1 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
3 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
5 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
6 (BLOCK_INLINE_ENTRY_LABEL): New.
7 (dwarf2out_var_location): Disregard inline entry markers.
8 (inline_entry_data): New struct.
9 (inline_entry_data_hasher): New hashtable type.
10 (inline_entry_data_hasher::hash): New.
11 (inline_entry_data_hasher::equal): New.
12 (inline_entry_data_table): New variable.
13 (add_high_low_attributes): Add DW_AT_entry_pc and
14 DW_AT_GNU_entry_view attributes if a pending entry is found
15 in inline_entry_data_table. Add old entry_pc attribute only
16 if debug nonbinding markers are disabled.
17 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
19 (block_within_block_p, dwarf2out_inline_entry): New.
20 (dwarf2out_finish): Check that no entries remained in
21 inline_entry_data_table.
22 * final.c (reemit_insn_block_notes): Handle inline entry notes.
23 (final_scan_insn, notice_source_line): Likewise.
24 (rest_of_clean_state): Skip inline entry markers.
25 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
27 * gimple.c (gimple_build_debug_inline_entry): New.
28 * gimple.h (enum gimple_debug_subcode): Add
29 GIMPLE_DEBUG_INLINE_ENTRY.
30 (gimple_build_debug_inline_entry): Declare.
31 (gimple_debug_inline_entry_p): New.
32 (gimple_debug_nonbind_marker_p): Adjust.
33 * insn-notes.def (INLINE_ENTRY): New.
34 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
35 inline entry marker notes.
36 (print_insn): Likewise.
37 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
38 (INSN_DEBUG_MARKER_KIND): Likewise.
39 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
40 * tree-inline.c (expand_call_inline): Build and insert
41 debug_inline_entry stmt.
42 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
43 inline entry blocks early, if nonbind markers are enabled.
44 (dump_scope_block): Dump fragment info.
45 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
46 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
47 (gimple_build_debug_inline_entry): New.
48 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
49 Enable/disable inline entry points too.
50 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
51 (DEBUG_INSN): Describe inline entry markers.
53 * common.opt (gvariable-location-views): New.
54 (gvariable-location-views=incompat5): New.
57 * configure.ac: Test assembler for view support.
58 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
59 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
60 * dwarf2out.c (var_loc_view): New typedef.
61 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
62 (dwarf2out_locviews_in_attribute): New.
63 (dwarf2out_locviews_in_loclist): New.
64 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
65 (enum dw_line_info_opcode): Add LI_adv_address.
66 (struct dw_line_info_table): Add view.
67 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
68 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
69 (zero_view_p): New variable.
70 (ZERO_VIEW_P): New macro.
71 (output_asm_line_debug_info): New.
72 (struct var_loc_node): Add view.
73 (add_AT_view_list, AT_loc_list): New.
74 (add_var_loc_to_decl): Add view param. Test it against last.
75 (new_loc_list): Add view params. Record them.
76 (AT_loc_list_ptr): Handle loc and view lists.
77 (view_list_to_loc_list_val_node): New.
78 (print_dw_val): Handle dw_val_class_view_list.
79 (size_of_die): Likewise.
80 (value_format): Likewise.
81 (loc_list_has_views): New.
82 (gen_llsym): Set vl_symbol too.
83 (maybe_gen_llsym, skip_loc_list_entry): New.
84 (dwarf2out_maybe_output_loclist_view_pair): New.
85 (output_loc_list): Output view list or entries too.
86 (output_view_list_offset): New.
87 (output_die): Handle dw_val_class_view_list.
88 (output_dwarf_version): New.
89 (output_compilation_unit_header): Use it.
90 (output_skeleton_debug_sections): Likewise.
91 (output_rnglists, output_line_info): Likewise.
92 (output_pubnames, output_aranges): Update version comments.
93 (output_one_line_info_table): Output view numbers in asm comments.
94 (dw_loc_list): Determine current endview, pass it to new_loc_list.
96 (loc_list_from_tree_1): Adjust.
97 (add_AT_location_description): Create view list attribute if
98 needed, check it's absent otherwise.
99 (convert_cfa_to_fb_loc_list): Adjust.
100 (maybe_emit_file): Call output_asm_line_debug_info for test.
101 (dwarf2out_var_location): Reset views as needed. Precompute
102 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
104 (new_line_info_table): Reset next view.
105 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
106 (dwarf2out_source_line): Likewise. Output view resets and labels to
107 the assembler, or select appropriate line info opcodes.
108 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
109 (optimize_string_length): Catch it. Adjust.
110 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
111 dw_val_class_view_list, and remove it if no longer needed.
112 (hash_loc_list): Hash view numbers.
113 (loc_list_hasher::equal): Compare them.
114 (optimize_location_lists): Check whether a view list symbol is
115 needed, and whether the locview attribute is present, and
116 whether they match. Remove the locview attribute if no longer
118 (index_location_lists): Call skip_loc_list_entry for test.
119 (dwarf2out_finish): Call output_asm_line_debug_info for test.
120 Use output_dwarf_version.
121 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
122 (struct dw_val_node): Add val_view_list.
123 * final.c (SEEN_NEXT_VIEW): New.
124 (set_next_view_needed): New.
125 (clear_next_view_needed): New.
126 (maybe_output_next_view): New.
127 (final_start_function): Rename to...
128 (final_start_function_1): ... this. Take pointer to FIRST,
129 add SEEN parameter. Emit param bindings in the initial view.
130 (final_start_function): Reintroduce SEEN-less interface.
131 (final): Rename to...
132 (final_1): ... this. Take SEEN parameter. Output final pending
133 next view at the end.
134 (final): Reintroduce seen-less interface.
135 (final_scan_insn): Output pending next view before switching
136 sections or ending a block. Mark the next view as needed when
137 outputting variable locations. Notify debug backend of section
138 changes, and of location view changes.
139 (rest_of_handle_final): Adjust.
140 * toplev.c (process_options): Autodetect value for debug variable
141 location views option. Warn on incompat5 without -gdwarf-5.
142 * doc/invoke.texi (gvariable-location-views): New.
143 (gvariable-location-views=incompat5): New.
144 (gno-variable-location-views): New.
146 2018-02-08 David Malcolm <dmalcolm@redhat.com>
148 PR tree-optimization/84136
149 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
150 that the result of find_edge is non-NULL.
152 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
155 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
156 storing integer register in SImode. Fix cost of 256 and 512
157 byte aligned SSE register store.
159 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
161 * config/i386/i386.c (ix86_multiplication_cost): Fix
162 multiplication cost for TARGET_AVX512DQ.
164 2018-02-08 Marek Polacek <polacek@redhat.com>
166 PR tree-optimization/84238
167 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
170 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
172 PR tree-optimization/84265
173 * tree-vect-stmts.c (vectorizable_store): Don't treat
174 VMAT_CONTIGUOUS accesses as grouped.
175 (vectorizable_load): Likewise.
177 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
179 PR tree-optimization/81635
180 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
181 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
182 (test_round_for_mask): New functions.
183 (wide_int_cc_tests): Call test_round_for_mask.
184 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
185 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
186 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
187 range returned by get_range_info.
189 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
192 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
193 * symtab.c: Include builtins.h
194 (symtab_node::output_to_lto_symbol_table_p): Move here
195 from lto-streamer-out.c:output_symbol_p.
196 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
197 (output_symbol_p): Move all logic to symtab.c
198 (produce_symtab): Update.
200 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
202 * config/s390/s390-opts.h (enum indirect_branch): Define.
203 * config/s390/s390-protos.h (s390_return_addr_from_memory)
204 (s390_indirect_branch_via_thunk)
205 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
206 (enum s390_indirect_branch_type): Define.
207 * config/s390/s390.c (struct s390_frame_layout, struct
208 machine_function): Remove.
209 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
210 (indirect_branch_table_label_no, indirect_branch_table_name):
212 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
213 (enum s390_indirect_branch_option): Define.
214 (s390_return_addr_from_memory): New function.
215 (s390_handle_string_attribute): New function.
216 (s390_attribute_table): Add new attribute handler.
217 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
218 (s390_indirect_branch_via_thunk): New function.
219 (s390_indirect_branch_via_inline_thunk): New function.
220 (s390_function_ok_for_sibcall): When jumping via thunk disallow
221 sibling call optimization for non z10 compiles.
222 (s390_emit_call): Force indirect branch target to be a single
223 register. Add r1 clobber for non-z10 compiles.
224 (s390_emit_epilogue): Emit return jump via return_use expander.
225 (s390_reorg): Handle JUMP_INSNs as execute targets.
226 (s390_option_override_internal): Perform validity checks for the
227 new command line options.
228 (s390_indirect_branch_attrvalue): New function.
229 (s390_indirect_branch_settings): New function.
230 (s390_set_current_function): Invoke s390_indirect_branch_settings.
231 (s390_output_indirect_thunk_function): New function.
232 (s390_code_end): Implement target hook.
233 (s390_case_values_threshold): Implement target hook.
234 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
236 * config/s390/s390.h (struct s390_frame_layout)
237 (struct machine_function): Move here from s390.c.
238 (TARGET_INDIRECT_BRANCH_NOBP_RET)
239 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
240 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
241 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
242 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
243 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
244 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
245 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
246 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
247 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
248 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
249 (mnemonic attribute): Add values which aren't recognized
251 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
252 pattern for branch conversion. Fix mnemonic attribute.
253 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
254 indirect branch via thunk if requested.
255 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
256 ("*indirect_jump"): Disable for branch conversion using out of
258 ("indirect_jump_via_thunk<mode>_z10")
259 ("indirect_jump_via_thunk<mode>")
260 ("indirect_jump_via_inlinethunk<mode>_z10")
261 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
262 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
263 ("casesi_jump_via_inlinethunk<mode>_z10")
264 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
265 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
266 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
267 ("*indirect2_jump"): Disable for branch conversion.
268 ("casesi_jump"): Turn into expander and expand patterns for branch
270 ("return_use"): New expander.
271 ("*return"): Emit return via thunk and rename it to ...
272 ("*return<mode>"): ... this one.
273 * config/s390/s390.opt: Add new options and and enum for the
276 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
278 * lra-constraints.c (match_reload): Unconditionally use
279 gen_lowpart_SUBREG, rather than selecting between that
280 and equivalent gen_rtx_SUBREG code.
282 2018-02-08 Richard Biener <rguenther@suse.de>
284 PR tree-optimization/84233
285 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
286 changed flag instead of boguously re-using phi_inserted.
288 2018-02-08 Martin Jambor <mjambor@suse.cz>
290 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
291 static local variables.
293 2018-02-08 Richard Biener <rguenther@suse.de>
295 PR tree-optimization/84278
296 * tree-vect-stmts.c (vectorizable_store): When looking for
297 smaller vector types to perform grouped strided loads/stores
298 make sure the mode is supported by the target.
299 (vectorizable_load): Likewise.
301 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
303 * config/aarch64/aarch64.c (aarch64_components_for_bb):
304 Increase LDP/STP opportunities by adding adjacent callee-saves.
306 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
308 PR rtl-optimization/84068
309 PR rtl-optimization/83459
310 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
312 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
314 PR tree-optimization/84224
315 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
316 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
319 2018-02-07 Iain Sandoe <iain@codesourcery.com>
322 * config/rs6000/altivec.md (*restore_world): Remove LR use.
323 * config/rs6000/predicates.md (restore_world_operation): Adjust op
324 count, remove one USE.
326 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
328 * doc/install.texi (Configuration): Document the
329 --with-long-double-format={ibm,ieee} PowerPC configuration
333 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
334 Convert from define_expand to be define_insn_and_split. Rework
335 float/double/_Float128 conversions to QI/HI/SImode to work with
336 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
337 conversions to QI/HImode types did a store and then a load to
338 truncate the value. For conversions to VSX registers, don't split
339 the insn, instead emit the code directly. Use the code iterator
340 any_fix to combine signed and unsigned conversions.
341 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
342 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
343 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
344 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
345 (fix_<mode>di2_hw): Likewise.
346 (fixuns_<mode>di2_hw): Likewise.
347 (fix_<mode>si2_hw): Likewise.
348 (fixuns_<mode>si2_hw): Likewise.
349 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
350 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
351 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
352 fix<uns>_trunc<SFDF:mode>si2_p8.
353 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
355 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
356 (fix<uns>_<mode>_mem): Likewise.
357 (fctiw<u>z_<mode>_mem): Likewise.
358 (fix<uns>_<mode>_mem): Likewise.
359 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
360 the register allocator from doing a direct move to the GPRs to do
361 a store, and instead use the ISA 3.0 store byte/half-word from
362 vector register instruction. For IEEE 128-bit floating point,
363 also optimize stores of 32-bit ints.
364 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
366 2018-02-07 Alan Hayward <alan.hayward@arm.com>
368 * genextract.c (push_pathstr_operand): New function to support
370 (walk_rtx): Call push_pathstr_operand.
371 (print_path): Support [a-zA-Z].
373 2018-02-07 Richard Biener <rguenther@suse.de>
375 PR tree-optimization/84037
376 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
377 (cse_and_gimplify_to_preheader): Declare.
378 (vect_get_place_in_interleaving_chain): Likewise.
379 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
381 (_loop_vec_info::~_loop_vec_info): Delete it.
382 (cse_and_gimplify_to_preheader): New function.
383 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
384 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
385 (vectorizable_load): Likewise. For grouped stores always base
386 the IV on the first element.
387 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
388 condition before gimplifying.
390 2018-02-07 Jakub Jelinek <jakub@redhat.com>
392 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
393 *DIV_EXPR and *MOD_EXPR.
395 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
398 * config/i386/i386.c (ix86_option_override_internal): Mask out
399 the CF_SET bit when checking -fcf-protection.
401 2018-02-07 Tom de Vries <tom@codesourcery.com>
404 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
407 2018-02-07 Richard Biener <rguenther@suse.de>
409 PR tree-optimization/84204
410 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
413 PR tree-optimization/84205
414 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
415 special-case isl_ast_op_zdiv_r.
417 PR tree-optimization/84223
418 * graphite-scop-detection.c (gather_bbs::before_dom_children):
419 Only add conditions from within the region.
420 (gather_bbs::after_dom_children): Adjust.
422 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
425 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
426 * config/avr/avr.md: Only post-reload split REG-REG moves if
427 either register is REGERAL_REG_P.
429 2018-02-07 Jakub Jelinek <jakub@redhat.com>
431 PR tree-optimization/84235
432 * tree-ssa-scopedtables.c
433 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
434 if the subtraction is performed in floating point type where NaNs are
435 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
436 build 1. Formatting fix.
438 2018-02-06 Jakub Jelinek <jakub@redhat.com>
441 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
442 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
443 and skip it regardless of bb boundaries. Use CALL_P macro,
444 don't test INSN_P (insn) together with CALL_P or JUMP_P check
445 unnecessarily, formatting fix.
447 2018-02-06 Michael Collison <michael.collison@arm.com>
449 * config/arm/thumb2.md:
450 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
451 (*thumb_mov_notscc): Ditto.
453 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
456 * config/rs6000/rs6000.md (su code attribute): Use "u" for
457 unsigned_fix, not "s".
459 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
461 * configure.ac (gcc_fn_eh_frame_ro): New function.
462 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
463 correct .eh_frame permissions.
464 * configure: Regenerate.
466 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
468 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
471 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
473 * config/rs6000/rs6000.c (rs6000_option_override_internal):
474 Display warning message for -mno-speculate-indirect-jumps.
476 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
478 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
480 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
482 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
484 PR tree-optimization/84225
485 * tree-eh.c (find_trapping_overflow): Only call
486 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
488 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
491 * config/i386/i386.c: Reimplement the check of possible options
492 -mibt/-mshstk conbination. Change error messages.
493 * doc/invoke.texi: Fix a typo: remove extra '='.
495 2018-02-06 Marek Polacek <polacek@redhat.com>
497 PR tree-optimization/84228
498 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
500 2018-02-06 Tamar Christina <tamar.christina@arm.com>
503 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
504 emitted arch directives.
505 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
506 __ARM_FEATURE_COPROC before changing architectures.
508 2018-02-06 Richard Biener <rguenther@suse.de>
510 * config/i386/i386.c (print_reg): Fix typo.
511 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
513 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
515 * configure: Regenerate.
517 2018-02-05 Martin Sebor <msebor@redhat.com>
519 PR tree-optimization/83369
520 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
523 2018-02-05 Martin Liska <mliska@suse.cz>
525 * doc/invoke.texi: Cherry-pick upstream r323995.
527 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
529 * ira.c (ira_init_register_move_cost): Adjust comment.
531 2018-02-05 Martin Liska <mliska@suse.cz>
533 PR gcov-profile/84137
534 * doc/gcov.texi: Fix typo in documentation.
536 2018-02-05 Martin Liska <mliska@suse.cz>
538 PR gcov-profile/83879
539 * doc/gcov.texi: Document necessity of --dynamic-list-data when
540 using dlopen functionality.
542 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
544 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
545 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
546 _mm_maskz_range_ss, _mm_mask_range_round_ss,
547 _mm_maskz_range_round_ss): New intrinsics.
548 (__builtin_ia32_rangesd128_round)
549 (__builtin_ia32_rangess128_round): Remove.
550 (__builtin_ia32_rangesd128_mask_round,
551 __builtin_ia32_rangess128_mask_round): New builtins.
552 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
553 __builtin_ia32_rangess128_round): Remove.
554 (__builtin_ia32_rangesd128_mask_round,
555 __builtin_ia32_rangess128_mask_round): New builtins.
556 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
557 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
558 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
559 "<round_saeonly_constraint>")): Changed to ...
560 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
561 "<round_saeonly_scalar_constraint>")): ... this.
562 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
563 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
564 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
565 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
566 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
568 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
570 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
572 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
573 Remove all values except native, 8540 and 8548.
575 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
577 * config/i386/i386.c (ix86_output_function_return): Pass
578 INVALID_REGNUM, instead of -1, as invalid register number to
579 indirect_thunk_name and output_indirect_thunk.
581 2018-02-02 Julia Koval <julia.koval@intel.com>
583 * config.gcc: Add -march=icelake.
584 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
585 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
586 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
587 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
588 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
589 (processor_target_table): Add icelake.
590 (ix86_option_override_internal): Handle new PTAs.
591 (get_builtin_code_for_version): Handle icelake.
592 (M_INTEL_COREI7_ICELAKE): New.
593 (fold_builtin_cpu): Handle icelake.
594 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
595 * doc/invoke.texi: Add -march=icelake.
597 2018-02-02 Julia Koval <julia.koval@intel.com>
599 * config/i386/i386.c (ix86_option_override_internal): Change flags type
601 * wide-int-bitmask.h: New.
603 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
606 * config/i386/i386.md: Replace Pmode with word_mode in
607 builtin_setjmp_setup and builtin_longjmp to support x32.
609 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
613 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
615 (rs6000_supported_cpu_names): New static variable.
616 (linux_cpu_translation_table): Likewise.
617 (elf_platform) <cpu>: Define new static variable and use it.
618 Translate kernel AT_PLATFORM name to canonical name if needed.
619 Error if platform name is unknown.
621 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
624 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
626 2018-02-01 Jeff Law <law@redhat.com>
629 * config/i386/i386.c (release_scratch_register_on_entry): Add new
630 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
631 the scratch if RELEASE_VIA_POP is false.
632 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
633 If we have to save a temporary register, decrement SIZE appropriately.
634 Pass new arguments to release_scratch_register_on_entry.
635 (ix86_adjust_stack_and_probe): Likewise.
636 (ix86_emit_probe_stack_range): Pass new arguments to
637 release_scratch_register_on_entry.
639 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
641 PR rtl-optimization/84157
642 * combine.c (change_zero_ext): Use REG_P predicate in
643 front of HARD_REGISTER_P predicate.
645 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
647 * config/avr/avr.c (avr_option_override): Move disabling of
648 -fdelete-null-pointer-checks to...
649 * common/config/avr/avr-common.c (avr_option_optimization_table):
652 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
654 PR tree-optimization/81635
655 * tree-data-ref.c (split_constant_offset_1): For types that
656 wrap on overflow, try to use range info to prove that wrapping
659 2018-02-01 Renlin Li <renlin.li@arm.com>
662 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
664 (aarch64_register_move_cost): Likewise.
665 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
667 (REG_CLASS_NAMES): Likewise.
668 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
669 TAILCALL_ADDR_REGS. Remove IP registers.
670 * config/aarch64/aarch64.md (Ucs): Update register constraint.
672 2018-02-01 Richard Biener <rguenther@suse.de>
674 * domwalk.h (dom_walker::dom_walker): Add additional constructor
675 for specifying RPO order and allow NULL for that.
676 * domwalk.c (dom_walker::dom_walker): Likewise.
677 (dom_walker::walk): Handle NULL RPO order.
678 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
680 (rewrite_update_dom_walker): Likewise.
681 (mark_def_dom_walker): Likewise.
683 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
685 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
686 (aarch64_maybe_expand_sve_subreg_move): Declare.
687 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
688 * config/aarch64/predicates.md (aarch64_any_register_operand): New
690 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
691 that are semantically a reverse operation.
692 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
693 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
694 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
696 (aarch64_can_change_mode_class): For big-endian, forbid changes
697 between two SVE modes if they have different element sizes.
699 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
701 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
702 the TImode handling for big-endian targets.
704 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
706 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
707 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
709 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
710 Remove BSWAP handing for big-endian targets and use the form of
711 LD1RQ appropariate for the mode.
713 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
715 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
716 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
719 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
722 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
723 check for operands that need to go through aarch64_sve_reload_be.
725 2018-02-01 Jakub Jelinek <jakub@redhat.com>
727 PR tree-optimization/81661
728 PR tree-optimization/84117
729 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
730 * tree-eh.c: Include gimplify.h.
731 (find_trapping_overflow, replace_trapping_overflow,
732 rewrite_to_non_trapping_overflow): New functions.
733 * tree-vect-loop.c: Include tree-eh.h.
734 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
735 * tree-data-ref.c: Include tree-eh.h.
736 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
738 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
740 PR rtl-optimization/84123
741 * combine.c (change_zero_ext): Check if hard register satisfies
742 can_change_dest_mode before calling gen_lowpart_SUBREG.
744 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
747 * ira.c (ira_init_register_move_cost): Remove assert.
749 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
751 PR rtl-optimization/84071
752 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
753 * doc/tm.texi: Regenerate.
755 2018-01-31 Richard Biener <rguenther@suse.de>
757 PR tree-optimization/84132
758 * tree-data-ref.c (analyze_miv_subscript): Properly
759 check whether evolution_function_is_affine_multivariate_p
760 before calling gcd_of_steps_may_divide_p.
762 2018-01-31 Julia Koval <julia.koval@intel.com>
765 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
766 * config/i386/i386.md (rdpid_rex64) New.
767 (rdpid): Make 32bit only.
769 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
772 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
773 an IDENTIFIER_NODE for FUNCTION_TYPE's.
775 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
778 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
780 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
782 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
784 PR rtl-optimization/84071
785 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
786 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
788 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
790 * config/arc/arc.c (arc_handle_aux_attribute): New function.
791 (arc_attribute_table): Add 'aux' attribute.
792 (arc_in_small_data_p): Consider aux like variables.
793 (arc_is_aux_reg_p): New function.
794 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
795 (arc_get_aux_arg): New function.
796 (prepare_move_operands): Handle aux-register access.
797 (arc_handle_aux_attribute): New function.
798 * doc/extend.texi (ARC Variable attributes): Add subsection.
800 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
802 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
803 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
804 (arc_attribute_table): Add 'uncached' attribute.
