPEC_BARRIER.
3774 (rs6000_init_builtins): Likewise.
3775 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3777 (speculation_barrier): New define_insn.
3778 * doc/extend.texi: Document __builtin_speculation_barrier.
3780 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3783 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3784 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3785 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3787 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3788 integral modes instead of "ss" and "sd".
3789 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3790 vectors with 32-bit and 64-bit elements.
3791 (vecdupssescalarmodesuffix): New mode attribute.
3792 (vec_dup<mode>): Use it.
3794 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3797 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3798 frame if argument is passed on stack.
3800 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3803 * ree.c (combine_reaching_defs): Optimize also
3804 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3805 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3807 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3810 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3812 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3815 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3816 after they are computed.
3818 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3820 PR tree-optimization/83695
3821 * gimple-loop-linterchange.cc
3822 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3823 reset cached scev information after interchange.
3824 (pass_linterchange::execute): Remove call to scev_reset_htab.
3826 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3828 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3829 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3830 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3831 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3832 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3833 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3834 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3835 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3836 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3837 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3838 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3839 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3840 (V_lane_reg): Likewise.
3841 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3843 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3844 (vfmal_lane_low<mode>_intrinsic,
3845 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3846 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3847 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3848 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3849 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3850 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3852 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3854 * config/arm/arm-cpus.in (fp16fml): New feature.
3855 (ALL_SIMD): Add fp16fml.
3856 (armv8.2-a): Add fp16fml as an option.
3857 (armv8.3-a): Likewise.
3858 (armv8.4-a): Add fp16fml as part of fp16.
3859 * config/arm/arm.h (TARGET_FP16FML): Define.
3860 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3862 * config/arm/arm-modes.def (V2HF): Define.
3863 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3864 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3865 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3866 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3867 vfmsl_low, vfmsl_high): New set of builtins.
3868 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3869 (vfml_op): New code attribute.
3870 (VFMLHALVES): New int iterator.
3871 (VFML, VFMLSEL): New mode attributes.
3872 (V_reg): Define mapping for V2HF.
3873 (V_hi, V_lo): New mode attributes.
3874 (VF_constraint): Likewise.
3875 (vfml_half, vfml_half_selector): New int attributes.
3876 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3878 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3879 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3881 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3882 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3883 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3884 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3886 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3887 Document new effective target and option set.
3889 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3891 * config/arm/arm-cpus.in (armv8_4): New feature.
3892 (ARMv8_4a): New fgroup.
3893 (armv8.4-a): New arch.
3894 * config/arm/arm-tables.opt: Regenerate.
3895 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3896 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3897 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3898 Add matching rules for -march=armv8.4-a and extensions.
3899 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3901 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3904 * config/rx/rx.md (BW): New mode attribute.
3905 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3907 2018-01-11 Richard Biener <rguenther@suse.de>
3909 PR tree-optimization/83435
3910 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3911 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3912 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3914 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3915 Alan Hayward <alan.hayward@arm.com>
3916 David Sherwood <david.sherwood@arm.com>
3918 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3920 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3921 (aarch64_print_address_internal): Use it to check for a zero offset.
3923 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3924 Alan Hayward <alan.hayward@arm.com>
3925 David Sherwood <david.sherwood@arm.com>
3927 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3928 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3929 Return a poly_int64 rather than a HOST_WIDE_INT.
3930 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3931 rather than a HOST_WIDE_INT.
3932 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3933 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3934 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3935 final_offset from HOST_WIDE_INT to poly_int64.
3936 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3937 to_constant when getting the number of units in an Advanced SIMD
3939 (aarch64_builtin_vectorized_function): Check for a constant number
3941 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3943 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3944 attribute instead of GET_MODE_NUNITS.
3945 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3946 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3947 GET_MODE_SIZE for fixed-size registers.
3948 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3949 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3950 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3951 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3952 (aarch64_print_operand, aarch64_print_address_internal)
3953 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3954 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3955 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3956 Handle polynomial GET_MODE_SIZE.
3957 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3958 wider than SImode without modification.
3959 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3960 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3961 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3962 passing and returning SVE modes.
3963 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3964 rather than GEN_INT.
3965 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3966 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3967 (aarch64_allocate_and_probe_stack_space): Likewise.
3968 (aarch64_layout_frame): Cope with polynomial offsets.
3969 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3970 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3972 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3973 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3974 poly_int64 rather than a HOST_WIDE_INT.
3975 (aarch64_get_separate_components, aarch64_process_components)
3976 (aarch64_expand_prologue, aarch64_expand_epilogue)
3977 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3978 (aarch64_anchor_offset): New function, split out from...
3979 (aarch64_legitimize_address): ...here.
3980 (aarch64_builtin_vectorization_cost): Handle polynomial
3981 TYPE_VECTOR_SUBPARTS.
3982 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3984 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3985 number of elements from the PARALLEL rather than the mode.
3986 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3987 rather than GET_MODE_BITSIZE.
3988 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3989 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3990 (aarch64_expand_vec_perm_const_1): Handle polynomial
3991 d->perm.length () and d->perm elements.
3992 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3993 Apply to_constant to d->perm elements.
3994 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3995 polynomial CONST_VECTOR_NUNITS.
3996 (aarch64_move_pointer): Take amount as a poly_int64 rather
3998 (aarch64_progress_pointer): Avoid temporary variable.
3999 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4000 the mode attribute instead of GET_MODE.
4002 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4003 Alan Hayward <alan.hayward@arm.com>
4004 David Sherwood <david.sherwood@arm.com>
4006 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4007 x exists before using it.
4008 (aarch64_add_constant_internal): Rename to...
4009 (aarch64_add_offset_1): ...this. Replace regnum with separate
4010 src and dest rtxes. Handle the case in which they're different,
4011 including when the offset is zero. Replace scratchreg with an rtx.
4012 Use 2 additions if there is no spare register into which we can
4013 move a 16-bit constant.
4014 (aarch64_add_constant): Delete.
4015 (aarch64_add_offset): Replace reg with separate src and dest
4016 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4017 Use aarch64_add_offset_1.
4018 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4019 an rtx rather than an int. Take the delta as a poly_int64
4020 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4021 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4022 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4023 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4024 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4026 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4027 aarch64_add_constant.
4029 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4031 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4032 Use scalar_float_mode.
4034 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4036 * config/aarch64/aarch64-simd.md
4037 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4038 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4039 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4040 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4041 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4042 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4043 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4044 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4045 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4046 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4048 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4051 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4052 targ_options->x_arm_arch_string is non NULL.
4054 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4056 * config/aarch64/aarch64.h
4057 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4059 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4062 * expmed.c (emit_store_flag_force): Swap if const op0
4063 and change VOIDmode to mode of op0.
4065 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4067 PR rtl-optimization/83761
4068 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4069 than bytes to mode_for_size.
4071 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4074 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4075 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4078 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4081 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4082 when in layout mode.
4083 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4084 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4087 2018-01-10 Michael Collison <michael.collison@arm.com>
4089 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4090 * config/aarch64/aarch64-option-extension.def: Add
4091 AARCH64_OPT_EXTENSION of 'fp16fml'.
4092 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4093 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4094 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4095 * config/aarch64/constraints.md (Ui7): New constraint.
4096 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4097 (VFMLA_SEL_W): Ditto.
4100 (VFMLA16_LOW): New int iterator.
4101 (VFMLA16_HIGH): Ditto.
4102 (UNSPEC_FMLAL): New unspec.
4103 (UNSPEC_FMLSL): Ditto.
4104 (UNSPEC_FMLAL2): Ditto.
4105 (UNSPEC_FMLSL2): Ditto.
4106 (f16mac): New code attribute.
4107 * config/aarch64/aarch64-simd-builtins.def
4108 (aarch64_fmlal_lowv2sf): Ditto.
4109 (aarch64_fmlsl_lowv2sf): Ditto.
4110 (aarch64_fmlalq_lowv4sf): Ditto.
4111 (aarch64_fmlslq_lowv4sf): Ditto.
4112 (aarch64_fmlal_highv2sf): Ditto.
4113 (aarch64_fmlsl_highv2sf): Ditto.
4114 (aarch64_fmlalq_highv4sf): Ditto.
4115 (aarch64_fmlslq_highv4sf): Ditto.
4116 (aarch64_fmlal_lane_lowv2sf): Ditto.
4117 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4118 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4119 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4120 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4121 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4122 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4123 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4124 (aarch64_fmlal_lane_highv2sf): Ditto.
4125 (aarch64_fmlsl_lane_highv2sf): Ditto.
4126 (aarch64_fmlal_laneq_highv2sf): Ditto.
4127 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4128 (aarch64_fmlalq_lane_highv4sf): Ditto.
4129 (aarch64_fmlsl_lane_highv4sf): Ditto.
4130 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4131 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4132 * config/aarch64/aarch64-simd.md:
4133 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4134 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4135 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4136 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4137 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4138 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4139 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4140 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4141 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4142 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4143 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4144 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4145 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4146 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4147 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4148 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4149 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4150 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4151 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4152 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4153 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4154 (vfmlsl_low_u32): Ditto.
4155 (vfmlalq_low_u32): Ditto.
4156 (vfmlslq_low_u32): Ditto.
4157 (vfmlal_high_u32): Ditto.
4158 (vfmlsl_high_u32): Ditto.
4159 (vfmlalq_high_u32): Ditto.
4160 (vfmlslq_high_u32): Ditto.
4161 (vfmlal_lane_low_u32): Ditto.
4162 (vfmlsl_lane_low_u32): Ditto.
4163 (vfmlal_laneq_low_u32): Ditto.
4164 (vfmlsl_laneq_low_u32): Ditto.
4165 (vfmlalq_lane_low_u32): Ditto.
4166 (vfmlslq_lane_low_u32): Ditto.
4167 (vfmlalq_laneq_low_u32): Ditto.
4168 (vfmlslq_laneq_low_u32): Ditto.
4169 (vfmlal_lane_high_u32): Ditto.
4170 (vfmlsl_lane_high_u32): Ditto.
4171 (vfmlal_laneq_high_u32): Ditto.
4172 (vfmlsl_laneq_high_u32): Ditto.
4173 (vfmlalq_lane_high_u32): Ditto.
4174 (vfmlslq_lane_high_u32): Ditto.
4175 (vfmlalq_laneq_high_u32): Ditto.
