1 2018-02-09 Jakub Jelinek <jakub@redhat.com>
4 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
5 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
9 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
10 PARALLEL incoming that failed vt_get_decl_and_offset check.
13 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
14 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
16 (get_variable_section): For decls in named .bss* sections pass true as
17 second argument to bss_initializer_p.
19 2018-02-09 Marek Polacek <polacek@redhat.com>
20 Jakub Jelinek <jakub@redhat.com>
23 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
24 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
25 Sync some changes from cxx_fold_indirect_ref.
27 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
29 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
31 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
32 (BLOCK_INLINE_ENTRY_LABEL): New.
33 (dwarf2out_var_location): Disregard inline entry markers.
34 (inline_entry_data): New struct.
35 (inline_entry_data_hasher): New hashtable type.
36 (inline_entry_data_hasher::hash): New.
37 (inline_entry_data_hasher::equal): New.
38 (inline_entry_data_table): New variable.
39 (add_high_low_attributes): Add DW_AT_entry_pc and
40 DW_AT_GNU_entry_view attributes if a pending entry is found
41 in inline_entry_data_table. Add old entry_pc attribute only
42 if debug nonbinding markers are disabled.
43 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
45 (block_within_block_p, dwarf2out_inline_entry): New.
46 (dwarf2out_finish): Check that no entries remained in
47 inline_entry_data_table.
48 * final.c (reemit_insn_block_notes): Handle inline entry notes.
49 (final_scan_insn, notice_source_line): Likewise.
50 (rest_of_clean_state): Skip inline entry markers.
51 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
53 * gimple.c (gimple_build_debug_inline_entry): New.
54 * gimple.h (enum gimple_debug_subcode): Add
55 GIMPLE_DEBUG_INLINE_ENTRY.
56 (gimple_build_debug_inline_entry): Declare.
57 (gimple_debug_inline_entry_p): New.
58 (gimple_debug_nonbind_marker_p): Adjust.
59 * insn-notes.def (INLINE_ENTRY): New.
60 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
61 inline entry marker notes.
62 (print_insn): Likewise.
63 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
64 (INSN_DEBUG_MARKER_KIND): Likewise.
65 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
66 * tree-inline.c (expand_call_inline): Build and insert
67 debug_inline_entry stmt.
68 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
69 inline entry blocks early, if nonbind markers are enabled.
70 (dump_scope_block): Dump fragment info.
71 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
72 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
73 (gimple_build_debug_inline_entry): New.
74 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
75 Enable/disable inline entry points too.
76 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
77 (DEBUG_INSN): Describe inline entry markers.
79 * common.opt (gvariable-location-views): New.
80 (gvariable-location-views=incompat5): New.
83 * configure.ac: Test assembler for view support.
84 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
85 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
86 * dwarf2out.c (var_loc_view): New typedef.
87 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
88 (dwarf2out_locviews_in_attribute): New.
89 (dwarf2out_locviews_in_loclist): New.
90 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
91 (enum dw_line_info_opcode): Add LI_adv_address.
92 (struct dw_line_info_table): Add view.
93 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
94 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
95 (zero_view_p): New variable.
96 (ZERO_VIEW_P): New macro.
97 (output_asm_line_debug_info): New.
98 (struct var_loc_node): Add view.
99 (add_AT_view_list, AT_loc_list): New.
100 (add_var_loc_to_decl): Add view param. Test it against last.
101 (new_loc_list): Add view params. Record them.
102 (AT_loc_list_ptr): Handle loc and view lists.
103 (view_list_to_loc_list_val_node): New.
104 (print_dw_val): Handle dw_val_class_view_list.
105 (size_of_die): Likewise.
106 (value_format): Likewise.
107 (loc_list_has_views): New.
108 (gen_llsym): Set vl_symbol too.
109 (maybe_gen_llsym, skip_loc_list_entry): New.
110 (dwarf2out_maybe_output_loclist_view_pair): New.
111 (output_loc_list): Output view list or entries too.
112 (output_view_list_offset): New.
113 (output_die): Handle dw_val_class_view_list.
114 (output_dwarf_version): New.
115 (output_compilation_unit_header): Use it.
116 (output_skeleton_debug_sections): Likewise.
117 (output_rnglists, output_line_info): Likewise.
118 (output_pubnames, output_aranges): Update version comments.
119 (output_one_line_info_table): Output view numbers in asm comments.
120 (dw_loc_list): Determine current endview, pass it to new_loc_list.
121 Call maybe_gen_llsym.
122 (loc_list_from_tree_1): Adjust.
123 (add_AT_location_description): Create view list attribute if
124 needed, check it's absent otherwise.
125 (convert_cfa_to_fb_loc_list): Adjust.
126 (maybe_emit_file): Call output_asm_line_debug_info for test.
127 (dwarf2out_var_location): Reset views as needed. Precompute
128 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
130 (new_line_info_table): Reset next view.
131 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
132 (dwarf2out_source_line): Likewise. Output view resets and labels to
133 the assembler, or select appropriate line info opcodes.
134 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
135 (optimize_string_length): Catch it. Adjust.
136 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
137 dw_val_class_view_list, and remove it if no longer needed.
138 (hash_loc_list): Hash view numbers.
139 (loc_list_hasher::equal): Compare them.
140 (optimize_location_lists): Check whether a view list symbol is
141 needed, and whether the locview attribute is present, and
142 whether they match. Remove the locview attribute if no longer
144 (index_location_lists): Call skip_loc_list_entry for test.
145 (dwarf2out_finish): Call output_asm_line_debug_info for test.
146 Use output_dwarf_version.
147 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
148 (struct dw_val_node): Add val_view_list.
149 * final.c (SEEN_NEXT_VIEW): New.
150 (set_next_view_needed): New.
151 (clear_next_view_needed): New.
152 (maybe_output_next_view): New.
153 (final_start_function): Rename to...
154 (final_start_function_1): ... this. Take pointer to FIRST,
155 add SEEN parameter. Emit param bindings in the initial view.
156 (final_start_function): Reintroduce SEEN-less interface.
157 (final): Rename to...
158 (final_1): ... this. Take SEEN parameter. Output final pending
159 next view at the end.
160 (final): Reintroduce seen-less interface.
161 (final_scan_insn): Output pending next view before switching
162 sections or ending a block. Mark the next view as needed when
163 outputting variable locations. Notify debug backend of section
164 changes, and of location view changes.
165 (rest_of_handle_final): Adjust.
166 * toplev.c (process_options): Autodetect value for debug variable
167 location views option. Warn on incompat5 without -gdwarf-5.
168 * doc/invoke.texi (gvariable-location-views): New.
169 (gvariable-location-views=incompat5): New.
170 (gno-variable-location-views): New.
172 2018-02-08 David Malcolm <dmalcolm@redhat.com>
174 PR tree-optimization/84136
175 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
176 that the result of find_edge is non-NULL.
178 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
181 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
182 storing integer register in SImode. Fix cost of 256 and 512
183 byte aligned SSE register store.
185 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
187 * config/i386/i386.c (ix86_multiplication_cost): Fix
188 multiplication cost for TARGET_AVX512DQ.
190 2018-02-08 Marek Polacek <polacek@redhat.com>
192 PR tree-optimization/84238
193 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
196 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
198 PR tree-optimization/84265
199 * tree-vect-stmts.c (vectorizable_store): Don't treat
200 VMAT_CONTIGUOUS accesses as grouped.
201 (vectorizable_load): Likewise.
203 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
205 PR tree-optimization/81635
206 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
207 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
208 (test_round_for_mask): New functions.
209 (wide_int_cc_tests): Call test_round_for_mask.
210 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
211 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
212 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
213 range returned by get_range_info.
215 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
218 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
219 * symtab.c: Include builtins.h
220 (symtab_node::output_to_lto_symbol_table_p): Move here
221 from lto-streamer-out.c:output_symbol_p.
222 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
223 (output_symbol_p): Move all logic to symtab.c
224 (produce_symtab): Update.
226 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
228 * config/s390/s390-opts.h (enum indirect_branch): Define.
229 * config/s390/s390-protos.h (s390_return_addr_from_memory)
230 (s390_indirect_branch_via_thunk)
231 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
232 (enum s390_indirect_branch_type): Define.
233 * config/s390/s390.c (struct s390_frame_layout, struct
234 machine_function): Remove.
235 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
236 (indirect_branch_table_label_no, indirect_branch_table_name):
238 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
239 (enum s390_indirect_branch_option): Define.
240 (s390_return_addr_from_memory): New function.
241 (s390_handle_string_attribute): New function.
242 (s390_attribute_table): Add new attribute handler.
243 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
244 (s390_indirect_branch_via_thunk): New function.
245 (s390_indirect_branch_via_inline_thunk): New function.
246 (s390_function_ok_for_sibcall): When jumping via thunk disallow
247 sibling call optimization for non z10 compiles.
248 (s390_emit_call): Force indirect branch target to be a single
249 register. Add r1 clobber for non-z10 compiles.
250 (s390_emit_epilogue): Emit return jump via return_use expander.
251 (s390_reorg): Handle JUMP_INSNs as execute targets.
252 (s390_option_override_internal): Perform validity checks for the
253 new command line options.
254 (s390_indirect_branch_attrvalue): New function.
255 (s390_indirect_branch_settings): New function.
256 (s390_set_current_function): Invoke s390_indirect_branch_settings.
257 (s390_output_indirect_thunk_function): New function.
258 (s390_code_end): Implement target hook.
259 (s390_case_values_threshold): Implement target hook.
260 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
262 * config/s390/s390.h (struct s390_frame_layout)
263 (struct machine_function): Move here from s390.c.
264 (TARGET_INDIRECT_BRANCH_NOBP_RET)
265 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
266 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
267 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
268 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
269 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
270 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
271 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
272 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
273 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
274 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
275 (mnemonic attribute): Add values which aren't recognized
277 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
278 pattern for branch conversion. Fix mnemonic attribute.
279 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
280 indirect branch via thunk if requested.
281 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
282 ("*indirect_jump"): Disable for branch conversion using out of
284 ("indirect_jump_via_thunk<mode>_z10")
285 ("indirect_jump_via_thunk<mode>")
286 ("indirect_jump_via_inlinethunk<mode>_z10")
287 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
288 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
289 ("casesi_jump_via_inlinethunk<mode>_z10")
290 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
291 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
292 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
293 ("*indirect2_jump"): Disable for branch conversion.
294 ("casesi_jump"): Turn into expander and expand patterns for branch
296 ("return_use"): New expander.
297 ("*return"): Emit return via thunk and rename it to ...
298 ("*return<mode>"): ... this one.
299 * config/s390/s390.opt: Add new options and and enum for the
302 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
304 * lra-constraints.c (match_reload): Unconditionally use
305 gen_lowpart_SUBREG, rather than selecting between that
306 and equivalent gen_rtx_SUBREG code.
308 2018-02-08 Richard Biener <rguenther@suse.de>
310 PR tree-optimization/84233
311 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
312 changed flag instead of boguously re-using phi_inserted.
314 2018-02-08 Martin Jambor <mjambor@suse.cz>
316 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
317 static local variables.
319 2018-02-08 Richard Biener <rguenther@suse.de>
321 PR tree-optimization/84278
322 * tree-vect-stmts.c (vectorizable_store): When looking for
323 smaller vector types to perform grouped strided loads/stores
324 make sure the mode is supported by the target.
325 (vectorizable_load): Likewise.
327 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
329 * config/aarch64/aarch64.c (aarch64_components_for_bb):
330 Increase LDP/STP opportunities by adding adjacent callee-saves.
332 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
334 PR rtl-optimization/84068
335 PR rtl-optimization/83459
336 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
338 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
340 PR tree-optimization/84224
341 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
342 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
345 2018-02-07 Iain Sandoe <iain@codesourcery.com>
348 * config/rs6000/altivec.md (*restore_world): Remove LR use.
349 * config/rs6000/predicates.md (restore_world_operation): Adjust op
350 count, remove one USE.
352 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
354 * doc/install.texi (Configuration): Document the
355 --with-long-double-format={ibm,ieee} PowerPC configuration
359 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
360 Convert from define_expand to be define_insn_and_split. Rework
361 float/double/_Float128 conversions to QI/HI/SImode to work with
362 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
363 conversions to QI/HImode types did a store and then a load to
364 truncate the value. For conversions to VSX registers, don't split
365 the insn, instead emit the code directly. Use the code iterator
366 any_fix to combine signed and unsigned conversions.
367 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
368 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
369 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
370 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
371 (fix_<mode>di2_hw): Likewise.
372 (fixuns_<mode>di2_hw): Likewise.
373 (fix_<mode>si2_hw): Likewise.