805 (arc_print_operand): Print '.di' flag for uncached memory
807 (arc_in_small_data_p): Do not consider for small data the uncached
809 (arc_is_uncached_mem_p): New function.
810 * config/arc/predicates.md (compact_store_memory_operand): Check
811 for uncached memory accesses.
812 (nonvol_nonimm_operand): Likewise.
813 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
815 2018-01-31 Jakub Jelinek <jakub@redhat.com>
818 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
819 falign-loops=): Add Optimization flag.
821 2018-01-30 Jeff Law <law@redhat.com>
824 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
825 INT_REGISTERS_SAVED. Check it prior to calling
826 get_scratch_register_on_entry.
827 (ix86_adjust_stack_and_probe): Similarly.
828 (ix86_emit_probe_stack_range): Similarly.
829 (ix86_expand_prologue): Corresponding changes.
831 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
834 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
835 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
837 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
840 * lra-constraints.c (curr_insn_transform): Process AND in the
843 2018-01-30 Jakub Jelinek <jakub@redhat.com>
845 PR rtl-optimization/83986
846 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
847 dependence against last_pending_memory_flush in addition to
850 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
852 PR tree-optimization/81611
853 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
856 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
858 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
861 2018-01-30 Richard Biener <rguenther@suse.de>
862 Jakub Jelinek <jakub@redhat.com>
864 PR tree-optimization/84111
865 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
866 inner loops added during recursion, as they don't have up-to-date
869 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
872 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
873 (can_inline_edge_by_limits_p): ... here.
874 (can_early_inline_edge_p, check_callers,
875 update_caller_keys, update_callee_keys, recursive_inlining,
876 add_new_edges_to_heap, speculation_useful_p,
877 inline_small_functions,
878 inline_small_functions, flatten_function,
879 inline_to_all_callers_1): Update.
881 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
883 * profile-count.c (profile_count::combine_with_ipa_count): Handle
886 2018-01-30 Richard Biener <rguenther@suse.de>
888 PR tree-optimization/83008
889 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
890 invariant and constant vector uses in stmts when they need
893 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
896 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
897 * configure: Regenerate.
899 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
901 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
903 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
904 Use gen_rtx_REG rather than gen_lowpart.
906 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
908 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
909 rather than 0 when creating partial subregs.
911 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
913 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
916 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
919 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
920 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
921 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
922 flags. This restores the settings used before the 2017-07-24.
923 Turning off pre increment/decrement/modify allows IVOPTS to
924 optimize DF/SF loops where the index is an int.
926 2018-01-29 Richard Biener <rguenther@suse.de>
927 Kelvin Nilsen <kelvin@gcc.gnu.org>
930 * tree-vect-stmts.c (vectorizable_call): Don't call
931 targetm.vectorize_builtin_md_vectorized_function if callee is
934 2018-01-22 Carl Love <cel@us.ibm.com>
936 * doc/extend.tex: Fix typo in second arg in
937 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
939 2018-01-29 Richard Biener <rguenther@suse.de>
941 PR tree-optimization/84086
942 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
943 (flush_ssaname_freelist): When SSA names were released reset
946 2018-01-29 Richard Biener <rguenther@suse.de>
948 PR tree-optimization/84057
949 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
950 removed paths when removing edges.
952 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
954 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
955 -mfunction-return=@var{choice}.
957 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
960 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
962 (layout::print_source_line): Likewise.
963 (test_get_line_width_without_trailing_whitespace): Add test cases.
965 2018-01-27 Jakub Jelinek <jakub@redhat.com>
968 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
971 2018-01-26 Jim Wilson <jimw@sifive.com>
973 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
975 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
978 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
980 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
981 and CMP + SUB-immediate -> SUBS.
983 2018-01-26 Martin Sebor <msebor@redhat.com>
985 PR tree-optimization/83896
986 * tree-ssa-strlen.c (get_string_len): Rename...
987 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
988 Avoid assuming length is constant.
989 (handle_char_store): Use HOST_WIDE_INT for string length.
991 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
994 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
995 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
997 2018-01-26 Richard Biener <rguenther@suse.de>
999 PR rtl-optimization/84003
1000 * dse.c (record_store): Only record redundant stores when
1001 the earlier store aliases at least all accesses the later one does.
1003 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1005 PR rtl-optimization/83985
1006 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1007 REG_CFA_RESTORE insns.
1008 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1009 REG_CFA_RESTORE insns here.
1012 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1013 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1015 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1017 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1018 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1019 (arc_init): Likewise.
1020 (arc_override_options): Likewise.
1021 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1023 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1025 * config/arc/arc.h (TARGET_DBNZ): Define.
1026 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1027 properly set the tune attribute.
1028 (dbnz): Use TARGET_DBNZ guard.
1029 * config/arc/arc.opt (mtune): Add core3 option.
1031 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1033 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1034 recognize new pic like addresses.
1035 (arc_delegitimize_address): Clean up.
1037 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1039 * config/arc/arc-arches.def: Option mrf16 valid for all
1041 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1042 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1043 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1044 * config/arc/arc-tables.opt: Regenerate.
1045 * config/arc/arc.c (arc_conditional_register_usage): Handle
1046 reduced register file case.
1047 (arc_file_start): Set must have build attributes.
1048 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1050 * config/arc/arc.opt (mrf16): Add new option.
1051 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1052 * config/arc/genmultilib.awk: Handle new mrf16 option.
1053 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1054 * config/arc/t-multilib: Regenerate.
1055 * doc/invoke.texi (ARC Options): Document mrf16 option.
1057 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1059 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1060 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1061 (arc_attribute_table): Add 'secure_call' attribute.
1062 (arc_print_operand): Print secure call operand.
1063 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1065 (arc_is_secure_call_p): New function. * config/arc/arc.md
1066 (call_i): Add support for sjli instruction.
1067 (call_value_i): Likewise.
1068 * config/arc/constraints.md (Csc): New constraint.
1070 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1071 John Eric Martin <John.Martin@emmicro-us.com>
1073 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1074 * config/arc/arc.c (_arc_jli_section): New struct.
1075 (arc_jli_section): New type.
1076 (rc_jli_sections): New static variable.
1077 (arc_handle_jli_attribute): New function.
1078 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1079 (arc_file_end): New function.
1080 (TARGET_ASM_FILE_END): Define.
1081 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1082 (arc_add_jli_section): New function.
1083 (jli_call_scan): Likewise.
1084 (arc_reorg): Call jli_call_scan.
1085 (arc_output_addsi): Remove 'S' from printing asm operand.
1086 (arc_is_jli_call_p): New function.
1087 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1089 (movhi_insn): Likewise.
1090 (movsi_insn): Likewise.
1091 (movsi_set_cc_insn): Likewise.
1092 (loadqi_update): Likewise.
1093 (load_zeroextendqisi_update): Likewise.
1094 (load_signextendqisi_update): Likewise.
1095 (loadhi_update): Likewise.
1096 (load_zeroextendhisi_update): Likewise.
1097 (load_signextendhisi_update): Likewise.
1098 (loadsi_update): Likewise.
1099 (loadsf_update): Likewise.
1100 (movsicc_insn): Likewise.
1101 (bset_insn): Likewise.
1102 (bxor_insn): Likewise.
1103 (bclr_insn): Likewise.
1104 (bmsk_insn): Likewise.
1105 (bicsi3_insn): Likewise.
1106 (cmpsi_cc_c_insn): Likewise.
1107 (movsi_ne): Likewise.
1108 (movsi_cond_exec): Likewise.
1109 (clrsbsi2): Likewise.
1120 (call_i): Remove 'S' asm letter, add jli instruction.
1121 (call_value_i): Likewise.
1122 * config/arc/arc.op (mjli-always): New option.
1123 * config/arc/constraints.md (Cji): New constraint.
1124 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1126 (subsf3_fpx): Likewise.
1127 (mulsf3_fpx): Likewise.
1128 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1130 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1132 * doc/invoke.texi (ARC): Document mjli-always option.
1134 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1136 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1137 and use incw and decw where possible
1138 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1140 2018-01-26 Richard Biener <rguenther@suse.de>
1142 PR tree-optimization/81082
1143 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1144 association if it requires casting to unsigned.
1145 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1146 from fold_plusminus_mult_expr to catch important cases late when
1147 range info is available.
1149 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1151 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1152 * configure.ac (hidden_linkonce): New test.
1153 * configure: Regenerate.
1154 * config.in: Regenerate.
1156 2018-01-26 Julia Koval <julia.koval@intel.com>
1158 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1159 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1160 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1161 _mm_mask_bitshuffle_epi64_mask): Fix type.
1162 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1163 USI_FTYPE_V4DI_V4DI_USI): Remove.
1164 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1165 __builtin_ia32_vpshufbitqmb256_mask,
1166 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1167 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1168 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1170 2018-01-26 Alan Modra <amodra@gmail.com>
1173 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1174 UNSPEC_VBPERMQ. Sort other unspecs.
1176 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1178 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1180 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1183 * predict.c (drop_profile): Do not push/pop cfun; update also
1185 (handle_missing_profiles): Fix logic looking for zero profiles.
1187 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1190 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1191 on functions with #pragma omp declare simd or functions with simd
1193 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1194 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1195 Remove trailing \n from warning_at calls.
1197 2018-01-25 Tom de Vries <tom@codesourcery.com>
1200 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1201 for neutered workers.
1203 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1206 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1207 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1209 2018-01-24 Jeff Law <law@redhat.com>
1212 * i386.c (get_probe_interval): Move to earlier point.
1213 (ix86_compute_frame_layout): If -fstack-clash-protection and
1214 the frame is larger than the probe interval, then use pushes
1215 to save registers rather than reg->mem moves.
1216 (ix86_expand_prologue): Remove conditional for int_registers_saved
1219 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1222 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1223 min/max for never referenced object.
1225 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1228 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1230 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1231 attributes from DECL_ATTRIBUTES (decl) without affecting
1232 DECL_ATTRIBUTES (current_function_decl).
1233 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1234 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1236 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1238 PR tree-optimization/83979
1239 * fold-const.c (fold_comparison): Use constant_boolean_node
1240 instead of boolean_{true,false}_node.
1242 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1244 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1247 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1249 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1250 Simplify the clause that sets the length attribute.
1251 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1252 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1253 clause that sets the length attribute.
1254 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1256 2018-01-24 Tom de Vries <tom@codesourcery.com>
1259 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1260 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1261 Add strict parameter.
1262 (prevent_branch_around_nothing): Insert dummy insn between branch to
1263 label and label with no ptx insn inbetween.
1264 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1266 2018-01-24 Tom de Vries <tom@codesourcery.com>
1269 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1270 for neutered threads in warp.
1271 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1273 2018-01-24 Richard Biener <rguenther@suse.de>
1275 PR tree-optimization/83176
1276 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1279 2018-01-24 Richard Biener <rguenther@suse.de>
1281 PR tree-optimization/82819
1282 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1283 code generating pluses that are no-ops in the target precision.
1285 2018-01-24 Richard Biener <rguenther@suse.de>
1288 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1290 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1292 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1293 to merge probabilities.
1294 * predict.c (probably_never_executed): Also mark as cold functions
1295 with global 0 profile and guessed local profile.
1296 * profile-count.c (profile_probability::combine_with_count): New
1298 * profile-count.h (profile_probability::operator*,
1299 profile_probability::operator*=, profile_probability::operator/,
1300 profile_probability::operator/=): Reduce precision to adjusted
1301 and set value to guessed on contradictory divisions.
1302 (profile_probability::combine_with_freq): Remove.
1303 (profile_probability::combine_wiht_count): Declare.
1304 (profile_count::force_nonzero):: Set to adjusted.
1305 (profile_count::probability_in):: Set quality to adjusted.
1306 * tree-ssa-tail-merge.c (replace_block_by): Use
1309 2018-01-23 Andrew Waterman <andrew@sifive.com>
1310 Jim Wilson <jimw@sifive.com>
1312 * config/riscv/riscv.c (riscv_stack_boundary): New.
1313 (riscv_option_override): Set riscv_stack_boundary. Handle
1314 riscv_preferred_stack_boundary_arg.
1315 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1316 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1317 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1318 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1319 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1320 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1322 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1325 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1326 of struct ix86_frame.
1327 (ix86_expand_epilogue): Likewise. Add a local variable for
1328 the reg_save_offset field in struct ix86_frame.
1330 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1332 PR tree-optimization/82604
1333 * tree-loop-distribution.c (enum partition_kind): New enum item
1334 PKIND_PARTIAL_MEMSET.
1335 (partition_builtin_p): Support above new enum item.
1336 (generate_code_for_partition): Ditto.
1337 (compute_access_range): Differentiate cases that equality can be
1338 proven at all loops, the innermost loops or no loops.
1339 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1340 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1341 (finalize_partitions, distribute_loop): Don't fuse partition of
1342 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1343 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1346 2018-01-23 Martin Liska <mliska@suse.cz>
1348 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1349 order to ignore the predictor.
1350 (PRED_POLYMORPHIC_CALL): Likewise.
1351 (PRED_RECURSIVE_CALL): Likewise.
1353 2018-01-23 Martin Liska <mliska@suse.cz>
1355 * tree-profile.c (tree_profiling): Print function header to
1356 aware reader which function we are working on.
1357 * value-prof.c (gimple_find_values_to_profile): Do not print
1358 not interesting value histograms.
1360 2018-01-23 Martin Liska <mliska@suse.cz>
1362 * profile-count.h (enum profile_quality): Add
1363 profile_uninitialized as the first value. Do not number values
1364 as they are zero based.
1365 (profile_count::verify): Update sanity check.
1366 (profile_probability::verify): Likewise.
1368 2018-01-23 Nathan Sidwell <nathan@acm.org>
1370 * doc/invoke.texi (ffor-scope): Deprecate.
1372 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1374 PR tree-optimization/83510
1375 * domwalk.c (set_all_edges_as_executable): New function.
1376 (dom_walker::dom_walker): Convert bool param
1377 "skip_unreachable_blocks" to enum reachability. Move setup of
1378 edge flags to set_all_edges_as_executable and only do it when
1379 reachability is REACHABLE_BLOCKS.
1380 * domwalk.h (enum dom_walker::reachability): New enum.
1381 (dom_walker::dom_walker): Convert bool param
1382 "skip_unreachable_blocks" to enum reachability.
1383 (set_all_edges_as_executable): New decl.
1384 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1385 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1387 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1388 but converting true to REACHABLE_BLOCKS.
1389 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1391 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1392 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1393 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1395 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1396 if check_all_array_refs will be called.
1398 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1400 * tree.c (selftest::test_location_wrappers): Add more test
1403 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1405 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1406 (selftest::test_bit_in_range): Likewise.
1408 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1411 * doc/sourcebuild.texi (vect_float): Say that the selector
1412 only describes the situation when -funsafe-math-optimizations is on.
1413 (vect_float_strict): Document.
1415 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1417 PR tree-optimization/83965
1418 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1419 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1420 instead of checking only for a reduction.
1421 (vect_recog_widen_sum_pattern): Likewise.
1423 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1425 * predict.c (probably_never_executed): Only use precise profile info.
1426 (compute_function_frequency): Skip after inlining hack since we now
1427 have quality checking.
1429 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1431 * profile-count.h (profile_probability::very_unlikely,
1432 profile_probability::unlikely, profile_probability::even): Set
1433 precision to guessed.
1435 2018-01-23 Richard Biener <rguenther@suse.de>
1437 PR tree-optimization/83963
1438 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1439 Properly terminate dominator walk when crossing the exit edge not
1440 when visiting its source block.
1442 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1445 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1446 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1448 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1450 PR tree-optimization/83957
1451 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1452 semicolon after for body surrounded by braces.
1454 PR tree-optimization/83081
1455 * profile-count.h (profile_probability::split): New method.
1456 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1457 Use profile_probability::split.
1458 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1459 when splitting a single conditional jump into 2.
1461 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1463 PR tree-optimization/69452
1464 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1467 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1469 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1470 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1471 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1473 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1475 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1476 * config/rl78/rl78.md: New define_expand "movdi"
1477 * config/rl78/rl78.c: New function definition rl78_split_movdi
1479 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1482 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1484 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1485 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1486 128-bit to produce an UNSPEC move to get the double word with the
1487 signbit and then a shift directly to do signbit.
1488 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1489 implementation with a new version that just does either a direct
1490 move or a regular move. Move memory interface to separate insns.
1491 Move insns so they are next to the expander.
1492 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1493 with signbit move. Split big and little endian case.
1494 (signbit<mode>2_dm_mem_le): Likewise.
1495 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1496 (signbit<mode>2_dm2): Likewise.
1498 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1500 * config/rl78/rl78.md: New define_expand "anddi3".
1502 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1504 * config/rl78/rl78.md: New define_expand "umindi3".
1506 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1508 * config/rl78/rl78.md: New define_expand "smindi3".
1510 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1512 * config/rl78/rl78.md: New define_expand "smaxdi3".
1514 2018-01-22 Carl Love <cel@us.ibm.com>
1516 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1517 LVX_V1TI): Add macro expansion.
1518 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1519 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1520 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1521 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1522 Change check to determine if the instruction is a byte reversing
1523 entry. Fix typo in comment.
1524 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1525 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1526 Add def_builtin calls for new builtins.
1527 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1528 Add define_insn expansion.
1530 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1532 * config/rl78/rl78.md: New define_expand "umaxdi3".
1534 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1536 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1537 for non-QImode registers
1539 2018-01-22 Richard Biener <rguenther@suse.de>
1541 PR tree-optimization/83963
1542 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1543 including the loop exit block.
1544 (scop_detection::merge_sese): Likewise.
1545 (scop_detection::add_scop): Do it here instead.
1547 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1549 * doc/sourcebuild.texi (arm_softfloat): Document.
1551 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1554 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1555 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1556 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1558 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1559 David Edelsohn <dje.gcc@gmail.com>
1562 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1563 Change "crset eq" to "crset 2".
1564 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1565 (*call_indirect_aix<mode>_nospec): Likewise.
1566 (*call_value_indirect_aix<mode>_nospec): Likewise.
1567 (*call_indirect_elfv2<mode>_nospec): Likewise.
1568 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1569 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1570 change assembly output from . to $.
1571 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1572 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1573 (*tablejump<mode>_internal1_nospec): Likewise.
1575 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1578 * config/sh/sh_optimize_sett_clrt.cc:
1579 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1581 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1583 PR tree-optimization/83940
1584 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1585 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1586 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1587 use it to pass back the definition type.
1588 (vect_check_store_rhs): Likewise rhs_dt_out.
1589 (vect_build_gather_load_calls): Add a mask_dt argument and use
1590 it instead of a call to vect_is_simple_use.
1591 (vectorizable_store): Update calls to vect_check_load_store_mask
1592 and vect_check_store_rhs. Use the dt returned by the latter instead
1593 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1594 instead of calls to vect_is_simple_use. Pass the scalar rather
1595 than the vector operand to vect_is_simple_use when handling
1596 second and subsequent copies of an rhs value.
1597 (vectorizable_load): Update calls to vect_check_load_store_mask
1598 and vect_build_gather_load_calls. Use the cached mask_dt and
1599 gs_info.offset_dt instead of calls to vect_is_simple_use.
1601 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1604 * tree-emutls.c: Include gimplify.h.
1605 (lower_emutls_2): New function.
1606 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1607 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1608 it before further processing.
1611 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1612 UINTVAL (trueop1) instead of INTVAL (op1).
1614 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1618 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1619 INCOMING_FRAME_SP_OFFSET if not defined.
1620 (scan_trace): Add ENTRY argument. If true and
1621 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1622 emit a note to adjust the CFA offset.
1623 (create_cfi_notes): Adjust scan_trace callers.
1624 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1625 INCOMING_FRAME_SP_OFFSET in the CIE.
1626 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1627 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1629 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1630 * doc/tm.texi: Regenerated.
1632 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1634 PR rtl-optimization/83147
1635 * lra-constraints.c (remove_inheritance_pseudos): Use
1636 lra_substitute_pseudo_within_insn.
1638 2018-01-19 Tom de Vries <tom@codesourcery.com>
1639 Cesar Philippidis <cesar@codesourcery.com>
1642 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1644 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1647 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1648 spaces for function labels.
1650 2018-01-19 Martin Liska <mliska@suse.cz>
1652 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1653 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1654 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1655 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1656 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1657 (PRED_CONST_RETURN): Change from 69 to 65.
1658 (PRED_NULL_RETURN): Change from 91 to 71.
1659 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1660 (PRED_LOOP_GUARD): Change from 66 to 73.
1662 2018-01-19 Martin Liska <mliska@suse.cz>
1664 * predict.c (predict_insn_def): Add new assert.
1665 (struct branch_predictor): Change type to signed integer.
1666 (test_prediction_value_range): Amend test to cover
1668 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1669 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1670 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1671 (PRED_LOOP_IV_COMPARE): Likewise.
1672 * predict.h (PROB_UNINITIALIZED): Define new constant.
1674 2018-01-19 Martin Liska <mliska@suse.cz>
1676 * predict.c (dump_prediction): Add new format for
1677 analyze_brprob.py script which is enabled with -details
1679 * profile-count.h (precise_p): New function.
1681 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1683 PR tree-optimization/83922
1684 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1685 there are no statements that need masking.
1686 (vect_active_double_reduction_p): New function.
1687 (vect_analyze_loop_operations): Use it when handling phis that
1688 are not in the loop header.
1690 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1692 PR tree-optimization/83914
1693 * tree-vect-loop.c (vectorizable_induction): Don't convert
1694 init_expr or apply the peeling adjustment for inductions
1695 that are nested within the vectorized loop.
1697 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1699 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1702 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1706 * function.h (gimplify_parameters): Add gimple_seq * argument.
1707 * function.c: Include gimple.h and options.h.
1708 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1709 for the added local temporaries if needed.
1710 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1711 if there are any parameter cleanups, wrap whole body into a
1712 try/finally with the cleanups.
1714 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1717 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1718 Use GET_MODE_CLASS for scalar floating point.
1720 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1724 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1725 Fix call of call_cgraph_insertion_hooks.
1727 2018-01-18 Martin Sebor <msebor@redhat.com>
1729 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1731 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1734 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1737 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1740 * common.opt: (-ffile-prefix-map): New option.
1741 * opts.c (common_handle_option): Defer it.
1742 * opts-global.c (handle_common_deferred_options): Handle it.
1743 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1744 * file-prefix-map.h: New file.
1745 (remap_debug_filename, add_debug_prefix_map): ...here.
1746 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1747 * final.c (debug_prefix_map, add_debug_prefix_map
1748 remap_debug_filename): Move to...
1749 * file-prefix-map.c: New file.
1750 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1751 generalize, get rid of alloca(), use strrchr() instead of strchr().
1752 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1753 Implement in terms of add_prefix_map().
1754 (remap_macro_filename, remap_debug_filename): Implement in term of
1756 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1757 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1758 * dbxout.c: Include file-prefix-map.h.
1759 * varasm.c: Likewise.
1760 * vmsdbgout.c: Likewise.
1761 * xcoffout.c: Likewise.
1762 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1763 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1764 * doc/invoke.texi (-ffile-prefix-map): Document.
1765 (-fdebug-prefix-map): Update description.
1767 2018-01-18 Martin Liska <mliska@suse.cz>
1769 * config/i386/i386.c (indirect_thunk_name): Document that also
1771 (output_indirect_thunk): Document why both instructions
1772 (pause and lfence) are generated.