4176 (vfmlslq_laneq_high_u32): Ditto.
4177 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4178 (AARCH64_FL_FOR_ARCH8_4): New.
4179 (AARCH64_ISA_F16FML): New ISA flag.
4180 (TARGET_F16FML): New feature flag for fp16fml.
4181 (doc/invoke.texi): Document new fp16fml option.
4183 2018-01-10 Michael Collison <michael.collison@arm.com>
4185 * config/aarch64/aarch64-builtins.c:
4186 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4187 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4188 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4189 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4190 (AARCH64_ISA_SHA3): New ISA flag.
4191 (TARGET_SHA3): New feature flag for sha3.
4192 * config/aarch64/iterators.md (sha512_op): New int attribute.
4193 (CRYPTO_SHA512): New int iterator.
4194 (UNSPEC_SHA512H): New unspec.
4195 (UNSPEC_SHA512H2): Ditto.
4196 (UNSPEC_SHA512SU0): Ditto.
4197 (UNSPEC_SHA512SU1): Ditto.
4198 * config/aarch64/aarch64-simd-builtins.def
4199 (aarch64_crypto_sha512hqv2di): New builtin.
4200 (aarch64_crypto_sha512h2qv2di): Ditto.
4201 (aarch64_crypto_sha512su0qv2di): Ditto.
4202 (aarch64_crypto_sha512su1qv2di): Ditto.
4203 (aarch64_eor3qv8hi): Ditto.
4204 (aarch64_rax1qv2di): Ditto.
4205 (aarch64_xarqv2di): Ditto.
4206 (aarch64_bcaxqv8hi): Ditto.
4207 * config/aarch64/aarch64-simd.md:
4208 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4209 (aarch64_crypto_sha512su0qv2di): Ditto.
4210 (aarch64_crypto_sha512su1qv2di): Ditto.
4211 (aarch64_eor3qv8hi): Ditto.
4212 (aarch64_rax1qv2di): Ditto.
4213 (aarch64_xarqv2di): Ditto.
4214 (aarch64_bcaxqv8hi): Ditto.
4215 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4216 (vsha512h2q_u64): Ditto.
4217 (vsha512su0q_u64): Ditto.
4218 (vsha512su1q_u64): Ditto.
4219 (veor3q_u16): Ditto.
4220 (vrax1q_u64): Ditto.
4222 (vbcaxq_u16): Ditto.
4223 * config/arm/types.md (crypto_sha512): New type attribute.
4224 (crypto_sha3): Ditto.
4225 (doc/invoke.texi): Document new sha3 option.
4227 2018-01-10 Michael Collison <michael.collison@arm.com>
4229 * config/aarch64/aarch64-builtins.c:
4230 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4231 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4232 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4233 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4234 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4235 (AARCH64_ISA_SM4): New ISA flag.
4236 (TARGET_SM4): New feature flag for sm4.
4237 * config/aarch64/aarch64-simd-builtins.def
4238 (aarch64_sm3ss1qv4si): Ditto.
4239 (aarch64_sm3tt1aq4si): Ditto.
4240 (aarch64_sm3tt1bq4si): Ditto.
4241 (aarch64_sm3tt2aq4si): Ditto.
4242 (aarch64_sm3tt2bq4si): Ditto.
4243 (aarch64_sm3partw1qv4si): Ditto.
4244 (aarch64_sm3partw2qv4si): Ditto.
4245 (aarch64_sm4eqv4si): Ditto.
4246 (aarch64_sm4ekeyqv4si): Ditto.
4247 * config/aarch64/aarch64-simd.md:
4248 (aarch64_sm3ss1qv4si): Ditto.
4249 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4250 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4251 (aarch64_sm4eqv4si): Ditto.
4252 (aarch64_sm4ekeyqv4si): Ditto.
4253 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4254 (sm3part_op): Ditto.
4255 (CRYPTO_SM3TT): Ditto.
4256 (CRYPTO_SM3PART): Ditto.
4257 (UNSPEC_SM3SS1): New unspec.
4258 (UNSPEC_SM3TT1A): Ditto.
4259 (UNSPEC_SM3TT1B): Ditto.
4260 (UNSPEC_SM3TT2A): Ditto.
4261 (UNSPEC_SM3TT2B): Ditto.
4262 (UNSPEC_SM3PARTW1): Ditto.
4263 (UNSPEC_SM3PARTW2): Ditto.
4264 (UNSPEC_SM4E): Ditto.
4265 (UNSPEC_SM4EKEY): Ditto.
4266 * config/aarch64/constraints.md (Ui2): New constraint.
4267 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4268 * config/arm/types.md (crypto_sm3): New type attribute.
4269 (crypto_sm4): Ditto.
4270 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4271 (vsm3tt1aq_u32): Ditto.
4272 (vsm3tt1bq_u32): Ditto.
4273 (vsm3tt2aq_u32): Ditto.
4274 (vsm3tt2bq_u32): Ditto.
4275 (vsm3partw1q_u32): Ditto.
4276 (vsm3partw2q_u32): Ditto.
4277 (vsm4eq_u32): Ditto.
4278 (vsm4ekeyq_u32): Ditto.
4279 (doc/invoke.texi): Document new sm4 option.
4281 2018-01-10 Michael Collison <michael.collison@arm.com>
4283 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4284 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4285 (AARCH64_FL_FOR_ARCH8_4): New.
4286 (AARCH64_FL_V8_4): New flag.
4287 (doc/invoke.texi): Document new armv8.4-a option.
4289 2018-01-10 Michael Collison <michael.collison@arm.com>
4291 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4292 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4293 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4294 * config/aarch64/aarch64-option-extension.def: Add
4295 AARCH64_OPT_EXTENSION of 'sha2'.
4296 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4297 (crypto): Disable sha2 and aes if crypto disabled.
4298 (crypto): Enable aes and sha2 if enabled.
4299 (simd): Disable sha2 and aes if simd disabled.
4300 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4302 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4303 (TARGET_SHA2): New feature flag for sha2.
4304 (TARGET_AES): New feature flag for aes.
4305 * config/aarch64/aarch64-simd.md:
4306 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4307 conditional on TARGET_AES.
4308 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4309 (aarch64_crypto_sha1hsi): Make pattern conditional
4311 (aarch64_crypto_sha1hv4si): Ditto.
4312 (aarch64_be_crypto_sha1hv4si): Ditto.
4313 (aarch64_crypto_sha1su1v4si): Ditto.
4314 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4315 (aarch64_crypto_sha1su0v4si): Ditto.
4316 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4317 (aarch64_crypto_sha256su0v4si): Ditto.
4318 (aarch64_crypto_sha256su1v4si): Ditto.
4319 (doc/invoke.texi): Document new aes and sha2 options.
4321 2018-01-10 Martin Sebor <msebor@redhat.com>
4323 PR tree-optimization/83781
4324 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4327 2018-01-11 Martin Sebor <msebor@gmail.com>
4328 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4330 PR tree-optimization/83501
4331 PR tree-optimization/81703
4333 * tree-ssa-strlen.c (get_string_cst): Rename...
4334 (get_string_len): ...to this. Handle global constants.
4335 (handle_char_store): Adjust.
4337 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4338 Jim Wilson <jimw@sifive.com>
4340 * config/riscv/riscv-protos.h (riscv_output_return): New.
4341 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4342 (riscv_attribute_table, riscv_output_return),
4343 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4344 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4345 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4347 (riscv_expand_prologue): Add early return for naked function.
4348 (riscv_expand_epilogue): Likewise.
4349 (riscv_function_ok_for_sibcall): Return false for naked function.
4350 (riscv_set_current_function): New.
4351 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4352 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4353 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4354 * doc/extend.texi (RISC-V Function Attributes): New.
4356 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4358 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4359 check for 128-bit long double before checking TCmode.
4360 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4361 128-bit long doubles before checking TFmode or TCmode.
4362 (FLOAT128_IBM_P): Likewise.
4364 2018-01-10 Martin Sebor <msebor@redhat.com>
4366 PR tree-optimization/83671
4367 * builtins.c (c_strlen): Unconditionally return zero for the empty
4369 Use -Warray-bounds for warnings.
4370 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4371 for non-constant array indices with COMPONENT_REF, arrays of
4372 arrays, and pointers to arrays.
4373 (gimple_fold_builtin_strlen): Determine and set length range for
4374 non-constant character arrays.
4376 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4379 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4382 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4384 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4386 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4389 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4390 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4391 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4392 indexed_or_indirect_operand predicate.
4393 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4394 (*vsx_le_perm_load_v8hi): Likewise.
4395 (*vsx_le_perm_load_v16qi): Likewise.
4396 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4397 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4398 (*vsx_le_perm_store_v8hi): Likewise.
4399 (*vsx_le_perm_store_v16qi): Likewise.
4400 (eight unnamed splitters): Likewise.
4402 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4404 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4405 * config/rs6000/emmintrin.h: Likewise.
4406 * config/rs6000/mmintrin.h: Likewise.
4407 * config/rs6000/xmmintrin.h: Likewise.
4409 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4412 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4414 * tree.c (tree_nop_conversion): Return true for location wrapper
4416 (maybe_wrap_with_location): New function.
4417 (selftest::check_strip_nops): New function.
4418 (selftest::test_location_wrappers): New function.
4419 (selftest::tree_c_tests): Call it.
4420 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4421 (maybe_wrap_with_location): New decl.
4422 (EXPR_LOCATION_WRAPPER_P): New macro.
4423 (location_wrapper_p): New inline function.
4424 (tree_strip_any_location_wrapper): New inline function.
4426 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4429 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4430 stack_realign_offset for the largest alignment of stack slot
4432 (ix86_find_max_used_stack_alignment): New function.
4433 (ix86_finalize_stack_frame_flags): Use it. Set
4434 max_used_stack_alignment if we don't realign stack.
4435 * config/i386/i386.h (machine_function): Add
4436 max_used_stack_alignment.
4438 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4440 * config/arm/arm.opt (-mbranch-cost): New option.
4441 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4444 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4447 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4448 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4450 2018-01-10 Richard Biener <rguenther@suse.de>
4453 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4454 early out so it also covers the case where we have a non-NULL
4457 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4459 PR tree-optimization/83753
4460 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4461 for non-strided grouped accesses if the number of elements is 1.
4463 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4466 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4467 * i386.h (TARGET_USE_GATHER): Define.