374 (fixuns_<mode>si2_hw): Likewise.
375 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
376 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
377 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
378 fix<uns>_trunc<SFDF:mode>si2_p8.
379 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
381 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
382 (fix<uns>_<mode>_mem): Likewise.
383 (fctiw<u>z_<mode>_mem): Likewise.
384 (fix<uns>_<mode>_mem): Likewise.
385 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
386 the register allocator from doing a direct move to the GPRs to do
387 a store, and instead use the ISA 3.0 store byte/half-word from
388 vector register instruction. For IEEE 128-bit floating point,
389 also optimize stores of 32-bit ints.
390 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
392 2018-02-07 Alan Hayward <alan.hayward@arm.com>
394 * genextract.c (push_pathstr_operand): New function to support
396 (walk_rtx): Call push_pathstr_operand.
397 (print_path): Support [a-zA-Z].
399 2018-02-07 Richard Biener <rguenther@suse.de>
401 PR tree-optimization/84037
402 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
403 (cse_and_gimplify_to_preheader): Declare.
404 (vect_get_place_in_interleaving_chain): Likewise.
405 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
407 (_loop_vec_info::~_loop_vec_info): Delete it.
408 (cse_and_gimplify_to_preheader): New function.
409 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
410 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
411 (vectorizable_load): Likewise. For grouped stores always base
412 the IV on the first element.
413 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
414 condition before gimplifying.
416 2018-02-07 Jakub Jelinek <jakub@redhat.com>
418 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
419 *DIV_EXPR and *MOD_EXPR.
421 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
424 * config/i386/i386.c (ix86_option_override_internal): Mask out
425 the CF_SET bit when checking -fcf-protection.
427 2018-02-07 Tom de Vries <tom@codesourcery.com>
430 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
433 2018-02-07 Richard Biener <rguenther@suse.de>
435 PR tree-optimization/84204
436 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
439 PR tree-optimization/84205
440 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
441 special-case isl_ast_op_zdiv_r.
443 PR tree-optimization/84223
444 * graphite-scop-detection.c (gather_bbs::before_dom_children):
445 Only add conditions from within the region.
446 (gather_bbs::after_dom_children): Adjust.
448 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
451 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
452 * config/avr/avr.md: Only post-reload split REG-REG moves if
453 either register is REGERAL_REG_P.
455 2018-02-07 Jakub Jelinek <jakub@redhat.com>
457 PR tree-optimization/84235
458 * tree-ssa-scopedtables.c
459 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
460 if the subtraction is performed in floating point type where NaNs are
461 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
462 build 1. Formatting fix.
464 2018-02-06 Jakub Jelinek <jakub@redhat.com>
467 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
468 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
469 and skip it regardless of bb boundaries. Use CALL_P macro,
470 don't test INSN_P (insn) together with CALL_P or JUMP_P check
471 unnecessarily, formatting fix.
473 2018-02-06 Michael Collison <michael.collison@arm.com>
475 * config/arm/thumb2.md:
476 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
477 (*thumb_mov_notscc): Ditto.
479 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
482 * config/rs6000/rs6000.md (su code attribute): Use "u" for
483 unsigned_fix, not "s".
485 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
487 * configure.ac (gcc_fn_eh_frame_ro): New function.
488 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
489 correct .eh_frame permissions.
490 * configure: Regenerate.
492 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
494 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
497 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
499 * config/rs6000/rs6000.c (rs6000_option_override_internal):
500 Display warning message for -mno-speculate-indirect-jumps.
502 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
504 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
506 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
508 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
510 PR tree-optimization/84225
511 * tree-eh.c (find_trapping_overflow): Only call
512 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
514 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
517 * config/i386/i386.c: Reimplement the check of possible options
518 -mibt/-mshstk conbination. Change error messages.
519 * doc/invoke.texi: Fix a typo: remove extra '='.
521 2018-02-06 Marek Polacek <polacek@redhat.com>
523 PR tree-optimization/84228
524 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
526 2018-02-06 Tamar Christina <tamar.christina@arm.com>
529 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
530 emitted arch directives.
531 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
532 __ARM_FEATURE_COPROC before changing architectures.
534 2018-02-06 Richard Biener <rguenther@suse.de>
536 * config/i386/i386.c (print_reg): Fix typo.
537 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
539 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
541 * configure: Regenerate.
543 2018-02-05 Martin Sebor <msebor@redhat.com>
545 PR tree-optimization/83369
546 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
549 2018-02-05 Martin Liska <mliska@suse.cz>
551 * doc/invoke.texi: Cherry-pick upstream r323995.
553 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
555 * ira.c (ira_init_register_move_cost): Adjust comment.
557 2018-02-05 Martin Liska <mliska@suse.cz>
559 PR gcov-profile/84137
560 * doc/gcov.texi: Fix typo in documentation.
562 2018-02-05 Martin Liska <mliska@suse.cz>
564 PR gcov-profile/83879
565 * doc/gcov.texi: Document necessity of --dynamic-list-data when
566 using dlopen functionality.
568 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
570 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
571 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
572 _mm_maskz_range_ss, _mm_mask_range_round_ss,
573 _mm_maskz_range_round_ss): New intrinsics.
574 (__builtin_ia32_rangesd128_round)
575 (__builtin_ia32_rangess128_round): Remove.
576 (__builtin_ia32_rangesd128_mask_round,
577 __builtin_ia32_rangess128_mask_round): New builtins.
578 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
579 __builtin_ia32_rangess128_round): Remove.
580 (__builtin_ia32_rangesd128_mask_round,
581 __builtin_ia32_rangess128_mask_round): New builtins.
582 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
583 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
584 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
585 "<round_saeonly_constraint>")): Changed to ...
586 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
587 "<round_saeonly_scalar_constraint>")): ... this.
588 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
589 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
590 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
591 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
592 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
594 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
596 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
598 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
599 Remove all values except native, 8540 and 8548.
601 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
603 * config/i386/i386.c (ix86_output_function_return): Pass
604 INVALID_REGNUM, instead of -1, as invalid register number to
605 indirect_thunk_name and output_indirect_thunk.
607 2018-02-02 Julia Koval <julia.koval@intel.com>
609 * config.gcc: Add -march=icelake.
610 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
611 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
612 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
613 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
614 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
615 (processor_target_table): Add icelake.
616 (ix86_option_override_internal): Handle new PTAs.
617 (get_builtin_code_for_version): Handle icelake.
618 (M_INTEL_COREI7_ICELAKE): New.
619 (fold_builtin_cpu): Handle icelake.
620 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
621 * doc/invoke.texi: Add -march=icelake.
623 2018-02-02 Julia Koval <julia.koval@intel.com>
625 * config/i386/i386.c (ix86_option_override_internal): Change flags type
627 * wide-int-bitmask.h: New.
629 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
632 * config/i386/i386.md: Replace Pmode with word_mode in
633 builtin_setjmp_setup and builtin_longjmp to support x32.
635 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
639 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
641 (rs6000_supported_cpu_names): New static variable.
642 (linux_cpu_translation_table): Likewise.
643 (elf_platform) <cpu>: Define new static variable and use it.
644 Translate kernel AT_PLATFORM name to canonical name if needed.
645 Error if platform name is unknown.
647 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
650 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
652 2018-02-01 Jeff Law <law@redhat.com>
655 * config/i386/i386.c (release_scratch_register_on_entry): Add new
656 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
657 the scratch if RELEASE_VIA_POP is false.
658 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
659 If we have to save a temporary register, decrement SIZE appropriately.
660 Pass new arguments to release_scratch_register_on_entry.
661 (ix86_adjust_stack_and_probe): Likewise.
662 (ix86_emit_probe_stack_range): Pass new arguments to
663 release_scratch_register_on_entry.
665 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
667 PR rtl-optimization/84157
668 * combine.c (change_zero_ext): Use REG_P predicate in
669 front of HARD_REGISTER_P predicate.
671 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
673 * config/avr/avr.c (avr_option_override): Move disabling of
674 -fdelete-null-pointer-checks to...
675 * common/config/avr/avr-common.c (avr_option_optimization_table):
678 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
680 PR tree-optimization/81635
681 * tree-data-ref.c (split_constant_offset_1): For types that
682 wrap on overflow, try to use range info to prove that wrapping
685 2018-02-01 Renlin Li <renlin.li@arm.com>
688 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
690 (aarch64_register_move_cost): Likewise.
691 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
693 (REG_CLASS_NAMES): Likewise.
694 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
695 TAILCALL_ADDR_REGS. Remove IP registers.
696 * config/aarch64/aarch64.md (Ucs): Update register constraint.
698 2018-02-01 Richard Biener <rguenther@suse.de>
700 * domwalk.h (dom_walker::dom_walker): Add additional constructor
701 for specifying RPO order and allow NULL for that.
702 * domwalk.c (dom_walker::dom_walker): Likewise.
703 (dom_walker::walk): Handle NULL RPO order.
704 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
706 (rewrite_update_dom_walker): Likewise.
707 (mark_def_dom_walker): Likewise.
709 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
711 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
712 (aarch64_maybe_expand_sve_subreg_move): Declare.
713 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
714 * config/aarch64/predicates.md (aarch64_any_register_operand): New
716 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
717 that are semantically a reverse operation.
718 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
719 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
720 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
722 (aarch64_can_change_mode_class): For big-endian, forbid changes
723 between two SVE modes if they have different element sizes.
725 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
727 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
728 the TImode handling for big-endian targets.
730 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
732 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
733 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
735 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
736 Remove BSWAP handing for big-endian targets and use the form of
737 LD1RQ appropariate for the mode.
739 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
741 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
742 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
745 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
748 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
749 check for operands that need to go through aarch64_sve_reload_be.
751 2018-02-01 Jakub Jelinek <jakub@redhat.com>
753 PR tree-optimization/81661
754 PR tree-optimization/84117
755 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
756 * tree-eh.c: Include gimplify.h.
757 (find_trapping_overflow, replace_trapping_overflow,
758 rewrite_to_non_trapping_overflow): New functions.
759 * tree-vect-loop.c: Include tree-eh.h.
760 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
761 * tree-data-ref.c: Include tree-eh.h.
762 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
764 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
766 PR rtl-optimization/84123
767 * combine.c (change_zero_ext): Check if hard register satisfies
768 can_change_dest_mode before calling gen_lowpart_SUBREG.
770 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
773 * ira.c (ira_init_register_move_cost): Remove assert.
775 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
777 PR rtl-optimization/84071
778 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
779 * doc/tm.texi: Regenerate.
781 2018-01-31 Richard Biener <rguenther@suse.de>
783 PR tree-optimization/84132
784 * tree-data-ref.c (analyze_miv_subscript): Properly
785 check whether evolution_function_is_affine_multivariate_p
786 before calling gcd_of_steps_may_divide_p.
788 2018-01-31 Julia Koval <julia.koval@intel.com>
791 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
792 * config/i386/i386.md (rdpid_rex64) New.
793 (rdpid): Make 32bit only.
795 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
798 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
799 an IDENTIFIER_NODE for FUNCTION_TYPE's.
801 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
804 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
806 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
808 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
810 PR rtl-optimization/84071
811 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
812 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
814 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
816 * config/arc/arc.c (arc_handle_aux_attribute): New function.
817 (arc_attribute_table): Add 'aux' attribute.
818 (arc_in_small_data_p): Consider aux like variables.
819 (arc_is_aux_reg_p): New function.
820 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
821 (arc_get_aux_arg): New function.
822 (prepare_move_operands): Handle aux-register access.
823 (arc_handle_aux_attribute): New function.
824 * doc/extend.texi (ARC Variable attributes): Add subsection.
826 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
828 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
829 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
830 (arc_attribute_table): Add 'uncached' attribute.
831 (arc_print_operand): Print '.di' flag for uncached memory
833 (arc_in_small_data_p): Do not consider for small data the uncached
835 (arc_is_uncached_mem_p): New function.
836 * config/arc/predicates.md (compact_store_memory_operand): Check
837 for uncached memory accesses.
838 (nonvol_nonimm_operand): Likewise.
839 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
841 2018-01-31 Jakub Jelinek <jakub@redhat.com>
844 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
845 falign-loops=): Add Optimization flag.
847 2018-01-30 Jeff Law <law@redhat.com>
850 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
851 INT_REGISTERS_SAVED. Check it prior to calling
852 get_scratch_register_on_entry.
853 (ix86_adjust_stack_and_probe): Similarly.
854 (ix86_emit_probe_stack_range): Similarly.
855 (ix86_expand_prologue): Corresponding changes.