1774 2018-01-18 Richard Biener <rguenther@suse.de>
1776 PR tree-optimization/83887
1777 * graphite-scop-detection.c
1778 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1779 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1780 (scop_detection::merge_sese): Re-implement with a flood-fill
1781 algorithm that properly finds a SESE region if it exists.
1783 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1786 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1787 pointer_diff optimizations use view_convert instead of convert.
1789 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1791 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1792 Generate different code for -mno-speculate-indirect-jumps.
1793 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1794 (*call_indirect_aix<mode>): Disable for
1795 -mno-speculate-indirect-jumps.
1796 (*call_indirect_aix<mode>_nospec): New define_insn.
1797 (*call_value_indirect_aix<mode>): Disable for
1798 -mno-speculate-indirect-jumps.
1799 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1800 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1801 -mno-speculate-indirect-jumps.
1802 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1804 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1806 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1807 long double type, set the flags for noting the default long double
1808 type, even if we don't pass or return a long double type.
1810 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1813 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1816 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1818 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1819 support for merge[hl].
1820 (fold_mergehl_helper): New helper function.
1821 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1822 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1823 (altivec_vmrglw_direct): Add xxmrglw insn.
1825 2018-01-17 Andrew Waterman <andrew@sifive.com>
1827 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1828 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1830 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1833 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1834 call the lto_location_cache before reading the
1835 DECL_SOURCE_LOCATION of the types.
1837 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1838 Richard Sandiford <richard.sandiford@linaro.org>
1840 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1841 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1842 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1843 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1844 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1846 * config/aarch64/constraints.md (aarch64_movti_operand):
1848 * config/aarch64/predicates.md (Uti): Add new constraint.
1850 2018-01-17 Carl Love <cel@us.ibm.com>
1851 * config/rs6000/vsx.md (define_expand xl_len_r,
1852 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1853 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1855 (define_expand, define_insn): Move the shift left from the
1856 define_insn to the define_expand for lxvl and stxvl instructions.
1857 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1858 and XL_LEN_R definitions to PURE.
1860 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1862 * config/i386/i386.c (indirect_thunk_name): Declare regno
1863 as unsigned int. Compare regno with INVALID_REGNUM.
1864 (output_indirect_thunk): Ditto.
1865 (output_indirect_thunk_function): Ditto.
1866 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1867 in the call to output_indirect_thunk_function.
1869 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1872 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1873 rather than the size of inner_type to determine the stack slot size
1874 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1876 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1879 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1882 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1884 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1885 endian Linux systems to optionally enable multilibs for selecting
1886 the long double type if the user configured an explicit type.
1887 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1888 have no long double multilibs if not defined.
1889 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1890 warn if the user used -mabi={ieee,ibm}longdouble and we built
1891 multilibs for long double.
1892 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1893 appropriate multilib option.
1894 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1896 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1897 for building long double multilibs.
1898 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1900 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1902 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1905 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1907 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1910 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1913 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1916 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1918 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1919 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1921 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1923 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1924 different rtl trees depending on TARGET_64BIT.
1925 (rs6000_gen_lvx): Likewise.
1927 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1929 * config/visium/visium.md (nop): Tweak comment.
1930 (hazard_nop): Likewise.
1932 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1934 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1935 -mspeculate-indirect-jumps.
1936 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1937 for -mno-speculate-indirect-jumps.
1938 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1939 (*call_value_indirect_elfv2<mode>): Disable for
1940 -mno-speculate-indirect-jumps.
1941 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1942 (indirect_jump): Emit different RTL for
1943 -mno-speculate-indirect-jumps.
1944 (*indirect_jump<mode>): Disable for
1945 -mno-speculate-indirect-jumps.
1946 (*indirect_jump<mode>_nospec): New define_insn.
1947 (tablejump): Emit different RTL for
1948 -mno-speculate-indirect-jumps.
1949 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1950 (tablejumpsi_nospec): New define_expand.
1951 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1952 (tablejumpdi_nospec): New define_expand.
1953 (*tablejump<mode>_internal1): Disable for
1954 -mno-speculate-indirect-jumps.
1955 (*tablejump<mode>_internal1_nospec): New define_insn.
1956 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1959 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1961 * caller-save.c (insert_save): Drop unnecessary parameter. All
1964 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1965 Richard Biener <rguenth@suse.de>
1968 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1969 return early, inline manually is_gimple_sizepos. Make sure if we
1970 call gimplify_expr we don't end up with a gimple constant.
1971 * tree.c (variably_modified_type_p): Don't return true for
1972 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1973 * gimplify.h (is_gimple_sizepos): Remove.
1975 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1977 PR tree-optimization/83857
1978 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1979 vectorizable_live_operation for pure SLP statements.
1980 (vectorizable_live_operation): Handle PHIs.
1982 2018-01-16 Richard Biener <rguenther@suse.de>
1984 PR tree-optimization/83867
1985 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1986 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1988 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1991 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1992 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1993 If off is not INTEGER_CST, issue a may not be aligned warning
1994 rather than isn't aligned. Use isn%'t rather than isn't.
1995 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1997 <case MULT_EXPR>: Improve the case when bottom and one of the
1998 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1999 operand, in that case check if the other operand is multiple of
2000 bottom divided by the INTEGER_CST operand.
2002 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2005 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2006 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2007 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2008 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2009 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2010 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2011 (pa_function_arg_size): New function.
2013 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2015 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2016 in a separate statement.
2018 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2020 PR tree-optimization/83847
2021 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2022 group gathers and scatters.
2024 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2026 PR rtl-optimization/86620
2027 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2029 PR rtl-optimization/83213
2030 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2031 to last if both are JUMP_INSNs.
2033 PR tree-optimization/83843
2034 * gimple-ssa-store-merging.c
2035 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2036 store_immediate_info for bswap/nop orig_stores.
2038 2018-01-15 Andrew Waterman <andrew@sifive.com>
2040 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2042 <UDIV>: Increase cost if !TARGET_DIV.
2044 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2046 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2047 (define_attr "cr_logical_3op"): New.
2048 (cceq_ior_compare): Adjust.
2049 (cceq_ior_compare_complement): Adjust.
2050 (*cceq_rev_compare): Adjust.
2051 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2052 (is_cracked_insn): Adjust.
2053 (insn_must_be_first_in_group): Adjust.
2054 * config/rs6000/40x.md: Adjust.
2055 * config/rs6000/440.md: Adjust.
2056 * config/rs6000/476.md: Adjust.
2057 * config/rs6000/601.md: Adjust.
2058 * config/rs6000/603.md: Adjust.
2059 * config/rs6000/6xx.md: Adjust.
2060 * config/rs6000/7450.md: Adjust.
2061 * config/rs6000/7xx.md: Adjust.
2062 * config/rs6000/8540.md: Adjust.
2063 * config/rs6000/cell.md: Adjust.
2064 * config/rs6000/e300c2c3.md: Adjust.
2065 * config/rs6000/e500mc.md: Adjust.
2066 * config/rs6000/e500mc64.md: Adjust.
2067 * config/rs6000/e5500.md: Adjust.
2068 * config/rs6000/e6500.md: Adjust.
2069 * config/rs6000/mpc.md: Adjust.
2070 * config/rs6000/power4.md: Adjust.
2071 * config/rs6000/power5.md: Adjust.
2072 * config/rs6000/power6.md: Adjust.
2073 * config/rs6000/power7.md: Adjust.
2074 * config/rs6000/power8.md: Adjust.
2075 * config/rs6000/power9.md: Adjust.
2076 * config/rs6000/rs64.md: Adjust.
2077 * config/rs6000/titan.md: Adjust.
2079 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2081 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2082 ix86_indirect_branch_register logic.
2084 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2086 * config/i386/constraints.md (Bs): Update
2087 ix86_indirect_branch_register check. Don't check
2088 ix86_indirect_branch_register with GOT_memory_operand.
2090 * config/i386/predicates.md (GOT_memory_operand): Don't check
2091 ix86_indirect_branch_register here.
2092 (GOT32_symbol_operand): Likewise.
2094 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2096 * config/i386/predicates.md (constant_call_address_operand):
2097 Rewrite ix86_indirect_branch_register logic.
2098 (sibcall_insn_operand): Likewise.
2100 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2102 * config/i386/constraints.md (Bs): Replace
2103 ix86_indirect_branch_thunk_register with
2104 ix86_indirect_branch_register.
2106 * config/i386/i386.md (indirect_jump): Likewise.
2107 (tablejump): Likewise.
2108 (*sibcall_memory): Likewise.
2109 (*sibcall_value_memory): Likewise.
2110 Peepholes of indirect call and jump via memory: Likewise.
2111 * config/i386/i386.opt: Likewise.
2112 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2113 (GOT_memory_operand): Likewise.
2114 (call_insn_operand): Likewise.
2115 (sibcall_insn_operand): Likewise.
2116 (GOT32_symbol_operand): Likewise.
2118 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2121 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2122 type rather than type addr's type points to.
2123 (expand_omp_atomic_mutex): Likewise.
2124 (expand_omp_atomic): Likewise.
2126 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2129 * config/i386/i386.c (output_indirect_thunk_function): Use
2130 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2131 for __x86_return_thunk.
2133 2018-01-15 Richard Biener <rguenther@suse.de>
2136 * expmed.c (extract_bit_field_1): Fix typo.
2138 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2141 * config/arm/iterators.md (VF): New mode iterator.
2142 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2143 Remove integer-related logic from pattern.
2144 (neon_vabd<mode>_3): Likewise.
2146 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2149 * common.opt (fstrict-overflow): No longer an alias.
2150 (fwrapv-pointer): New option.
2151 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2152 also for pointer types based on flag_wrapv_pointer.
2153 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2154 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2155 opts->x_flag_wrapv got set.
2156 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2157 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2158 POINTER_TYPE_OVERFLOW_UNDEFINED.
2159 * match.pd: Likewise in address comparison pattern.
2160 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2162 2018-01-15 Richard Biener <rguenther@suse.de>
2165 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2166 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2167 Reset type names to their identifier if their TYPE_DECL doesn't
2168 have linkage (and thus is used for ODR and devirt).
2169 (save_debug_info_for_decl): Remove.
2170 (save_debug_info_for_type): Likewise.
2171 (add_tree_to_fld_list): Adjust.
2172 * tree-pretty-print.c (dump_generic_node): Make dumping of
2173 type names more robust.
2175 2018-01-15 Richard Biener <rguenther@suse.de>
2177 * BASE-VER: Bump to 8.0.1.
2179 2018-01-14 Martin Sebor <msebor@redhat.com>
2182 * builtins.c (check_access): Avoid warning when the no-warning bit
2185 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2187 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2188 * ira-color (allocno_hard_regs_compare): Likewise.
2190 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2193 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2194 Use .pushsection/.popsection.
2196 2018-01-14 Martin Sebor <msebor@redhat.com>
2199 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2201 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2203 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2204 entry from extra_headers.
2205 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2206 extra_headers, make the list bitwise identical to the i?86-*-* one.
2208 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2210 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2211 -mcmodel=large with -mindirect-branch=thunk,
2212 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2213 -mfunction-return=thunk-extern.
2214 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2215 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2216 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2218 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2220 * config/i386/i386.c (print_reg): Print the name of the full
2221 integer register without '%'.
2222 (ix86_print_operand): Handle 'V'.
2223 * doc/extend.texi: Document 'V' modifier.
2225 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2227 * config/i386/constraints.md (Bs): Disallow memory operand for
2228 -mindirect-branch-register.
2230 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2231 (GOT_memory_operand): Likewise.
2232 (call_insn_operand): Likewise.
2233 (sibcall_insn_operand): Likewise.
2234 (GOT32_symbol_operand): Likewise.
2235 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2236 for -mindirect-branch-register.
2237 (tablejump): Likewise.
2238 (*sibcall_memory): Likewise.
2239 (*sibcall_value_memory): Likewise.
2240 Disallow peepholes of indirect call and jump via memory for
2241 -mindirect-branch-register.
2242 (*call_pop): Replace m with Bw.
2243 (*call_value_pop): Likewise.
2244 (*sibcall_pop_memory): Replace m with Bs.
2245 * config/i386/i386.opt (mindirect-branch-register): New option.
2246 * doc/invoke.texi: Document -mindirect-branch-register option.
2248 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2250 * config/i386/i386-protos.h (ix86_output_function_return): New.
2251 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2252 set function_return_type.
2253 (indirect_thunk_name): Add ret_p to indicate thunk for function
2255 (output_indirect_thunk_function): Pass false to
2256 indirect_thunk_name.
2257 (ix86_output_indirect_branch_via_reg): Likewise.
2258 (ix86_output_indirect_branch_via_push): Likewise.
2259 (output_indirect_thunk_function): Create alias for function
2260 return thunk if regno < 0.
2261 (ix86_output_function_return): New function.
2262 (ix86_handle_fndecl_attribute): Handle function_return.
2263 (ix86_attribute_table): Add function_return.
2264 * config/i386/i386.h (machine_function): Add
2265 function_return_type.
2266 * config/i386/i386.md (simple_return_internal): Use
2267 ix86_output_function_return.
2268 (simple_return_internal_long): Likewise.
2269 * config/i386/i386.opt (mfunction-return=): New option.
2270 (indirect_branch): Mention -mfunction-return=.
2271 * doc/extend.texi: Document function_return function attribute.
2272 * doc/invoke.texi: Document -mfunction-return= option.
2274 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2276 * config/i386/i386-opts.h (indirect_branch): New.
2277 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2278 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2279 with local indirect jump when converting indirect call and jump.
2280 (ix86_set_indirect_branch_type): New.
2281 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2282 (indirectlabelno): New.
2283 (indirect_thunk_needed): Likewise.
2284 (indirect_thunk_bnd_needed): Likewise.
2285 (indirect_thunks_used): Likewise.
2286 (indirect_thunks_bnd_used): Likewise.
2287 (INDIRECT_LABEL): Likewise.
2288 (indirect_thunk_name): Likewise.
2289 (output_indirect_thunk): Likewise.
2290 (output_indirect_thunk_function): Likewise.
2291 (ix86_output_indirect_branch_via_reg): Likewise.
2292 (ix86_output_indirect_branch_via_push): Likewise.
2293 (ix86_output_indirect_branch): Likewise.
2294 (ix86_output_indirect_jmp): Likewise.
2295 (ix86_code_end): Call output_indirect_thunk_function if needed.
2296 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2298 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2299 (ix86_attribute_table): Add indirect_branch.
2300 * config/i386/i386.h (machine_function): Add indirect_branch_type
2301 and has_local_indirect_jump.
2302 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2304 (tablejump): Likewise.
2305 (*indirect_jump): Use ix86_output_indirect_jmp.
2306 (*tablejump_1): Likewise.
2307 (simple_return_indirect_internal): Likewise.
2308 * config/i386/i386.opt (mindirect-branch=): New option.
2309 (indirect_branch): New.
2312 (thunk-inline): Likewise.
2313 (thunk-extern): Likewise.
2314 * doc/extend.texi: Document indirect_branch function attribute.
2315 * doc/invoke.texi: Document -mindirect-branch= option.
2317 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2320 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2322 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2324 * ipa-inline.c (want_inline_small_function_p): Return false if
2325 inlining has already failed with CIF_FINAL_ERROR.
2326 (update_caller_keys): Call want_inline_small_function_p before
2328 (update_callee_keys): Likewise.
2330 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2332 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2334 (rs6000_quadword_masked_address_p): Likewise.
2335 (quad_aligned_load_p): Likewise.
2336 (quad_aligned_store_p): Likewise.
2337 (const_load_sequence_p): Add comment to describe the outer-most loop.
2338 (mimic_memory_attributes_and_flags): New function.
2339 (rs6000_gen_stvx): Likewise.
2340 (replace_swapped_aligned_store): Likewise.
2341 (rs6000_gen_lvx): Likewise.
2342 (replace_swapped_aligned_load): Likewise.
2343 (replace_swapped_load_constant): Capitalize argument name in
2344 comment describing this function.
2345 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2346 and stores that access quad-word aligned addresses and replace
2347 with stvx or lvx instructions when appropriate.
2348 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2349 New function prototype.
2350 (rs6000_quadword_masked_address_p): Likewise.
2351 (rs6000_gen_lvx): Likewise.
2352 (rs6000_gen_stvx): Likewise.
2353 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2354 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2355 when memory address is aligned.
2356 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2357 this split to select lvx instruction when memory address is aligned.
2358 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2359 instruction when memory address is aligned.
2360 (*vsx_le_perm_load_v16qi): Likewise.
2361 (four unnamed splitters): Modify to select the stvx instruction
2362 when memory is aligned.
2364 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2366 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2367 which appears in the queue multiple times.
2369 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2370 Alan Hayward <alan.hayward@arm.com>
2371 David Sherwood <david.sherwood@arm.com>
2373 * tree-vectorizer.h (vec_lower_bound): New structure.
2374 (_loop_vec_info): Add check_nonzero and lower_bounds.
2375 (LOOP_VINFO_CHECK_NONZERO): New macro.
2376 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2377 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2378 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2379 fields. Make seg_len the distance travelled, not including the
2381 (dr_direction_indicator): Declare.
2382 (dr_zero_step_indicator): Likewise.
2383 (dr_known_forward_stride_p): Likewise.
2384 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2386 (runtime_alias_check_p): Allow runtime alias checks with
2388 (operator ==): Compare access_size and align.
2389 (prune_runtime_alias_test_list): Rework for new distinction between
2390 the access_size and seg_len.
2391 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2393 (get_segment_min_max): New function.
2394 (create_intersect_range_checks): Use it.
2395 (dr_step_indicator): New function.
2396 (dr_direction_indicator): Likewise.
2397 (dr_zero_step_indicator): Likewise.
2398 (dr_known_forward_stride_p): Likewise.
2399 * tree-loop-distribution.c (data_ref_segment_size): Return
2400 DR_STEP * (niters - 1).
2401 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2403 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2404 (vect_preserves_scalar_order_p): New function, split out from...
2405 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2406 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2407 (vect_vfa_access_size): New function.
2408 (vect_vfa_align): Likewise.
2409 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2410 (dump_lower_bound): New function.
2411 (vect_check_lower_bound): Likewise.
2412 (vect_small_gap_p): Likewise.
2413 (vectorizable_with_step_bound_p): Likewise.
2414 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2415 depencies if the vectorization factor is 1. Convert the checks
2416 for nonzero steps into checks on the bounds of DR_STEP. Try using
2417 a bunds check for variable steps if the minimum required step is
2418 relatively small. Update calls to the dr_with_seg_len
2419 constructor and to vect_compile_time_alias.
2420 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2422 (vect_loop_versioning): Call it.
2423 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2425 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2427 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2428 Alan Hayward <alan.hayward@arm.com>
2429 David Sherwood <david.sherwood@arm.com>
2431 * doc/sourcebuild.texi (vect_scatter_store): Document.
2432 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2434 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2436 * genopinit.c (main): Add supports_vec_scatter_store and
2437 supports_vec_scatter_store_cached to target_optabs.
2438 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2439 IFN_MASK_SCATTER_STORE.
2440 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2442 * internal-fn.h (internal_store_fn_p): Declare.
2443 (internal_fn_stored_value_index): Likewise.
2444 * internal-fn.c (scatter_store_direct): New macro.
2445 (expand_scatter_store_optab_fn): New function.
2446 (direct_scatter_store_optab_supported_p): New macro.
2447 (internal_store_fn_p): New function.
2448 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2449 IFN_MASK_SCATTER_STORE.
2450 (internal_fn_mask_index): Likewise.
2451 (internal_fn_stored_value_index): New function.
2452 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2454 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2455 * optabs-query.c (supports_vec_scatter_store_p): New function.
2456 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2457 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2458 true for scatter stores.
2459 (vect_gather_scatter_fn_p): Handle scatter stores too.
2460 (vect_check_gather_scatter): Consider using scatter stores if
2461 supports_vec_scatter_store_p.
2462 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2464 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2465 internal_fn_stored_value_index.
2466 (check_load_store_masking): Handle scatter stores too.
2467 (vect_get_store_rhs): Make public.
2468 (vectorizable_call): Use internal_store_fn_p.
2469 (vectorizable_store): Handle scatter store internal functions.
2470 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2471 when deciding whether the end of the group has been reached.
2472 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2473 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2474 (mask_scatter_store<mode>): New insns.
2476 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2477 Alan Hayward <alan.hayward@arm.com>
2478 David Sherwood <david.sherwood@arm.com>
2480 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2481 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2482 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2484 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2485 Use vect_truncate_gather_scatter_offset if we can't treat the
2486 operation as a normal gather load or scatter store.
2487 (get_group_load_store_type): Take the gather_scatter_info
2488 as argument. Try using a gather load or scatter store for
2489 single-element groups.
2490 (get_load_store_type): Update calls to get_group_load_store_type
2491 and vect_use_strided_gather_scatters_p.
2493 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2494 Alan Hayward <alan.hayward@arm.com>
2495 David Sherwood <david.sherwood@arm.com>
2497 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2498 optional tree argument.
2499 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2501 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2502 but continue to use the current value as a fallback.
2503 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2504 to compare the updates.
2505 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2506 (get_load_store_type): Use it when handling a strided access.
2507 (vect_get_strided_load_store_ops): New function.
2508 (vect_get_data_ptr_increment): Likewise.
2509 (vectorizable_load): Handle strided gather loads. Always pass
2510 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2512 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2513 Alan Hayward <alan.hayward@arm.com>
2514 David Sherwood <david.sherwood@arm.com>
2516 * doc/md.texi (gather_load@var{m}): Document.
2517 (mask_gather_load@var{m}): Likewise.
2518 * genopinit.c (main): Add supports_vec_gather_load and
2519 supports_vec_gather_load_cached to target_optabs.
2520 * optabs-tree.c (init_tree_optimization_optabs): Use
2521 ggc_cleared_alloc to allocate target_optabs.
2522 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2523 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2525 * internal-fn.h (internal_load_fn_p): Declare.
2526 (internal_gather_scatter_fn_p): Likewise.
2527 (internal_fn_mask_index): Likewise.
2528 (internal_gather_scatter_fn_supported_p): Likewise.
2529 * internal-fn.c (gather_load_direct): New macro.
2530 (expand_gather_load_optab_fn): New function.
2531 (direct_gather_load_optab_supported_p): New macro.
2532 (direct_internal_fn_optab): New function.
2533 (internal_load_fn_p): Likewise.
2534 (internal_gather_scatter_fn_p): Likewise.
2535 (internal_fn_mask_index): Likewise.
2536 (internal_gather_scatter_fn_supported_p): Likewise.
2537 * optabs-query.c (supports_at_least_one_mode_p): New function.
2538 (supports_vec_gather_load_p): Likewise.
2539 * optabs-query.h (supports_vec_gather_load_p): Declare.
2540 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2541 and memory_type field.
2542 (NUM_PATTERNS): Bump to 15.
2543 * tree-vect-data-refs.c: Include internal-fn.h.
2544 (vect_gather_scatter_fn_p): New function.
2545 (vect_describe_gather_scatter_call): Likewise.
2546 (vect_check_gather_scatter): Try using internal functions for
2547 gather loads. Recognize existing calls to a gather load function.
2548 (vect_analyze_data_refs): Consider using gather loads if
2549 supports_vec_gather_load_p.
2550 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2551 (vect_get_gather_scatter_offset_type): Likewise.
2552 (vect_convert_mask_for_vectype): Likewise.
2553 (vect_add_conversion_to_patterm): Likewise.
2554 (vect_try_gather_scatter_pattern): Likewise.
2555 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2556 (vect_vect_recog_func_ptrs): Add it.