4468 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4470 2018-01-10 Martin Liska <mliska@suse.cz>
4473 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4474 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4476 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4477 CLEANUP_NO_PARTITIONING is not set.
4479 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4481 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4482 for vectors, as a partial revert of r254296.
4483 * rtl.h (const_vec_p): Delete.
4484 (const_vec_duplicate_p): Don't test for vector CONSTs.
4485 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4486 * expmed.c (make_tree): Likewise.
4489 * common.md (E, F): Use CONSTANT_P instead of checking for
4491 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4492 checking for CONST_VECTOR.
4494 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4497 * predict.c (force_edge_cold): Handle in more sane way edges
4500 2018-01-09 Carl Love <cel@us.ibm.com>
4502 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4504 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4505 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4506 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4507 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4508 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4509 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4510 * config/rs6000/rs6000-protos.h: Add extern defition for
4511 rs6000_generate_float2_double_code.
4512 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4514 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4515 (float2_v2df): Add define_expand.
4517 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4520 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4521 op_mode in the force_to_mode call.
4523 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4525 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4526 instead of checking each element individually.
4527 (aarch64_evpc_uzp): Likewise.
4528 (aarch64_evpc_zip): Likewise.
4529 (aarch64_evpc_ext): Likewise.
4530 (aarch64_evpc_rev): Likewise.
4531 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4532 instead of checking each element individually. Return true without
4534 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4535 whether all selected elements come from the same input, instead of
4536 checking each element individually. Remove calls to gen_rtx_REG,
4537 start_sequence and end_sequence and instead assert that no rtl is
4540 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4542 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4543 order of HIGH and CONST checks.
4545 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4547 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4548 if the destination isn't an SSA_NAME.
4550 2018-01-09 Richard Biener <rguenther@suse.de>
4552 PR tree-optimization/83668
4553 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4555 (canonicalize_loop_form): ... here, renamed from ...
4556 (canonicalize_loop_closed_ssa_form): ... this and amended to
4557 swap successor edges for loop exit blocks to make us use
4558 the RPO order we need for initial schedule generation.
4560 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4562 PR tree-optimization/64811
4563 * match.pd: When optimizing comparisons with Inf, avoid
4564 introducing or losing exceptions from comparisons with NaN.
4566 2018-01-09 Martin Liska <mliska@suse.cz>
4569 * asan.c (shadow_mem_size): Add gcc_assert.
4571 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4573 Don't save registers in main().
4576 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4577 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4578 * config/avr/avr.c (avr_set_current_function): Don't error if
4579 naked, OS_task or OS_main are specified at the same time.
4580 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4582 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4584 * common/config/avr/avr-common.c (avr_option_optimization_table):
4585 Switch on -mmain-is-OS_task for optimizing compilations.
4587 2018-01-09 Richard Biener <rguenther@suse.de>
4589 PR tree-optimization/83572
4590 * graphite.c: Include cfganal.h.
4591 (graphite_transform_loops): Connect infinite loops to exit
4592 and remove fake edges at the end.
4594 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4596 * ipa-inline.c (edge_badness): Revert accidental checkin.
4598 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4601 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4602 symbols; not inline clones.
4604 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4607 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4608 hard registers. Formatting fixes.
4610 PR preprocessor/83722
4611 * gcc.c (try_generate_repro): Pass
4612 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4613 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4616 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4617 Kito Cheng <kito.cheng@gmail.com>
4619 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4620 (riscv_leaf_function_p): Delete.
4621 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4623 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4625 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4627 (do_ifelse): New function.
4628 (do_isel): New function.
4629 (do_sub3): New function.
4630 (do_add3): New function.
4631 (do_load_mask_compare): New function.
4632 (do_overlap_load_compare): New function.
4633 (expand_compare_loop): New function.
4634 (expand_block_compare): Call expand_compare_loop() when appropriate.
4635 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4637 (-mblock-compare-inline-loop-limit): New option.
4639 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4642 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4643 Reverse order of second and third operands in first alternative.
4644 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4645 of first and second elements in UNSPEC_VPERMR vector.
4646 (altivec_expand_vec_perm_le): Likewise.
4648 2018-01-08 Jeff Law <law@redhat.com>
4650 PR rtl-optimizatin/81308
4651 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4652 (process_switch): If group_case_labels makes a change, then set
4654 (pass_convert_switch::execute): If a switch is converted, then
4655 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4657 PR rtl-optimization/81308
4658 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4661 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4663 PR target/83663 - Revert r255946
4664 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4665 generation for cases where splatting a value is not useful.
4666 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4667 across a vec_duplicate and a paradoxical subreg forming a vector
4668 mode to a vec_concat.
4670 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4672 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4673 -march=armv8.3-a variants.
4674 * config/arm/t-multilib: Likewise.
4675 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4677 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4679 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4681 (cceq_ior_compare_complement): Give it a name so I can use it, and
4682 change boolean_or_operator predicate to boolean_operator so it can
4683 be used to generate a crand.
4684 (eqne): New code iterator.
4685 (bd/bd_neg): New code_attrs.
4686 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4687 a single define_insn.
4688 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4689 decrement (bdnzt/bdnzf/bdzt/bdzf).
4690 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4691 with the new names of the branch decrement patterns, and added the
4692 names of the branch decrement conditional patterns.
4694 2018-01-08 Richard Biener <rguenther@suse.de>
4696 PR tree-optimization/83563
4697 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4700 2018-01-08 Richard Biener <rguenther@suse.de>
4703 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4705 2018-01-08 Richard Biener <rguenther@suse.de>
4707 PR tree-optimization/83685
4708 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4709 references to abnormals.
4711 2018-01-08 Richard Biener <rguenther@suse.de>
4714 * dwarf2out.c (output_indirect_strings): Handle empty
4715 skeleton_debug_str_hash.
4716 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4718 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4720 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4721 (emit_store_direct): Likewise.
4722 (arc_trampoline_adjust_address): Likewise.
4723 (arc_asm_trampoline_template): New function.
4724 (arc_initialize_trampoline): Use asm_trampoline_template.
4725 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4726 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4727 * config/arc/arc.md (flush_icache): Delete pattern.
4729 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4731 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4732 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4735 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4738 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4739 by not USED_FOR_TARGET.
4740 (make_pass_resolve_sw_modes): Likewise.
4742 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4744 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4747 2018-01-08 Richard Biener <rguenther@suse.de>
4750 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4752 2018-01-08 Richard Biener <rguenther@suse.de>
4755 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4757 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4760 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4761 basic blocks with a small number of successors.
4762 (convert_control_dep_chain_into_preds): Improve handling of
4764 (dump_predicates): Split apart into...
4765 (dump_pred_chain): ...here...
4766 (dump_pred_info): ...and here.
4767 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4768 (can_chain_union_be_invalidated_p): Improve check for invalidation
4770 (uninit_uses_cannot_happen): Avoid unnecessary if
4771 convert_control_dep_chain_into_preds yielded nothing.
4773 2018-01-06 Martin Sebor <msebor@redhat.com>
4775 PR tree-optimization/83640
4776 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4777 subtracting negative offset from size.
4778 (builtin_access::overlap): Adjust offset bounds of the access to fall
4779 within the size of the object if possible.
4781 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4783 PR rtl-optimization/83699
4784 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4785 extract_bit_field_as_subreg to cases in which the extracted
4786 value is also a vector.
4788 * lra-constraints.c (process_alt_operands): Test for the equivalence
4789 substitutions when detecting a possible reload cycle.
4791 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4794 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4795 by default if flag_selective_schedling{,2}. Formatting fixes.
4797 PR rtl-optimization/83682
4798 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4799 if it has non-VECTOR_MODE element mode.
4800 (vec_duplicate_p): Likewise.
4803 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4804 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4806 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4809 * config/i386/i386-builtin.def
4810 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4811 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4812 Require also OPTION_MASK_ISA_AVX512F in addition to
4813 OPTION_MASK_ISA_GFNI.
4814 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4815 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4816 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4817 to OPTION_MASK_ISA_GFNI.
4818 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4819 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4820 OPTION_MASK_ISA_AVX512BW.
4821 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4822 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4823 addition to OPTION_MASK_ISA_GFNI.
4824 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4825 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4826 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4827 to OPTION_MASK_ISA_GFNI.
4828 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4829 a requirement for all ISAs rather than any of them with a few
4831 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4833 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4834 bitmasks to be enabled with 3 exceptions, instead of requiring any
4835 enabled ISA with lots of exceptions.
4836 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4837 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4838 Change avx512bw in isa attribute to avx512f.
4839 * config/i386/sgxintrin.h: Add license boilerplate.
4840 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4841 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4842 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4843 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4845 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4846 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4847 temporarily sse2 rather than sse if not enabled already.
4850 * config/i386/sse.md (VI248_VLBW): Rename to ...
4851 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4852 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4853 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4854 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4855 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4856 mode iterator instead of VI248_VLBW.
4858 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4860 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4861 (record_modified): Skip clobbers; add debug output.
4862 (param_change_prob): Use sreal frequencies.
4864 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4866 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4867 punt for user-aligned variables.
4869 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4871 * tree-chrec.c (chrec_contains_symbols): Return true for
4874 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4877 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4878 of (x|y) == x for BICS pattern.
4880 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4882 PR tree-optimization/83605
4883 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4884 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4887 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4889 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4890 * config/epiphany/rtems.h: New file.
4892 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4893 Uros Bizjak <ubizjak@gmail.com>
4896 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4897 QIreg_operand instead of register_operand predicate.
4898 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4899 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4900 comments instead of -fmitigate[-_]rop.
4902 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4905 * cgraphunit.c (symbol_table::compile): Switch to text_section
4906 before calling assembly_start debug hook.
4907 * run-rtl-passes.c (run_rtl_passes): Likewise.
4910 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4912 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4913 range_int_cst_p rather than !symbolic_range_p before calling
4914 extract_range_from_multiplicative_op_1.
4916 2018-01-04 Jeff Law <law@redhat.com>
4918 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4919 redundant test in assertion.
4921 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4923 * doc/rtl.texi: Document machine_mode wrapper classes.
4925 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4927 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4930 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4932 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4933 the VEC_PERM_EXPR fold to fail.
4935 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4938 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4939 to switched_sections.
4941 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4944 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4947 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4950 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4951 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4953 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4956 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4957 is BLKmode and bitpos not zero or mode change is needed.