857 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
860 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
861 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
863 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
866 * lra-constraints.c (curr_insn_transform): Process AND in the
869 2018-01-30 Jakub Jelinek <jakub@redhat.com>
871 PR rtl-optimization/83986
872 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
873 dependence against last_pending_memory_flush in addition to
876 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
878 PR tree-optimization/81611
879 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
882 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
884 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
887 2018-01-30 Richard Biener <rguenther@suse.de>
888 Jakub Jelinek <jakub@redhat.com>
890 PR tree-optimization/84111
891 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
892 inner loops added during recursion, as they don't have up-to-date
895 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
898 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
899 (can_inline_edge_by_limits_p): ... here.
900 (can_early_inline_edge_p, check_callers,
901 update_caller_keys, update_callee_keys, recursive_inlining,
902 add_new_edges_to_heap, speculation_useful_p,
903 inline_small_functions,
904 inline_small_functions, flatten_function,
905 inline_to_all_callers_1): Update.
907 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
909 * profile-count.c (profile_count::combine_with_ipa_count): Handle
912 2018-01-30 Richard Biener <rguenther@suse.de>
914 PR tree-optimization/83008
915 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
916 invariant and constant vector uses in stmts when they need
919 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
922 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
923 * configure: Regenerate.
925 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
927 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
929 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
930 Use gen_rtx_REG rather than gen_lowpart.
932 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
934 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
935 rather than 0 when creating partial subregs.
937 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
939 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
942 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
945 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
946 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
947 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
948 flags. This restores the settings used before the 2017-07-24.
949 Turning off pre increment/decrement/modify allows IVOPTS to
950 optimize DF/SF loops where the index is an int.
952 2018-01-29 Richard Biener <rguenther@suse.de>
953 Kelvin Nilsen <kelvin@gcc.gnu.org>
956 * tree-vect-stmts.c (vectorizable_call): Don't call
957 targetm.vectorize_builtin_md_vectorized_function if callee is
960 2018-01-22 Carl Love <cel@us.ibm.com>
962 * doc/extend.tex: Fix typo in second arg in
963 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
965 2018-01-29 Richard Biener <rguenther@suse.de>
967 PR tree-optimization/84086
968 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
969 (flush_ssaname_freelist): When SSA names were released reset
972 2018-01-29 Richard Biener <rguenther@suse.de>
974 PR tree-optimization/84057
975 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
976 removed paths when removing edges.
978 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
980 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
981 -mfunction-return=@var{choice}.
983 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
986 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
988 (layout::print_source_line): Likewise.
989 (test_get_line_width_without_trailing_whitespace): Add test cases.
991 2018-01-27 Jakub Jelinek <jakub@redhat.com>
994 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
997 2018-01-26 Jim Wilson <jimw@sifive.com>
999 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1001 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1004 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1006 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1007 and CMP + SUB-immediate -> SUBS.
1009 2018-01-26 Martin Sebor <msebor@redhat.com>
1011 PR tree-optimization/83896
1012 * tree-ssa-strlen.c (get_string_len): Rename...
1013 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1014 Avoid assuming length is constant.
1015 (handle_char_store): Use HOST_WIDE_INT for string length.
1017 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1020 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1021 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1023 2018-01-26 Richard Biener <rguenther@suse.de>
1025 PR rtl-optimization/84003
1026 * dse.c (record_store): Only record redundant stores when
1027 the earlier store aliases at least all accesses the later one does.
1029 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1031 PR rtl-optimization/83985
1032 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1033 REG_CFA_RESTORE insns.
1034 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1035 REG_CFA_RESTORE insns here.
1038 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1039 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1041 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1043 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1044 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1045 (arc_init): Likewise.
1046 (arc_override_options): Likewise.
1047 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1049 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1051 * config/arc/arc.h (TARGET_DBNZ): Define.
1052 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1053 properly set the tune attribute.
1054 (dbnz): Use TARGET_DBNZ guard.
1055 * config/arc/arc.opt (mtune): Add core3 option.
1057 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1059 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1060 recognize new pic like addresses.
1061 (arc_delegitimize_address): Clean up.
1063 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1065 * config/arc/arc-arches.def: Option mrf16 valid for all
1067 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1068 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1069 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1070 * config/arc/arc-tables.opt: Regenerate.
1071 * config/arc/arc.c (arc_conditional_register_usage): Handle
1072 reduced register file case.
1073 (arc_file_start): Set must have build attributes.
1074 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1076 * config/arc/arc.opt (mrf16): Add new option.
1077 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1078 * config/arc/genmultilib.awk: Handle new mrf16 option.
1079 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1080 * config/arc/t-multilib: Regenerate.
1081 * doc/invoke.texi (ARC Options): Document mrf16 option.
1083 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1085 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1086 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1087 (arc_attribute_table): Add 'secure_call' attribute.
1088 (arc_print_operand): Print secure call operand.
1089 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1091 (arc_is_secure_call_p): New function. * config/arc/arc.md
1092 (call_i): Add support for sjli instruction.
1093 (call_value_i): Likewise.
1094 * config/arc/constraints.md (Csc): New constraint.
1096 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1097 John Eric Martin <John.Martin@emmicro-us.com>
1099 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1100 * config/arc/arc.c (_arc_jli_section): New struct.
1101 (arc_jli_section): New type.
1102 (rc_jli_sections): New static variable.
1103 (arc_handle_jli_attribute): New function.
1104 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1105 (arc_file_end): New function.
1106 (TARGET_ASM_FILE_END): Define.
1107 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1108 (arc_add_jli_section): New function.
1109 (jli_call_scan): Likewise.
1110 (arc_reorg): Call jli_call_scan.
1111 (arc_output_addsi): Remove 'S' from printing asm operand.
1112 (arc_is_jli_call_p): New function.
1113 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1115 (movhi_insn): Likewise.
1116 (movsi_insn): Likewise.
1117 (movsi_set_cc_insn): Likewise.
1118 (loadqi_update): Likewise.
1119 (load_zeroextendqisi_update): Likewise.
1120 (load_signextendqisi_update): Likewise.
1121 (loadhi_update): Likewise.
1122 (load_zeroextendhisi_update): Likewise.
1123 (load_signextendhisi_update): Likewise.
1124 (loadsi_update): Likewise.
1125 (loadsf_update): Likewise.
1126 (movsicc_insn): Likewise.
1127 (bset_insn): Likewise.
1128 (bxor_insn): Likewise.
1129 (bclr_insn): Likewise.
1130 (bmsk_insn): Likewise.
1131 (bicsi3_insn): Likewise.
1132 (cmpsi_cc_c_insn): Likewise.
1133 (movsi_ne): Likewise.
1134 (movsi_cond_exec): Likewise.
1135 (clrsbsi2): Likewise.
1146 (call_i): Remove 'S' asm letter, add jli instruction.
1147 (call_value_i): Likewise.
1148 * config/arc/arc.op (mjli-always): New option.
1149 * config/arc/constraints.md (Cji): New constraint.
1150 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1152 (subsf3_fpx): Likewise.
1153 (mulsf3_fpx): Likewise.
1154 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1156 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1158 * doc/invoke.texi (ARC): Document mjli-always option.
1160 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1162 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1163 and use incw and decw where possible
1164 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1166 2018-01-26 Richard Biener <rguenther@suse.de>
1168 PR tree-optimization/81082
1169 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1170 association if it requires casting to unsigned.
1171 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1172 from fold_plusminus_mult_expr to catch important cases late when
1173 range info is available.
1175 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1177 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1178 * configure.ac (hidden_linkonce): New test.
1179 * configure: Regenerate.
1180 * config.in: Regenerate.
1182 2018-01-26 Julia Koval <julia.koval@intel.com>
1184 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1185 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1186 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1187 _mm_mask_bitshuffle_epi64_mask): Fix type.
1188 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1189 USI_FTYPE_V4DI_V4DI_USI): Remove.
1190 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1191 __builtin_ia32_vpshufbitqmb256_mask,
1192 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1193 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1194 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1196 2018-01-26 Alan Modra <amodra@gmail.com>
1199 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1200 UNSPEC_VBPERMQ. Sort other unspecs.
1202 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1204 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1206 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1209 * predict.c (drop_profile): Do not push/pop cfun; update also
1211 (handle_missing_profiles): Fix logic looking for zero profiles.
1213 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1216 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1217 on functions with #pragma omp declare simd or functions with simd
1219 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1220 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1221 Remove trailing \n from warning_at calls.
1223 2018-01-25 Tom de Vries <tom@codesourcery.com>
1226 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1227 for neutered workers.
1229 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1232 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1233 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1235 2018-01-24 Jeff Law <law@redhat.com>
1238 * i386.c (get_probe_interval): Move to earlier point.
1239 (ix86_compute_frame_layout): If -fstack-clash-protection and
1240 the frame is larger than the probe interval, then use pushes
1241 to save registers rather than reg->mem moves.
1242 (ix86_expand_prologue): Remove conditional for int_registers_saved
1245 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1248 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1249 min/max for never referenced object.
1251 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1254 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1256 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1257 attributes from DECL_ATTRIBUTES (decl) without affecting
1258 DECL_ATTRIBUTES (current_function_decl).
1259 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1260 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1262 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1264 PR tree-optimization/83979
1265 * fold-const.c (fold_comparison): Use constant_boolean_node
1266 instead of boolean_{true,false}_node.
1268 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1270 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1273 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1275 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1276 Simplify the clause that sets the length attribute.
1277 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1278 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1279 clause that sets the length attribute.
1280 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1282 2018-01-24 Tom de Vries <tom@codesourcery.com>
1285 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1286 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1287 Add strict parameter.
1288 (prevent_branch_around_nothing): Insert dummy insn between branch to
1289 label and label with no ptx insn inbetween.
1290 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1292 2018-01-24 Tom de Vries <tom@codesourcery.com>
1295 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1296 for neutered threads in warp.
1297 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1299 2018-01-24 Richard Biener <rguenther@suse.de>
1301 PR tree-optimization/83176
1302 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1305 2018-01-24 Richard Biener <rguenther@suse.de>
1307 PR tree-optimization/82819
1308 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1309 code generating pluses that are no-ops in the target precision.
1311 2018-01-24 Richard Biener <rguenther@suse.de>
1314 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1316 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1318 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1319 to merge probabilities.
1320 * predict.c (probably_never_executed): Also mark as cold functions
1321 with global 0 profile and guessed local profile.
1322 * profile-count.c (profile_probability::combine_with_count): New
1324 * profile-count.h (profile_probability::operator*,
1325 profile_probability::operator*=, profile_probability::operator/,
1326 profile_probability::operator/=): Reduce precision to adjusted
1327 and set value to guessed on contradictory divisions.
1328 (profile_probability::combine_with_freq): Remove.
1329 (profile_probability::combine_wiht_count): Declare.
1330 (profile_count::force_nonzero):: Set to adjusted.
1331 (profile_count::probability_in):: Set quality to adjusted.
1332 * tree-ssa-tail-merge.c (replace_block_by): Use
1335 2018-01-23 Andrew Waterman <andrew@sifive.com>
1336 Jim Wilson <jimw@sifive.com>
1338 * config/riscv/riscv.c (riscv_stack_boundary): New.
1339 (riscv_option_override): Set riscv_stack_boundary. Handle
1340 riscv_preferred_stack_boundary_arg.
1341 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1342 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1343 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1344 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1345 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1346 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1348 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1351 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1352 of struct ix86_frame.
1353 (ix86_expand_epilogue): Likewise. Add a local variable for
1354 the reg_save_offset field in struct ix86_frame.
1356 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1358 PR tree-optimization/82604
1359 * tree-loop-distribution.c (enum partition_kind): New enum item
1360 PKIND_PARTIAL_MEMSET.
1361 (partition_builtin_p): Support above new enum item.
1362 (generate_code_for_partition): Ditto.
1363 (compute_access_range): Differentiate cases that equality can be
1364 proven at all loops, the innermost loops or no loops.
1365 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1366 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1367 (finalize_partitions, distribute_loop): Don't fuse partition of
1368 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1369 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1372 2018-01-23 Martin Liska <mliska@suse.cz>
1374 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1375 order to ignore the predictor.
1376 (PRED_POLYMORPHIC_CALL): Likewise.
1377 (PRED_RECURSIVE_CALL): Likewise.
1379 2018-01-23 Martin Liska <mliska@suse.cz>
1381 * tree-profile.c (tree_profiling): Print function header to
1382 aware reader which function we are working on.
1383 * value-prof.c (gimple_find_values_to_profile): Do not print
1384 not interesting value histograms.
1386 2018-01-23 Martin Liska <mliska@suse.cz>
1388 * profile-count.h (enum profile_quality): Add
1389 profile_uninitialized as the first value. Do not number values
1390 as they are zero based.
1391 (profile_count::verify): Update sanity check.