2557 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2558 internal_fn_mask_index and internal_gather_scatter_fn_p.
2559 (check_load_store_masking): Take the gather_scatter_info as an
2560 argument and handle gather loads.
2561 (vect_get_gather_scatter_ops): New function.
2562 (vectorizable_call): Check internal_load_fn_p.
2563 (vectorizable_load): Likewise. Handle gather load internal
2565 (vectorizable_store): Update call to check_load_store_masking.
2566 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2567 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2568 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2569 (aarch64_gather_scale_operand_d): New predicates.
2570 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2571 (mask_gather_load<mode>): New insns.
2573 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2574 Alan Hayward <alan.hayward@arm.com>
2575 David Sherwood <david.sherwood@arm.com>
2577 * optabs.def (fold_left_plus_optab): New optab.
2578 * doc/md.texi (fold_left_plus_@var{m}): Document.
2579 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2580 * internal-fn.c (fold_left_direct): Define.
2581 (expand_fold_left_optab_fn): Likewise.
2582 (direct_fold_left_optab_supported_p): Likewise.
2583 * fold-const-call.c (fold_const_fold_left): New function.
2584 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2585 * tree-parloops.c (valid_reduction_p): New function.
2586 (gather_scalar_reductions): Use it.
2587 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2588 (vect_finish_replace_stmt): Declare.
2589 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2590 (needs_fold_left_reduction_p): New function, split out from...
2591 (vect_is_simple_reduction): ...here. Accept reductions that
2592 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2593 (vect_force_simple_reduction): Also store the reduction type in
2594 the assignment's STMT_VINFO_REDUC_TYPE.
2595 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2596 (merge_with_identity): New function.
2597 (vect_expand_fold_left): Likewise.
2598 (vectorize_fold_left_reduction): Likewise.
2599 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2600 scalar phi in place for it. Check for target support and reject
2601 cases that would reassociate the operation. Defer the transform
2602 phase to vectorize_fold_left_reduction.
2603 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2604 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2605 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2607 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2609 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2610 call to ifc_temp_var.
2612 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2613 Alan Hayward <alan.hayward@arm.com>
2614 David Sherwood <david.sherwood@arm.com>
2616 * target.def (legitimize_address_displacement): Take the original
2617 offset as a poly_int.
2618 * targhooks.h (default_legitimize_address_displacement): Update
2620 * targhooks.c (default_legitimize_address_displacement): Likewise.
2621 * doc/tm.texi: Regenerate.
2622 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2623 as an argument, moving assert of ad->disp == ad->disp_term to...
2624 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2625 Try calling targetm.legitimize_address_displacement before expanding
2626 the address rather than afterwards, and adjust for the new interface.
2627 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2628 Match the new hook interface. Handle SVE addresses.
2629 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2632 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2634 * Makefile.in (OBJS): Add early-remat.o.
2635 * target.def (select_early_remat_modes): New hook.
2636 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2637 * doc/tm.texi: Regenerate.
2638 * targhooks.h (default_select_early_remat_modes): Declare.
2639 * targhooks.c (default_select_early_remat_modes): New function.
2640 * timevar.def (TV_EARLY_REMAT): New timevar.
2641 * passes.def (pass_early_remat): New pass.
2642 * tree-pass.h (make_pass_early_remat): Declare.
2643 * early-remat.c: New file.
2644 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2646 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
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 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2653 vfm1 with a bound_epilog parameter.
2654 (vect_do_peeling): Update calls accordingly, and move the prologue
2655 call earlier in the function. Treat the base bound_epilog as 0 for
2656 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2657 this base when peeling for gaps.
2658 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2659 with fully-masked loops.
2660 (vect_estimate_min_profitable_iters): Handle the single peeled
2661 iteration in that case.
2663 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2664 Alan Hayward <alan.hayward@arm.com>
2665 David Sherwood <david.sherwood@arm.com>
2667 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2668 single-element interleaving even if the size is not a power of 2.
2669 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2670 accesses for single-element interleaving if the group size is
2673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2674 Alan Hayward <alan.hayward@arm.com>
2675 David Sherwood <david.sherwood@arm.com>
2677 * doc/md.texi (fold_extract_last_@var{m}): Document.
2678 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2679 * optabs.def (fold_extract_last_optab): New optab.
2680 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2681 * internal-fn.c (fold_extract_direct): New macro.
2682 (expand_fold_extract_optab_fn): Likewise.
2683 (direct_fold_extract_optab_supported_p): Likewise.
2684 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2685 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2686 EXTRACT_LAST_REDUCTION.
2687 (get_initial_def_for_reduction): Do not create an initial vector
2688 for EXTRACT_LAST_REDUCTION reductions.
2689 (vectorizable_reduction): Leave the scalar phi in place for
2690 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2691 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2692 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2693 transform phase to vectorizable_condition.
2694 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2696 (vect_finish_stmt_generation): ...here.
2697 (vect_finish_replace_stmt): New function.
2698 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2699 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2701 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2704 Alan Hayward <alan.hayward@arm.com>
2705 David Sherwood <david.sherwood@arm.com>
2707 * doc/md.texi (extract_last_@var{m}): Document.
2708 * optabs.def (extract_last_optab): New optab.
2709 * internal-fn.def (EXTRACT_LAST): New internal function.
2710 * internal-fn.c (cond_unary_direct): New macro.
2711 (expand_cond_unary_optab_fn): Likewise.
2712 (direct_cond_unary_optab_supported_p): Likewise.
2713 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2714 loops using EXTRACT_LAST.
2715 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2716 (extract_last_<mode>): ...this optab.
2717 (vec_extract<mode><Vel>): Update accordingly.
2719 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2720 Alan Hayward <alan.hayward@arm.com>
2721 David Sherwood <david.sherwood@arm.com>
2723 * target.def (empty_mask_is_expensive): New hook.
2724 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2725 * doc/tm.texi: Regenerate.
2726 * targhooks.h (default_empty_mask_is_expensive): Declare.
2727 * targhooks.c (default_empty_mask_is_expensive): New function.
2728 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2729 if the target says that empty masks are expensive.
2730 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2732 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2734 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2735 Alan Hayward <alan.hayward@arm.com>
2736 David Sherwood <david.sherwood@arm.com>
2738 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2739 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2740 (vect_use_loop_mask_for_alignment_p): New function.
2741 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2742 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2743 niters_skip argument. Make sure that the first niters_skip elements
2744 of the first iteration are inactive.
2745 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2746 Update call to vect_set_loop_masks_directly.
2747 (get_misalign_in_elems): New function, split out from...
2748 (vect_gen_prolog_loop_niters): ...here.
2749 (vect_update_init_of_dr): Take a code argument that specifies whether
2750 the adjustment should be added or subtracted.
2751 (vect_update_init_of_drs): Likewise.
2752 (vect_prepare_for_masked_peels): New function.
2753 (vect_do_peeling): Skip prologue peeling if we're using a mask
2754 instead. Update call to vect_update_inits_of_drs.
2755 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2757 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2758 alignment. Do not include the number of peeled iterations in
2759 the minimum threshold in that case.
2760 (vectorizable_induction): Adjust the start value down by
2761 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2762 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2763 Take the number of skipped iterations into account when calculating
2765 * tree-vect-stmts.c (vect_gen_while_not): New function.
2767 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2768 Alan Hayward <alan.hayward@arm.com>
2769 David Sherwood <david.sherwood@arm.com>
2771 * doc/sourcebuild.texi (vect_fully_masked): Document.
2772 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2774 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2776 (vect_analyze_loop_2): ...here. Don't check the vectorization
2777 factor against the number of loop iterations if the loop is
2780 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2781 Alan Hayward <alan.hayward@arm.com>
2782 David Sherwood <david.sherwood@arm.com>
2784 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2785 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2786 (dump_groups): Update accordingly.
2787 (iv_use::mem_type): New member variable.
2788 (address_p): New function.
2789 (record_use): Add a mem_type argument and initialize the new
2791 (record_group_use): Add a mem_type argument. Use address_p.
2792 Remove obsolete null checks of base_object. Update call to record_use.
2793 (find_interesting_uses_op): Update call to record_group_use.
2794 (find_interesting_uses_cond): Likewise.
2795 (find_interesting_uses_address): Likewise.
2796 (get_mem_type_for_internal_fn): New function.
2797 (find_address_like_use): Likewise.
2798 (find_interesting_uses_stmt): Try find_address_like_use before
2799 calling find_interesting_uses_op.
2800 (addr_offset_valid_p): Use the iv mem_type field as the type
2801 of the addressed memory.
2802 (add_autoinc_candidates): Likewise.
2803 (get_address_cost): Likewise.
2804 (split_small_address_groups_p): Use address_p.
2805 (split_address_groups): Likewise.
2806 (add_iv_candidate_for_use): Likewise.
2807 (autoinc_possible_for_pair): Likewise.
2808 (rewrite_groups): Likewise.
2809 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2810 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2811 (get_alias_ptr_type_for_ptr_address): New function.
2812 (rewrite_use_address): Rewrite address uses in calls that were
2813 identified by find_address_like_use.
2815 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2816 Alan Hayward <alan.hayward@arm.com>
2817 David Sherwood <david.sherwood@arm.com>
2819 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2821 * gimple-expr.h (is_gimple_addressable: Likewise.
2822 * gimple-expr.c (is_gimple_address): Likewise.
2823 * internal-fn.c (expand_call_mem_ref): New function.
2824 (expand_mask_load_optab_fn): Use it.
2825 (expand_mask_store_optab_fn): Likewise.
2827 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2828 Alan Hayward <alan.hayward@arm.com>
2829 David Sherwood <david.sherwood@arm.com>
2831 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2832 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2833 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2834 (cond_umax@var{mode}): Document.
2835 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2836 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2837 (cond_umin_optab, cond_umax_optab): New optabs.
2838 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2839 (COND_IOR, COND_XOR): New internal functions.
2840 * internal-fn.h (get_conditional_internal_fn): Declare.
2841 * internal-fn.c (cond_binary_direct): New macro.
2842 (expand_cond_binary_optab_fn): Likewise.
2843 (direct_cond_binary_optab_supported_p): Likewise.
2844 (get_conditional_internal_fn): New function.
2845 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2846 Cope with reduction statements that are vectorized as calls rather
2848 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2849 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2850 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2851 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2852 (UNSPEC_COND_EOR): New unspecs.
2853 (optab): Add mappings for them.
2854 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2855 (sve_int_op, sve_fp_op): New int attributes.
2857 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2858 Alan Hayward <alan.hayward@arm.com>
2859 David Sherwood <david.sherwood@arm.com>
2861 * optabs.def (while_ult_optab): New optab.
2862 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2863 * internal-fn.def (WHILE_ULT): New internal function.
2864 * internal-fn.h (direct_internal_fn_supported_p): New override
2865 that takes two types as argument.
2866 * internal-fn.c (while_direct): New macro.
2867 (expand_while_optab_fn): New function.
2868 (convert_optab_supported_p): Likewise.
2869 (direct_while_optab_supported_p): New macro.
2870 * wide-int.h (wi::udiv_ceil): New function.
2871 * tree-vectorizer.h (rgroup_masks): New structure.
2872 (vec_loop_masks): New typedef.
2873 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2875 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2876 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2877 (vect_max_vf): New function.
2878 (slpeel_make_loop_iterate_ntimes): Delete.
2879 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2880 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2881 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2882 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2883 internal-fn.h, stor-layout.h and optabs-query.h.
2884 (vect_set_loop_mask): New function.
2885 (add_preheader_seq): Likewise.
2886 (add_header_seq): Likewise.
2887 (interleave_supported_p): Likewise.
2888 (vect_maybe_permute_loop_masks): Likewise.
2889 (vect_set_loop_masks_directly): Likewise.
2890 (vect_set_loop_condition_masked): Likewise.
2891 (vect_set_loop_condition_unmasked): New function, split out from
2892 slpeel_make_loop_iterate_ntimes.
2893 (slpeel_make_loop_iterate_ntimes): Rename to..
2894 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2895 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2896 (vect_do_peeling): Update call accordingly.
2897 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2899 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2900 mask_compare_type, can_fully_mask_p and fully_masked_p.
2901 (release_vec_loop_masks): New function.
2902 (_loop_vec_info): Use it to free the loop masks.
2903 (can_produce_all_loop_masks_p): New function.
2904 (vect_get_max_nscalars_per_iter): Likewise.
2905 (vect_verify_full_masking): Likewise.
2906 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2907 retries, and free the mask rgroups before retrying. Check loop-wide
2908 reasons for disallowing fully-masked loops. Make the final decision
2909 about whether use a fully-masked loop or not.
2910 (vect_estimate_min_profitable_iters): Do not assume that peeling
2911 for the number of iterations will be needed for fully-masked loops.
2912 (vectorizable_reduction): Disable fully-masked loops.
2913 (vectorizable_live_operation): Likewise.
2914 (vect_halve_mask_nunits): New function.
2915 (vect_double_mask_nunits): Likewise.
2916 (vect_record_loop_mask): Likewise.
2917 (vect_get_loop_mask): Likewise.
2918 (vect_transform_loop): Handle the case in which the final loop
2919 iteration might handle a partial vector. Call vect_set_loop_condition
2920 instead of slpeel_make_loop_iterate_ntimes.
2921 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2922 (check_load_store_masking): New function.
2923 (prepare_load_store_mask): Likewise.
2924 (vectorizable_store): Handle fully-masked loops.
2925 (vectorizable_load): Likewise.
2926 (supportable_widening_operation): Use vect_halve_mask_nunits for
2928 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2929 (vect_gen_while): New function.
2930 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2931 (aarch64_uqdec<mode>): New insn.
2933 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2934 Alan Hayward <alan.hayward@arm.com>
2935 David Sherwood <david.sherwood@arm.com>
2937 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2938 (reduc_xor_scal_optab): New optabs.
2939 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2940 (reduc_xor_scal_@var{m}): Document.
2941 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2942 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2944 * fold-const-call.c (fold_const_call): Handle them.
2945 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2946 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2947 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2948 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2949 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2950 (UNSPEC_XORV): New unspecs.
2951 (optab): Add entries for them.
2952 (BITWISEV): New int iterator.
2953 (bit_reduc_op): New int attributes.
2955 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2956 Alan Hayward <alan.hayward@arm.com>
2957 David Sherwood <david.sherwood@arm.com>
2959 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2960 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2961 * optabs.def (vec_shl_insert_optab): New optab.
2962 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2963 (duplicate_and_interleave): Likewise.
2964 * tree-vect-loop.c: Include internal-fn.h.
2965 (neutral_op_for_slp_reduction): New function, split out from
2966 get_initial_defs_for_reduction.
2967 (get_initial_def_for_reduction): Handle option 2 for variable-length
2968 vectors by loading the neutral value into a vector and then shifting
2969 the initial value into element 0.
2970 (get_initial_defs_for_reduction): Replace the code argument with
2971 the neutral value calculated by neutral_op_for_slp_reduction.
2972 Use gimple_build_vector for constant-length vectors.
2973 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2974 but the first group_size elements have a neutral value.
2975 Use duplicate_and_interleave otherwise.
2976 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2977 Update call to get_initial_defs_for_reduction. Handle SLP
2978 reductions for variable-length vectors by creating one vector
2979 result for each scalar result, with the elements associated
2980 with other scalar results stubbed out with the neutral value.
2981 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2982 Require IFN_VEC_SHL_INSERT for double reductions on
2983 variable-length vectors, or SLP reductions that have
2984 a neutral value. Require can_duplicate_and_interleave_p
2985 support for variable-length unchained SLP reductions if there
2986 is no neutral value, such as for MIN/MAX reductions. Also require
2987 the number of vector elements to be a multiple of the number of
2988 SLP statements when doing variable-length unchained SLP reductions.
2989 Update call to vect_create_epilog_for_reduction.
2990 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2991 and remove initial values.
2992 (duplicate_and_interleave): Make public.
2993 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2994 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2996 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2997 Alan Hayward <alan.hayward@arm.com>
2998 David Sherwood <david.sherwood@arm.com>
3000 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3001 (can_duplicate_and_interleave_p): New function.
3002 (vect_get_and_check_slp_defs): Take the vector of statements
3003 rather than just the current one. Remove excess parentheses.
3004 Restriction rejectinon of vect_constant_def and vect_external_def
3005 for variable-length vectors to boolean types, or types for which
3006 can_duplicate_and_interleave_p is false.
3007 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3008 (duplicate_and_interleave): New function.
3009 (vect_get_constant_vectors): Use gimple_build_vector for
3010 constant-length vectors and suitable variable-length constant
3011 vectors. Use duplicate_and_interleave for other variable-length
3012 vectors. Don't defer the update when inserting new statements.
3014 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3015 Alan Hayward <alan.hayward@arm.com>
3016 David Sherwood <david.sherwood@arm.com>
3018 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3019 min_profitable_iters doesn't go negative.
3021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3022 Alan Hayward <alan.hayward@arm.com>
3023 David Sherwood <david.sherwood@arm.com>
3025 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3026 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3027 * optabs.def (vec_mask_load_lanes_optab): New optab.
3028 (vec_mask_store_lanes_optab): Likewise.
3029 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3030 (MASK_STORE_LANES): Likewise.
3031 * internal-fn.c (mask_load_lanes_direct): New macro.
3032 (mask_store_lanes_direct): Likewise.
3033 (expand_mask_load_optab_fn): Handle masked operations.
3034 (expand_mask_load_lanes_optab_fn): New macro.
3035 (expand_mask_store_optab_fn): Handle masked operations.
3036 (expand_mask_store_lanes_optab_fn): New macro.
3037 (direct_mask_load_lanes_optab_supported_p): Likewise.
3038 (direct_mask_store_lanes_optab_supported_p): Likewise.
3039 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3041 (vect_load_lanes_supported): Likewise.
3042 * tree-vect-data-refs.c (strip_conversion): New function.
3043 (can_group_stmts_p): Likewise.
3044 (vect_analyze_data_ref_accesses): Use it instead of checking
3045 for a pair of assignments.
3046 (vect_store_lanes_supported): Take a masked_p parameter.
3047 (vect_load_lanes_supported): Likewise.
3048 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3049 vect_store_lanes_supported and vect_load_lanes_supported.
3050 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3051 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3052 parameter. Don't allow gaps for masked accesses.
3053 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3054 and vect_load_lanes_supported.
3055 (get_load_store_type): Take a masked_p parameter and update
3056 call to get_group_load_store_type.
3057 (vectorizable_store): Update call to get_load_store_type.
3058 Handle IFN_MASK_STORE_LANES.
3059 (vectorizable_load): Update call to get_load_store_type.
3060 Handle IFN_MASK_LOAD_LANES.
3062 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3063 Alan Hayward <alan.hayward@arm.com>
3064 David Sherwood <david.sherwood@arm.com>
3066 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3068 * config/aarch64/aarch64-protos.h
3069 (aarch64_sve_struct_memory_operand_p): Declare.
3070 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3071 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3072 (VPRED, vpred): Handle SVE structure modes.
3073 * config/aarch64/constraints.md (Utx): New constraint.
3074 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3075 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3076 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3077 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3078 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3079 structure modes. Split into pieces after RA.
3080 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3081 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3083 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3084 SVE structure modes.
3085 (aarch64_classify_address): Likewise.
3086 (sizetochar): Move earlier in file.
3087 (aarch64_print_operand): Handle SVE register lists.
3088 (aarch64_array_mode): New function.
3089 (aarch64_sve_struct_memory_operand_p): Likewise.
3090 (TARGET_ARRAY_MODE): Redefine.
3092 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3093 Alan Hayward <alan.hayward@arm.com>
3094 David Sherwood <david.sherwood@arm.com>
3096 * target.def (array_mode): New target hook.
3097 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3098 * doc/tm.texi: Regenerate.
3099 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3100 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3101 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3103 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3106 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3107 Alan Hayward <alan.hayward@arm.com>
3108 David Sherwood <david.sherwood@arm.com>
3110 * fold-const.c (fold_binary_loc): Check the argument types
3111 rather than the result type when testing for a vector operation.
3113 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3115 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3116 * doc/tm.texi: Regenerate.
3118 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3119 Alan Hayward <alan.hayward@arm.com>
3120 David Sherwood <david.sherwood@arm.com>
3122 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3123 (sve): Document new AArch64 extension.
3124 * doc/md.texi (w): Extend the description of the AArch64
3125 constraint to include SVE vectors.
3126 (Upl, Upa): Document new AArch64 predicate constraints.
3127 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3129 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3130 (msve-vector-bits=): New option.
3131 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3132 SVE when these are disabled.
3133 (sve): New extension.
3134 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3135 modes. Adjust their number of units based on aarch64_sve_vg.
3136 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3137 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3138 aarch64_addr_query_type.
3139 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3140 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3141 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3142 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3143 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3144 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3145 (aarch64_simd_imm_zero_p): Delete.
3146 (aarch64_check_zero_based_sve_index_immediate): Declare.
3147 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3148 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3149 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3150 (aarch64_sve_float_mul_immediate_p): Likewise.
3151 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3153 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3154 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3155 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3156 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3157 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3158 (aarch64_regmode_natural_size): Likewise.
3159 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3160 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3162 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3163 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3164 for VG and the SVE predicate registers.
3165 (V_ALIASES): Add a "z"-prefixed alias.
3166 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3167 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3168 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3169 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3170 (REG_CLASS_NAMES): Add entries for them.
3171 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3172 and the predicate registers.
3173 (aarch64_sve_vg): Declare.
3174 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3175 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3176 (REGMODE_NATURAL_SIZE): Define.
3177 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3179 * config/aarch64/aarch64.c: Include cfgrtl.h.
3180 (simd_immediate_info): Add a constructor for series vectors,
3181 and an associated step field.
3182 (aarch64_sve_vg): New variable.
3183 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3184 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3185 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3186 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3187 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3188 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3189 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3190 (aarch64_get_mask_mode): New functions.
3191 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3192 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3193 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3194 predicate modes and predicate registers. Explicitly restrict
3195 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3196 to store a vector mode if it is recognized by
3197 aarch64_classify_vector_mode.
3198 (aarch64_regmode_natural_size): New function.
3199 (aarch64_hard_regno_caller_save_mode): Return the original mode
3201 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3202 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3203 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3204 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3206 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3207 does not overlap dest if the function is frame-related. Handle
3209 (aarch64_split_add_offset): New function.
3210 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3211 them aarch64_add_offset.
3212 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3213 and update call to aarch64_sub_sp.
3214 (aarch64_add_cfa_expression): New function.
3215 (aarch64_expand_prologue): Pass extra temporary registers to the
3216 functions above. Handle the case in which we need to emit new
3217 DW_CFA_expressions for registers that were originally saved
3218 relative to the stack pointer, but now have to be expressed
3219 relative to the frame pointer.
3220 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3222 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3223 IP0 and IP1 values for SVE frames.
3224 (aarch64_expand_vec_series): New function.
3225 (aarch64_expand_sve_widened_duplicate): Likewise.
3226 (aarch64_expand_sve_const_vector): Likewise.
3227 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3228 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3229 into the register, rather than emitting a SET directly.
3230 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3231 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3232 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3233 (offset_9bit_signed_scaled_p): New functions.
3234 (aarch64_replicate_bitmask_imm): New function.
3235 (aarch64_bitmask_imm): Use it.
3236 (aarch64_cannot_force_const_mem): Reject expressions involving
3237 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3238 (aarch64_classify_index): Handle SVE indices, by requiring
3239 a plain register index with a scale that matches the element size.
3240 (aarch64_classify_address): Handle SVE addresses. Assert that
3241 the mode of the address is VOIDmode or an integer mode.