4959 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4962 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4965 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4968 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4969 instead of MULT rtx. Update all corresponding splitters.
4971 (*ssub<modesuffix>): Ditto.
4973 (*cmp_sadd_di): Update split patterns.
4974 (*cmp_sadd_si): Ditto.
4975 (*cmp_sadd_sidi): Ditto.
4976 (*cmp_ssub_di): Ditto.
4977 (*cmp_ssub_si): Ditto.
4978 (*cmp_ssub_sidi): Ditto.
4979 * config/alpha/predicates.md (const23_operand): New predicate.
4980 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4981 Look for ASHIFT, not MULT inner operand.
4982 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4984 2018-01-04 Martin Liska <mliska@suse.cz>
4986 PR gcov-profile/83669
4987 * gcov.c (output_intermediate_file): Add version to intermediate
4989 * doc/gcov.texi: Document new field 'version' in intermediate
4990 file format. Fix location of '-k' option of gcov command.
4992 2018-01-04 Martin Liska <mliska@suse.cz>
4995 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4997 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4999 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5001 2018-01-03 Martin Sebor <msebor@redhat.com>
5003 PR tree-optimization/83655
5004 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5005 checking calls with invalid arguments.
5007 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5009 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5010 (vectorizable_mask_load_store): Delete.
5011 (vectorizable_call): Return false for masked loads and stores.
5012 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5013 instead of gimple_assign_rhs1.
5014 (vectorizable_load): Handle IFN_MASK_LOAD.
5015 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5017 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5019 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5021 (vectorizable_mask_load_store): ...here.
5022 (vectorizable_load): ...and here.
5024 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5026 * tree-vect-stmts.c (vect_build_all_ones_mask)
5027 (vect_build_zero_merge_argument): New functions, split out from...
5028 (vectorizable_load): ...here.
5030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5032 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5034 (vectorizable_mask_load_store): ...here.
5035 (vectorizable_store): ...and here.
5037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5039 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5041 (vectorizable_mask_load_store): ...here.
5043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5045 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5046 (vect_model_store_cost): Take a vec_load_store_type instead of a
5048 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5049 (vect_model_store_cost): Take a vec_load_store_type instead of a
5051 (vectorizable_mask_load_store): Update accordingly.
5052 (vectorizable_store): Likewise.
5053 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5057 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5058 IFN_MASK_LOAD calls here rather than...
5059 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5062 Alan Hayward <alan.hayward@arm.com>
5063 David Sherwood <david.sherwood@arm.com>
5065 * expmed.c (extract_bit_field_1): For vector extracts,
5066 fall back to extract_bit_field_as_subreg if vec_extract
5069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5070 Alan Hayward <alan.hayward@arm.com>
5071 David Sherwood <david.sherwood@arm.com>
5073 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5074 they are variable or constant sized.
5075 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5076 slots for constant-sized data.
5078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5079 Alan Hayward <alan.hayward@arm.com>
5080 David Sherwood <david.sherwood@arm.com>
5082 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5083 handling COND_EXPRs with boolean comparisons, try to find a better
5084 basis for the mask type than the boolean itself.
5086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5088 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5089 is calculated and how it can be overridden.
5090 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5091 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5093 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5097 Alan Hayward <alan.hayward@arm.com>
5098 David Sherwood <david.sherwood@arm.com>
5100 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5101 Remove the mode argument.
5102 (aarch64_simd_valid_immediate): Remove the mode and inverse
5104 * config/aarch64/iterators.md (bitsize): New iterator.
5105 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5106 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5107 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5108 aarch64_simd_valid_immediate.
5109 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5110 (aarch64_reg_or_bic_imm): Likewise.
5111 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5112 with an insn_type enum and msl with a modifier_type enum.
5113 Replace element_width with a scalar_mode. Change the shift
5114 to unsigned int. Add constructors for scalar_float_mode and
5115 scalar_int_mode elements.
5116 (aarch64_vect_float_const_representable_p): Delete.
5117 (aarch64_can_const_movi_rtx_p)
5118 (aarch64_simd_scalar_immediate_valid_for_move)
5119 (aarch64_simd_make_constant): Update call to
5120 aarch64_simd_valid_immediate.
5121 (aarch64_advsimd_valid_immediate_hs): New function.
5122 (aarch64_advsimd_valid_immediate): Likewise.
5123 (aarch64_simd_valid_immediate): Remove mode and inverse
5124 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5125 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5126 and aarch64_float_const_representable_p on the result.
5127 (aarch64_output_simd_mov_immediate): Remove mode argument.
5128 Update call to aarch64_simd_valid_immediate and use of
5129 simd_immediate_info.
5130 (aarch64_output_scalar_simd_mov_immediate): Update call
5133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5134 Alan Hayward <alan.hayward@arm.com>
5135 David Sherwood <david.sherwood@arm.com>
5137 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5138 (mode_nunits): Likewise CONST_MODE_NUNITS.
5139 * machmode.def (ADJUST_NUNITS): Document.
5140 * genmodes.c (mode_data::need_nunits_adj): New field.
5141 (blank_mode): Update accordingly.
5142 (adj_nunits): New variable.
5143 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5145 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5146 listed in adj_nunits.
5147 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5148 listed in adj_nunits. Don't emit case statements for such modes.
5149 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5150 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5151 nothing if adj_nunits is nonnull.
5152 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5153 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5154 (emit_mode_fbit): Update use of print_maybe_const_decl.
5155 (emit_move_size): Likewise. Treat the array as non-const
5157 (emit_mode_adjustments): Handle adj_nunits.
5159 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5161 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5162 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5163 (VECTOR_MODES): Use it.
5164 (make_vector_modes): Take the prefix as an argument.
5166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5167 Alan Hayward <alan.hayward@arm.com>
5168 David Sherwood <david.sherwood@arm.com>
5170 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5171 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5172 for MODE_VECTOR_BOOL.
5173 * machmode.def (VECTOR_BOOL_MODE): Document.
5174 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5175 (make_vector_bool_mode): New function.
5176 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5178 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5179 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5181 * stor-layout.c (int_mode_for_mode): Likewise.
5182 * tree.c (build_vector_type_for_mode): Likewise.
5183 * varasm.c (output_constant_pool_2): Likewise.
5184 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5185 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5186 for MODE_VECTOR_BOOL.
5187 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5188 of mode class checks.
5189 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5190 instead of a list of mode class checks.
5191 (expand_vector_scalar_condition): Likewise.
5192 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5195 Alan Hayward <alan.hayward@arm.com>
5196 David Sherwood <david.sherwood@arm.com>
5198 * machmode.h (mode_size): Change from unsigned short to
5200 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5201 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5202 or if measurement_type is not polynomial.
5203 (fixed_size_mode::includes_p): Check for constant-sized modes.
5204 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5205 return a poly_uint16 rather than an unsigned short.
5206 (emit_mode_size): Change the type of mode_size from unsigned short
5207 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5208 (emit_mode_adjustments): Cope with polynomial vector sizes.
5209 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5211 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5213 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5214 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5215 * caller-save.c (setup_save_areas): Likewise.
5216 (replace_reg_with_saved_mem): Likewise.
5217 * calls.c (emit_library_call_value_1): Likewise.
5218 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5219 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5220 (gen_lowpart_for_combine): Likewise.
5221 * convert.c (convert_to_integer_1): Likewise.
5222 * cse.c (equiv_constant, cse_insn): Likewise.
5223 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5224 (cselib_subst_to_values): Likewise.
5225 * dce.c (word_dce_process_block): Likewise.
5226 * df-problems.c (df_word_lr_mark_ref): Likewise.
5227 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5228 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5229 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5230 (rtl_for_decl_location): Likewise.
5231 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5232 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5233 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5234 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5235 (expand_expr_real_1): Likewise.
5236 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5237 (pad_below): Likewise.
5238 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5239 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5240 * ira.c (get_subreg_tracking_sizes): Likewise.
5241 * ira-build.c (ira_create_allocno_objects): Likewise.
5242 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5243 (ira_sort_regnos_for_alter_reg): Likewise.
5244 * ira-costs.c (record_operand_costs): Likewise.
5245 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5246 (resolve_simple_move): Likewise.
5247 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5248 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5249 (lra_constraints): Likewise.
5250 (CONST_POOL_OK_P): Reject variable-sized modes.
5251 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5252 (add_pseudo_to_slot, lra_spill): Likewise.
5253 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5254 * optabs-query.c (get_best_extraction_insn): Likewise.
5255 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5256 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5257 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5258 * recog.c (offsettable_address_addr_space_p): Likewise.
5259 * regcprop.c (maybe_mode_change): Likewise.
5260 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5261 * regrename.c (build_def_use): Likewise.
5262 * regstat.c (dump_reg_info): Likewise.
5263 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5264 (find_reloads, find_reloads_subreg_address): Likewise.
5265 * reload1.c (eliminate_regs_1): Likewise.
5266 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5267 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5268 (simplify_binary_operation_1, simplify_subreg): Likewise.
5269 * targhooks.c (default_function_arg_padding): Likewise.
5270 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5271 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5272 (verify_gimple_assign_ternary): Likewise.
5273 * tree-inline.c (estimate_move_cost): Likewise.
5274 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5275 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5276 (get_address_cost_ainc): Likewise.
5277 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5278 (vect_supportable_dr_alignment): Likewise.
5279 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5280 (vectorizable_reduction): Likewise.
5281 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5282 (vectorizable_operation, vectorizable_load): Likewise.
5283 * tree.c (build_same_sized_truth_vector_type): Likewise.
5284 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5285 * var-tracking.c (emit_note_insn_var_location): Likewise.
5286 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5287 (ADDR_VEC_ALIGN): Likewise.
5289 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5290 Alan Hayward <alan.hayward@arm.com>
5291 David Sherwood <david.sherwood@arm.com>
5293 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5295 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5296 or if measurement_type is polynomial.
5297 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5298 * combine.c (make_extraction): Likewise.
5299 * dse.c (find_shift_sequence): Likewise.
5300 * dwarf2out.c (mem_loc_descriptor): Likewise.
5301 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5302 (extract_bit_field, extract_low_bits): Likewise.
5303 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5304 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5305 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5306 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5307 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5308 * reload.c (find_reloads): Likewise.
5309 * reload1.c (alter_reg): Likewise.