1392 (profile_probability::verify): Likewise.
1394 2018-01-23 Nathan Sidwell <nathan@acm.org>
1396 * doc/invoke.texi (ffor-scope): Deprecate.
1398 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1400 PR tree-optimization/83510
1401 * domwalk.c (set_all_edges_as_executable): New function.
1402 (dom_walker::dom_walker): Convert bool param
1403 "skip_unreachable_blocks" to enum reachability. Move setup of
1404 edge flags to set_all_edges_as_executable and only do it when
1405 reachability is REACHABLE_BLOCKS.
1406 * domwalk.h (enum dom_walker::reachability): New enum.
1407 (dom_walker::dom_walker): Convert bool param
1408 "skip_unreachable_blocks" to enum reachability.
1409 (set_all_edges_as_executable): New decl.
1410 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1411 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1413 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1414 but converting true to REACHABLE_BLOCKS.
1415 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1417 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1418 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1419 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1421 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1422 if check_all_array_refs will be called.
1424 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1426 * tree.c (selftest::test_location_wrappers): Add more test
1429 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1431 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1432 (selftest::test_bit_in_range): Likewise.
1434 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1437 * doc/sourcebuild.texi (vect_float): Say that the selector
1438 only describes the situation when -funsafe-math-optimizations is on.
1439 (vect_float_strict): Document.
1441 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1443 PR tree-optimization/83965
1444 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1445 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1446 instead of checking only for a reduction.
1447 (vect_recog_widen_sum_pattern): Likewise.
1449 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1451 * predict.c (probably_never_executed): Only use precise profile info.
1452 (compute_function_frequency): Skip after inlining hack since we now
1453 have quality checking.
1455 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1457 * profile-count.h (profile_probability::very_unlikely,
1458 profile_probability::unlikely, profile_probability::even): Set
1459 precision to guessed.
1461 2018-01-23 Richard Biener <rguenther@suse.de>
1463 PR tree-optimization/83963
1464 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1465 Properly terminate dominator walk when crossing the exit edge not
1466 when visiting its source block.
1468 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1471 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1472 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1474 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1476 PR tree-optimization/83957
1477 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1478 semicolon after for body surrounded by braces.
1480 PR tree-optimization/83081
1481 * profile-count.h (profile_probability::split): New method.
1482 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1483 Use profile_probability::split.
1484 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1485 when splitting a single conditional jump into 2.
1487 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1489 PR tree-optimization/69452
1490 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1493 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1495 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1496 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1497 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1499 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1501 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1502 * config/rl78/rl78.md: New define_expand "movdi"
1503 * config/rl78/rl78.c: New function definition rl78_split_movdi
1505 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1508 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1510 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1511 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1512 128-bit to produce an UNSPEC move to get the double word with the
1513 signbit and then a shift directly to do signbit.
1514 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1515 implementation with a new version that just does either a direct
1516 move or a regular move. Move memory interface to separate insns.
1517 Move insns so they are next to the expander.
1518 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1519 with signbit move. Split big and little endian case.
1520 (signbit<mode>2_dm_mem_le): Likewise.
1521 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1522 (signbit<mode>2_dm2): Likewise.
1524 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1526 * config/rl78/rl78.md: New define_expand "anddi3".
1528 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1530 * config/rl78/rl78.md: New define_expand "umindi3".
1532 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1534 * config/rl78/rl78.md: New define_expand "smindi3".
1536 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1538 * config/rl78/rl78.md: New define_expand "smaxdi3".
1540 2018-01-22 Carl Love <cel@us.ibm.com>
1542 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1543 LVX_V1TI): Add macro expansion.
1544 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1545 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1546 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1547 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1548 Change check to determine if the instruction is a byte reversing
1549 entry. Fix typo in comment.
1550 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1551 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1552 Add def_builtin calls for new builtins.
1553 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1554 Add define_insn expansion.
1556 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1558 * config/rl78/rl78.md: New define_expand "umaxdi3".
1560 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1562 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1563 for non-QImode registers
1565 2018-01-22 Richard Biener <rguenther@suse.de>
1567 PR tree-optimization/83963
1568 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1569 including the loop exit block.
1570 (scop_detection::merge_sese): Likewise.
1571 (scop_detection::add_scop): Do it here instead.
1573 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1575 * doc/sourcebuild.texi (arm_softfloat): Document.
1577 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1580 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1581 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1582 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1584 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1585 David Edelsohn <dje.gcc@gmail.com>
1588 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1589 Change "crset eq" to "crset 2".
1590 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1591 (*call_indirect_aix<mode>_nospec): Likewise.
1592 (*call_value_indirect_aix<mode>_nospec): Likewise.
1593 (*call_indirect_elfv2<mode>_nospec): Likewise.
1594 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1595 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1596 change assembly output from . to $.
1597 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1598 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1599 (*tablejump<mode>_internal1_nospec): Likewise.
1601 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1604 * config/sh/sh_optimize_sett_clrt.cc:
1605 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1607 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1609 PR tree-optimization/83940
1610 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1611 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1612 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1613 use it to pass back the definition type.
1614 (vect_check_store_rhs): Likewise rhs_dt_out.
1615 (vect_build_gather_load_calls): Add a mask_dt argument and use
1616 it instead of a call to vect_is_simple_use.
1617 (vectorizable_store): Update calls to vect_check_load_store_mask
1618 and vect_check_store_rhs. Use the dt returned by the latter instead
1619 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1620 instead of calls to vect_is_simple_use. Pass the scalar rather
1621 than the vector operand to vect_is_simple_use when handling
1622 second and subsequent copies of an rhs value.
1623 (vectorizable_load): Update calls to vect_check_load_store_mask
1624 and vect_build_gather_load_calls. Use the cached mask_dt and
1625 gs_info.offset_dt instead of calls to vect_is_simple_use.
1627 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1630 * tree-emutls.c: Include gimplify.h.
1631 (lower_emutls_2): New function.
1632 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1633 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1634 it before further processing.
1637 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1638 UINTVAL (trueop1) instead of INTVAL (op1).
1640 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1644 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1645 INCOMING_FRAME_SP_OFFSET if not defined.
1646 (scan_trace): Add ENTRY argument. If true and
1647 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1648 emit a note to adjust the CFA offset.
1649 (create_cfi_notes): Adjust scan_trace callers.
1650 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1651 INCOMING_FRAME_SP_OFFSET in the CIE.
1652 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1653 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1655 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1656 * doc/tm.texi: Regenerated.
1658 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1660 PR rtl-optimization/83147
1661 * lra-constraints.c (remove_inheritance_pseudos): Use
1662 lra_substitute_pseudo_within_insn.
1664 2018-01-19 Tom de Vries <tom@codesourcery.com>
1665 Cesar Philippidis <cesar@codesourcery.com>
1668 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1670 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1673 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1674 spaces for function labels.
1676 2018-01-19 Martin Liska <mliska@suse.cz>
1678 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1679 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1680 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1681 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1682 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1683 (PRED_CONST_RETURN): Change from 69 to 65.
1684 (PRED_NULL_RETURN): Change from 91 to 71.
1685 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1686 (PRED_LOOP_GUARD): Change from 66 to 73.
1688 2018-01-19 Martin Liska <mliska@suse.cz>
1690 * predict.c (predict_insn_def): Add new assert.
1691 (struct branch_predictor): Change type to signed integer.
1692 (test_prediction_value_range): Amend test to cover
1694 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1695 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1696 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1697 (PRED_LOOP_IV_COMPARE): Likewise.
1698 * predict.h (PROB_UNINITIALIZED): Define new constant.
1700 2018-01-19 Martin Liska <mliska@suse.cz>
1702 * predict.c (dump_prediction): Add new format for
1703 analyze_brprob.py script which is enabled with -details
1705 * profile-count.h (precise_p): New function.
1707 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1709 PR tree-optimization/83922
1710 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1711 there are no statements that need masking.
1712 (vect_active_double_reduction_p): New function.
1713 (vect_analyze_loop_operations): Use it when handling phis that
1714 are not in the loop header.
1716 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1718 PR tree-optimization/83914
1719 * tree-vect-loop.c (vectorizable_induction): Don't convert
1720 init_expr or apply the peeling adjustment for inductions
1721 that are nested within the vectorized loop.
1723 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1725 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1728 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1732 * function.h (gimplify_parameters): Add gimple_seq * argument.
1733 * function.c: Include gimple.h and options.h.
1734 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1735 for the added local temporaries if needed.
1736 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1737 if there are any parameter cleanups, wrap whole body into a
1738 try/finally with the cleanups.
1740 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1743 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1744 Use GET_MODE_CLASS for scalar floating point.
1746 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1750 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1751 Fix call of call_cgraph_insertion_hooks.
1753 2018-01-18 Martin Sebor <msebor@redhat.com>
1755 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1757 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1760 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1763 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1766 * common.opt: (-ffile-prefix-map): New option.
1767 * opts.c (common_handle_option): Defer it.
1768 * opts-global.c (handle_common_deferred_options): Handle it.
1769 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1770 * file-prefix-map.h: New file.
1771 (remap_debug_filename, add_debug_prefix_map): ...here.
1772 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1773 * final.c (debug_prefix_map, add_debug_prefix_map
1774 remap_debug_filename): Move to...
1775 * file-prefix-map.c: New file.
1776 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1777 generalize, get rid of alloca(), use strrchr() instead of strchr().
1778 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1779 Implement in terms of add_prefix_map().
1780 (remap_macro_filename, remap_debug_filename): Implement in term of
1782 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1783 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1784 * dbxout.c: Include file-prefix-map.h.
1785 * varasm.c: Likewise.
1786 * vmsdbgout.c: Likewise.
1787 * xcoffout.c: Likewise.
1788 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1789 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1790 * doc/invoke.texi (-ffile-prefix-map): Document.
1791 (-fdebug-prefix-map): Update description.
1793 2018-01-18 Martin Liska <mliska@suse.cz>
1795 * config/i386/i386.c (indirect_thunk_name): Document that also
1797 (output_indirect_thunk): Document why both instructions
1798 (pause and lfence) are generated.
1800 2018-01-18 Richard Biener <rguenther@suse.de>
1802 PR tree-optimization/83887
1803 * graphite-scop-detection.c
1804 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1805 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1806 (scop_detection::merge_sese): Re-implement with a flood-fill
1807 algorithm that properly finds a SESE region if it exists.
1809 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1812 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1813 pointer_diff optimizations use view_convert instead of convert.
1815 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1817 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1818 Generate different code for -mno-speculate-indirect-jumps.
1819 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1820 (*call_indirect_aix<mode>): Disable for
1821 -mno-speculate-indirect-jumps.
1822 (*call_indirect_aix<mode>_nospec): New define_insn.
1823 (*call_value_indirect_aix<mode>): Disable for
1824 -mno-speculate-indirect-jumps.
1825 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1826 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1827 -mno-speculate-indirect-jumps.
1828 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1830 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1832 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1833 long double type, set the flags for noting the default long double
1834 type, even if we don't pass or return a long double type.
1836 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1839 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1842 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1844 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1845 support for merge[hl].
1846 (fold_mergehl_helper): New helper function.
1847 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1848 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1849 (altivec_vmrglw_direct): Add xxmrglw insn.
1851 2018-01-17 Andrew Waterman <andrew@sifive.com>
1853 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1854 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1856 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1859 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1860 call the lto_location_cache before reading the
1861 DECL_SOURCE_LOCATION of the types.
1863 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1864 Richard Sandiford <richard.sandiford@linaro.org>
1866 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1867 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1868 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1869 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1870 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1872 * config/aarch64/constraints.md (aarch64_movti_operand):
1874 * config/aarch64/predicates.md (Uti): Add new constraint.
1876 2018-01-17 Carl Love <cel@us.ibm.com>
1877 * config/rs6000/vsx.md (define_expand xl_len_r,
1878 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1879 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1881 (define_expand, define_insn): Move the shift left from the
1882 define_insn to the define_expand for lxvl and stxvl instructions.
1883 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1884 and XL_LEN_R definitions to PURE.
1886 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1888 * config/i386/i386.c (indirect_thunk_name): Declare regno
1889 as unsigned int. Compare regno with INVALID_REGNUM.
1890 (output_indirect_thunk): Ditto.
1891 (output_indirect_thunk_function): Ditto.
1892 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1893 in the call to output_indirect_thunk_function.
1895 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1898 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1899 rather than the size of inner_type to determine the stack slot size
1900 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1902 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1905 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1908 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1910 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1911 endian Linux systems to optionally enable multilibs for selecting
1912 the long double type if the user configured an explicit type.
1913 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1914 have no long double multilibs if not defined.
1915 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1916 warn if the user used -mabi={ieee,ibm}longdouble and we built
1917 multilibs for long double.