3242 Update call to aarch64_classify_symbol.
3243 (aarch64_classify_symbolic_expression): Update call to
3244 aarch64_classify_symbol.
3245 (aarch64_const_vec_all_in_range_p): New function.
3246 (aarch64_print_vector_float_operand): Likewise.
3247 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3248 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3249 and the FP immediates 1.0 and 0.5.
3250 (aarch64_print_address_internal): Handle SVE addresses.
3251 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3252 (aarch64_regno_regclass): Handle predicate registers.
3253 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3255 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3256 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3257 (aarch64_convert_sve_vector_bits): New function.
3258 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3259 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3261 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3262 Handle SVE vector and predicate modes. Accept VL-based constants
3263 that need only one temporary register, and VL offsets that require
3264 no temporary registers.
3265 (aarch64_conditional_register_usage): Mark the predicate registers
3266 as fixed if SVE isn't available.
3267 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3268 Return true for SVE vector and predicate modes.
3269 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3270 rather than an unsigned int. Handle SVE modes.
3271 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3273 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3275 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3276 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3277 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3278 (aarch64_sve_float_mul_immediate_p): New functions.
3279 (aarch64_sve_valid_immediate): New function.
3280 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3281 Explicitly reject structure modes. Check for INDEX constants.
3282 Handle PTRUE and PFALSE constants.
3283 (aarch64_check_zero_based_sve_index_immediate): New function.
3284 (aarch64_simd_imm_zero_p): Delete.
3285 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3286 vector modes. Accept constants in the range of CNT[BHWD].
3287 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3288 ask for an Advanced SIMD mode.
3289 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3290 (aarch64_simd_vector_alignment): Handle SVE predicates.
3291 (aarch64_vectorize_preferred_vector_alignment): New function.
3292 (aarch64_simd_vector_alignment_reachable): Use it instead of
3294 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3295 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3297 (MAX_VECT_LEN): Delete.
3298 (expand_vec_perm_d): Add a vec_flags field.
3299 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3300 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3301 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3303 (aarch64_evpc_rev): Rename to...
3304 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3305 (aarch64_evpc_rev_global): New function.
3306 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3307 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3309 (aarch64_evpc_sve_tbl): New function.
3310 (aarch64_expand_vec_perm_const_1): Update after rename of
3311 aarch64_evpc_rev. Handle SVE permutes too, trying
3312 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3313 than aarch64_evpc_tbl.
3314 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3315 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3316 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3317 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3318 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3319 (aarch64_expand_sve_vcond): New functions.
3320 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3321 of aarch64_vector_mode_p.
3322 (aarch64_dwarf_poly_indeterminate_value): New function.
3323 (aarch64_compute_pressure_classes): Likewise.
3324 (aarch64_can_change_mode_class): Likewise.
3325 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3326 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3327 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3328 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3329 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3330 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3331 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3332 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3334 (Dn, Dl, Dr): Accept const as well as const_vector.
3335 (Dz): Likewise. Compare against CONST0_RTX.
3336 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3337 of "vector" where appropriate.
3338 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3339 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3340 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3341 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3342 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3343 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3344 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3345 (v_int_equiv): Extend to SVE modes.
3346 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3348 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3349 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3350 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3351 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3352 (SVE_COND_FP_CMP): New int iterators.
3353 (perm_hilo): Handle the new unpack unspecs.
3354 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3356 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3357 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3358 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3359 (aarch64_equality_operator, aarch64_constant_vector_operand)
3360 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3361 (aarch64_sve_nonimmediate_operand): Likewise.
3362 (aarch64_sve_general_operand): Likewise.
3363 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3364 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3365 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3366 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3367 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3368 (aarch64_sve_float_arith_immediate): Likewise.
3369 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3370 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3371 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3372 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3373 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3374 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3375 (aarch64_sve_float_arith_operand): Likewise.
3376 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3377 (aarch64_sve_float_mul_operand): Likewise.
3378 (aarch64_sve_vec_perm_operand): Likewise.
3379 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3380 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3381 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3382 as well as const_vector.
3383 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3384 in file. Use CONST0_RTX and CONSTM1_RTX.
3385 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3386 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3387 Use aarch64_simd_imm_zero.
3388 * config/aarch64/aarch64-sve.md: New file.
3389 * config/aarch64/aarch64.md: Include it.
3390 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3391 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3392 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3393 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3394 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3395 (sve): New attribute.
3396 (enabled): Disable instructions with the sve attribute unless
3398 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3399 aarch64_expand_mov_immediate.
3400 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3401 CNT[BHSD] immediates.
3402 (movti): Split CONST_POLY_INT moves into two halves.
3403 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3404 Split additions that need a temporary here if the destination
3405 is the stack pointer.
3406 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3407 (*add<mode>3_poly_1): New instruction.
3408 (set_clobber_cc): New expander.
3410 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3412 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3413 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3414 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3415 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3416 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3417 Change innermode from fixed_mode_size to machine_mode.
3418 (simplify_subreg): Update call accordingly. Handle a constant-sized
3419 subreg of a variable-length CONST_VECTOR.
3421 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3422 Alan Hayward <alan.hayward@arm.com>
3423 David Sherwood <david.sherwood@arm.com>
3425 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3426 (add_offset_to_base): New function, split out from...
3427 (create_mem_ref): ...here. When handling a scale other than 1,
3428 check first whether the address is valid without the offset.
3429 Add it into the base if so, leaving the index and scale as-is.
3431 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3434 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3435 fold_for_warn before checking if arg2 is INTEGER_CST.
3437 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3439 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3440 (store_multiple_operation): Delete.
3441 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3442 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3443 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3444 guarded by TARGET_STRING.
3445 (rs6000_output_load_multiple): Delete.
3446 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3447 OPTION_MASK_STRING / TARGET_STRING handling.
3448 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3449 (const rs6000_opt_masks) <"string">: Change mask to 0.
3450 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3451 (MASK_STRING): Delete.
3452 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3454 (load_multiple): Delete.
3461 (store_multiple): Delete.
3468 (movmemsi_8reg): Delete.
3469 (corresponding unnamed define_insn): Delete.
3470 (movmemsi_6reg): Delete.
3471 (corresponding unnamed define_insn): Delete.
3472 (movmemsi_4reg): Delete.
3473 (corresponding unnamed define_insn): Delete.
3474 (movmemsi_2reg): Delete.
3475 (corresponding unnamed define_insn): Delete.
3476 (movmemsi_1reg): Delete.
3477 (corresponding unnamed define_insn): Delete.
3478 * config/rs6000/rs6000.opt (mno-string): New.
3479 (mstring): Replace by deprecation warning stub.
3480 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3482 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3484 * regrename.c (regrename_do_replace): If replacing the same
3485 reg multiple times, try to reuse last created gen_raw_REG.
3488 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3489 main to workaround a bug in GDB.
3491 2018-01-12 Tom de Vries <tom@codesourcery.com>
3494 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3496 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3498 PR rtl-optimization/80481
3499 * ira-color.c (get_cap_member): New function.
3500 (allocnos_conflict_by_live_ranges_p): Use it.
3501 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3502 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3504 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3507 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3508 (*saddl_se_1): Ditto.
3510 (*ssubl_se_1): Ditto.
3512 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3514 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3515 rather than wi::to_widest for DR_INITs.
3516 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3517 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3518 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3520 (vect_analyze_group_access_1): Note that here.
3522 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3524 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3525 polynomial type sizes.
3527 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3529 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3530 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3531 (gimple_add_tmp_var): Likewise.
3533 2018-01-12 Martin Liska <mliska@suse.cz>
3535 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3536 (gimple_alloc_sizes): Likewise.
3537 (dump_gimple_statistics): Use PRIu64 in printf format.
3538 * gimple.h: Change uint64_t to int.
3540 2018-01-12 Martin Liska <mliska@suse.cz>
3542 * tree-core.h: Use uint64_t instead of int.
3543 * tree.c (tree_node_counts): Likewise.
3544 (tree_node_sizes): Likewise.
3545 (dump_tree_statistics): Use PRIu64 in printf format.
3547 2018-01-12 Martin Liska <mliska@suse.cz>
3549 * Makefile.in: As qsort_chk is implemented in vec.c, add
3550 vec.o to linkage of gencfn-macros.
3551 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3552 passing the info to record_node_allocation_statistics.
3553 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3555 * ggc-common.c (struct ggc_usage): Add operator== and use
3556 it in operator< and compare function.
3557 * mem-stats.h (struct mem_usage): Likewise.
3558 * vec.c (struct vec_usage): Remove operator< and compare
3559 function. Can be simply inherited.
3561 2018-01-12 Martin Jambor <mjambor@suse.cz>
3564 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3565 * tree-ssa-math-opts.c: Include domwalk.h.
3566 (convert_mult_to_fma_1): New function.
3567 (fma_transformation_info): New type.
3568 (fma_deferring_state): Likewise.
3569 (cancel_fma_deferring): New function.
3570 (result_of_phi): Likewise.
3571 (last_fma_candidate_feeds_initial_phi): Likewise.
3572 (convert_mult_to_fma): Added deferring logic, split actual
3573 transformation to convert_mult_to_fma_1.
3574 (math_opts_dom_walker): New type.
3575 (math_opts_dom_walker::after_dom_children): New method, body moved
3576 here from pass_optimize_widening_mul::execute, added deferring logic
3578 (pass_optimize_widening_mul::execute): Moved most of code to
3579 math_opts_dom_walker::after_dom_children.
3580 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3581 * config/i386/i386.c (ix86_option_override_internal): Added
3582 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3584 2018-01-12 Richard Biener <rguenther@suse.de>
3587 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3588 inline instance vars.
3590 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3593 * config/rx/rx.c (rx_is_restricted_memory_address):
3596 2018-01-12 Richard Biener <rguenther@suse.de>
3598 PR tree-optimization/80846
3599 * target.def (split_reduction): New target hook.
3600 * targhooks.c (default_split_reduction): New function.
3601 * targhooks.h (default_split_reduction): Declare.
3602 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3603 target requests first reduce vectors by combining low and high
3605 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3606 (get_vectype_for_scalar_type_and_size): Export.
3607 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3608 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3609 * doc/tm.texi: Regenerate.
3610 * config/i386/i386.c (ix86_split_reduction): Implement
3611 TARGET_VECTORIZE_SPLIT_REDUCTION.
3613 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3616 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3617 in PIC mode except for TARGET_VXWORKS_RTP.
3618 * config/sparc/sparc.c: Include cfgrtl.h.
3619 (TARGET_INIT_PIC_REG): Define.
3620 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3621 (sparc_pic_register_p): New predicate.
3622 (sparc_legitimate_address_p): Use it.
3623 (sparc_legitimize_pic_address): Likewise.
3624 (sparc_delegitimize_address): Likewise.
3625 (sparc_mode_dependent_address_p): Likewise.
3626 (gen_load_pcrel_sym): Remove 4th parameter.
3627 (load_got_register): Adjust call to above. Remove obsolete stuff.
3628 (sparc_expand_prologue): Do not call load_got_register here.
3629 (sparc_flat_expand_prologue): Likewise.
3630 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3631 (sparc_use_pseudo_pic_reg): New function.
3632 (sparc_init_pic_reg): Likewise.
3633 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3634 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3636 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3638 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3639 Add item for branch_cost.
3641 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3643 PR rtl-optimization/83565
3644 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3645 not extend the result to a larger mode for rotate operations.
3646 (num_sign_bit_copies1): Likewise.
3648 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3651 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3653 Use values-Xc.o for -pedantic.
3654 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3656 2018-01-12 Martin Liska <mliska@suse.cz>
3659 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3661 (possible_polymorphic_call_targets): Use it.
3662 (ipa_devirt): Likewise.
3664 2018-01-12 Martin Liska <mliska@suse.cz>
3666 * profile-count.h (enum profile_quality): Use 0 as invalid
3667 enum value of profile_quality.
3669 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3671 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3672 -mext-string options.
3674 2018-01-12 Richard Biener <rguenther@suse.de>
3676 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3677 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3678 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3680 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3682 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3684 * configure.ac (--with-long-double-format): Add support for the
3685 configuration option to change the default long double format on
3687 * config.gcc (powerpc*-linux*-*): Likewise.
3688 * configure: Regenerate.
3689 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3690 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3691 used without modification.
3693 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3695 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3696 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3697 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3698 MISC_BUILTIN_SPEC_BARRIER.
3699 (rs6000_init_builtins): Likewise.
3700 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3702 (speculation_barrier): New define_insn.
3703 * doc/extend.texi: Document __builtin_speculation_barrier.
3705 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3708 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3709 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3710 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3712 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3713 integral modes instead of "ss" and "sd".
3714 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3715 vectors with 32-bit and 64-bit elements.
3716 (vecdupssescalarmodesuffix): New mode attribute.
3717 (vec_dup<mode>): Use it.
3719 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3722 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3723 frame if argument is passed on stack.
3725 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3728 * ree.c (combine_reaching_defs): Optimize also
3729 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3730 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3732 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3735 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3737 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3740 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3741 after they are computed.
3743 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3745 PR tree-optimization/83695
3746 * gimple-loop-linterchange.cc
3747 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3748 reset cached scev information after interchange.
3749 (pass_linterchange::execute): Remove call to scev_reset_htab.
3751 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3753 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3754 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3755 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3756 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3757 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3758 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3759 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3760 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3761 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3762 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3763 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3764 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3765 (V_lane_reg): Likewise.
3766 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3768 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3769 (vfmal_lane_low<mode>_intrinsic,
3770 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3771 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3772 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3773 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3774 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3775 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3777 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3779 * config/arm/arm-cpus.in (fp16fml): New feature.
3780 (ALL_SIMD): Add fp16fml.
3781 (armv8.2-a): Add fp16fml as an option.
3782 (armv8.3-a): Likewise.
3783 (armv8.4-a): Add fp16fml as part of fp16.
3784 * config/arm/arm.h (TARGET_FP16FML): Define.
3785 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3787 * config/arm/arm-modes.def (V2HF): Define.
3788 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3789 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3790 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3791 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3792 vfmsl_low, vfmsl_high): New set of builtins.
3793 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3794 (vfml_op): New code attribute.
3795 (VFMLHALVES): New int iterator.
3796 (VFML, VFMLSEL): New mode attributes.
3797 (V_reg): Define mapping for V2HF.
3798 (V_hi, V_lo): New mode attributes.
3799 (VF_constraint): Likewise.
3800 (vfml_half, vfml_half_selector): New int attributes.
3801 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3803 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3804 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3806 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3807 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3808 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3809 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3811 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3812 Document new effective target and option set.
3814 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3816 * config/arm/arm-cpus.in (armv8_4): New feature.
3817 (ARMv8_4a): New fgroup.
3818 (armv8.4-a): New arch.
3819 * config/arm/arm-tables.opt: Regenerate.
3820 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3821 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3822 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3823 Add matching rules for -march=armv8.4-a and extensions.
3824 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3826 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3829 * config/rx/rx.md (BW): New mode attribute.
3830 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3832 2018-01-11 Richard Biener <rguenther@suse.de>
3834 PR tree-optimization/83435
3835 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3836 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3837 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3839 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3840 Alan Hayward <alan.hayward@arm.com>
3841 David Sherwood <david.sherwood@arm.com>
3843 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3845 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3846 (aarch64_print_address_internal): Use it to check for a zero offset.
3848 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3849 Alan Hayward <alan.hayward@arm.com>
3850 David Sherwood <david.sherwood@arm.com>
3852 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3853 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3854 Return a poly_int64 rather than a HOST_WIDE_INT.
3855 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3856 rather than a HOST_WIDE_INT.
3857 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3858 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3859 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3860 final_offset from HOST_WIDE_INT to poly_int64.
3861 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3862 to_constant when getting the number of units in an Advanced SIMD
3864 (aarch64_builtin_vectorized_function): Check for a constant number
3866 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3868 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3869 attribute instead of GET_MODE_NUNITS.
3870 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3871 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3872 GET_MODE_SIZE for fixed-size registers.
3873 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3874 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3875 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3876 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3877 (aarch64_print_operand, aarch64_print_address_internal)
3878 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3879 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3880 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3881 Handle polynomial GET_MODE_SIZE.
3882 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3883 wider than SImode without modification.
3884 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3885 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3886 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3887 passing and returning SVE modes.
3888 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3889 rather than GEN_INT.
3890 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3891 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3892 (aarch64_allocate_and_probe_stack_space): Likewise.
3893 (aarch64_layout_frame): Cope with polynomial offsets.
3894 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3895 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3897 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3898 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3899 poly_int64 rather than a HOST_WIDE_INT.
3900 (aarch64_get_separate_components, aarch64_process_components)
3901 (aarch64_expand_prologue, aarch64_expand_epilogue)
3902 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3903 (aarch64_anchor_offset): New function, split out from...
3904 (aarch64_legitimize_address): ...here.
3905 (aarch64_builtin_vectorization_cost): Handle polynomial
3906 TYPE_VECTOR_SUBPARTS.
3907 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3909 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3910 number of elements from the PARALLEL rather than the mode.
3911 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3912 rather than GET_MODE_BITSIZE.
3913 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3914 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3915 (aarch64_expand_vec_perm_const_1): Handle polynomial
3916 d->perm.length () and d->perm elements.
3917 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3918 Apply to_constant to d->perm elements.
3919 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3920 polynomial CONST_VECTOR_NUNITS.
3921 (aarch64_move_pointer): Take amount as a poly_int64 rather
3923 (aarch64_progress_pointer): Avoid temporary variable.
3924 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3925 the mode attribute instead of GET_MODE.
3927 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3928 Alan Hayward <alan.hayward@arm.com>
3929 David Sherwood <david.sherwood@arm.com>
3931 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3932 x exists before using it.
3933 (aarch64_add_constant_internal): Rename to...
3934 (aarch64_add_offset_1): ...this. Replace regnum with separate
3935 src and dest rtxes. Handle the case in which they're different,
3936 including when the offset is zero. Replace scratchreg with an rtx.
3937 Use 2 additions if there is no spare register into which we can
3938 move a 16-bit constant.
3939 (aarch64_add_constant): Delete.
3940 (aarch64_add_offset): Replace reg with separate src and dest
3941 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3942 Use aarch64_add_offset_1.
3943 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3944 an rtx rather than an int. Take the delta as a poly_int64
3945 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3946 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3947 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3948 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3949 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3951 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3952 aarch64_add_constant.
3954 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3956 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3957 Use scalar_float_mode.
3959 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3961 * config/aarch64/aarch64-simd.md
3962 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3963 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3964 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3965 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3966 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3967 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3968 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3969 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3970 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3971 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3973 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3976 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3977 targ_options->x_arm_arch_string is non NULL.
3979 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3981 * config/aarch64/aarch64.h
3982 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3984 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3987 * expmed.c (emit_store_flag_force): Swap if const op0
3988 and change VOIDmode to mode of op0.
3990 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3992 PR rtl-optimization/83761
3993 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3994 than bytes to mode_for_size.
3996 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3999 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4000 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4003 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4006 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4007 when in layout mode.
4008 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4009 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4012 2018-01-10 Michael Collison <michael.collison@arm.com>
4014 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4015 * config/aarch64/aarch64-option-extension.def: Add
4016 AARCH64_OPT_EXTENSION of 'fp16fml'.
4017 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4018 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4019 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4020 * config/aarch64/constraints.md (Ui7): New constraint.
4021 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4022 (VFMLA_SEL_W): Ditto.
4025 (VFMLA16_LOW): New int iterator.
4026 (VFMLA16_HIGH): Ditto.
4027 (UNSPEC_FMLAL): New unspec.
4028 (UNSPEC_FMLSL): Ditto.
4029 (UNSPEC_FMLAL2): Ditto.
4030 (UNSPEC_FMLSL2): Ditto.
4031 (f16mac): New code attribute.
4032 * config/aarch64/aarch64-simd-builtins.def
4033 (aarch64_fmlal_lowv2sf): Ditto.
4034 (aarch64_fmlsl_lowv2sf): Ditto.
4035 (aarch64_fmlalq_lowv4sf): Ditto.
4036 (aarch64_fmlslq_lowv4sf): Ditto.
4037 (aarch64_fmlal_highv2sf): Ditto.
4038 (aarch64_fmlsl_highv2sf): Ditto.
4039 (aarch64_fmlalq_highv4sf): Ditto.
4040 (aarch64_fmlslq_highv4sf): Ditto.
4041 (aarch64_fmlal_lane_lowv2sf): Ditto.
4042 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4043 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4044 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4045 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4046 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4047 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4048 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4049 (aarch64_fmlal_lane_highv2sf): Ditto.
4050 (aarch64_fmlsl_lane_highv2sf): Ditto.
4051 (aarch64_fmlal_laneq_highv2sf): Ditto.
4052 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4053 (aarch64_fmlalq_lane_highv4sf): Ditto.
4054 (aarch64_fmlsl_lane_highv4sf): Ditto.
4055 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4056 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4057 * config/aarch64/aarch64-simd.md:
4058 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4059 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4060 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4061 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4062 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4063 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4064 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4065 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4066 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4067 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4068 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4069 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4070 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4071 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4072 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4073 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4074 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4075 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4076 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4077 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4078 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4079 (vfmlsl_low_u32): Ditto.
4080 (vfmlalq_low_u32): Ditto.
4081 (vfmlslq_low_u32): Ditto.
4082 (vfmlal_high_u32): Ditto.
4083 (vfmlsl_high_u32): Ditto.
4084 (vfmlalq_high_u32): Ditto.
4085 (vfmlslq_high_u32): Ditto.
4086 (vfmlal_lane_low_u32): Ditto.
4087 (vfmlsl_lane_low_u32): Ditto.
4088 (vfmlal_laneq_low_u32): Ditto.
4089 (vfmlsl_laneq_low_u32): Ditto.
4090 (vfmlalq_lane_low_u32): Ditto.
4091 (vfmlslq_lane_low_u32): Ditto.
4092 (vfmlalq_laneq_low_u32): Ditto.
4093 (vfmlslq_laneq_low_u32): Ditto.
4094 (vfmlal_lane_high_u32): Ditto.
4095 (vfmlsl_lane_high_u32): Ditto.
4096 (vfmlal_laneq_high_u32): Ditto.
4097 (vfmlsl_laneq_high_u32): Ditto.
4098 (vfmlalq_lane_high_u32): Ditto.
4099 (vfmlslq_lane_high_u32): Ditto.
4100 (vfmlalq_laneq_high_u32): Ditto.
4101 (vfmlslq_laneq_high_u32): Ditto.
4102 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4103 (AARCH64_FL_FOR_ARCH8_4): New.
4104 (AARCH64_ISA_F16FML): New ISA flag.
4105 (TARGET_F16FML): New feature flag for fp16fml.
4106 (doc/invoke.texi): Document new fp16fml option.
4108 2018-01-10 Michael Collison <michael.collison@arm.com>
4110 * config/aarch64/aarch64-builtins.c:
4111 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4112 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4113 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4114 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4115 (AARCH64_ISA_SHA3): New ISA flag.
4116 (TARGET_SHA3): New feature flag for sha3.
4117 * config/aarch64/iterators.md (sha512_op): New int attribute.
4118 (CRYPTO_SHA512): New int iterator.
4119 (UNSPEC_SHA512H): New unspec.
4120 (UNSPEC_SHA512H2): Ditto.
4121 (UNSPEC_SHA512SU0): Ditto.
4122 (UNSPEC_SHA512SU1): Ditto.