5310 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5311 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5312 * tree-if-conv.c (predicate_mem_writes): Likewise.
5313 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5314 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5315 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5316 * valtrack.c (dead_debug_insert_temp): Likewise.
5317 * varasm.c (mergeable_constant_section): Likewise.
5318 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5321 Alan Hayward <alan.hayward@arm.com>
5322 David Sherwood <david.sherwood@arm.com>
5324 * expr.c (expand_assignment): Cope with polynomial mode sizes
5325 when assigning to a CONCAT.
5327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5328 Alan Hayward <alan.hayward@arm.com>
5329 David Sherwood <david.sherwood@arm.com>
5331 * machmode.h (mode_precision): Change from unsigned short to
5333 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5335 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5336 or if measurement_type is not polynomial.
5337 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5338 in which the mode is already known to be a scalar_int_mode.
5339 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5340 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5342 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5343 for GET_MODE_PRECISION.
5344 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5345 for GET_MODE_PRECISION.
5346 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5348 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5349 (expand_field_assignment, make_extraction): Likewise.
5350 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5351 (get_last_value): Likewise.
5352 * convert.c (convert_to_integer_1): Likewise.
5353 * cse.c (cse_insn): Likewise.
5354 * expr.c (expand_expr_real_1): Likewise.
5355 * lra-constraints.c (simplify_operand_subreg): Likewise.
5356 * optabs-query.c (can_atomic_load_p): Likewise.
5357 * optabs.c (expand_atomic_load): Likewise.
5358 (expand_atomic_store): Likewise.
5359 * ree.c (combine_reaching_defs): Likewise.
5360 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5361 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5362 * tree.h (type_has_mode_precision_p): Likewise.
5363 * ubsan.c (instrument_si_overflow): Likewise.
5365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5366 Alan Hayward <alan.hayward@arm.com>
5367 David Sherwood <david.sherwood@arm.com>
5369 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5370 polynomial numbers of units.
5371 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5372 (valid_vector_subparts_p): New function.
5373 (build_vector_type): Remove temporary shim and take the number
5374 of units as a poly_uint64 rather than an int.
5375 (build_opaque_vector_type): Take the number of units as a
5376 poly_uint64 rather than an int.
5377 * tree.c (build_vector_from_ctor): Handle polynomial
5378 TYPE_VECTOR_SUBPARTS.
5379 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5380 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5381 (build_vector_from_val): If the number of units is variable,
5382 use build_vec_duplicate_cst for constant operands and
5383 VEC_DUPLICATE_EXPR otherwise.
5384 (make_vector_type): Remove temporary is_constant ().
5385 (build_vector_type, build_opaque_vector_type): Take the number of
5386 units as a poly_uint64 rather than an int.
5387 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5389 * cfgexpand.c (expand_debug_expr): Likewise.
5390 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5391 (store_constructor, expand_expr_real_1): Likewise.
5392 (const_scalar_mask_from_tree): Likewise.
5393 * fold-const-call.c (fold_const_reduction): Likewise.
5394 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5395 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5396 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5397 (fold_relational_const): Likewise.
5398 (native_interpret_vector): Likewise. Change the size from an
5399 int to an unsigned int.
5400 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5401 TYPE_VECTOR_SUBPARTS.
5402 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5403 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5404 duplicating a non-constant operand into a variable-length vector.
5405 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5406 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5407 * ipa-icf.c (sem_variable::equals): Likewise.
5408 * match.pd: Likewise.
5409 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5410 * print-tree.c (print_node): Likewise.
5411 * stor-layout.c (layout_type): Likewise.
5412 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5413 * tree-cfg.c (verify_gimple_comparison): Likewise.
5414 (verify_gimple_assign_binary): Likewise.
5415 (verify_gimple_assign_ternary): Likewise.
5416 (verify_gimple_assign_single): Likewise.
5417 * tree-pretty-print.c (dump_generic_node): Likewise.
5418 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5419 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5420 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5421 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5422 (vect_shift_permute_load_chain): Likewise.
5423 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5424 (expand_vector_condition, optimize_vector_constructor): Likewise.
5425 (lower_vec_perm, get_compute_type): Likewise.
5426 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5427 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5428 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5429 (vect_recog_mask_conversion_pattern): Likewise.
5430 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5431 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5432 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5433 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5434 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5435 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5436 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5437 (supportable_widening_operation): Likewise.
5438 (supportable_narrowing_operation): Likewise.
5439 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5441 * varasm.c (output_constant): Likewise.
5443 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5444 Alan Hayward <alan.hayward@arm.com>
5445 David Sherwood <david.sherwood@arm.com>
5447 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5448 so that both the length == 3 and length != 3 cases set up their
5449 own permute vectors. Add comments explaining why we know the
5450 number of elements is constant.
5451 (vect_permute_load_chain): Likewise.
5453 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5454 Alan Hayward <alan.hayward@arm.com>
5455 David Sherwood <david.sherwood@arm.com>
5457 * machmode.h (mode_nunits): Change from unsigned char to
5459 (ONLY_FIXED_SIZE_MODES): New macro.
5460 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5461 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5462 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5464 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5465 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5466 or if measurement_type is not polynomial.
5467 * genmodes.c (ZERO_COEFFS): New macro.
5468 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5470 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5471 Use ZERO_COEFFS when emitting initializers.
5472 * data-streamer.h (bp_pack_poly_value): New function.
5473 (bp_unpack_poly_value): Likewise.
5474 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5475 for GET_MODE_NUNITS.
5476 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5477 for GET_MODE_NUNITS.
5478 * tree.c (make_vector_type): Remove temporary shim and make
5479 the real function take the number of units as a poly_uint64
5481 (build_vector_type_for_mode): Handle polynomial nunits.
5482 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5483 * emit-rtl.c (const_vec_series_p_1): Likewise.
5484 (gen_rtx_CONST_VECTOR): Likewise.
5485 * fold-const.c (test_vec_duplicate_folding): Likewise.
5486 * genrecog.c (validate_pattern): Likewise.
5487 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5488 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5489 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5490 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5491 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5492 * rtlanal.c (subreg_get_info): Likewise.
5493 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5494 (vect_grouped_load_supported): Likewise.
5495 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5496 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5497 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5498 (simplify_const_unary_operation, simplify_binary_operation_1)
5499 (simplify_const_binary_operation, simplify_ternary_operation)
5500 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5501 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5502 instead of CONST_VECTOR_NUNITS.
5503 * varasm.c (output_constant_pool_2): Likewise.
5504 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5505 explicit-encoded elements in the XVEC for variable-length vectors.
5507 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5509 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5512 Alan Hayward <alan.hayward@arm.com>
5513 David Sherwood <david.sherwood@arm.com>
5515 * coretypes.h (fixed_size_mode): Declare.
5516 (fixed_size_mode_pod): New typedef.
5517 * builtins.h (target_builtins::x_apply_args_mode)
5518 (target_builtins::x_apply_result_mode): Change type to
5519 fixed_size_mode_pod.
5520 * builtins.c (apply_args_size, apply_result_size, result_vector)
5521 (expand_builtin_apply_args_1, expand_builtin_apply)
5522 (expand_builtin_return): Update accordingly.
5524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5526 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5527 * cselib.c (cselib_hash_rtx): Likewise.
5528 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5529 CONST_VECTOR encoding.
5531 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5532 Jeff Law <law@redhat.com>
5535 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5536 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5537 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5538 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5541 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5542 explicitly probe *sp in a noreturn function if there were any callee
5543 register saves or frame pointer is needed.
5545 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5548 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5549 BLKmode for ternary, binary or unary expressions.
5552 * var-tracking.c (delete_vta_debug_insn): New inline function.
5553 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5554 insns from get_insns () to NULL instead of each bb separately.
5555 Use delete_vta_debug_insn. No longer static.
5556 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5557 delete_vta_debug_insns callers.
5558 * rtl.h (delete_vta_debug_insns): Declare.
5559 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5560 instead of variable_tracking_main.
5562 2018-01-03 Martin Sebor <msebor@redhat.com>
5564 PR tree-optimization/83603
5565 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5566 arguments past the endof the argument list in functions declared
5567 without a prototype.
5568 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5569 Avoid checking when arguments are null.
5571 2018-01-03 Martin Sebor <msebor@redhat.com>
5574 * doc/extend.texi (attribute const): Fix a typo.
5575 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5576 issuing -Wsuggest-attribute for void functions.
5578 2018-01-03 Martin Sebor <msebor@redhat.com>
5580 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5581 offset_int::from instead of wide_int::to_shwi.
5582 (maybe_diag_overlap): Remove assertion.
5583 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5584 * gimple-ssa-sprintf.c (format_directive): Same.
5585 (parse_directive): Same.
5586 (sprintf_dom_walker::compute_format_length): Same.
5587 (try_substitute_return_value): Same.
5589 2018-01-03 Jeff Law <law@redhat.com>
5592 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5593 non-constant residual for zero at runtime and avoid probing in
5594 that case. Reorganize code for trailing problem to mirror handling
5597 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5599 PR tree-optimization/83501
5600 * tree-ssa-strlen.c (get_string_cst): New.
5601 (handle_char_store): Call get_string_cst.
5603 2018-01-03 Martin Liska <mliska@suse.cz>
5605 PR tree-optimization/83593
5606 * tree-ssa-strlen.c: Include tree-cfg.h.
5607 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5608 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5609 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5611 (strlen_dom_walker::before_dom_children): Call
5612 gimple_purge_dead_eh_edges. Dump tranformation with details
5614 (strlen_dom_walker::before_dom_children): Update call by adding
5615 new argument cleanup_eh.
5616 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5618 2018-01-03 Martin Liska <mliska@suse.cz>
5621 * cif-code.def (VARIADIC_THUNK): New enum value.
5622 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5625 2018-01-03 Jan Beulich <jbeulich@suse.com>
5627 * sse.md (mov<mode>_internal): Tighten condition for when to use
5628 vmovdqu<ssescalarsize> for TI and OI modes.
5630 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5632 Update copyright years.
5634 2018-01-03 Martin Liska <mliska@suse.cz>
5637 * ipa-visibility.c (function_and_variable_visibility): Skip
5638 functions with noipa attribure.
5640 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5642 * gcc.c (process_command): Update copyright notice dates.
5643 * gcov-dump.c (print_version): Ditto.
5644 * gcov.c (print_version): Ditto.