1918 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1919 appropriate multilib option.
1920 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1922 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1923 for building long double multilibs.
1924 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1926 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1928 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1931 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1933 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1936 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1939 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1942 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1944 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1945 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1947 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1949 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1950 different rtl trees depending on TARGET_64BIT.
1951 (rs6000_gen_lvx): Likewise.
1953 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1955 * config/visium/visium.md (nop): Tweak comment.
1956 (hazard_nop): Likewise.
1958 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1960 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1961 -mspeculate-indirect-jumps.
1962 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1963 for -mno-speculate-indirect-jumps.
1964 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1965 (*call_value_indirect_elfv2<mode>): Disable for
1966 -mno-speculate-indirect-jumps.
1967 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1968 (indirect_jump): Emit different RTL for
1969 -mno-speculate-indirect-jumps.
1970 (*indirect_jump<mode>): Disable for
1971 -mno-speculate-indirect-jumps.
1972 (*indirect_jump<mode>_nospec): New define_insn.
1973 (tablejump): Emit different RTL for
1974 -mno-speculate-indirect-jumps.
1975 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1976 (tablejumpsi_nospec): New define_expand.
1977 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1978 (tablejumpdi_nospec): New define_expand.
1979 (*tablejump<mode>_internal1): Disable for
1980 -mno-speculate-indirect-jumps.
1981 (*tablejump<mode>_internal1_nospec): New define_insn.
1982 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1985 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1987 * caller-save.c (insert_save): Drop unnecessary parameter. All
1990 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1991 Richard Biener <rguenth@suse.de>
1994 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1995 return early, inline manually is_gimple_sizepos. Make sure if we
1996 call gimplify_expr we don't end up with a gimple constant.
1997 * tree.c (variably_modified_type_p): Don't return true for
1998 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1999 * gimplify.h (is_gimple_sizepos): Remove.
2001 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2003 PR tree-optimization/83857
2004 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2005 vectorizable_live_operation for pure SLP statements.
2006 (vectorizable_live_operation): Handle PHIs.
2008 2018-01-16 Richard Biener <rguenther@suse.de>
2010 PR tree-optimization/83867
2011 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2012 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2014 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2017 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2018 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2019 If off is not INTEGER_CST, issue a may not be aligned warning
2020 rather than isn't aligned. Use isn%'t rather than isn't.
2021 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2023 <case MULT_EXPR>: Improve the case when bottom and one of the
2024 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2025 operand, in that case check if the other operand is multiple of
2026 bottom divided by the INTEGER_CST operand.
2028 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2031 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2032 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2033 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2034 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2035 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2036 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2037 (pa_function_arg_size): New function.
2039 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2041 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2042 in a separate statement.
2044 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2046 PR tree-optimization/83847
2047 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2048 group gathers and scatters.
2050 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2052 PR rtl-optimization/86620
2053 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2055 PR rtl-optimization/83213
2056 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2057 to last if both are JUMP_INSNs.
2059 PR tree-optimization/83843
2060 * gimple-ssa-store-merging.c
2061 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2062 store_immediate_info for bswap/nop orig_stores.
2064 2018-01-15 Andrew Waterman <andrew@sifive.com>
2066 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2068 <UDIV>: Increase cost if !TARGET_DIV.
2070 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2072 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2073 (define_attr "cr_logical_3op"): New.
2074 (cceq_ior_compare): Adjust.
2075 (cceq_ior_compare_complement): Adjust.
2076 (*cceq_rev_compare): Adjust.
2077 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2078 (is_cracked_insn): Adjust.
2079 (insn_must_be_first_in_group): Adjust.
2080 * config/rs6000/40x.md: Adjust.
2081 * config/rs6000/440.md: Adjust.
2082 * config/rs6000/476.md: Adjust.
2083 * config/rs6000/601.md: Adjust.
2084 * config/rs6000/603.md: Adjust.
2085 * config/rs6000/6xx.md: Adjust.
2086 * config/rs6000/7450.md: Adjust.
2087 * config/rs6000/7xx.md: Adjust.
2088 * config/rs6000/8540.md: Adjust.
2089 * config/rs6000/cell.md: Adjust.
2090 * config/rs6000/e300c2c3.md: Adjust.
2091 * config/rs6000/e500mc.md: Adjust.
2092 * config/rs6000/e500mc64.md: Adjust.
2093 * config/rs6000/e5500.md: Adjust.
2094 * config/rs6000/e6500.md: Adjust.
2095 * config/rs6000/mpc.md: Adjust.
2096 * config/rs6000/power4.md: Adjust.
2097 * config/rs6000/power5.md: Adjust.
2098 * config/rs6000/power6.md: Adjust.
2099 * config/rs6000/power7.md: Adjust.
2100 * config/rs6000/power8.md: Adjust.
2101 * config/rs6000/power9.md: Adjust.
2102 * config/rs6000/rs64.md: Adjust.
2103 * config/rs6000/titan.md: Adjust.
2105 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2107 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2108 ix86_indirect_branch_register logic.
2110 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2112 * config/i386/constraints.md (Bs): Update
2113 ix86_indirect_branch_register check. Don't check
2114 ix86_indirect_branch_register with GOT_memory_operand.
2116 * config/i386/predicates.md (GOT_memory_operand): Don't check
2117 ix86_indirect_branch_register here.
2118 (GOT32_symbol_operand): Likewise.
2120 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2122 * config/i386/predicates.md (constant_call_address_operand):
2123 Rewrite ix86_indirect_branch_register logic.
2124 (sibcall_insn_operand): Likewise.
2126 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2128 * config/i386/constraints.md (Bs): Replace
2129 ix86_indirect_branch_thunk_register with
2130 ix86_indirect_branch_register.
2132 * config/i386/i386.md (indirect_jump): Likewise.
2133 (tablejump): Likewise.
2134 (*sibcall_memory): Likewise.
2135 (*sibcall_value_memory): Likewise.
2136 Peepholes of indirect call and jump via memory: Likewise.
2137 * config/i386/i386.opt: Likewise.
2138 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2139 (GOT_memory_operand): Likewise.
2140 (call_insn_operand): Likewise.
2141 (sibcall_insn_operand): Likewise.
2142 (GOT32_symbol_operand): Likewise.
2144 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2147 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2148 type rather than type addr's type points to.
2149 (expand_omp_atomic_mutex): Likewise.
2150 (expand_omp_atomic): Likewise.
2152 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2155 * config/i386/i386.c (output_indirect_thunk_function): Use
2156 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2157 for __x86_return_thunk.
2159 2018-01-15 Richard Biener <rguenther@suse.de>
2162 * expmed.c (extract_bit_field_1): Fix typo.
2164 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2167 * config/arm/iterators.md (VF): New mode iterator.
2168 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2169 Remove integer-related logic from pattern.
2170 (neon_vabd<mode>_3): Likewise.
2172 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2175 * common.opt (fstrict-overflow): No longer an alias.
2176 (fwrapv-pointer): New option.
2177 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2178 also for pointer types based on flag_wrapv_pointer.
2179 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2180 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2181 opts->x_flag_wrapv got set.
2182 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2183 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2184 POINTER_TYPE_OVERFLOW_UNDEFINED.
2185 * match.pd: Likewise in address comparison pattern.
2186 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2188 2018-01-15 Richard Biener <rguenther@suse.de>
2191 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2192 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2193 Reset type names to their identifier if their TYPE_DECL doesn't
2194 have linkage (and thus is used for ODR and devirt).
2195 (save_debug_info_for_decl): Remove.
2196 (save_debug_info_for_type): Likewise.
2197 (add_tree_to_fld_list): Adjust.
2198 * tree-pretty-print.c (dump_generic_node): Make dumping of
2199 type names more robust.
2201 2018-01-15 Richard Biener <rguenther@suse.de>
2203 * BASE-VER: Bump to 8.0.1.
2205 2018-01-14 Martin Sebor <msebor@redhat.com>
2208 * builtins.c (check_access): Avoid warning when the no-warning bit
2211 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2213 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2214 * ira-color (allocno_hard_regs_compare): Likewise.
2216 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2219 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2220 Use .pushsection/.popsection.
2222 2018-01-14 Martin Sebor <msebor@redhat.com>
2225 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2227 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2229 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2230 entry from extra_headers.
2231 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2232 extra_headers, make the list bitwise identical to the i?86-*-* one.
2234 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2236 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2237 -mcmodel=large with -mindirect-branch=thunk,
2238 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2239 -mfunction-return=thunk-extern.
2240 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2241 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2242 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2244 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2246 * config/i386/i386.c (print_reg): Print the name of the full
2247 integer register without '%'.
2248 (ix86_print_operand): Handle 'V'.
2249 * doc/extend.texi: Document 'V' modifier.
2251 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2253 * config/i386/constraints.md (Bs): Disallow memory operand for
2254 -mindirect-branch-register.
2256 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2257 (GOT_memory_operand): Likewise.
2258 (call_insn_operand): Likewise.
2259 (sibcall_insn_operand): Likewise.
2260 (GOT32_symbol_operand): Likewise.
2261 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2262 for -mindirect-branch-register.
2263 (tablejump): Likewise.
2264 (*sibcall_memory): Likewise.
2265 (*sibcall_value_memory): Likewise.
2266 Disallow peepholes of indirect call and jump via memory for
2267 -mindirect-branch-register.
2268 (*call_pop): Replace m with Bw.
2269 (*call_value_pop): Likewise.
2270 (*sibcall_pop_memory): Replace m with Bs.
2271 * config/i386/i386.opt (mindirect-branch-register): New option.
2272 * doc/invoke.texi: Document -mindirect-branch-register option.
2274 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2276 * config/i386/i386-protos.h (ix86_output_function_return): New.
2277 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2278 set function_return_type.
2279 (indirect_thunk_name): Add ret_p to indicate thunk for function
2281 (output_indirect_thunk_function): Pass false to
2282 indirect_thunk_name.
2283 (ix86_output_indirect_branch_via_reg): Likewise.
2284 (ix86_output_indirect_branch_via_push): Likewise.
2285 (output_indirect_thunk_function): Create alias for function
2286 return thunk if regno < 0.
2287 (ix86_output_function_return): New function.
2288 (ix86_handle_fndecl_attribute): Handle function_return.
2289 (ix86_attribute_table): Add function_return.
2290 * config/i386/i386.h (machine_function): Add
2291 function_return_type.
2292 * config/i386/i386.md (simple_return_internal): Use
2293 ix86_output_function_return.
2294 (simple_return_internal_long): Likewise.
2295 * config/i386/i386.opt (mfunction-return=): New option.
2296 (indirect_branch): Mention -mfunction-return=.
2297 * doc/extend.texi: Document function_return function attribute.
2298 * doc/invoke.texi: Document -mfunction-return= option.
2300 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2302 * config/i386/i386-opts.h (indirect_branch): New.
2303 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2304 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2305 with local indirect jump when converting indirect call and jump.
2306 (ix86_set_indirect_branch_type): New.
2307 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2308 (indirectlabelno): New.
2309 (indirect_thunk_needed): Likewise.
2310 (indirect_thunk_bnd_needed): Likewise.
2311 (indirect_thunks_used): Likewise.
2312 (indirect_thunks_bnd_used): Likewise.
2313 (INDIRECT_LABEL): Likewise.
2314 (indirect_thunk_name): Likewise.
2315 (output_indirect_thunk): Likewise.
2316 (output_indirect_thunk_function): Likewise.
2317 (ix86_output_indirect_branch_via_reg): Likewise.
2318 (ix86_output_indirect_branch_via_push): Likewise.
2319 (ix86_output_indirect_branch): Likewise.
2320 (ix86_output_indirect_jmp): Likewise.
2321 (ix86_code_end): Call output_indirect_thunk_function if needed.
2322 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2324 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2325 (ix86_attribute_table): Add indirect_branch.
2326 * config/i386/i386.h (machine_function): Add indirect_branch_type
2327 and has_local_indirect_jump.
2328 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2330 (tablejump): Likewise.
2331 (*indirect_jump): Use ix86_output_indirect_jmp.
2332 (*tablejump_1): Likewise.
2333 (simple_return_indirect_internal): Likewise.
2334 * config/i386/i386.opt (mindirect-branch=): New option.
2335 (indirect_branch): New.
2338 (thunk-inline): Likewise.
2339 (thunk-extern): Likewise.
2340 * doc/extend.texi: Document indirect_branch function attribute.
2341 * doc/invoke.texi: Document -mindirect-branch= option.
2343 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2346 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2348 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2350 * ipa-inline.c (want_inline_small_function_p): Return false if
2351 inlining has already failed with CIF_FINAL_ERROR.