4123 * config/aarch64/aarch64-simd-builtins.def
4124 (aarch64_crypto_sha512hqv2di): New builtin.
4125 (aarch64_crypto_sha512h2qv2di): Ditto.
4126 (aarch64_crypto_sha512su0qv2di): Ditto.
4127 (aarch64_crypto_sha512su1qv2di): Ditto.
4128 (aarch64_eor3qv8hi): Ditto.
4129 (aarch64_rax1qv2di): Ditto.
4130 (aarch64_xarqv2di): Ditto.
4131 (aarch64_bcaxqv8hi): Ditto.
4132 * config/aarch64/aarch64-simd.md:
4133 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4134 (aarch64_crypto_sha512su0qv2di): Ditto.
4135 (aarch64_crypto_sha512su1qv2di): Ditto.
4136 (aarch64_eor3qv8hi): Ditto.
4137 (aarch64_rax1qv2di): Ditto.
4138 (aarch64_xarqv2di): Ditto.
4139 (aarch64_bcaxqv8hi): Ditto.
4140 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4141 (vsha512h2q_u64): Ditto.
4142 (vsha512su0q_u64): Ditto.
4143 (vsha512su1q_u64): Ditto.
4144 (veor3q_u16): Ditto.
4145 (vrax1q_u64): Ditto.
4147 (vbcaxq_u16): Ditto.
4148 * config/arm/types.md (crypto_sha512): New type attribute.
4149 (crypto_sha3): Ditto.
4150 (doc/invoke.texi): Document new sha3 option.
4152 2018-01-10 Michael Collison <michael.collison@arm.com>
4154 * config/aarch64/aarch64-builtins.c:
4155 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4156 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4157 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4158 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4159 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4160 (AARCH64_ISA_SM4): New ISA flag.
4161 (TARGET_SM4): New feature flag for sm4.
4162 * config/aarch64/aarch64-simd-builtins.def
4163 (aarch64_sm3ss1qv4si): Ditto.
4164 (aarch64_sm3tt1aq4si): Ditto.
4165 (aarch64_sm3tt1bq4si): Ditto.
4166 (aarch64_sm3tt2aq4si): Ditto.
4167 (aarch64_sm3tt2bq4si): Ditto.
4168 (aarch64_sm3partw1qv4si): Ditto.
4169 (aarch64_sm3partw2qv4si): Ditto.
4170 (aarch64_sm4eqv4si): Ditto.
4171 (aarch64_sm4ekeyqv4si): Ditto.
4172 * config/aarch64/aarch64-simd.md:
4173 (aarch64_sm3ss1qv4si): Ditto.
4174 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4175 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4176 (aarch64_sm4eqv4si): Ditto.
4177 (aarch64_sm4ekeyqv4si): Ditto.
4178 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4179 (sm3part_op): Ditto.
4180 (CRYPTO_SM3TT): Ditto.
4181 (CRYPTO_SM3PART): Ditto.
4182 (UNSPEC_SM3SS1): New unspec.
4183 (UNSPEC_SM3TT1A): Ditto.
4184 (UNSPEC_SM3TT1B): Ditto.
4185 (UNSPEC_SM3TT2A): Ditto.
4186 (UNSPEC_SM3TT2B): Ditto.
4187 (UNSPEC_SM3PARTW1): Ditto.
4188 (UNSPEC_SM3PARTW2): Ditto.
4189 (UNSPEC_SM4E): Ditto.
4190 (UNSPEC_SM4EKEY): Ditto.
4191 * config/aarch64/constraints.md (Ui2): New constraint.
4192 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4193 * config/arm/types.md (crypto_sm3): New type attribute.
4194 (crypto_sm4): Ditto.
4195 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4196 (vsm3tt1aq_u32): Ditto.
4197 (vsm3tt1bq_u32): Ditto.
4198 (vsm3tt2aq_u32): Ditto.
4199 (vsm3tt2bq_u32): Ditto.
4200 (vsm3partw1q_u32): Ditto.
4201 (vsm3partw2q_u32): Ditto.
4202 (vsm4eq_u32): Ditto.
4203 (vsm4ekeyq_u32): Ditto.
4204 (doc/invoke.texi): Document new sm4 option.
4206 2018-01-10 Michael Collison <michael.collison@arm.com>
4208 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4209 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4210 (AARCH64_FL_FOR_ARCH8_4): New.
4211 (AARCH64_FL_V8_4): New flag.
4212 (doc/invoke.texi): Document new armv8.4-a option.
4214 2018-01-10 Michael Collison <michael.collison@arm.com>
4216 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4217 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4218 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4219 * config/aarch64/aarch64-option-extension.def: Add
4220 AARCH64_OPT_EXTENSION of 'sha2'.
4221 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4222 (crypto): Disable sha2 and aes if crypto disabled.
4223 (crypto): Enable aes and sha2 if enabled.
4224 (simd): Disable sha2 and aes if simd disabled.
4225 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4227 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4228 (TARGET_SHA2): New feature flag for sha2.
4229 (TARGET_AES): New feature flag for aes.
4230 * config/aarch64/aarch64-simd.md:
4231 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4232 conditional on TARGET_AES.
4233 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4234 (aarch64_crypto_sha1hsi): Make pattern conditional
4236 (aarch64_crypto_sha1hv4si): Ditto.
4237 (aarch64_be_crypto_sha1hv4si): Ditto.
4238 (aarch64_crypto_sha1su1v4si): Ditto.
4239 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4240 (aarch64_crypto_sha1su0v4si): Ditto.
4241 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4242 (aarch64_crypto_sha256su0v4si): Ditto.
4243 (aarch64_crypto_sha256su1v4si): Ditto.
4244 (doc/invoke.texi): Document new aes and sha2 options.
4246 2018-01-10 Martin Sebor <msebor@redhat.com>
4248 PR tree-optimization/83781
4249 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4252 2018-01-11 Martin Sebor <msebor@gmail.com>
4253 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4255 PR tree-optimization/83501
4256 PR tree-optimization/81703
4258 * tree-ssa-strlen.c (get_string_cst): Rename...
4259 (get_string_len): ...to this. Handle global constants.
4260 (handle_char_store): Adjust.
4262 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4263 Jim Wilson <jimw@sifive.com>
4265 * config/riscv/riscv-protos.h (riscv_output_return): New.
4266 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4267 (riscv_attribute_table, riscv_output_return),
4268 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4269 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4270 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4272 (riscv_expand_prologue): Add early return for naked function.
4273 (riscv_expand_epilogue): Likewise.
4274 (riscv_function_ok_for_sibcall): Return false for naked function.
4275 (riscv_set_current_function): New.
4276 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4277 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4278 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4279 * doc/extend.texi (RISC-V Function Attributes): New.
4281 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4283 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4284 check for 128-bit long double before checking TCmode.
4285 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4286 128-bit long doubles before checking TFmode or TCmode.
4287 (FLOAT128_IBM_P): Likewise.
4289 2018-01-10 Martin Sebor <msebor@redhat.com>
4291 PR tree-optimization/83671
4292 * builtins.c (c_strlen): Unconditionally return zero for the empty
4294 Use -Warray-bounds for warnings.
4295 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4296 for non-constant array indices with COMPONENT_REF, arrays of
4297 arrays, and pointers to arrays.
4298 (gimple_fold_builtin_strlen): Determine and set length range for
4299 non-constant character arrays.
4301 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4304 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4307 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4309 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4311 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4314 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4315 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4316 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4317 indexed_or_indirect_operand predicate.
4318 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4319 (*vsx_le_perm_load_v8hi): Likewise.
4320 (*vsx_le_perm_load_v16qi): Likewise.
4321 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4322 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4323 (*vsx_le_perm_store_v8hi): Likewise.
4324 (*vsx_le_perm_store_v16qi): Likewise.
4325 (eight unnamed splitters): Likewise.
4327 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4329 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4330 * config/rs6000/emmintrin.h: Likewise.
4331 * config/rs6000/mmintrin.h: Likewise.
4332 * config/rs6000/xmmintrin.h: Likewise.
4334 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4337 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4339 * tree.c (tree_nop_conversion): Return true for location wrapper
4341 (maybe_wrap_with_location): New function.
4342 (selftest::check_strip_nops): New function.
4343 (selftest::test_location_wrappers): New function.
4344 (selftest::tree_c_tests): Call it.
4345 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4346 (maybe_wrap_with_location): New decl.
4347 (EXPR_LOCATION_WRAPPER_P): New macro.
4348 (location_wrapper_p): New inline function.
4349 (tree_strip_any_location_wrapper): New inline function.
4351 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4354 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4355 stack_realign_offset for the largest alignment of stack slot
4357 (ix86_find_max_used_stack_alignment): New function.
4358 (ix86_finalize_stack_frame_flags): Use it. Set
4359 max_used_stack_alignment if we don't realign stack.
4360 * config/i386/i386.h (machine_function): Add
4361 max_used_stack_alignment.
4363 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4365 * config/arm/arm.opt (-mbranch-cost): New option.
4366 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4369 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4372 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4373 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4375 2018-01-10 Richard Biener <rguenther@suse.de>
4378 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4379 early out so it also covers the case where we have a non-NULL
4382 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4384 PR tree-optimization/83753
4385 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4386 for non-strided grouped accesses if the number of elements is 1.
4388 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4391 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4392 * i386.h (TARGET_USE_GATHER): Define.
4393 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4395 2018-01-10 Martin Liska <mliska@suse.cz>
4398 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4399 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4401 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4402 CLEANUP_NO_PARTITIONING is not set.
4404 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4406 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4407 for vectors, as a partial revert of r254296.
4408 * rtl.h (const_vec_p): Delete.
4409 (const_vec_duplicate_p): Don't test for vector CONSTs.
4410 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4411 * expmed.c (make_tree): Likewise.
4414 * common.md (E, F): Use CONSTANT_P instead of checking for
4416 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4417 checking for CONST_VECTOR.
4419 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4422 * predict.c (force_edge_cold): Handle in more sane way edges
4425 2018-01-09 Carl Love <cel@us.ibm.com>
4427 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4429 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4430 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4431 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4432 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4433 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4434 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4435 * config/rs6000/rs6000-protos.h: Add extern defition for
4436 rs6000_generate_float2_double_code.
4437 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4439 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4440 (float2_v2df): Add define_expand.
4442 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4445 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4446 op_mode in the force_to_mode call.
4448 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4450 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4451 instead of checking each element individually.
4452 (aarch64_evpc_uzp): Likewise.
4453 (aarch64_evpc_zip): Likewise.
4454 (aarch64_evpc_ext): Likewise.
4455 (aarch64_evpc_rev): Likewise.
4456 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4457 instead of checking each element individually. Return true without
4459 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4460 whether all selected elements come from the same input, instead of
4461 checking each element individually. Remove calls to gen_rtx_REG,
4462 start_sequence and end_sequence and instead assert that no rtl is
4465 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4467 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4468 order of HIGH and CONST checks.
4470 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4472 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4473 if the destination isn't an SSA_NAME.
4475 2018-01-09 Richard Biener <rguenther@suse.de>
4477 PR tree-optimization/83668
4478 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4480 (canonicalize_loop_form): ... here, renamed from ...
4481 (canonicalize_loop_closed_ssa_form): ... this and amended to
4482 swap successor edges for loop exit blocks to make us use
4483 the RPO order we need for initial schedule generation.
4485 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4487 PR tree-optimization/64811
4488 * match.pd: When optimizing comparisons with Inf, avoid
4489 introducing or losing exceptions from comparisons with NaN.
4491 2018-01-09 Martin Liska <mliska@suse.cz>
4494 * asan.c (shadow_mem_size): Add gcc_assert.
4496 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4498 Don't save registers in main().
4501 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4502 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4503 * config/avr/avr.c (avr_set_current_function): Don't error if
4504 naked, OS_task or OS_main are specified at the same time.
4505 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4507 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4509 * common/config/avr/avr-common.c (avr_option_optimization_table):
4510 Switch on -mmain-is-OS_task for optimizing compilations.
4512 2018-01-09 Richard Biener <rguenther@suse.de>
4514 PR tree-optimization/83572
4515 * graphite.c: Include cfganal.h.
4516 (graphite_transform_loops): Connect infinite loops to exit
4517 and remove fake edges at the end.
4519 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4521 * ipa-inline.c (edge_badness): Revert accidental checkin.
4523 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4526 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4527 symbols; not inline clones.
4529 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4532 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4533 hard registers. Formatting fixes.
4535 PR preprocessor/83722
4536 * gcc.c (try_generate_repro): Pass
4537 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4538 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4541 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4542 Kito Cheng <kito.cheng@gmail.com>
4544 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4545 (riscv_leaf_function_p): Delete.
4546 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4548 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4550 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4552 (do_ifelse): New function.
4553 (do_isel): New function.
4554 (do_sub3): New function.
4555 (do_add3): New function.
4556 (do_load_mask_compare): New function.
4557 (do_overlap_load_compare): New function.
4558 (expand_compare_loop): New function.
4559 (expand_block_compare): Call expand_compare_loop() when appropriate.
4560 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4562 (-mblock-compare-inline-loop-limit): New option.
4564 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4567 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4568 Reverse order of second and third operands in first alternative.
4569 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4570 of first and second elements in UNSPEC_VPERMR vector.
4571 (altivec_expand_vec_perm_le): Likewise.
4573 2018-01-08 Jeff Law <law@redhat.com>
4575 PR rtl-optimizatin/81308
4576 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4577 (process_switch): If group_case_labels makes a change, then set
4579 (pass_convert_switch::execute): If a switch is converted, then
4580 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4582 PR rtl-optimization/81308
4583 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4586 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4588 PR target/83663 - Revert r255946
4589 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4590 generation for cases where splatting a value is not useful.
4591 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4592 across a vec_duplicate and a paradoxical subreg forming a vector
4593 mode to a vec_concat.
4595 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4597 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4598 -march=armv8.3-a variants.
4599 * config/arm/t-multilib: Likewise.
4600 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4602 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4604 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4606 (cceq_ior_compare_complement): Give it a name so I can use it, and
4607 change boolean_or_operator predicate to boolean_operator so it can
4608 be used to generate a crand.
4609 (eqne): New code iterator.
4610 (bd/bd_neg): New code_attrs.
4611 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4612 a single define_insn.
4613 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4614 decrement (bdnzt/bdnzf/bdzt/bdzf).
4615 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4616 with the new names of the branch decrement patterns, and added the
4617 names of the branch decrement conditional patterns.
4619 2018-01-08 Richard Biener <rguenther@suse.de>
4621 PR tree-optimization/83563
4622 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4625 2018-01-08 Richard Biener <rguenther@suse.de>
4628 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4630 2018-01-08 Richard Biener <rguenther@suse.de>
4632 PR tree-optimization/83685
4633 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4634 references to abnormals.
4636 2018-01-08 Richard Biener <rguenther@suse.de>
4639 * dwarf2out.c (output_indirect_strings): Handle empty
4640 skeleton_debug_str_hash.
4641 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4643 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4645 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4646 (emit_store_direct): Likewise.
4647 (arc_trampoline_adjust_address): Likewise.
4648 (arc_asm_trampoline_template): New function.
4649 (arc_initialize_trampoline): Use asm_trampoline_template.
4650 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4651 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4652 * config/arc/arc.md (flush_icache): Delete pattern.
4654 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4656 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4657 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4660 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4663 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4664 by not USED_FOR_TARGET.
4665 (make_pass_resolve_sw_modes): Likewise.
4667 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4669 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4672 2018-01-08 Richard Biener <rguenther@suse.de>
4675 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4677 2018-01-08 Richard Biener <rguenther@suse.de>
4680 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4682 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4685 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4686 basic blocks with a small number of successors.
4687 (convert_control_dep_chain_into_preds): Improve handling of
4689 (dump_predicates): Split apart into...
4690 (dump_pred_chain): ...here...
4691 (dump_pred_info): ...and here.
4692 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4693 (can_chain_union_be_invalidated_p): Improve check for invalidation
4695 (uninit_uses_cannot_happen): Avoid unnecessary if
4696 convert_control_dep_chain_into_preds yielded nothing.
4698 2018-01-06 Martin Sebor <msebor@redhat.com>
4700 PR tree-optimization/83640
4701 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4702 subtracting negative offset from size.
4703 (builtin_access::overlap): Adjust offset bounds of the access to fall
4704 within the size of the object if possible.
4706 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4708 PR rtl-optimization/83699
4709 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4710 extract_bit_field_as_subreg to cases in which the extracted
4711 value is also a vector.
4713 * lra-constraints.c (process_alt_operands): Test for the equivalence
4714 substitutions when detecting a possible reload cycle.
4716 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4719 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4720 by default if flag_selective_schedling{,2}. Formatting fixes.
4722 PR rtl-optimization/83682
4723 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4724 if it has non-VECTOR_MODE element mode.
4725 (vec_duplicate_p): Likewise.
4728 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4729 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4731 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4734 * config/i386/i386-builtin.def
4735 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4736 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4737 Require also OPTION_MASK_ISA_AVX512F in addition to
4738 OPTION_MASK_ISA_GFNI.
4739 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4740 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4741 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4742 to OPTION_MASK_ISA_GFNI.
4743 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4744 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4745 OPTION_MASK_ISA_AVX512BW.
4746 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4747 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4748 addition to OPTION_MASK_ISA_GFNI.
4749 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4750 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4751 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4752 to OPTION_MASK_ISA_GFNI.
4753 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4754 a requirement for all ISAs rather than any of them with a few
4756 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4758 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4759 bitmasks to be enabled with 3 exceptions, instead of requiring any
4760 enabled ISA with lots of exceptions.
4761 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4762 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4763 Change avx512bw in isa attribute to avx512f.
4764 * config/i386/sgxintrin.h: Add license boilerplate.
4765 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4766 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4767 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4768 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4770 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4771 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4772 temporarily sse2 rather than sse if not enabled already.
4775 * config/i386/sse.md (VI248_VLBW): Rename to ...
4776 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4777 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4778 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4779 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4780 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4781 mode iterator instead of VI248_VLBW.
4783 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4785 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4786 (record_modified): Skip clobbers; add debug output.
4787 (param_change_prob): Use sreal frequencies.
4789 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4791 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4792 punt for user-aligned variables.
4794 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4796 * tree-chrec.c (chrec_contains_symbols): Return true for
4799 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4802 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4803 of (x|y) == x for BICS pattern.
4805 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4807 PR tree-optimization/83605
4808 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4809 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4812 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4814 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4815 * config/epiphany/rtems.h: New file.
4817 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4818 Uros Bizjak <ubizjak@gmail.com>
4821 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4822 QIreg_operand instead of register_operand predicate.
4823 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4824 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4825 comments instead of -fmitigate[-_]rop.
4827 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4830 * cgraphunit.c (symbol_table::compile): Switch to text_section
4831 before calling assembly_start debug hook.
4832 * run-rtl-passes.c (run_rtl_passes): Likewise.
4835 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4837 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4838 range_int_cst_p rather than !symbolic_range_p before calling
4839 extract_range_from_multiplicative_op_1.
4841 2018-01-04 Jeff Law <law@redhat.com>
4843 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4844 redundant test in assertion.
4846 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4848 * doc/rtl.texi: Document machine_mode wrapper classes.
4850 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4852 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4855 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4857 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4858 the VEC_PERM_EXPR fold to fail.
4860 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4863 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4864 to switched_sections.
4866 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4869 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4872 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4875 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4876 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4878 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4881 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4882 is BLKmode and bitpos not zero or mode change is needed.
4884 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4887 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4890 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4893 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4894 instead of MULT rtx. Update all corresponding splitters.
4896 (*ssub<modesuffix>): Ditto.
4898 (*cmp_sadd_di): Update split patterns.
4899 (*cmp_sadd_si): Ditto.
4900 (*cmp_sadd_sidi): Ditto.
4901 (*cmp_ssub_di): Ditto.
4902 (*cmp_ssub_si): Ditto.
4903 (*cmp_ssub_sidi): Ditto.
4904 * config/alpha/predicates.md (const23_operand): New predicate.
4905 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4906 Look for ASHIFT, not MULT inner operand.
4907 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4909 2018-01-04 Martin Liska <mliska@suse.cz>
4911 PR gcov-profile/83669
4912 * gcov.c (output_intermediate_file): Add version to intermediate
4914 * doc/gcov.texi: Document new field 'version' in intermediate
4915 file format. Fix location of '-k' option of gcov command.
4917 2018-01-04 Martin Liska <mliska@suse.cz>
4920 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4922 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4924 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4926 2018-01-03 Martin Sebor <msebor@redhat.com>
4928 PR tree-optimization/83655
4929 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4930 checking calls with invalid arguments.
4932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4934 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4935 (vectorizable_mask_load_store): Delete.
4936 (vectorizable_call): Return false for masked loads and stores.
4937 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4938 instead of gimple_assign_rhs1.
4939 (vectorizable_load): Handle IFN_MASK_LOAD.
4940 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4944 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4946 (vectorizable_mask_load_store): ...here.
4947 (vectorizable_load): ...and here.
4949 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4951 * tree-vect-stmts.c (vect_build_all_ones_mask)
4952 (vect_build_zero_merge_argument): New functions, split out from...
4953 (vectorizable_load): ...here.
4955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4957 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4959 (vectorizable_mask_load_store): ...here.
4960 (vectorizable_store): ...and here.
4962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4964 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4966 (vectorizable_mask_load_store): ...here.
4968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4970 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4971 (vect_model_store_cost): Take a vec_load_store_type instead of a
4973 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4974 (vect_model_store_cost): Take a vec_load_store_type instead of a
4976 (vectorizable_mask_load_store): Update accordingly.
4977 (vectorizable_store): Likewise.
4978 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4980 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4982 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4983 IFN_MASK_LOAD calls here rather than...
4984 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4986 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4987 Alan Hayward <alan.hayward@arm.com>
4988 David Sherwood <david.sherwood@arm.com>
4990 * expmed.c (extract_bit_field_1): For vector extracts,
4991 fall back to extract_bit_field_as_subreg if vec_extract
4994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4995 Alan Hayward <alan.hayward@arm.com>
4996 David Sherwood <david.sherwood@arm.com>
4998 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4999 they are variable or constant sized.
5000 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5001 slots for constant-sized data.
5003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5004 Alan Hayward <alan.hayward@arm.com>
5005 David Sherwood <david.sherwood@arm.com>
5007 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5008 handling COND_EXPRs with boolean comparisons, try to find a better
5009 basis for the mask type than the boolean itself.
5011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5013 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5014 is calculated and how it can be overridden.
5015 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5016 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5018 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5022 Alan Hayward <alan.hayward@arm.com>
5023 David Sherwood <david.sherwood@arm.com>
5025 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5026 Remove the mode argument.
5027 (aarch64_simd_valid_immediate): Remove the mode and inverse
5029 * config/aarch64/iterators.md (bitsize): New iterator.
5030 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5031 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5032 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5033 aarch64_simd_valid_immediate.
5034 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5035 (aarch64_reg_or_bic_imm): Likewise.
5036 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5037 with an insn_type enum and msl with a modifier_type enum.
5038 Replace element_width with a scalar_mode. Change the shift
5039 to unsigned int. Add constructors for scalar_float_mode and
5040 scalar_int_mode elements.
5041 (aarch64_vect_float_const_representable_p): Delete.
5042 (aarch64_can_const_movi_rtx_p)
5043 (aarch64_simd_scalar_immediate_valid_for_move)
5044 (aarch64_simd_make_constant): Update call to
5045 aarch64_simd_valid_immediate.
5046 (aarch64_advsimd_valid_immediate_hs): New function.