5645 * gcov-tool.c (print_version): Ditto.
5646 * gengtype.c (create_file): Ditto.
5647 * doc/cpp.texi: Bump @copying's copyright year.
5648 * doc/cppinternals.texi: Ditto.
5649 * doc/gcc.texi: Ditto.
5650 * doc/gccint.texi: Ditto.
5651 * doc/gcov.texi: Ditto.
5652 * doc/install.texi: Ditto.
5653 * doc/invoke.texi: Ditto.
5655 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5657 * vector-builder.h (vector_builder::m_full_nelts): Change from
5658 unsigned int to poly_uint64.
5659 (vector_builder::full_nelts): Update prototype accordingly.
5660 (vector_builder::new_vector): Likewise.
5661 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5662 (vector_builder::operator ==): Likewise.
5663 (vector_builder::finalize): Likewise.
5664 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5665 Take the number of elements as a poly_uint64 rather than an
5667 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5668 from unsigned int to poly_uint64.
5669 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5670 (vec_perm_indices::new_vector): Likewise.
5671 (vec_perm_indices::length): Likewise.
5672 (vec_perm_indices::nelts_per_input): Likewise.
5673 (vec_perm_indices::input_nelts): Likewise.
5674 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5675 number of elements per input as a poly_uint64 rather than an
5676 unsigned int. Use the original encoding for variable-length
5677 vectors, rather than clamping each individual element.
5678 For the second and subsequent elements in each pattern,
5679 clamp the step and base before clamping their sum.
5680 (vec_perm_indices::series_p): Handle polynomial element counts.
5681 (vec_perm_indices::all_in_range_p): Likewise.
5682 (vec_perm_indices_to_tree): Likewise.
5683 (vec_perm_indices_to_rtx): Likewise.
5684 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5685 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5686 (tree_vector_builder::new_binary_operation): Handle polynomial
5687 element counts. Return false if we need to know the number
5688 of elements at compile time.
5689 * fold-const.c (fold_vec_perm): Punt if the number of elements
5690 isn't known at compile time.
5692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5694 * vec-perm-indices.h (vec_perm_builder): Change element type
5695 from HOST_WIDE_INT to poly_int64.
5696 (vec_perm_indices::element_type): Update accordingly.
5697 (vec_perm_indices::clamp): Handle polynomial element_types.
5698 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5699 (vec_perm_indices::all_in_range_p): Likewise.
5700 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5702 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5703 polynomial vec_perm_indices element types.
5704 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5705 * fold-const.c (fold_vec_perm): Likewise.
5706 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5707 * tree-vect-generic.c (lower_vec_perm): Likewise.
5708 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5709 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5710 element type to HOST_WIDE_INT.
5712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5713 Alan Hayward <alan.hayward@arm.com>
5714 David Sherwood <david.sherwood@arm.com>
5716 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5717 rather than an int. Use plus_constant.
5718 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5719 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5721 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5722 Alan Hayward <alan.hayward@arm.com>
5723 David Sherwood <david.sherwood@arm.com>
5725 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5726 a HOST_WIDE_INT to a poly_int64.
5728 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5729 Alan Hayward <alan.hayward@arm.com>
5730 David Sherwood <david.sherwood@arm.com>
5732 * calls.c (load_register_parameters): Cope with polynomial
5733 mode sizes. Require a constant size for BLKmode parameters
5734 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5735 forces a parameter to be padded at the lsb end in order to
5736 fill a complete number of words, require the parameter size
5737 to be ordered wrt UNITS_PER_WORD.
5739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5740 Alan Hayward <alan.hayward@arm.com>
5741 David Sherwood <david.sherwood@arm.com>
5743 * reload1.c (spill_stack_slot_width): Change element type
5744 from unsigned int to poly_uint64_pod.
5745 (alter_reg): Treat mode sizes as polynomial.
5747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5748 Alan Hayward <alan.hayward@arm.com>
5749 David Sherwood <david.sherwood@arm.com>
5751 * reload.c (complex_word_subreg_p): New function.
5752 (reload_inner_reg_of_subreg, push_reload): Use it.
5754 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5755 Alan Hayward <alan.hayward@arm.com>
5756 David Sherwood <david.sherwood@arm.com>
5758 * lra-constraints.c (process_alt_operands): Reject matched
5759 operands whose sizes aren't ordered.
5760 (match_reload): Refer to this check here.
5762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5763 Alan Hayward <alan.hayward@arm.com>
5764 David Sherwood <david.sherwood@arm.com>
5766 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5767 that the mode size is in the set {1, 2, 4, 8, 16}.
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 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5774 Use plus_constant instead of gen_rtx_PLUS.
5776 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5777 Alan Hayward <alan.hayward@arm.com>
5778 David Sherwood <david.sherwood@arm.com>
5780 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5781 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5782 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5783 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5784 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5785 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5786 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5787 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5788 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5789 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5791 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5792 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5793 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5794 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5795 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5796 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5797 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5798 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5799 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5800 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5802 * expr.c (emit_move_resolve_push): Treat the input and result
5803 of PUSH_ROUNDING as a poly_int64.
5804 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5805 (emit_push_insn): Likewise.
5806 * lra-eliminations.c (mark_not_eliminable): Likewise.
5807 * recog.c (push_operand): Likewise.
5808 * reload1.c (elimination_effects): Likewise.
5809 * rtlanal.c (nonzero_bits1): Likewise.
5810 * calls.c (store_one_arg): Likewise. Require the padding to be
5811 known at compile time.
5813 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5814 Alan Hayward <alan.hayward@arm.com>
5815 David Sherwood <david.sherwood@arm.com>
5817 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5818 Use plus_constant instead of gen_rtx_PLUS.
5820 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5821 Alan Hayward <alan.hayward@arm.com>
5822 David Sherwood <david.sherwood@arm.com>
5824 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5827 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5828 Alan Hayward <alan.hayward@arm.com>
5829 David Sherwood <david.sherwood@arm.com>
5831 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5832 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5833 via stack temporaries. Treat the mode size as polynomial too.
5835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5836 Alan Hayward <alan.hayward@arm.com>
5837 David Sherwood <david.sherwood@arm.com>
5839 * expr.c (expand_expr_real_2): When handling conversions involving
5840 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5841 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5842 as a poly_uint64 too.
5844 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5845 Alan Hayward <alan.hayward@arm.com>
5846 David Sherwood <david.sherwood@arm.com>
5848 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
5855 REGMODE_NATURAL_SIZE.
5856 * expmed.c (store_bit_field_1): Likewise.
5857 * expr.c (store_constructor): Likewise.
5858 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5859 and polynomial REGMODE_NATURAL_SIZE.
5860 (gen_lowpart_common): Likewise.
5861 * reginfo.c (record_subregs_of_mode): Likewise.
5862 * rtlanal.c (read_modify_subreg_p): Likewise.
5864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5865 Alan Hayward <alan.hayward@arm.com>
5866 David Sherwood <david.sherwood@arm.com>
5868 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5869 numbers of elements.
5871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5872 Alan Hayward <alan.hayward@arm.com>
5873 David Sherwood <david.sherwood@arm.com>
5875 * match.pd: Cope with polynomial numbers of vector elements.
5877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5878 Alan Hayward <alan.hayward@arm.com>
5879 David Sherwood <david.sherwood@arm.com>
5881 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5882 in a POINTER_PLUS_EXPR.
5884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5885 Alan Hayward <alan.hayward@arm.com>
5886 David Sherwood <david.sherwood@arm.com>
5888 * omp-simd-clone.c (simd_clone_subparts): New function.
5889 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5890 (ipa_simd_modify_function_body): Likewise.
5892 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5893 Alan Hayward <alan.hayward@arm.com>
5894 David Sherwood <david.sherwood@arm.com>
5896 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5897 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5898 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5899 (expand_vector_condition, vector_element): Likewise.
5900 (subparts_gt): New function.
5901 (get_compute_type): Use subparts_gt.
5902 (count_type_subparts): Delete.
5903 (expand_vector_operations_1): Use subparts_gt instead of
5904 count_type_subparts.
5906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5907 Alan Hayward <alan.hayward@arm.com>
5908 David Sherwood <david.sherwood@arm.com>
5910 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5911 (vect_compile_time_alias): ...this new function. Do the calculation
5912 on poly_ints rather than trees.
5913 (vect_prune_runtime_alias_test_list): Update call accordingly.
5915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5916 Alan Hayward <alan.hayward@arm.com>
5917 David Sherwood <david.sherwood@arm.com>
5919 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5921 (vect_schedule_slp_instance): Likewise.
5923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5924 Alan Hayward <alan.hayward@arm.com>
5925 David Sherwood <david.sherwood@arm.com>
5927 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5928 constant and extern definitions for variable-length vectors.
5929 (vect_get_constant_vectors): Note that the number of units
5930 is known to be constant.
5932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5933 Alan Hayward <alan.hayward@arm.com>
5934 David Sherwood <david.sherwood@arm.com>
5936 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5937 of units as polynomial. Choose between WIDE and NARROW based
5940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5941 Alan Hayward <alan.hayward@arm.com>
5942 David Sherwood <david.sherwood@arm.com>
5944 * tree-vect-stmts.c (simd_clone_subparts): New function.
5945 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5947 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5948 Alan Hayward <alan.hayward@arm.com>
5949 David Sherwood <david.sherwood@arm.com>
5951 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5952 vectors as polynomial. Use build_index_vector for
5955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5956 Alan Hayward <alan.hayward@arm.com>
5957 David Sherwood <david.sherwood@arm.com>
5959 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5960 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5961 for variable-length vectors.
5962 (vectorizable_mask_load_store): Treat the number of units as
5963 polynomial, asserting that it is constant if the condition has
5964 already been enforced.
5965 (vectorizable_store, vectorizable_load): Likewise.
5967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5968 Alan Hayward <alan.hayward@arm.com>
5969 David Sherwood <david.sherwood@arm.com>
5971 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5972 of units as polynomial. Punt if we can't tell at compile time
5973 which vector contains the final result.
5975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5976 Alan Hayward <alan.hayward@arm.com>
5977 David Sherwood <david.sherwood@arm.com>
5979 * tree-vect-loop.c (vectorizable_induction): Treat the number
5980 of units as polynomial. Punt on SLP inductions. Use an integer
5981 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5982 cast of such a series for variable-length floating-point
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.h (build_index_vector): Declare.