2352 (update_caller_keys): Call want_inline_small_function_p before
2354 (update_callee_keys): Likewise.
2356 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2358 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2360 (rs6000_quadword_masked_address_p): Likewise.
2361 (quad_aligned_load_p): Likewise.
2362 (quad_aligned_store_p): Likewise.
2363 (const_load_sequence_p): Add comment to describe the outer-most loop.
2364 (mimic_memory_attributes_and_flags): New function.
2365 (rs6000_gen_stvx): Likewise.
2366 (replace_swapped_aligned_store): Likewise.
2367 (rs6000_gen_lvx): Likewise.
2368 (replace_swapped_aligned_load): Likewise.
2369 (replace_swapped_load_constant): Capitalize argument name in
2370 comment describing this function.
2371 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2372 and stores that access quad-word aligned addresses and replace
2373 with stvx or lvx instructions when appropriate.
2374 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2375 New function prototype.
2376 (rs6000_quadword_masked_address_p): Likewise.
2377 (rs6000_gen_lvx): Likewise.
2378 (rs6000_gen_stvx): Likewise.
2379 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2380 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2381 when memory address is aligned.
2382 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2383 this split to select lvx instruction when memory address is aligned.
2384 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2385 instruction when memory address is aligned.
2386 (*vsx_le_perm_load_v16qi): Likewise.
2387 (four unnamed splitters): Modify to select the stvx instruction
2388 when memory is aligned.
2390 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2392 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2393 which appears in the queue multiple times.
2395 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2396 Alan Hayward <alan.hayward@arm.com>
2397 David Sherwood <david.sherwood@arm.com>
2399 * tree-vectorizer.h (vec_lower_bound): New structure.
2400 (_loop_vec_info): Add check_nonzero and lower_bounds.
2401 (LOOP_VINFO_CHECK_NONZERO): New macro.
2402 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2403 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2404 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2405 fields. Make seg_len the distance travelled, not including the
2407 (dr_direction_indicator): Declare.
2408 (dr_zero_step_indicator): Likewise.
2409 (dr_known_forward_stride_p): Likewise.
2410 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2412 (runtime_alias_check_p): Allow runtime alias checks with
2414 (operator ==): Compare access_size and align.
2415 (prune_runtime_alias_test_list): Rework for new distinction between
2416 the access_size and seg_len.
2417 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2419 (get_segment_min_max): New function.
2420 (create_intersect_range_checks): Use it.
2421 (dr_step_indicator): New function.
2422 (dr_direction_indicator): Likewise.
2423 (dr_zero_step_indicator): Likewise.
2424 (dr_known_forward_stride_p): Likewise.
2425 * tree-loop-distribution.c (data_ref_segment_size): Return
2426 DR_STEP * (niters - 1).
2427 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2429 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2430 (vect_preserves_scalar_order_p): New function, split out from...
2431 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2432 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2433 (vect_vfa_access_size): New function.
2434 (vect_vfa_align): Likewise.
2435 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2436 (dump_lower_bound): New function.
2437 (vect_check_lower_bound): Likewise.
2438 (vect_small_gap_p): Likewise.
2439 (vectorizable_with_step_bound_p): Likewise.
2440 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2441 depencies if the vectorization factor is 1. Convert the checks
2442 for nonzero steps into checks on the bounds of DR_STEP. Try using
2443 a bunds check for variable steps if the minimum required step is
2444 relatively small. Update calls to the dr_with_seg_len
2445 constructor and to vect_compile_time_alias.
2446 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2448 (vect_loop_versioning): Call it.
2449 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2451 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2453 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2454 Alan Hayward <alan.hayward@arm.com>
2455 David Sherwood <david.sherwood@arm.com>
2457 * doc/sourcebuild.texi (vect_scatter_store): Document.
2458 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2460 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2462 * genopinit.c (main): Add supports_vec_scatter_store and
2463 supports_vec_scatter_store_cached to target_optabs.
2464 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2465 IFN_MASK_SCATTER_STORE.
2466 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2468 * internal-fn.h (internal_store_fn_p): Declare.
2469 (internal_fn_stored_value_index): Likewise.
2470 * internal-fn.c (scatter_store_direct): New macro.
2471 (expand_scatter_store_optab_fn): New function.
2472 (direct_scatter_store_optab_supported_p): New macro.
2473 (internal_store_fn_p): New function.
2474 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2475 IFN_MASK_SCATTER_STORE.
2476 (internal_fn_mask_index): Likewise.
2477 (internal_fn_stored_value_index): New function.
2478 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2480 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2481 * optabs-query.c (supports_vec_scatter_store_p): New function.
2482 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2483 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2484 true for scatter stores.
2485 (vect_gather_scatter_fn_p): Handle scatter stores too.
2486 (vect_check_gather_scatter): Consider using scatter stores if
2487 supports_vec_scatter_store_p.
2488 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2490 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2491 internal_fn_stored_value_index.
2492 (check_load_store_masking): Handle scatter stores too.
2493 (vect_get_store_rhs): Make public.
2494 (vectorizable_call): Use internal_store_fn_p.
2495 (vectorizable_store): Handle scatter store internal functions.
2496 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2497 when deciding whether the end of the group has been reached.
2498 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2499 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2500 (mask_scatter_store<mode>): New insns.
2502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2503 Alan Hayward <alan.hayward@arm.com>
2504 David Sherwood <david.sherwood@arm.com>
2506 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2507 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2508 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2510 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2511 Use vect_truncate_gather_scatter_offset if we can't treat the
2512 operation as a normal gather load or scatter store.
2513 (get_group_load_store_type): Take the gather_scatter_info
2514 as argument. Try using a gather load or scatter store for
2515 single-element groups.
2516 (get_load_store_type): Update calls to get_group_load_store_type
2517 and vect_use_strided_gather_scatters_p.
2519 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2520 Alan Hayward <alan.hayward@arm.com>
2521 David Sherwood <david.sherwood@arm.com>
2523 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2524 optional tree argument.
2525 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2527 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2528 but continue to use the current value as a fallback.
2529 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2530 to compare the updates.
2531 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2532 (get_load_store_type): Use it when handling a strided access.
2533 (vect_get_strided_load_store_ops): New function.
2534 (vect_get_data_ptr_increment): Likewise.
2535 (vectorizable_load): Handle strided gather loads. Always pass
2536 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2538 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2539 Alan Hayward <alan.hayward@arm.com>
2540 David Sherwood <david.sherwood@arm.com>
2542 * doc/md.texi (gather_load@var{m}): Document.
2543 (mask_gather_load@var{m}): Likewise.
2544 * genopinit.c (main): Add supports_vec_gather_load and
2545 supports_vec_gather_load_cached to target_optabs.
2546 * optabs-tree.c (init_tree_optimization_optabs): Use
2547 ggc_cleared_alloc to allocate target_optabs.
2548 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2549 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2551 * internal-fn.h (internal_load_fn_p): Declare.
2552 (internal_gather_scatter_fn_p): Likewise.
2553 (internal_fn_mask_index): Likewise.
2554 (internal_gather_scatter_fn_supported_p): Likewise.
2555 * internal-fn.c (gather_load_direct): New macro.
2556 (expand_gather_load_optab_fn): New function.
2557 (direct_gather_load_optab_supported_p): New macro.
2558 (direct_internal_fn_optab): New function.
2559 (internal_load_fn_p): Likewise.
2560 (internal_gather_scatter_fn_p): Likewise.
2561 (internal_fn_mask_index): Likewise.
2562 (internal_gather_scatter_fn_supported_p): Likewise.
2563 * optabs-query.c (supports_at_least_one_mode_p): New function.
2564 (supports_vec_gather_load_p): Likewise.
2565 * optabs-query.h (supports_vec_gather_load_p): Declare.
2566 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2567 and memory_type field.
2568 (NUM_PATTERNS): Bump to 15.
2569 * tree-vect-data-refs.c: Include internal-fn.h.
2570 (vect_gather_scatter_fn_p): New function.
2571 (vect_describe_gather_scatter_call): Likewise.
2572 (vect_check_gather_scatter): Try using internal functions for
2573 gather loads. Recognize existing calls to a gather load function.
2574 (vect_analyze_data_refs): Consider using gather loads if
2575 supports_vec_gather_load_p.
2576 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2577 (vect_get_gather_scatter_offset_type): Likewise.
2578 (vect_convert_mask_for_vectype): Likewise.
2579 (vect_add_conversion_to_patterm): Likewise.
2580 (vect_try_gather_scatter_pattern): Likewise.
2581 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2582 (vect_vect_recog_func_ptrs): Add it.
2583 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2584 internal_fn_mask_index and internal_gather_scatter_fn_p.
2585 (check_load_store_masking): Take the gather_scatter_info as an
2586 argument and handle gather loads.
2587 (vect_get_gather_scatter_ops): New function.
2588 (vectorizable_call): Check internal_load_fn_p.
2589 (vectorizable_load): Likewise. Handle gather load internal
2591 (vectorizable_store): Update call to check_load_store_masking.
2592 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2593 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2594 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2595 (aarch64_gather_scale_operand_d): New predicates.
2596 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2597 (mask_gather_load<mode>): New insns.
2599 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2600 Alan Hayward <alan.hayward@arm.com>
2601 David Sherwood <david.sherwood@arm.com>
2603 * optabs.def (fold_left_plus_optab): New optab.
2604 * doc/md.texi (fold_left_plus_@var{m}): Document.
2605 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2606 * internal-fn.c (fold_left_direct): Define.
2607 (expand_fold_left_optab_fn): Likewise.
2608 (direct_fold_left_optab_supported_p): Likewise.
2609 * fold-const-call.c (fold_const_fold_left): New function.
2610 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2611 * tree-parloops.c (valid_reduction_p): New function.
2612 (gather_scalar_reductions): Use it.
2613 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2614 (vect_finish_replace_stmt): Declare.
2615 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2616 (needs_fold_left_reduction_p): New function, split out from...
2617 (vect_is_simple_reduction): ...here. Accept reductions that
2618 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2619 (vect_force_simple_reduction): Also store the reduction type in
2620 the assignment's STMT_VINFO_REDUC_TYPE.
2621 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2622 (merge_with_identity): New function.
2623 (vect_expand_fold_left): Likewise.
2624 (vectorize_fold_left_reduction): Likewise.
2625 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2626 scalar phi in place for it. Check for target support and reject
2627 cases that would reassociate the operation. Defer the transform
2628 phase to vectorize_fold_left_reduction.
2629 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2630 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2631 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2633 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2635 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2636 call to ifc_temp_var.
2638 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2639 Alan Hayward <alan.hayward@arm.com>
2640 David Sherwood <david.sherwood@arm.com>
2642 * target.def (legitimize_address_displacement): Take the original
2643 offset as a poly_int.
2644 * targhooks.h (default_legitimize_address_displacement): Update
2646 * targhooks.c (default_legitimize_address_displacement): Likewise.
2647 * doc/tm.texi: Regenerate.
2648 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2649 as an argument, moving assert of ad->disp == ad->disp_term to...
2650 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2651 Try calling targetm.legitimize_address_displacement before expanding
2652 the address rather than afterwards, and adjust for the new interface.
2653 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2654 Match the new hook interface. Handle SVE addresses.
2655 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2658 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2660 * Makefile.in (OBJS): Add early-remat.o.
2661 * target.def (select_early_remat_modes): New hook.
2662 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2663 * doc/tm.texi: Regenerate.
2664 * targhooks.h (default_select_early_remat_modes): Declare.
2665 * targhooks.c (default_select_early_remat_modes): New function.
2666 * timevar.def (TV_EARLY_REMAT): New timevar.
2667 * passes.def (pass_early_remat): New pass.
2668 * tree-pass.h (make_pass_early_remat): Declare.
2669 * early-remat.c: New file.
2670 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2672 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2674 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2675 Alan Hayward <alan.hayward@arm.com>
2676 David Sherwood <david.sherwood@arm.com>
2678 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2679 vfm1 with a bound_epilog parameter.
2680 (vect_do_peeling): Update calls accordingly, and move the prologue
2681 call earlier in the function. Treat the base bound_epilog as 0 for
2682 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2683 this base when peeling for gaps.
2684 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2685 with fully-masked loops.
2686 (vect_estimate_min_profitable_iters): Handle the single peeled
2687 iteration in that case.
2689 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2690 Alan Hayward <alan.hayward@arm.com>
2691 David Sherwood <david.sherwood@arm.com>
2693 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2694 single-element interleaving even if the size is not a power of 2.
2695 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2696 accesses for single-element interleaving if the group size is
2699 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2700 Alan Hayward <alan.hayward@arm.com>
2701 David Sherwood <david.sherwood@arm.com>
2703 * doc/md.texi (fold_extract_last_@var{m}): Document.