5047 (aarch64_advsimd_valid_immediate): Likewise.
5048 (aarch64_simd_valid_immediate): Remove mode and inverse
5049 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5050 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5051 and aarch64_float_const_representable_p on the result.
5052 (aarch64_output_simd_mov_immediate): Remove mode argument.
5053 Update call to aarch64_simd_valid_immediate and use of
5054 simd_immediate_info.
5055 (aarch64_output_scalar_simd_mov_immediate): Update call
5058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5059 Alan Hayward <alan.hayward@arm.com>
5060 David Sherwood <david.sherwood@arm.com>
5062 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5063 (mode_nunits): Likewise CONST_MODE_NUNITS.
5064 * machmode.def (ADJUST_NUNITS): Document.
5065 * genmodes.c (mode_data::need_nunits_adj): New field.
5066 (blank_mode): Update accordingly.
5067 (adj_nunits): New variable.
5068 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5070 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5071 listed in adj_nunits.
5072 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5073 listed in adj_nunits. Don't emit case statements for such modes.
5074 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5075 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5076 nothing if adj_nunits is nonnull.
5077 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5078 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5079 (emit_mode_fbit): Update use of print_maybe_const_decl.
5080 (emit_move_size): Likewise. Treat the array as non-const
5082 (emit_mode_adjustments): Handle adj_nunits.
5084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5086 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5087 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5088 (VECTOR_MODES): Use it.
5089 (make_vector_modes): Take the prefix as an argument.
5091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5092 Alan Hayward <alan.hayward@arm.com>
5093 David Sherwood <david.sherwood@arm.com>
5095 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5096 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5097 for MODE_VECTOR_BOOL.
5098 * machmode.def (VECTOR_BOOL_MODE): Document.
5099 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5100 (make_vector_bool_mode): New function.
5101 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5103 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5104 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5106 * stor-layout.c (int_mode_for_mode): Likewise.
5107 * tree.c (build_vector_type_for_mode): Likewise.
5108 * varasm.c (output_constant_pool_2): Likewise.
5109 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5110 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5111 for MODE_VECTOR_BOOL.
5112 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5113 of mode class checks.
5114 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5115 instead of a list of mode class checks.
5116 (expand_vector_scalar_condition): Likewise.
5117 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5120 Alan Hayward <alan.hayward@arm.com>
5121 David Sherwood <david.sherwood@arm.com>
5123 * machmode.h (mode_size): Change from unsigned short to
5125 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5126 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5127 or if measurement_type is not polynomial.
5128 (fixed_size_mode::includes_p): Check for constant-sized modes.
5129 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5130 return a poly_uint16 rather than an unsigned short.
5131 (emit_mode_size): Change the type of mode_size from unsigned short
5132 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5133 (emit_mode_adjustments): Cope with polynomial vector sizes.
5134 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5136 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5138 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5139 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5140 * caller-save.c (setup_save_areas): Likewise.
5141 (replace_reg_with_saved_mem): Likewise.
5142 * calls.c (emit_library_call_value_1): Likewise.
5143 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5144 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5145 (gen_lowpart_for_combine): Likewise.
5146 * convert.c (convert_to_integer_1): Likewise.
5147 * cse.c (equiv_constant, cse_insn): Likewise.
5148 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5149 (cselib_subst_to_values): Likewise.
5150 * dce.c (word_dce_process_block): Likewise.
5151 * df-problems.c (df_word_lr_mark_ref): Likewise.
5152 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5153 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5154 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5155 (rtl_for_decl_location): Likewise.
5156 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5157 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5158 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5159 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5160 (expand_expr_real_1): Likewise.
5161 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5162 (pad_below): Likewise.
5163 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5164 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5165 * ira.c (get_subreg_tracking_sizes): Likewise.
5166 * ira-build.c (ira_create_allocno_objects): Likewise.
5167 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5168 (ira_sort_regnos_for_alter_reg): Likewise.
5169 * ira-costs.c (record_operand_costs): Likewise.
5170 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5171 (resolve_simple_move): Likewise.
5172 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5173 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5174 (lra_constraints): Likewise.
5175 (CONST_POOL_OK_P): Reject variable-sized modes.
5176 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5177 (add_pseudo_to_slot, lra_spill): Likewise.
5178 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5179 * optabs-query.c (get_best_extraction_insn): Likewise.
5180 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5181 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5182 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5183 * recog.c (offsettable_address_addr_space_p): Likewise.
5184 * regcprop.c (maybe_mode_change): Likewise.
5185 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5186 * regrename.c (build_def_use): Likewise.
5187 * regstat.c (dump_reg_info): Likewise.
5188 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5189 (find_reloads, find_reloads_subreg_address): Likewise.
5190 * reload1.c (eliminate_regs_1): Likewise.
5191 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5192 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5193 (simplify_binary_operation_1, simplify_subreg): Likewise.
5194 * targhooks.c (default_function_arg_padding): Likewise.
5195 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5196 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5197 (verify_gimple_assign_ternary): Likewise.
5198 * tree-inline.c (estimate_move_cost): Likewise.
5199 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5200 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5201 (get_address_cost_ainc): Likewise.
5202 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5203 (vect_supportable_dr_alignment): Likewise.
5204 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5205 (vectorizable_reduction): Likewise.
5206 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5207 (vectorizable_operation, vectorizable_load): Likewise.
5208 * tree.c (build_same_sized_truth_vector_type): Likewise.
5209 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5210 * var-tracking.c (emit_note_insn_var_location): Likewise.
5211 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5212 (ADDR_VEC_ALIGN): Likewise.
5214 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5215 Alan Hayward <alan.hayward@arm.com>
5216 David Sherwood <david.sherwood@arm.com>
5218 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5220 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5221 or if measurement_type is polynomial.
5222 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5223 * combine.c (make_extraction): Likewise.
5224 * dse.c (find_shift_sequence): Likewise.
5225 * dwarf2out.c (mem_loc_descriptor): Likewise.
5226 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5227 (extract_bit_field, extract_low_bits): Likewise.
5228 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5229 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5230 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5231 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5232 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5233 * reload.c (find_reloads): Likewise.
5234 * reload1.c (alter_reg): Likewise.
5235 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5236 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5237 * tree-if-conv.c (predicate_mem_writes): Likewise.
5238 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5239 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5240 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5241 * valtrack.c (dead_debug_insert_temp): Likewise.
5242 * varasm.c (mergeable_constant_section): Likewise.
5243 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5245 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5246 Alan Hayward <alan.hayward@arm.com>
5247 David Sherwood <david.sherwood@arm.com>
5249 * expr.c (expand_assignment): Cope with polynomial mode sizes
5250 when assigning to a CONCAT.
5252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5253 Alan Hayward <alan.hayward@arm.com>
5254 David Sherwood <david.sherwood@arm.com>
5256 * machmode.h (mode_precision): Change from unsigned short to
5258 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5260 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5261 or if measurement_type is not polynomial.
5262 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5263 in which the mode is already known to be a scalar_int_mode.
5264 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5265 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5267 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5268 for GET_MODE_PRECISION.
5269 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5270 for GET_MODE_PRECISION.
5271 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5273 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5274 (expand_field_assignment, make_extraction): Likewise.
5275 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5276 (get_last_value): Likewise.
5277 * convert.c (convert_to_integer_1): Likewise.
5278 * cse.c (cse_insn): Likewise.
5279 * expr.c (expand_expr_real_1): Likewise.
5280 * lra-constraints.c (simplify_operand_subreg): Likewise.
5281 * optabs-query.c (can_atomic_load_p): Likewise.
5282 * optabs.c (expand_atomic_load): Likewise.
5283 (expand_atomic_store): Likewise.
5284 * ree.c (combine_reaching_defs): Likewise.
5285 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5286 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5287 * tree.h (type_has_mode_precision_p): Likewise.
5288 * ubsan.c (instrument_si_overflow): Likewise.
5290 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5291 Alan Hayward <alan.hayward@arm.com>
5292 David Sherwood <david.sherwood@arm.com>
5294 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5295 polynomial numbers of units.
5296 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5297 (valid_vector_subparts_p): New function.
5298 (build_vector_type): Remove temporary shim and take the number
5299 of units as a poly_uint64 rather than an int.
5300 (build_opaque_vector_type): Take the number of units as a
5301 poly_uint64 rather than an int.
5302 * tree.c (build_vector_from_ctor): Handle polynomial
5303 TYPE_VECTOR_SUBPARTS.
5304 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5305 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5306 (build_vector_from_val): If the number of units is variable,
5307 use build_vec_duplicate_cst for constant operands and
5308 VEC_DUPLICATE_EXPR otherwise.
5309 (make_vector_type): Remove temporary is_constant ().
5310 (build_vector_type, build_opaque_vector_type): Take the number of
5311 units as a poly_uint64 rather than an int.
5312 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5314 * cfgexpand.c (expand_debug_expr): Likewise.
5315 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5316 (store_constructor, expand_expr_real_1): Likewise.
5317 (const_scalar_mask_from_tree): Likewise.
5318 * fold-const-call.c (fold_const_reduction): Likewise.
5319 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5320 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5321 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5322 (fold_relational_const): Likewise.
5323 (native_interpret_vector): Likewise. Change the size from an
5324 int to an unsigned int.
5325 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5326 TYPE_VECTOR_SUBPARTS.
5327 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5328 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5329 duplicating a non-constant operand into a variable-length vector.
5330 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5331 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5332 * ipa-icf.c (sem_variable::equals): Likewise.
5333 * match.pd: Likewise.
5334 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5335 * print-tree.c (print_node): Likewise.
5336 * stor-layout.c (layout_type): Likewise.
5337 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5338 * tree-cfg.c (verify_gimple_comparison): Likewise.
5339 (verify_gimple_assign_binary): Likewise.
5340 (verify_gimple_assign_ternary): Likewise.
5341 (verify_gimple_assign_single): Likewise.
5342 * tree-pretty-print.c (dump_generic_node): Likewise.
5343 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5344 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5345 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5346 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5347 (vect_shift_permute_load_chain): Likewise.
5348 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5349 (expand_vector_condition, optimize_vector_constructor): Likewise.
5350 (lower_vec_perm, get_compute_type): Likewise.
5351 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5352 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5353 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5354 (vect_recog_mask_conversion_pattern): Likewise.
5355 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5356 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5357 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5358 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5359 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5360 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5361 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5362 (supportable_widening_operation): Likewise.
5363 (supportable_narrowing_operation): Likewise.
5364 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5366 * varasm.c (output_constant): Likewise.
5368 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5369 Alan Hayward <alan.hayward@arm.com>
5370 David Sherwood <david.sherwood@arm.com>
5372 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5373 so that both the length == 3 and length != 3 cases set up their
5374 own permute vectors. Add comments explaining why we know the
5375 number of elements is constant.
5376 (vect_permute_load_chain): Likewise.
5378 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5379 Alan Hayward <alan.hayward@arm.com>
5380 David Sherwood <david.sherwood@arm.com>
5382 * machmode.h (mode_nunits): Change from unsigned char to
5384 (ONLY_FIXED_SIZE_MODES): New macro.
5385 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5386 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5387 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5389 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5390 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5391 or if measurement_type is not polynomial.
5392 * genmodes.c (ZERO_COEFFS): New macro.
5393 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5395 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5396 Use ZERO_COEFFS when emitting initializers.
5397 * data-streamer.h (bp_pack_poly_value): New function.
5398 (bp_unpack_poly_value): Likewise.
5399 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5400 for GET_MODE_NUNITS.
5401 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5402 for GET_MODE_NUNITS.
5403 * tree.c (make_vector_type): Remove temporary shim and make
5404 the real function take the number of units as a poly_uint64
5406 (build_vector_type_for_mode): Handle polynomial nunits.
5407 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5408 * emit-rtl.c (const_vec_series_p_1): Likewise.
5409 (gen_rtx_CONST_VECTOR): Likewise.
5410 * fold-const.c (test_vec_duplicate_folding): Likewise.
5411 * genrecog.c (validate_pattern): Likewise.
5412 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5413 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5414 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5415 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5416 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5417 * rtlanal.c (subreg_get_info): Likewise.
5418 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5419 (vect_grouped_load_supported): Likewise.
5420 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5421 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5422 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5423 (simplify_const_unary_operation, simplify_binary_operation_1)
5424 (simplify_const_binary_operation, simplify_ternary_operation)
5425 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5426 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5427 instead of CONST_VECTOR_NUNITS.
5428 * varasm.c (output_constant_pool_2): Likewise.
5429 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5430 explicit-encoded elements in the XVEC for variable-length vectors.
5432 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5434 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5437 Alan Hayward <alan.hayward@arm.com>
5438 David Sherwood <david.sherwood@arm.com>
5440 * coretypes.h (fixed_size_mode): Declare.
5441 (fixed_size_mode_pod): New typedef.
5442 * builtins.h (target_builtins::x_apply_args_mode)
5443 (target_builtins::x_apply_result_mode): Change type to
5444 fixed_size_mode_pod.
5445 * builtins.c (apply_args_size, apply_result_size, result_vector)
5446 (expand_builtin_apply_args_1, expand_builtin_apply)
5447 (expand_builtin_return): Update accordingly.
5449 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5451 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5452 * cselib.c (cselib_hash_rtx): Likewise.
5453 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5454 CONST_VECTOR encoding.
5456 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5457 Jeff Law <law@redhat.com>
5460 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5461 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5462 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5463 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5466 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5467 explicitly probe *sp in a noreturn function if there were any callee
5468 register saves or frame pointer is needed.
5470 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5473 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5474 BLKmode for ternary, binary or unary expressions.
5477 * var-tracking.c (delete_vta_debug_insn): New inline function.
5478 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5479 insns from get_insns () to NULL instead of each bb separately.
5480 Use delete_vta_debug_insn. No longer static.
5481 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5482 delete_vta_debug_insns callers.
5483 * rtl.h (delete_vta_debug_insns): Declare.
5484 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5485 instead of variable_tracking_main.
5487 2018-01-03 Martin Sebor <msebor@redhat.com>
5489 PR tree-optimization/83603
5490 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5491 arguments past the endof the argument list in functions declared
5492 without a prototype.
5493 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5494 Avoid checking when arguments are null.
5496 2018-01-03 Martin Sebor <msebor@redhat.com>
5499 * doc/extend.texi (attribute const): Fix a typo.
5500 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5501 issuing -Wsuggest-attribute for void functions.
5503 2018-01-03 Martin Sebor <msebor@redhat.com>
5505 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5506 offset_int::from instead of wide_int::to_shwi.
5507 (maybe_diag_overlap): Remove assertion.
5508 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5509 * gimple-ssa-sprintf.c (format_directive): Same.
5510 (parse_directive): Same.
5511 (sprintf_dom_walker::compute_format_length): Same.
5512 (try_substitute_return_value): Same.
5514 2018-01-03 Jeff Law <law@redhat.com>
5517 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5518 non-constant residual for zero at runtime and avoid probing in
5519 that case. Reorganize code for trailing problem to mirror handling
5522 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5524 PR tree-optimization/83501
5525 * tree-ssa-strlen.c (get_string_cst): New.
5526 (handle_char_store): Call get_string_cst.
5528 2018-01-03 Martin Liska <mliska@suse.cz>
5530 PR tree-optimization/83593
5531 * tree-ssa-strlen.c: Include tree-cfg.h.
5532 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5533 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5534 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5536 (strlen_dom_walker::before_dom_children): Call
5537 gimple_purge_dead_eh_edges. Dump tranformation with details
5539 (strlen_dom_walker::before_dom_children): Update call by adding
5540 new argument cleanup_eh.
5541 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5543 2018-01-03 Martin Liska <mliska@suse.cz>
5546 * cif-code.def (VARIADIC_THUNK): New enum value.
5547 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5550 2018-01-03 Jan Beulich <jbeulich@suse.com>
5552 * sse.md (mov<mode>_internal): Tighten condition for when to use
5553 vmovdqu<ssescalarsize> for TI and OI modes.
5555 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5557 Update copyright years.
5559 2018-01-03 Martin Liska <mliska@suse.cz>
5562 * ipa-visibility.c (function_and_variable_visibility): Skip
5563 functions with noipa attribure.
5565 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5567 * gcc.c (process_command): Update copyright notice dates.
5568 * gcov-dump.c (print_version): Ditto.
5569 * gcov.c (print_version): Ditto.
5570 * gcov-tool.c (print_version): Ditto.
5571 * gengtype.c (create_file): Ditto.
5572 * doc/cpp.texi: Bump @copying's copyright year.
5573 * doc/cppinternals.texi: Ditto.
5574 * doc/gcc.texi: Ditto.
5575 * doc/gccint.texi: Ditto.
5576 * doc/gcov.texi: Ditto.
5577 * doc/install.texi: Ditto.
5578 * doc/invoke.texi: Ditto.
5580 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5582 * vector-builder.h (vector_builder::m_full_nelts): Change from
5583 unsigned int to poly_uint64.
5584 (vector_builder::full_nelts): Update prototype accordingly.
5585 (vector_builder::new_vector): Likewise.
5586 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5587 (vector_builder::operator ==): Likewise.
5588 (vector_builder::finalize): Likewise.
5589 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5590 Take the number of elements as a poly_uint64 rather than an
5592 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5593 from unsigned int to poly_uint64.
5594 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5595 (vec_perm_indices::new_vector): Likewise.
5596 (vec_perm_indices::length): Likewise.
5597 (vec_perm_indices::nelts_per_input): Likewise.
5598 (vec_perm_indices::input_nelts): Likewise.
5599 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5600 number of elements per input as a poly_uint64 rather than an
5601 unsigned int. Use the original encoding for variable-length
5602 vectors, rather than clamping each individual element.
5603 For the second and subsequent elements in each pattern,
5604 clamp the step and base before clamping their sum.
5605 (vec_perm_indices::series_p): Handle polynomial element counts.
5606 (vec_perm_indices::all_in_range_p): Likewise.
5607 (vec_perm_indices_to_tree): Likewise.
5608 (vec_perm_indices_to_rtx): Likewise.
5609 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5610 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5611 (tree_vector_builder::new_binary_operation): Handle polynomial
5612 element counts. Return false if we need to know the number
5613 of elements at compile time.
5614 * fold-const.c (fold_vec_perm): Punt if the number of elements
5615 isn't known at compile time.
5617 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5619 * vec-perm-indices.h (vec_perm_builder): Change element type
5620 from HOST_WIDE_INT to poly_int64.
5621 (vec_perm_indices::element_type): Update accordingly.
5622 (vec_perm_indices::clamp): Handle polynomial element_types.
5623 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5624 (vec_perm_indices::all_in_range_p): Likewise.
5625 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5627 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5628 polynomial vec_perm_indices element types.
5629 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5630 * fold-const.c (fold_vec_perm): Likewise.
5631 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5632 * tree-vect-generic.c (lower_vec_perm): Likewise.
5633 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5634 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5635 element type to HOST_WIDE_INT.
5637 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5638 Alan Hayward <alan.hayward@arm.com>
5639 David Sherwood <david.sherwood@arm.com>
5641 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5642 rather than an int. Use plus_constant.
5643 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5644 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5646 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5647 Alan Hayward <alan.hayward@arm.com>
5648 David Sherwood <david.sherwood@arm.com>
5650 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5651 a HOST_WIDE_INT to a poly_int64.
5653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5654 Alan Hayward <alan.hayward@arm.com>
5655 David Sherwood <david.sherwood@arm.com>
5657 * calls.c (load_register_parameters): Cope with polynomial
5658 mode sizes. Require a constant size for BLKmode parameters
5659 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5660 forces a parameter to be padded at the lsb end in order to
5661 fill a complete number of words, require the parameter size
5662 to be ordered wrt UNITS_PER_WORD.
5664 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5665 Alan Hayward <alan.hayward@arm.com>
5666 David Sherwood <david.sherwood@arm.com>
5668 * reload1.c (spill_stack_slot_width): Change element type
5669 from unsigned int to poly_uint64_pod.
5670 (alter_reg): Treat mode sizes as polynomial.
5672 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5673 Alan Hayward <alan.hayward@arm.com>
5674 David Sherwood <david.sherwood@arm.com>
5676 * reload.c (complex_word_subreg_p): New function.
5677 (reload_inner_reg_of_subreg, push_reload): Use it.
5679 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5680 Alan Hayward <alan.hayward@arm.com>
5681 David Sherwood <david.sherwood@arm.com>
5683 * lra-constraints.c (process_alt_operands): Reject matched
5684 operands whose sizes aren't ordered.
5685 (match_reload): Refer to this check here.
5687 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5688 Alan Hayward <alan.hayward@arm.com>
5689 David Sherwood <david.sherwood@arm.com>
5691 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5692 that the mode size is in the set {1, 2, 4, 8, 16}.
5694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5695 Alan Hayward <alan.hayward@arm.com>
5696 David Sherwood <david.sherwood@arm.com>
5698 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5699 Use plus_constant instead of gen_rtx_PLUS.
5701 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5702 Alan Hayward <alan.hayward@arm.com>
5703 David Sherwood <david.sherwood@arm.com>
5705 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5706 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5707 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5708 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5709 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5710 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5711 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5712 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5713 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5714 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5716 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5717 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5718 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5719 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5720 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5721 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5722 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5723 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5724 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5725 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5727 * expr.c (emit_move_resolve_push): Treat the input and result
5728 of PUSH_ROUNDING as a poly_int64.
5729 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5730 (emit_push_insn): Likewise.
5731 * lra-eliminations.c (mark_not_eliminable): Likewise.
5732 * recog.c (push_operand): Likewise.
5733 * reload1.c (elimination_effects): Likewise.
5734 * rtlanal.c (nonzero_bits1): Likewise.
5735 * calls.c (store_one_arg): Likewise. Require the padding to be
5736 known at compile time.
5738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5739 Alan Hayward <alan.hayward@arm.com>
5740 David Sherwood <david.sherwood@arm.com>
5742 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5743 Use plus_constant instead of gen_rtx_PLUS.
5745 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5746 Alan Hayward <alan.hayward@arm.com>
5747 David Sherwood <david.sherwood@arm.com>
5749 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5752 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5753 Alan Hayward <alan.hayward@arm.com>
5754 David Sherwood <david.sherwood@arm.com>
5756 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5757 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5758 via stack temporaries. Treat the mode size as polynomial too.
5760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5761 Alan Hayward <alan.hayward@arm.com>
5762 David Sherwood <david.sherwood@arm.com>
5764 * expr.c (expand_expr_real_2): When handling conversions involving
5765 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5766 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5767 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5775 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5776 Alan Hayward <alan.hayward@arm.com>
5777 David Sherwood <david.sherwood@arm.com>
5779 * combine.c (can_change_dest_mode): Handle polynomial
5780 REGMODE_NATURAL_SIZE.
5781 * expmed.c (store_bit_field_1): Likewise.
5782 * expr.c (store_constructor): Likewise.
5783 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5784 and polynomial REGMODE_NATURAL_SIZE.
5785 (gen_lowpart_common): Likewise.
5786 * reginfo.c (record_subregs_of_mode): Likewise.
5787 * rtlanal.c (read_modify_subreg_p): Likewise.
5789 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5790 Alan Hayward <alan.hayward@arm.com>
5791 David Sherwood <david.sherwood@arm.com>
5793 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5794 numbers of elements.
5796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5797 Alan Hayward <alan.hayward@arm.com>
5798 David Sherwood <david.sherwood@arm.com>
5800 * match.pd: Cope with polynomial numbers of vector elements.
5802 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5803 Alan Hayward <alan.hayward@arm.com>
5804 David Sherwood <david.sherwood@arm.com>
5806 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5807 in a POINTER_PLUS_EXPR.