5990 * tree.c (build_index_vector): New function.
5991 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5992 of units as polynomial, forcibly converting it to a constant if
5993 vectorizable_reduction has already enforced the condition.
5994 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5995 to create a {1,2,3,...} vector.
5996 (vectorizable_reduction): Treat the number of units as polynomial.
5997 Choose vectype_in based on the largest scalar element size rather
5998 than the smallest number of units. Enforce the restrictions
6001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6002 Alan Hayward <alan.hayward@arm.com>
6003 David Sherwood <david.sherwood@arm.com>
6005 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6006 number of units as polynomial.
6008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6009 Alan Hayward <alan.hayward@arm.com>
6010 David Sherwood <david.sherwood@arm.com>
6012 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6013 * target.def (autovectorize_vector_sizes): Return the vector sizes
6014 by pointer, using vector_sizes rather than a bitmask.
6015 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6016 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6017 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6019 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6020 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6021 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6022 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6023 * omp-general.c (omp_max_vf): Likewise.
6024 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6025 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6026 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6027 * tree-vect-slp.c (vect_slp_bb): Likewise.
6028 * doc/tm.texi: Regenerate.
6029 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6031 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6032 the vector size as a poly_uint64 rather than an unsigned int.
6033 (current_vector_size): Change from an unsigned int to a poly_uint64.
6034 (get_vectype_for_scalar_type): Update accordingly.
6035 * tree.h (build_truth_vector_type): Take the size and number of
6036 units as a poly_uint64 rather than an unsigned int.
6037 (build_vector_type): Add a temporary overload that takes
6038 the number of units as a poly_uint64 rather than an unsigned int.
6039 * tree.c (make_vector_type): Likewise.
6040 (build_truth_vector_type): Take the number of units as a poly_uint64
6041 rather than an unsigned int.
6043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6044 Alan Hayward <alan.hayward@arm.com>
6045 David Sherwood <david.sherwood@arm.com>
6047 * target.def (get_mask_mode): Take the number of units and length
6048 as poly_uint64s rather than unsigned ints.
6049 * targhooks.h (default_get_mask_mode): Update accordingly.
6050 * targhooks.c (default_get_mask_mode): Likewise.
6051 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6052 * doc/tm.texi: Regenerate.
6054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6055 Alan Hayward <alan.hayward@arm.com>
6056 David Sherwood <david.sherwood@arm.com>
6058 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6059 * omp-general.c (omp_max_vf): Likewise.
6060 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6061 (expand_omp_simd): Handle polynomial safelen.
6062 * omp-low.c (omplow_simd_context): Add a default constructor.
6063 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6064 (lower_rec_simd_input_clauses): Update accordingly.
6065 (lower_rec_input_clauses): Likewise.
6067 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6068 Alan Hayward <alan.hayward@arm.com>
6069 David Sherwood <david.sherwood@arm.com>
6071 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6072 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6073 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6074 (vect_analyze_slp_cost): Likewise.
6075 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6076 (vect_model_load_cost): Likewise.
6078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6079 Alan Hayward <alan.hayward@arm.com>
6080 David Sherwood <david.sherwood@arm.com>
6082 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6083 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6084 from an unsigned int * to a poly_uint64_pod *.
6085 (calculate_unrolling_factor): New function.
6086 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6089 Alan Hayward <alan.hayward@arm.com>
6090 David Sherwood <david.sherwood@arm.com>
6092 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6093 from an unsigned int to a poly_uint64.
6094 (_loop_vec_info::slp_unrolling_factor): Likewise.
6095 (_loop_vec_info::vectorization_factor): Change from an int
6097 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6098 (vect_get_num_vectors): New function.
6099 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6100 (vect_get_num_copies): Use vect_get_num_vectors.
6101 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6102 to an unsigned int *.
6103 (vect_analyze_data_refs): Change min_vf from an int * to a
6105 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6106 than an unsigned HOST_WIDE_INT.
6107 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6108 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6109 to an unsigned int *.
6110 (vect_analyze_data_ref_dependences): Likewise.
6111 (vect_compute_data_ref_alignment): Handle polynomial vf.
6112 (vect_enhance_data_refs_alignment): Likewise.
6113 (vect_prune_runtime_alias_test_list): Likewise.
6114 (vect_shift_permute_load_chain): Likewise.
6115 (vect_supportable_dr_alignment): Likewise.
6116 (dependence_distance_ge_vf): Take the vectorization factor as a
6117 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6118 (vect_analyze_data_refs): Change min_vf from an int * to a
6120 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6121 vfm1 as a poly_uint64 rather than an int. Make the same change
6122 for the returned bound_scalar.
6123 (vect_gen_vector_loop_niters): Handle polynomial vf.
6124 (vect_do_peeling): Likewise. Update call to
6125 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6126 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6128 * tree-vect-loop.c (vect_determine_vectorization_factor)
6129 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6130 (vect_get_known_peeling_cost): Likewise.
6131 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6132 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6133 (vect_transform_loop): Likewise. Use the lowest possible VF when
6134 updating the upper bounds of the loop.
6135 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6137 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6138 polynomial unroll factors.
6139 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6140 (vect_make_slp_decision): Likewise.
6141 (vect_supported_load_permutation_p): Likewise, and polynomial
6143 (vect_analyze_slp_cost): Handle polynomial vf.
6144 (vect_slp_analyze_node_operations): Likewise.
6145 (vect_slp_analyze_bb_1): Likewise.
6146 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6147 than an unsigned HOST_WIDE_INT.
6148 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6149 (vectorizable_load): Handle polynomial vf.
6150 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6152 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6155 Alan Hayward <alan.hayward@arm.com>
6156 David Sherwood <david.sherwood@arm.com>
6158 * match.pd: Handle bit operations involving three constants
6159 and try to fold one pair.
6161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6163 * tree-vect-loop-manip.c: Include gimple-fold.h.
6164 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6165 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6166 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6167 Add a path that uses a step of VF instead of 1, but disable it
6169 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6170 and niters_no_overflow parameters. Update calls to
6171 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6172 Create a new SSA name if the latter choses to use a ste other
6173 than zero, and return it via niters_vector_mult_vf_var.
6174 * tree-vect-loop.c (vect_transform_loop): Update calls to
6175 vect_do_peeling, vect_gen_vector_loop_niters and
6176 slpeel_make_loop_iterate_ntimes.
6177 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6178 (vect_gen_vector_loop_niters): Update declarations after above changes.
6180 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6182 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6183 128-bit round to integer instructions.
6184 (ceil<mode>2): Likewise.
6185 (btrunc<mode>2): Likewise.
6186 (round<mode>2): Likewise.
6188 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6190 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6191 unaligned VSX load/store on P8/P9.
6192 (expand_block_clear): Allow the use of unaligned VSX
6193 load/store on P8/P9.
6195 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6197 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6199 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6200 swap associated with both a load and a store.
6202 2018-01-02 Andrew Waterman <andrew@sifive.com>
6204 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6205 * config/riscv/riscv.md (clear_cache): Use it.
6207 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6209 * web.c: Remove out-of-date comment.
6211 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6213 * expr.c (fixup_args_size_notes): Check that any existing
6214 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6215 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6216 (emit_single_push_insn): ...here.
6218 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6220 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6221 (const_vector_encoded_nelts): New function.
6222 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6223 (const_vector_int_elt, const_vector_elt): Declare.
6224 * emit-rtl.c (const_vector_int_elt_1): New function.
6225 (const_vector_elt): Likewise.
6226 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6227 of CONST_VECTOR_ELT.
6229 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6231 * expr.c: Include rtx-vector-builder.h.
6232 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6233 directly on the tree encoding.
6234 (const_vector_from_tree): Likewise.
6235 * optabs.c: Include rtx-vector-builder.h.
6236 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6237 sequence of "u" values.
6238 * vec-perm-indices.c: Include rtx-vector-builder.h.
6239 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6240 directly on the vec_perm_indices encoding.
6242 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6244 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6245 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6246 * rtx-vector-builder.h: New file.
6247 * rtx-vector-builder.c: Likewise.
6248 * rtl.h (rtx_def::u2): Add a const_vector field.
6249 (CONST_VECTOR_NPATTERNS): New macro.
6250 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6251 (CONST_VECTOR_DUPLICATE_P): Likewise.
6252 (CONST_VECTOR_STEPPED_P): Likewise.
6253 (CONST_VECTOR_ENCODED_ELT): Likewise.
6254 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6255 (unwrap_const_vec_duplicate): Likewise.
6256 (const_vec_series_p): Check for a non-duplicated vector encoding.
6257 Say that the function only returns true for integer vectors.
6258 * emit-rtl.c: Include rtx-vector-builder.h.
6259 (gen_const_vec_duplicate_1): Delete.
6260 (gen_const_vector): Call gen_const_vec_duplicate instead of
6261 gen_const_vec_duplicate_1.
6262 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6263 (gen_const_vec_duplicate): Use rtx_vector_builder.
6264 (gen_const_vec_series): Likewise.
6265 (gen_rtx_CONST_VECTOR): Likewise.
6266 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6267 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6268 Build a new vector rather than modifying a CONST_VECTOR in-place.
6269 (handle_special_swappables): Update call accordingly.
6270 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6271 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6272 Build a new vector rather than modifying a CONST_VECTOR in-place.
6273 (handle_special_swappables): Update call accordingly.
6275 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6277 * simplify-rtx.c (simplify_const_binary_operation): Use
6278 CONST_VECTOR_ELT instead of XVECEXP.
6280 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6282 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6283 the selector elements to be different from the data elements
6284 if the selector is a VECTOR_CST.
6285 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6286 ssizetype for the selector.
6288 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6290 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6291 before testing each element individually.
6292 * tree-vect-generic.c (lower_vec_perm): Likewise.
6294 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6296 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6297 * selftest-run-tests.c (selftest::run_tests): Call it.
6298 * vector-builder.h (vector_builder::operator ==): New function.
6299 (vector_builder::operator !=): Likewise.
6300 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6301 (vec_perm_indices::all_from_input_p): New function.
6302 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6303 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6304 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6305 instead of reading the VECTOR_CST directly. Detect whether both
6306 vector inputs are the same before constructing the vec_perm_indices,
6307 and update the number of inputs argument accordingly. Use the
6308 utility functions added above. Only construct sel2 if we need to.