2704 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2705 * optabs.def (fold_extract_last_optab): New optab.
2706 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2707 * internal-fn.c (fold_extract_direct): New macro.
2708 (expand_fold_extract_optab_fn): Likewise.
2709 (direct_fold_extract_optab_supported_p): Likewise.
2710 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2711 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2712 EXTRACT_LAST_REDUCTION.
2713 (get_initial_def_for_reduction): Do not create an initial vector
2714 for EXTRACT_LAST_REDUCTION reductions.
2715 (vectorizable_reduction): Leave the scalar phi in place for
2716 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2717 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2718 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2719 transform phase to vectorizable_condition.
2720 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2722 (vect_finish_stmt_generation): ...here.
2723 (vect_finish_replace_stmt): New function.
2724 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2725 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2727 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2729 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2730 Alan Hayward <alan.hayward@arm.com>
2731 David Sherwood <david.sherwood@arm.com>
2733 * doc/md.texi (extract_last_@var{m}): Document.
2734 * optabs.def (extract_last_optab): New optab.
2735 * internal-fn.def (EXTRACT_LAST): New internal function.
2736 * internal-fn.c (cond_unary_direct): New macro.
2737 (expand_cond_unary_optab_fn): Likewise.
2738 (direct_cond_unary_optab_supported_p): Likewise.
2739 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2740 loops using EXTRACT_LAST.
2741 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2742 (extract_last_<mode>): ...this optab.
2743 (vec_extract<mode><Vel>): Update accordingly.
2745 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2746 Alan Hayward <alan.hayward@arm.com>
2747 David Sherwood <david.sherwood@arm.com>
2749 * target.def (empty_mask_is_expensive): New hook.
2750 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2751 * doc/tm.texi: Regenerate.
2752 * targhooks.h (default_empty_mask_is_expensive): Declare.
2753 * targhooks.c (default_empty_mask_is_expensive): New function.
2754 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2755 if the target says that empty masks are expensive.
2756 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2758 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2760 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2761 Alan Hayward <alan.hayward@arm.com>
2762 David Sherwood <david.sherwood@arm.com>
2764 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2765 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2766 (vect_use_loop_mask_for_alignment_p): New function.
2767 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2768 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2769 niters_skip argument. Make sure that the first niters_skip elements
2770 of the first iteration are inactive.
2771 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2772 Update call to vect_set_loop_masks_directly.
2773 (get_misalign_in_elems): New function, split out from...
2774 (vect_gen_prolog_loop_niters): ...here.
2775 (vect_update_init_of_dr): Take a code argument that specifies whether
2776 the adjustment should be added or subtracted.
2777 (vect_update_init_of_drs): Likewise.
2778 (vect_prepare_for_masked_peels): New function.
2779 (vect_do_peeling): Skip prologue peeling if we're using a mask
2780 instead. Update call to vect_update_inits_of_drs.
2781 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2783 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2784 alignment. Do not include the number of peeled iterations in
2785 the minimum threshold in that case.
2786 (vectorizable_induction): Adjust the start value down by
2787 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2788 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2789 Take the number of skipped iterations into account when calculating
2791 * tree-vect-stmts.c (vect_gen_while_not): New function.
2793 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2794 Alan Hayward <alan.hayward@arm.com>
2795 David Sherwood <david.sherwood@arm.com>
2797 * doc/sourcebuild.texi (vect_fully_masked): Document.
2798 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2800 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2802 (vect_analyze_loop_2): ...here. Don't check the vectorization
2803 factor against the number of loop iterations if the loop is
2806 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2807 Alan Hayward <alan.hayward@arm.com>
2808 David Sherwood <david.sherwood@arm.com>
2810 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2811 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2812 (dump_groups): Update accordingly.
2813 (iv_use::mem_type): New member variable.
2814 (address_p): New function.
2815 (record_use): Add a mem_type argument and initialize the new
2817 (record_group_use): Add a mem_type argument. Use address_p.
2818 Remove obsolete null checks of base_object. Update call to record_use.
2819 (find_interesting_uses_op): Update call to record_group_use.
2820 (find_interesting_uses_cond): Likewise.
2821 (find_interesting_uses_address): Likewise.
2822 (get_mem_type_for_internal_fn): New function.
2823 (find_address_like_use): Likewise.
2824 (find_interesting_uses_stmt): Try find_address_like_use before
2825 calling find_interesting_uses_op.
2826 (addr_offset_valid_p): Use the iv mem_type field as the type
2827 of the addressed memory.
2828 (add_autoinc_candidates): Likewise.
2829 (get_address_cost): Likewise.
2830 (split_small_address_groups_p): Use address_p.
2831 (split_address_groups): Likewise.
2832 (add_iv_candidate_for_use): Likewise.
2833 (autoinc_possible_for_pair): Likewise.
2834 (rewrite_groups): Likewise.
2835 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2836 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2837 (get_alias_ptr_type_for_ptr_address): New function.
2838 (rewrite_use_address): Rewrite address uses in calls that were
2839 identified by find_address_like_use.
2841 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2842 Alan Hayward <alan.hayward@arm.com>
2843 David Sherwood <david.sherwood@arm.com>
2845 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2847 * gimple-expr.h (is_gimple_addressable: Likewise.
2848 * gimple-expr.c (is_gimple_address): Likewise.
2849 * internal-fn.c (expand_call_mem_ref): New function.
2850 (expand_mask_load_optab_fn): Use it.
2851 (expand_mask_store_optab_fn): Likewise.
2853 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2854 Alan Hayward <alan.hayward@arm.com>
2855 David Sherwood <david.sherwood@arm.com>
2857 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2858 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2859 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2860 (cond_umax@var{mode}): Document.
2861 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2862 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2863 (cond_umin_optab, cond_umax_optab): New optabs.
2864 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2865 (COND_IOR, COND_XOR): New internal functions.
2866 * internal-fn.h (get_conditional_internal_fn): Declare.
2867 * internal-fn.c (cond_binary_direct): New macro.
2868 (expand_cond_binary_optab_fn): Likewise.
2869 (direct_cond_binary_optab_supported_p): Likewise.
2870 (get_conditional_internal_fn): New function.
2871 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2872 Cope with reduction statements that are vectorized as calls rather
2874 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2875 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2876 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2877 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2878 (UNSPEC_COND_EOR): New unspecs.
2879 (optab): Add mappings for them.
2880 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2881 (sve_int_op, sve_fp_op): New int attributes.
2883 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2884 Alan Hayward <alan.hayward@arm.com>
2885 David Sherwood <david.sherwood@arm.com>
2887 * optabs.def (while_ult_optab): New optab.
2888 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2889 * internal-fn.def (WHILE_ULT): New internal function.
2890 * internal-fn.h (direct_internal_fn_supported_p): New override
2891 that takes two types as argument.
2892 * internal-fn.c (while_direct): New macro.
2893 (expand_while_optab_fn): New function.
2894 (convert_optab_supported_p): Likewise.
2895 (direct_while_optab_supported_p): New macro.
2896 * wide-int.h (wi::udiv_ceil): New function.
2897 * tree-vectorizer.h (rgroup_masks): New structure.
2898 (vec_loop_masks): New typedef.
2899 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2901 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2902 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2903 (vect_max_vf): New function.
2904 (slpeel_make_loop_iterate_ntimes): Delete.
2905 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2906 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2907 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2908 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2909 internal-fn.h, stor-layout.h and optabs-query.h.
2910 (vect_set_loop_mask): New function.
2911 (add_preheader_seq): Likewise.
2912 (add_header_seq): Likewise.
2913 (interleave_supported_p): Likewise.
2914 (vect_maybe_permute_loop_masks): Likewise.
2915 (vect_set_loop_masks_directly): Likewise.
2916 (vect_set_loop_condition_masked): Likewise.
2917 (vect_set_loop_condition_unmasked): New function, split out from
2918 slpeel_make_loop_iterate_ntimes.
2919 (slpeel_make_loop_iterate_ntimes): Rename to..
2920 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2921 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2922 (vect_do_peeling): Update call accordingly.
2923 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2925 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2926 mask_compare_type, can_fully_mask_p and fully_masked_p.
2927 (release_vec_loop_masks): New function.
2928 (_loop_vec_info): Use it to free the loop masks.
2929 (can_produce_all_loop_masks_p): New function.
2930 (vect_get_max_nscalars_per_iter): Likewise.
2931 (vect_verify_full_masking): Likewise.
2932 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2933 retries, and free the mask rgroups before retrying. Check loop-wide
2934 reasons for disallowing fully-masked loops. Make the final decision
2935 about whether use a fully-masked loop or not.
2936 (vect_estimate_min_profitable_iters): Do not assume that peeling
2937 for the number of iterations will be needed for fully-masked loops.
2938 (vectorizable_reduction): Disable fully-masked loops.
2939 (vectorizable_live_operation): Likewise.
2940 (vect_halve_mask_nunits): New function.
2941 (vect_double_mask_nunits): Likewise.
2942 (vect_record_loop_mask): Likewise.
2943 (vect_get_loop_mask): Likewise.
2944 (vect_transform_loop): Handle the case in which the final loop
2945 iteration might handle a partial vector. Call vect_set_loop_condition
2946 instead of slpeel_make_loop_iterate_ntimes.
2947 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2948 (check_load_store_masking): New function.
2949 (prepare_load_store_mask): Likewise.
2950 (vectorizable_store): Handle fully-masked loops.
2951 (vectorizable_load): Likewise.
2952 (supportable_widening_operation): Use vect_halve_mask_nunits for
2954 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2955 (vect_gen_while): New function.
2956 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2957 (aarch64_uqdec<mode>): New insn.
2959 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2960 Alan Hayward <alan.hayward@arm.com>
2961 David Sherwood <david.sherwood@arm.com>
2963 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2964 (reduc_xor_scal_optab): New optabs.
2965 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2966 (reduc_xor_scal_@var{m}): Document.
2967 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2968 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2970 * fold-const-call.c (fold_const_call): Handle them.
2971 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2972 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2973 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2974 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2975 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2976 (UNSPEC_XORV): New unspecs.
2977 (optab): Add entries for them.
2978 (BITWISEV): New int iterator.
2979 (bit_reduc_op): New int attributes.
2981 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2982 Alan Hayward <alan.hayward@arm.com>
2983 David Sherwood <david.sherwood@arm.com>
2985 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2986 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2987 * optabs.def (vec_shl_insert_optab): New optab.
2988 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2989 (duplicate_and_interleave): Likewise.
2990 * tree-vect-loop.c: Include internal-fn.h.
2991 (neutral_op_for_slp_reduction): New function, split out from
2992 get_initial_defs_for_reduction.
2993 (get_initial_def_for_reduction): Handle option 2 for variable-length
2994 vectors by loading the neutral value into a vector and then shifting
2995 the initial value into element 0.
2996 (get_initial_defs_for_reduction): Replace the code argument with
2997 the neutral value calculated by neutral_op_for_slp_reduction.
2998 Use gimple_build_vector for constant-length vectors.
2999 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3000 but the first group_size elements have a neutral value.
3001 Use duplicate_and_interleave otherwise.
3002 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3003 Update call to get_initial_defs_for_reduction. Handle SLP
3004 reductions for variable-length vectors by creating one vector
3005 result for each scalar result, with the elements associated
3006 with other scalar results stubbed out with the neutral value.
3007 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3008 Require IFN_VEC_SHL_INSERT for double reductions on
3009 variable-length vectors, or SLP reductions that have
3010 a neutral value. Require can_duplicate_and_interleave_p
3011 support for variable-length unchained SLP reductions if there
3012 is no neutral value, such as for MIN/MAX reductions. Also require
3013 the number of vector elements to be a multiple of the number of
3014 SLP statements when doing variable-length unchained SLP reductions.
3015 Update call to vect_create_epilog_for_reduction.
3016 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3017 and remove initial values.
3018 (duplicate_and_interleave): Make public.
3019 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3020 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3022 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3023 Alan Hayward <alan.hayward@arm.com>
3024 David Sherwood <david.sherwood@arm.com>
3026 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3027 (can_duplicate_and_interleave_p): New function.
3028 (vect_get_and_check_slp_defs): Take the vector of statements
3029 rather than just the current one. Remove excess parentheses.
3030 Restriction rejectinon of vect_constant_def and vect_external_def
3031 for variable-length vectors to boolean types, or types for which
3032 can_duplicate_and_interleave_p is false.
3033 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3034 (duplicate_and_interleave): New function.
3035 (vect_get_constant_vectors): Use gimple_build_vector for
3036 constant-length vectors and suitable variable-length constant
3037 vectors. Use duplicate_and_interleave for other variable-length
3038 vectors. Don't defer the update when inserting new statements.