5809 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5810 Alan Hayward <alan.hayward@arm.com>
5811 David Sherwood <david.sherwood@arm.com>
5813 * omp-simd-clone.c (simd_clone_subparts): New function.
5814 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5815 (ipa_simd_modify_function_body): Likewise.
5817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5818 Alan Hayward <alan.hayward@arm.com>
5819 David Sherwood <david.sherwood@arm.com>
5821 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5822 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5823 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5824 (expand_vector_condition, vector_element): Likewise.
5825 (subparts_gt): New function.
5826 (get_compute_type): Use subparts_gt.
5827 (count_type_subparts): Delete.
5828 (expand_vector_operations_1): Use subparts_gt instead of
5829 count_type_subparts.
5831 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5832 Alan Hayward <alan.hayward@arm.com>
5833 David Sherwood <david.sherwood@arm.com>
5835 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5836 (vect_compile_time_alias): ...this new function. Do the calculation
5837 on poly_ints rather than trees.
5838 (vect_prune_runtime_alias_test_list): Update call accordingly.
5840 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5841 Alan Hayward <alan.hayward@arm.com>
5842 David Sherwood <david.sherwood@arm.com>
5844 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5846 (vect_schedule_slp_instance): Likewise.
5848 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5849 Alan Hayward <alan.hayward@arm.com>
5850 David Sherwood <david.sherwood@arm.com>
5852 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5853 constant and extern definitions for variable-length vectors.
5854 (vect_get_constant_vectors): Note that the number of units
5855 is known to be constant.
5857 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5858 Alan Hayward <alan.hayward@arm.com>
5859 David Sherwood <david.sherwood@arm.com>
5861 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5862 of units as polynomial. Choose between WIDE and NARROW based
5865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5866 Alan Hayward <alan.hayward@arm.com>
5867 David Sherwood <david.sherwood@arm.com>
5869 * tree-vect-stmts.c (simd_clone_subparts): New function.
5870 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5872 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5873 Alan Hayward <alan.hayward@arm.com>
5874 David Sherwood <david.sherwood@arm.com>
5876 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5877 vectors as polynomial. Use build_index_vector for
5880 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5881 Alan Hayward <alan.hayward@arm.com>
5882 David Sherwood <david.sherwood@arm.com>
5884 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5885 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5886 for variable-length vectors.
5887 (vectorizable_mask_load_store): Treat the number of units as
5888 polynomial, asserting that it is constant if the condition has
5889 already been enforced.
5890 (vectorizable_store, vectorizable_load): 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-loop.c (vectorizable_live_operation): Treat the number
5897 of units as polynomial. Punt if we can't tell at compile time
5898 which vector contains the final result.
5900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5901 Alan Hayward <alan.hayward@arm.com>
5902 David Sherwood <david.sherwood@arm.com>
5904 * tree-vect-loop.c (vectorizable_induction): Treat the number
5905 of units as polynomial. Punt on SLP inductions. Use an integer
5906 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5907 cast of such a series for variable-length floating-point
5910 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5911 Alan Hayward <alan.hayward@arm.com>
5912 David Sherwood <david.sherwood@arm.com>
5914 * tree.h (build_index_vector): Declare.
5915 * tree.c (build_index_vector): New function.
5916 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5917 of units as polynomial, forcibly converting it to a constant if
5918 vectorizable_reduction has already enforced the condition.
5919 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5920 to create a {1,2,3,...} vector.
5921 (vectorizable_reduction): Treat the number of units as polynomial.
5922 Choose vectype_in based on the largest scalar element size rather
5923 than the smallest number of units. Enforce the restrictions
5926 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5927 Alan Hayward <alan.hayward@arm.com>
5928 David Sherwood <david.sherwood@arm.com>
5930 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5931 number of units as polynomial.
5933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5934 Alan Hayward <alan.hayward@arm.com>
5935 David Sherwood <david.sherwood@arm.com>
5937 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5938 * target.def (autovectorize_vector_sizes): Return the vector sizes
5939 by pointer, using vector_sizes rather than a bitmask.
5940 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5941 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5942 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5944 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5945 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5946 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5947 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5948 * omp-general.c (omp_max_vf): Likewise.
5949 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5950 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5951 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5952 * tree-vect-slp.c (vect_slp_bb): Likewise.
5953 * doc/tm.texi: Regenerate.
5954 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5956 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5957 the vector size as a poly_uint64 rather than an unsigned int.
5958 (current_vector_size): Change from an unsigned int to a poly_uint64.
5959 (get_vectype_for_scalar_type): Update accordingly.
5960 * tree.h (build_truth_vector_type): Take the size and number of
5961 units as a poly_uint64 rather than an unsigned int.
5962 (build_vector_type): Add a temporary overload that takes
5963 the number of units as a poly_uint64 rather than an unsigned int.
5964 * tree.c (make_vector_type): Likewise.
5965 (build_truth_vector_type): Take the number of units as a poly_uint64
5966 rather than an unsigned int.
5968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5969 Alan Hayward <alan.hayward@arm.com>
5970 David Sherwood <david.sherwood@arm.com>
5972 * target.def (get_mask_mode): Take the number of units and length
5973 as poly_uint64s rather than unsigned ints.
5974 * targhooks.h (default_get_mask_mode): Update accordingly.
5975 * targhooks.c (default_get_mask_mode): Likewise.
5976 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5977 * doc/tm.texi: Regenerate.
5979 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5980 Alan Hayward <alan.hayward@arm.com>
5981 David Sherwood <david.sherwood@arm.com>
5983 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5984 * omp-general.c (omp_max_vf): Likewise.
5985 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5986 (expand_omp_simd): Handle polynomial safelen.
5987 * omp-low.c (omplow_simd_context): Add a default constructor.
5988 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5989 (lower_rec_simd_input_clauses): Update accordingly.
5990 (lower_rec_input_clauses): Likewise.
5992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5993 Alan Hayward <alan.hayward@arm.com>
5994 David Sherwood <david.sherwood@arm.com>
5996 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5997 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5998 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5999 (vect_analyze_slp_cost): Likewise.
6000 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6001 (vect_model_load_cost): Likewise.
6003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6004 Alan Hayward <alan.hayward@arm.com>
6005 David Sherwood <david.sherwood@arm.com>
6007 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6008 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6009 from an unsigned int * to a poly_uint64_pod *.
6010 (calculate_unrolling_factor): New function.
6011 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6014 Alan Hayward <alan.hayward@arm.com>
6015 David Sherwood <david.sherwood@arm.com>
6017 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6018 from an unsigned int to a poly_uint64.
6019 (_loop_vec_info::slp_unrolling_factor): Likewise.
6020 (_loop_vec_info::vectorization_factor): Change from an int
6022 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6023 (vect_get_num_vectors): New function.
6024 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6025 (vect_get_num_copies): Use vect_get_num_vectors.
6026 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6027 to an unsigned int *.
6028 (vect_analyze_data_refs): Change min_vf from an int * to a
6030 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6031 than an unsigned HOST_WIDE_INT.
6032 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6033 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6034 to an unsigned int *.
6035 (vect_analyze_data_ref_dependences): Likewise.
6036 (vect_compute_data_ref_alignment): Handle polynomial vf.
6037 (vect_enhance_data_refs_alignment): Likewise.
6038 (vect_prune_runtime_alias_test_list): Likewise.
6039 (vect_shift_permute_load_chain): Likewise.
6040 (vect_supportable_dr_alignment): Likewise.
6041 (dependence_distance_ge_vf): Take the vectorization factor as a
6042 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6043 (vect_analyze_data_refs): Change min_vf from an int * to a
6045 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6046 vfm1 as a poly_uint64 rather than an int. Make the same change
6047 for the returned bound_scalar.
6048 (vect_gen_vector_loop_niters): Handle polynomial vf.
6049 (vect_do_peeling): Likewise. Update call to
6050 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6051 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6053 * tree-vect-loop.c (vect_determine_vectorization_factor)
6054 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6055 (vect_get_known_peeling_cost): Likewise.
6056 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6057 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6058 (vect_transform_loop): Likewise. Use the lowest possible VF when
6059 updating the upper bounds of the loop.
6060 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6062 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6063 polynomial unroll factors.
6064 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6065 (vect_make_slp_decision): Likewise.
6066 (vect_supported_load_permutation_p): Likewise, and polynomial
6068 (vect_analyze_slp_cost): Handle polynomial vf.
6069 (vect_slp_analyze_node_operations): Likewise.
6070 (vect_slp_analyze_bb_1): Likewise.
6071 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6072 than an unsigned HOST_WIDE_INT.
6073 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6074 (vectorizable_load): Handle polynomial vf.
6075 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6077 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6080 Alan Hayward <alan.hayward@arm.com>
6081 David Sherwood <david.sherwood@arm.com>
6083 * match.pd: Handle bit operations involving three constants
6084 and try to fold one pair.
6086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6088 * tree-vect-loop-manip.c: Include gimple-fold.h.
6089 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6090 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6091 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6092 Add a path that uses a step of VF instead of 1, but disable it
6094 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6095 and niters_no_overflow parameters. Update calls to
6096 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6097 Create a new SSA name if the latter choses to use a ste other
6098 than zero, and return it via niters_vector_mult_vf_var.
6099 * tree-vect-loop.c (vect_transform_loop): Update calls to
6100 vect_do_peeling, vect_gen_vector_loop_niters and
6101 slpeel_make_loop_iterate_ntimes.
6102 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6103 (vect_gen_vector_loop_niters): Update declarations after above changes.
6105 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6107 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6108 128-bit round to integer instructions.
6109 (ceil<mode>2): Likewise.
6110 (btrunc<mode>2): Likewise.
6111 (round<mode>2): Likewise.
6113 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6115 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6116 unaligned VSX load/store on P8/P9.
6117 (expand_block_clear): Allow the use of unaligned VSX
6118 load/store on P8/P9.
6120 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6122 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6124 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6125 swap associated with both a load and a store.
6127 2018-01-02 Andrew Waterman <andrew@sifive.com>
6129 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6130 * config/riscv/riscv.md (clear_cache): Use it.
6132 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6134 * web.c: Remove out-of-date comment.
6136 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6138 * expr.c (fixup_args_size_notes): Check that any existing
6139 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6140 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6141 (emit_single_push_insn): ...here.
6143 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6145 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6146 (const_vector_encoded_nelts): New function.
6147 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6148 (const_vector_int_elt, const_vector_elt): Declare.
6149 * emit-rtl.c (const_vector_int_elt_1): New function.
6150 (const_vector_elt): Likewise.
6151 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6152 of CONST_VECTOR_ELT.
6154 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6156 * expr.c: Include rtx-vector-builder.h.
6157 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6158 directly on the tree encoding.
6159 (const_vector_from_tree): Likewise.
6160 * optabs.c: Include rtx-vector-builder.h.
6161 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6162 sequence of "u" values.
6163 * vec-perm-indices.c: Include rtx-vector-builder.h.
6164 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6165 directly on the vec_perm_indices encoding.
6167 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6169 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6170 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6171 * rtx-vector-builder.h: New file.
6172 * rtx-vector-builder.c: Likewise.
6173 * rtl.h (rtx_def::u2): Add a const_vector field.
6174 (CONST_VECTOR_NPATTERNS): New macro.
6175 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6176 (CONST_VECTOR_DUPLICATE_P): Likewise.
6177 (CONST_VECTOR_STEPPED_P): Likewise.
6178 (CONST_VECTOR_ENCODED_ELT): Likewise.
6179 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6180 (unwrap_const_vec_duplicate): Likewise.
6181 (const_vec_series_p): Check for a non-duplicated vector encoding.
6182 Say that the function only returns true for integer vectors.
6183 * emit-rtl.c: Include rtx-vector-builder.h.
6184 (gen_const_vec_duplicate_1): Delete.
6185 (gen_const_vector): Call gen_const_vec_duplicate instead of
6186 gen_const_vec_duplicate_1.
6187 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6188 (gen_const_vec_duplicate): Use rtx_vector_builder.
6189 (gen_const_vec_series): Likewise.
6190 (gen_rtx_CONST_VECTOR): Likewise.
6191 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6192 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6193 Build a new vector rather than modifying a CONST_VECTOR in-place.
6194 (handle_special_swappables): Update call accordingly.
6195 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6196 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6197 Build a new vector rather than modifying a CONST_VECTOR in-place.
6198 (handle_special_swappables): Update call accordingly.
6200 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6202 * simplify-rtx.c (simplify_const_binary_operation): Use
6203 CONST_VECTOR_ELT instead of XVECEXP.
6205 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6207 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6208 the selector elements to be different from the data elements
6209 if the selector is a VECTOR_CST.
6210 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6211 ssizetype for the selector.
6213 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6215 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6216 before testing each element individually.
6217 * tree-vect-generic.c (lower_vec_perm): Likewise.
6219 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6221 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6222 * selftest-run-tests.c (selftest::run_tests): Call it.
6223 * vector-builder.h (vector_builder::operator ==): New function.
6224 (vector_builder::operator !=): Likewise.
6225 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6226 (vec_perm_indices::all_from_input_p): New function.
6227 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6228 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6229 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6230 instead of reading the VECTOR_CST directly. Detect whether both
6231 vector inputs are the same before constructing the vec_perm_indices,
6232 and update the number of inputs argument accordingly. Use the
6233 utility functions added above. Only construct sel2 if we need to.
6235 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6237 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6238 the broadcast of the low byte.
6239 (expand_mult_highpart): Use an explicit encoding for the permutes.
6240 * optabs-query.c (can_mult_highpart_p): Likewise.
6241 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6242 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6243 (vectorizable_bswap): Likewise.
6244 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6245 explicit encoding for the power-of-2 permutes.
6246 (vect_permute_store_chain): Likewise.
6247 (vect_grouped_load_supported): Likewise.
6248 (vect_permute_load_chain): Likewise.
6250 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6252 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6253 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6254 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6255 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6256 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6257 (vect_gen_perm_mask_any): Likewise.
6259 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6261 * int-vector-builder.h: New file.
6262 * vec-perm-indices.h: Include int-vector-builder.h.
6263 (vec_perm_indices): Redefine as an int_vector_builder.
6264 (auto_vec_perm_indices): Delete.
6265 (vec_perm_builder): Redefine as a stand-alone class.
6266 (vec_perm_indices::vec_perm_indices): New function.
6267 (vec_perm_indices::clamp): Likewise.
6268 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6269 (vec_perm_indices::new_vector): New function.
6270 (vec_perm_indices::new_expanded_vector): Update for new
6271 vec_perm_indices class.
6272 (vec_perm_indices::rotate_inputs): New function.
6273 (vec_perm_indices::all_in_range_p): Operate directly on the
6274 encoded form, without computing elided elements.
6275 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6276 encoding. Update for new vec_perm_indices class.
6277 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6278 the given vec_perm_builder.
6279 (expand_vec_perm_var): Update vec_perm_builder constructor.
6280 (expand_mult_highpart): Use vec_perm_builder instead of
6281 auto_vec_perm_indices.
6282 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6283 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6284 or double series encoding as appropriate.
6285 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6286 vec_perm_indices instead of auto_vec_perm_indices.
6287 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6288 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6289 (vect_permute_store_chain): Likewise.
6290 (vect_grouped_load_supported): Likewise.
6291 (vect_permute_load_chain): Likewise.
6292 (vect_shift_permute_load_chain): Likewise.
6293 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6294 (vect_transform_slp_perm_load): Likewise.
6295 (vect_schedule_slp_instance): Likewise.
6296 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6297 (vectorizable_mask_load_store): Likewise.
6298 (vectorizable_bswap): Likewise.
6299 (vectorizable_store): Likewise.
6300 (vectorizable_load): Likewise.
6301 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6302 vec_perm_indices instead of auto_vec_perm_indices. Use
6303 tree_to_vec_perm_builder to read the vector from a tree.
6304 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6305 vec_perm_builder instead of a vec_perm_indices.
6306 (have_whole_vector_shift): Use vec_perm_builder and
6307 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6308 truncation to calc_vec_perm_mask_for_shift.
6309 (vect_create_epilog_for_reduction): Likewise.
6310 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6311 from auto_vec_perm_indices to vec_perm_indices.
6312 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6313 instead of changing individual elements.
6314 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6315 the vector in d.perm.
6316 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6317 from auto_vec_perm_indices to vec_perm_indices.
6318 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6319 instead of changing individual elements.
6320 (arm_vectorize_vec_perm_const): Use new_vector to install
6321 the vector in d.perm.
6322 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6323 Update vec_perm_builder constructor.
6324 (rs6000_expand_interleave): Likewise.
6325 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6326 (rs6000_expand_interleave): Likewise.
6328 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6330 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6331 to qimode could truncate the indices.
6332 * optabs.c (expand_vec_perm_var): Likewise.
6334 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6336 * Makefile.in (OBJS): Add vec-perm-indices.o.
6337 * vec-perm-indices.h: New file.
6338 * vec-perm-indices.c: Likewise.
6339 * target.h (vec_perm_indices): Replace with a forward class
6341 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6342 * optabs.h: Include vec-perm-indices.h.
6343 (expand_vec_perm): Delete.
6344 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6345 (expand_vec_perm_const): Declare.
6346 * target.def (vec_perm_const_ok): Replace with...
6347 (vec_perm_const): ...this new hook.
6348 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6349 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6350 * doc/tm.texi: Regenerate.
6351 * optabs.def (vec_perm_const): Delete.
6352 * doc/md.texi (vec_perm_const): Likewise.
6353 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6354 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6355 expand_vec_perm for constant permutation vectors. Assert that
6356 the mode of variable permutation vectors is the integer equivalent
6357 of the mode that is being permuted.
6358 * optabs-query.h (selector_fits_mode_p): Declare.
6359 * optabs-query.c: Include vec-perm-indices.h.
6360 (selector_fits_mode_p): New function.
6361 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6362 is defined, instead of checking whether the vec_perm_const_optab
6363 exists. Use targetm.vectorize.vec_perm_const instead of
6364 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6365 fit in the vector mode before using a variable permute.
6366 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6367 vec_perm_indices instead of an rtx.
6368 (expand_vec_perm): Replace with...
6369 (expand_vec_perm_const): ...this new function. Take the selector
6370 as a vec_perm_indices rather than an rtx. Also take the mode of
6371 the selector. Update call to shift_amt_for_vec_perm_mask.
6372 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6373 Use vec_perm_indices::new_expanded_vector to expand the original
6374 selector into bytes. Check whether the indices fit in the vector
6375 mode before using a variable permute.
6376 (expand_vec_perm_var): Make global.
6377 (expand_mult_highpart): Use expand_vec_perm_const.
6378 * fold-const.c: Includes vec-perm-indices.h.
6379 * tree-ssa-forwprop.c: Likewise.
6380 * tree-vect-data-refs.c: Likewise.
6381 * tree-vect-generic.c: Likewise.
6382 * tree-vect-loop.c: Likewise.
6383 * tree-vect-slp.c: Likewise.
6384 * tree-vect-stmts.c: Likewise.
6385 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6387 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6388 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6389 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6390 (aarch64_vectorize_vec_perm_const): ...this new function.
6391 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6392 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6393 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6394 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6395 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6396 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6397 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6399 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6400 check for NEON modes.
6401 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6402 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6403 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6404 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6406 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6407 the old VEC_PERM_CONST conditions.
6408 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6409 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6410 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6411 (ia64_vectorize_vec_perm_const_ok): Merge into...
6412 (ia64_vectorize_vec_perm_const): ...this new function.
6413 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6414 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6415 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6416 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6417 * config/mips/mips.c (mips_expand_vec_perm_const)
6418 (mips_vectorize_vec_perm_const_ok): Merge into...
6419 (mips_vectorize_vec_perm_const): ...this new function.
6420 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6421 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6422 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6423 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6424 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6425 (rs6000_expand_vec_perm_const): Delete.
6426 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6428 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6429 (altivec_expand_vec_perm_const_le): Take each operand individually.
6430 Operate on constant selectors rather than rtxes.
6431 (altivec_expand_vec_perm_const): Likewise. Update call to
6432 altivec_expand_vec_perm_const_le.
6433 (rs6000_expand_vec_perm_const): Delete.
6434 (rs6000_vectorize_vec_perm_const_ok): Delete.
6435 (rs6000_vectorize_vec_perm_const): New function.
6436 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6437 an element count and rtx array.
6438 (rs6000_expand_extract_even): Update call accordingly.
6439 (rs6000_expand_interleave): Likewise.
6440 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6441 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6442 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6443 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6444 (rs6000_expand_vec_perm_const): Delete.
6445 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6446 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6447 (altivec_expand_vec_perm_const_le): Take each operand individually.
6448 Operate on constant selectors rather than rtxes.
6449 (altivec_expand_vec_perm_const): Likewise. Update call to
6450 altivec_expand_vec_perm_const_le.
6451 (rs6000_expand_vec_perm_const): Delete.
6452 (rs6000_vectorize_vec_perm_const_ok): Delete.
6453 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6454 reference to the SPE evmerge intructions.
6455 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6456 an element count and rtx array.
6457 (rs6000_expand_extract_even): Update call accordingly.
6458 (rs6000_expand_interleave): Likewise.
6459 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6460 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6462 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6464 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6466 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6467 vector mode and that that mode matches the mode of the data
6469 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6470 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6471 directly using expand_vec_perm_1 when forcing selectors into
6473 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6475 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6477 * optabs-query.h (can_vec_perm_p): Delete.
6478 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6479 * optabs-query.c (can_vec_perm_p): Split into...
6480 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6481 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6482 particular selector is valid.
6483 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6484 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6485 (vect_grouped_load_supported): Likewise.
6486 (vect_shift_permute_load_chain): Likewise.
6487 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6488 (vect_transform_slp_perm_load): Likewise.
6489 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6490 (vectorizable_bswap): Likewise.
6491 (vect_gen_perm_mask_checked): Likewise.
6492 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6493 implementations of variable permutation vectors into account
6494 when deciding which selector to use.
6495 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6496 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6497 with a false third argument.
6498 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6499 to test whether the constant selector is valid and can_vec_perm_var_p
6500 to test whether a variable selector is valid.
6502 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6504 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6505 * optabs-query.c (can_vec_perm_p): Likewise.
6506 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6507 instead of vec_perm_indices.
6508 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6509 (vect_gen_perm_mask_checked): Likewise,
6510 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6511 (vect_gen_perm_mask_checked): Likewise,
6513 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6515 * optabs-query.h (qimode_for_vec_perm): Declare.
6516 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6517 (qimode_for_vec_perm): ...this new function.
6518 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6520 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6522 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6523 does not have a conditional at the top.
6525 2018-01-02 Richard Biener <rguenther@suse.de>
6527 * ipa-inline.c (big_speedup_p): Fix expression.
6529 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6532 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6535 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6539 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6540 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6541 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6542 cond_taken_branch_cost 3->4.
6544 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6546 PR tree-optimization/83581
6547 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6548 TODO_cleanup_cfg if any changes have been made.
6551 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6552 convert_modes if target mode has the right side, but different mode
6556 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6557 last argument when extracting from CONCAT. If either from_real or
6558 from_imag is NULL, use expansion through memory. If result is not
6559 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6560 the parts directly to inner mode, if even that fails, use expansion
6564 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6565 check for bswap in mode rather than HImode and use that in expand_unop
6568 Copyright (C) 2018 Free Software Foundation, Inc.
6570 Copying and distribution of this file, with or without modification,
6571 are permitted in any medium without royalty provided the copyright
6572 notice and this notice are preserved.