6310 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6312 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6313 the broadcast of the low byte.
6314 (expand_mult_highpart): Use an explicit encoding for the permutes.
6315 * optabs-query.c (can_mult_highpart_p): Likewise.
6316 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6317 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6318 (vectorizable_bswap): Likewise.
6319 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6320 explicit encoding for the power-of-2 permutes.
6321 (vect_permute_store_chain): Likewise.
6322 (vect_grouped_load_supported): Likewise.
6323 (vect_permute_load_chain): Likewise.
6325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6327 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6328 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6329 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6330 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6331 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6332 (vect_gen_perm_mask_any): Likewise.
6334 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6336 * int-vector-builder.h: New file.
6337 * vec-perm-indices.h: Include int-vector-builder.h.
6338 (vec_perm_indices): Redefine as an int_vector_builder.
6339 (auto_vec_perm_indices): Delete.
6340 (vec_perm_builder): Redefine as a stand-alone class.
6341 (vec_perm_indices::vec_perm_indices): New function.
6342 (vec_perm_indices::clamp): Likewise.
6343 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6344 (vec_perm_indices::new_vector): New function.
6345 (vec_perm_indices::new_expanded_vector): Update for new
6346 vec_perm_indices class.
6347 (vec_perm_indices::rotate_inputs): New function.
6348 (vec_perm_indices::all_in_range_p): Operate directly on the
6349 encoded form, without computing elided elements.
6350 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6351 encoding. Update for new vec_perm_indices class.
6352 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6353 the given vec_perm_builder.
6354 (expand_vec_perm_var): Update vec_perm_builder constructor.
6355 (expand_mult_highpart): Use vec_perm_builder instead of
6356 auto_vec_perm_indices.
6357 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6358 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6359 or double series encoding as appropriate.
6360 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6361 vec_perm_indices instead of auto_vec_perm_indices.
6362 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6363 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6364 (vect_permute_store_chain): Likewise.
6365 (vect_grouped_load_supported): Likewise.
6366 (vect_permute_load_chain): Likewise.
6367 (vect_shift_permute_load_chain): Likewise.
6368 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6369 (vect_transform_slp_perm_load): Likewise.
6370 (vect_schedule_slp_instance): Likewise.
6371 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6372 (vectorizable_mask_load_store): Likewise.
6373 (vectorizable_bswap): Likewise.
6374 (vectorizable_store): Likewise.
6375 (vectorizable_load): Likewise.
6376 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6377 vec_perm_indices instead of auto_vec_perm_indices. Use
6378 tree_to_vec_perm_builder to read the vector from a tree.
6379 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6380 vec_perm_builder instead of a vec_perm_indices.
6381 (have_whole_vector_shift): Use vec_perm_builder and
6382 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6383 truncation to calc_vec_perm_mask_for_shift.
6384 (vect_create_epilog_for_reduction): Likewise.
6385 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6386 from auto_vec_perm_indices to vec_perm_indices.
6387 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6388 instead of changing individual elements.
6389 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6390 the vector in d.perm.
6391 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6392 from auto_vec_perm_indices to vec_perm_indices.
6393 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6394 instead of changing individual elements.
6395 (arm_vectorize_vec_perm_const): Use new_vector to install
6396 the vector in d.perm.
6397 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6398 Update vec_perm_builder constructor.
6399 (rs6000_expand_interleave): Likewise.
6400 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6401 (rs6000_expand_interleave): Likewise.
6403 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6405 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6406 to qimode could truncate the indices.
6407 * optabs.c (expand_vec_perm_var): Likewise.
6409 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6411 * Makefile.in (OBJS): Add vec-perm-indices.o.
6412 * vec-perm-indices.h: New file.
6413 * vec-perm-indices.c: Likewise.
6414 * target.h (vec_perm_indices): Replace with a forward class
6416 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6417 * optabs.h: Include vec-perm-indices.h.
6418 (expand_vec_perm): Delete.
6419 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6420 (expand_vec_perm_const): Declare.
6421 * target.def (vec_perm_const_ok): Replace with...
6422 (vec_perm_const): ...this new hook.
6423 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6424 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6425 * doc/tm.texi: Regenerate.
6426 * optabs.def (vec_perm_const): Delete.
6427 * doc/md.texi (vec_perm_const): Likewise.
6428 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6429 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6430 expand_vec_perm for constant permutation vectors. Assert that
6431 the mode of variable permutation vectors is the integer equivalent
6432 of the mode that is being permuted.
6433 * optabs-query.h (selector_fits_mode_p): Declare.
6434 * optabs-query.c: Include vec-perm-indices.h.
6435 (selector_fits_mode_p): New function.
6436 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6437 is defined, instead of checking whether the vec_perm_const_optab
6438 exists. Use targetm.vectorize.vec_perm_const instead of
6439 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6440 fit in the vector mode before using a variable permute.
6441 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6442 vec_perm_indices instead of an rtx.
6443 (expand_vec_perm): Replace with...
6444 (expand_vec_perm_const): ...this new function. Take the selector
6445 as a vec_perm_indices rather than an rtx. Also take the mode of
6446 the selector. Update call to shift_amt_for_vec_perm_mask.
6447 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6448 Use vec_perm_indices::new_expanded_vector to expand the original
6449 selector into bytes. Check whether the indices fit in the vector
6450 mode before using a variable permute.
6451 (expand_vec_perm_var): Make global.
6452 (expand_mult_highpart): Use expand_vec_perm_const.
6453 * fold-const.c: Includes vec-perm-indices.h.
6454 * tree-ssa-forwprop.c: Likewise.
6455 * tree-vect-data-refs.c: Likewise.
6456 * tree-vect-generic.c: Likewise.
6457 * tree-vect-loop.c: Likewise.
6458 * tree-vect-slp.c: Likewise.
6459 * tree-vect-stmts.c: Likewise.
6460 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6462 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6463 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6464 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6465 (aarch64_vectorize_vec_perm_const): ...this new function.
6466 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6467 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6468 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6469 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6470 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6471 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6472 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6474 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6475 check for NEON modes.
6476 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6477 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6478 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6479 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6481 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6482 the old VEC_PERM_CONST conditions.
6483 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6484 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6485 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6486 (ia64_vectorize_vec_perm_const_ok): Merge into...
6487 (ia64_vectorize_vec_perm_const): ...this new function.
6488 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6489 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6490 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6491 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6492 * config/mips/mips.c (mips_expand_vec_perm_const)
6493 (mips_vectorize_vec_perm_const_ok): Merge into...
6494 (mips_vectorize_vec_perm_const): ...this new function.
6495 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6496 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6497 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6498 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6499 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6500 (rs6000_expand_vec_perm_const): Delete.
6501 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6503 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6504 (altivec_expand_vec_perm_const_le): Take each operand individually.
6505 Operate on constant selectors rather than rtxes.
6506 (altivec_expand_vec_perm_const): Likewise. Update call to
6507 altivec_expand_vec_perm_const_le.
6508 (rs6000_expand_vec_perm_const): Delete.
6509 (rs6000_vectorize_vec_perm_const_ok): Delete.
6510 (rs6000_vectorize_vec_perm_const): New function.
6511 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6512 an element count and rtx array.
6513 (rs6000_expand_extract_even): Update call accordingly.
6514 (rs6000_expand_interleave): Likewise.
6515 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6516 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6517 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6518 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6519 (rs6000_expand_vec_perm_const): Delete.
6520 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6521 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6522 (altivec_expand_vec_perm_const_le): Take each operand individually.
6523 Operate on constant selectors rather than rtxes.
6524 (altivec_expand_vec_perm_const): Likewise. Update call to
6525 altivec_expand_vec_perm_const_le.
6526 (rs6000_expand_vec_perm_const): Delete.
6527 (rs6000_vectorize_vec_perm_const_ok): Delete.
6528 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6529 reference to the SPE evmerge intructions.
6530 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6531 an element count and rtx array.
6532 (rs6000_expand_extract_even): Update call accordingly.
6533 (rs6000_expand_interleave): Likewise.
6534 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6535 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6537 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6539 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6541 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6542 vector mode and that that mode matches the mode of the data
6544 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6545 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6546 directly using expand_vec_perm_1 when forcing selectors into
6548 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6550 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6552 * optabs-query.h (can_vec_perm_p): Delete.
6553 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6554 * optabs-query.c (can_vec_perm_p): Split into...
6555 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6556 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6557 particular selector is valid.
6558 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6559 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6560 (vect_grouped_load_supported): Likewise.
6561 (vect_shift_permute_load_chain): Likewise.
6562 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6563 (vect_transform_slp_perm_load): Likewise.
6564 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6565 (vectorizable_bswap): Likewise.
6566 (vect_gen_perm_mask_checked): Likewise.
6567 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6568 implementations of variable permutation vectors into account
6569 when deciding which selector to use.
6570 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6571 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6572 with a false third argument.
6573 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6574 to test whether the constant selector is valid and can_vec_perm_var_p
6575 to test whether a variable selector is valid.
6577 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6579 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6580 * optabs-query.c (can_vec_perm_p): Likewise.
6581 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6582 instead of vec_perm_indices.
6583 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6584 (vect_gen_perm_mask_checked): Likewise,
6585 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6586 (vect_gen_perm_mask_checked): Likewise,
6588 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6590 * optabs-query.h (qimode_for_vec_perm): Declare.
6591 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6592 (qimode_for_vec_perm): ...this new function.
6593 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6595 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6597 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6598 does not have a conditional at the top.
6600 2018-01-02 Richard Biener <rguenther@suse.de>
6602 * ipa-inline.c (big_speedup_p): Fix expression.
6604 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6607 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6610 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6614 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6615 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6616 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6617 cond_taken_branch_cost 3->4.
6619 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6621 PR tree-optimization/83581
6622 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6623 TODO_cleanup_cfg if any changes have been made.
6626 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6627 convert_modes if target mode has the right side, but different mode
6631 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6632 last argument when extracting from CONCAT. If either from_real or
6633 from_imag is NULL, use expansion through memory. If result is not
6634 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6635 the parts directly to inner mode, if even that fails, use expansion
6639 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6640 check for bswap in mode rather than HImode and use that in expand_unop
6643 Copyright (C) 2018 Free Software Foundation, Inc.
6645 Copying and distribution of this file, with or without modification,
6646 are permitted in any medium without royalty provided the copyright
6647 notice and this notice are preserved.