3040 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3041 Alan Hayward <alan.hayward@arm.com>
3042 David Sherwood <david.sherwood@arm.com>
3044 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3045 min_profitable_iters doesn't go negative.
3047 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3048 Alan Hayward <alan.hayward@arm.com>
3049 David Sherwood <david.sherwood@arm.com>
3051 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3052 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3053 * optabs.def (vec_mask_load_lanes_optab): New optab.
3054 (vec_mask_store_lanes_optab): Likewise.
3055 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3056 (MASK_STORE_LANES): Likewise.
3057 * internal-fn.c (mask_load_lanes_direct): New macro.
3058 (mask_store_lanes_direct): Likewise.
3059 (expand_mask_load_optab_fn): Handle masked operations.
3060 (expand_mask_load_lanes_optab_fn): New macro.
3061 (expand_mask_store_optab_fn): Handle masked operations.
3062 (expand_mask_store_lanes_optab_fn): New macro.
3063 (direct_mask_load_lanes_optab_supported_p): Likewise.
3064 (direct_mask_store_lanes_optab_supported_p): Likewise.
3065 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3067 (vect_load_lanes_supported): Likewise.
3068 * tree-vect-data-refs.c (strip_conversion): New function.
3069 (can_group_stmts_p): Likewise.
3070 (vect_analyze_data_ref_accesses): Use it instead of checking
3071 for a pair of assignments.
3072 (vect_store_lanes_supported): Take a masked_p parameter.
3073 (vect_load_lanes_supported): Likewise.
3074 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3075 vect_store_lanes_supported and vect_load_lanes_supported.
3076 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3077 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3078 parameter. Don't allow gaps for masked accesses.
3079 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3080 and vect_load_lanes_supported.
3081 (get_load_store_type): Take a masked_p parameter and update
3082 call to get_group_load_store_type.
3083 (vectorizable_store): Update call to get_load_store_type.
3084 Handle IFN_MASK_STORE_LANES.
3085 (vectorizable_load): Update call to get_load_store_type.
3086 Handle IFN_MASK_LOAD_LANES.
3088 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3089 Alan Hayward <alan.hayward@arm.com>
3090 David Sherwood <david.sherwood@arm.com>
3092 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3094 * config/aarch64/aarch64-protos.h
3095 (aarch64_sve_struct_memory_operand_p): Declare.
3096 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3097 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3098 (VPRED, vpred): Handle SVE structure modes.
3099 * config/aarch64/constraints.md (Utx): New constraint.
3100 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3101 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3102 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3103 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3104 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3105 structure modes. Split into pieces after RA.
3106 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3107 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3109 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3110 SVE structure modes.
3111 (aarch64_classify_address): Likewise.
3112 (sizetochar): Move earlier in file.
3113 (aarch64_print_operand): Handle SVE register lists.
3114 (aarch64_array_mode): New function.
3115 (aarch64_sve_struct_memory_operand_p): Likewise.
3116 (TARGET_ARRAY_MODE): Redefine.
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 * target.def (array_mode): New target hook.
3123 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3124 * doc/tm.texi: Regenerate.
3125 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3126 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3127 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3129 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3132 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3133 Alan Hayward <alan.hayward@arm.com>
3134 David Sherwood <david.sherwood@arm.com>
3136 * fold-const.c (fold_binary_loc): Check the argument types
3137 rather than the result type when testing for a vector operation.
3139 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3141 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3142 * doc/tm.texi: Regenerate.
3144 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3145 Alan Hayward <alan.hayward@arm.com>
3146 David Sherwood <david.sherwood@arm.com>
3148 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3149 (sve): Document new AArch64 extension.
3150 * doc/md.texi (w): Extend the description of the AArch64
3151 constraint to include SVE vectors.
3152 (Upl, Upa): Document new AArch64 predicate constraints.
3153 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3155 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3156 (msve-vector-bits=): New option.
3157 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3158 SVE when these are disabled.
3159 (sve): New extension.
3160 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3161 modes. Adjust their number of units based on aarch64_sve_vg.
3162 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3163 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3164 aarch64_addr_query_type.
3165 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3166 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3167 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3168 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3169 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3170 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3171 (aarch64_simd_imm_zero_p): Delete.
3172 (aarch64_check_zero_based_sve_index_immediate): Declare.
3173 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3174 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3175 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3176 (aarch64_sve_float_mul_immediate_p): Likewise.
3177 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3179 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3180 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3181 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3182 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3183 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3184 (aarch64_regmode_natural_size): Likewise.
3185 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3186 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3188 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3189 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3190 for VG and the SVE predicate registers.
3191 (V_ALIASES): Add a "z"-prefixed alias.
3192 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3193 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3194 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3195 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3196 (REG_CLASS_NAMES): Add entries for them.
3197 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3198 and the predicate registers.
3199 (aarch64_sve_vg): Declare.
3200 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3201 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3202 (REGMODE_NATURAL_SIZE): Define.
3203 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3205 * config/aarch64/aarch64.c: Include cfgrtl.h.
3206 (simd_immediate_info): Add a constructor for series vectors,
3207 and an associated step field.
3208 (aarch64_sve_vg): New variable.
3209 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3210 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3211 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3212 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3213 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3214 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3215 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3216 (aarch64_get_mask_mode): New functions.
3217 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3218 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3219 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3220 predicate modes and predicate registers. Explicitly restrict
3221 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3222 to store a vector mode if it is recognized by
3223 aarch64_classify_vector_mode.
3224 (aarch64_regmode_natural_size): New function.
3225 (aarch64_hard_regno_caller_save_mode): Return the original mode
3227 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3228 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3229 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3230 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3232 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3233 does not overlap dest if the function is frame-related. Handle
3235 (aarch64_split_add_offset): New function.
3236 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3237 them aarch64_add_offset.
3238 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3239 and update call to aarch64_sub_sp.
3240 (aarch64_add_cfa_expression): New function.
3241 (aarch64_expand_prologue): Pass extra temporary registers to the
3242 functions above. Handle the case in which we need to emit new
3243 DW_CFA_expressions for registers that were originally saved
3244 relative to the stack pointer, but now have to be expressed
3245 relative to the frame pointer.
3246 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3248 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3249 IP0 and IP1 values for SVE frames.
3250 (aarch64_expand_vec_series): New function.
3251 (aarch64_expand_sve_widened_duplicate): Likewise.
3252 (aarch64_expand_sve_const_vector): Likewise.
3253 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3254 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3255 into the register, rather than emitting a SET directly.
3256 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3257 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3258 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3259 (offset_9bit_signed_scaled_p): New functions.
3260 (aarch64_replicate_bitmask_imm): New function.
3261 (aarch64_bitmask_imm): Use it.
3262 (aarch64_cannot_force_const_mem): Reject expressions involving
3263 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3264 (aarch64_classify_index): Handle SVE indices, by requiring
3265 a plain register index with a scale that matches the element size.
3266 (aarch64_classify_address): Handle SVE addresses. Assert that
3267 the mode of the address is VOIDmode or an integer mode.
3268 Update call to aarch64_classify_symbol.
3269 (aarch64_classify_symbolic_expression): Update call to
3270 aarch64_classify_symbol.
3271 (aarch64_const_vec_all_in_range_p): New function.
3272 (aarch64_print_vector_float_operand): Likewise.
3273 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3274 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3275 and the FP immediates 1.0 and 0.5.
3276 (aarch64_print_address_internal): Handle SVE addresses.
3277 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3278 (aarch64_regno_regclass): Handle predicate registers.
3279 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3281 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3282 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3283 (aarch64_convert_sve_vector_bits): New function.
3284 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3285 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3287 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3288 Handle SVE vector and predicate modes. Accept VL-based constants
3289 that need only one temporary register, and VL offsets that require
3290 no temporary registers.
3291 (aarch64_conditional_register_usage): Mark the predicate registers
3292 as fixed if SVE isn't available.
3293 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3294 Return true for SVE vector and predicate modes.
3295 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3296 rather than an unsigned int. Handle SVE modes.
3297 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3299 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3301 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3302 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3303 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3304 (aarch64_sve_float_mul_immediate_p): New functions.
3305 (aarch64_sve_valid_immediate): New function.
3306 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3307 Explicitly reject structure modes. Check for INDEX constants.
3308 Handle PTRUE and PFALSE constants.
3309 (aarch64_check_zero_based_sve_index_immediate): New function.
3310 (aarch64_simd_imm_zero_p): Delete.
3311 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3312 vector modes. Accept constants in the range of CNT[BHWD].
3313 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3314 ask for an Advanced SIMD mode.
3315 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3316 (aarch64_simd_vector_alignment): Handle SVE predicates.
3317 (aarch64_vectorize_preferred_vector_alignment): New function.
3318 (aarch64_simd_vector_alignment_reachable): Use it instead of
3320 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3321 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3323 (MAX_VECT_LEN): Delete.
3324 (expand_vec_perm_d): Add a vec_flags field.
3325 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3326 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3327 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3329 (aarch64_evpc_rev): Rename to...
3330 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3331 (aarch64_evpc_rev_global): New function.
3332 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3333 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3335 (aarch64_evpc_sve_tbl): New function.
3336 (aarch64_expand_vec_perm_const_1): Update after rename of
3337 aarch64_evpc_rev. Handle SVE permutes too, trying
3338 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3339 than aarch64_evpc_tbl.
3340 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3341 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3342 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3343 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3344 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3345 (aarch64_expand_sve_vcond): New functions.
3346 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3347 of aarch64_vector_mode_p.
3348 (aarch64_dwarf_poly_indeterminate_value): New function.
3349 (aarch64_compute_pressure_classes): Likewise.
3350 (aarch64_can_change_mode_class): Likewise.
3351 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3352 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3353 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3354 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3355 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3356 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3357 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3358 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3360 (Dn, Dl, Dr): Accept const as well as const_vector.
3361 (Dz): Likewise. Compare against CONST0_RTX.
3362 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3363 of "vector" where appropriate.
3364 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3365 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3366 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3367 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3368 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3369 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3370 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3371 (v_int_equiv): Extend to SVE modes.
3372 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3374 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3375 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3376 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3377 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3378 (SVE_COND_FP_CMP): New int iterators.
3379 (perm_hilo): Handle the new unpack unspecs.
3380 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3382 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3383 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3384 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3385 (aarch64_equality_operator, aarch64_constant_vector_operand)
3386 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3387 (aarch64_sve_nonimmediate_operand): Likewise.
3388 (aarch64_sve_general_operand): Likewise.
3389 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3390 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3391 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3392 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3393 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3394 (aarch64_sve_float_arith_immediate): Likewise.
3395 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3396 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3397 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3398 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3399 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3400 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3401 (aarch64_sve_float_arith_operand): Likewise.
3402 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3403 (aarch64_sve_float_mul_operand): Likewise.
3404 (aarch64_sve_vec_perm_operand): Likewise.
3405 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3406 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3407 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3408 as well as const_vector.
3409 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3410 in file. Use CONST0_RTX and CONSTM1_RTX.
3411 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3412 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3413 Use aarch64_simd_imm_zero.
3414 * config/aarch64/aarch64-sve.md: New file.
3415 * config/aarch64/aarch64.md: Include it.
3416 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3417 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3418 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3419 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3420 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3421 (sve): New attribute.
3422 (enabled): Disable instructions with the sve attribute unless
3424 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3425 aarch64_expand_mov_immediate.
3426 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3427 CNT[BHSD] immediates.
3428 (movti): Split CONST_POLY_INT moves into two halves.
3429 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3430 Split additions that need a temporary here if the destination
3431 is the stack pointer.
3432 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3433 (*add<mode>3_poly_1): New instruction.
3434 (set_clobber_cc): New expander.
3436 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3438 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3439 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3440 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3441 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3442 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3443 Change innermode from fixed_mode_size to machine_mode.
3444 (simplify_subreg): Update call accordingly. Handle a constant-sized
3445 subreg of a variable-length CONST_VECTOR.
3447 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3448 Alan Hayward <alan.hayward@arm.com>
3449 David Sherwood <david.sherwood@arm.com>
3451 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3452 (add_offset_to_base): New function, split out from...
3453 (create_mem_ref): ...here. When handling a scale other than 1,
3454 check first whether the address is valid without the offset.
3455 Add it into the base if so, leaving the index and scale as-is.
3457 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3460 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3461 fold_for_warn before checking if arg2 is INTEGER_CST.
3463 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3465 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3466 (store_multiple_operation): Delete.
3467 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3468 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3469 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3470 guarded by TARGET_STRING.
3471 (rs6000_output_load_multiple): Delete.
3472 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3473 OPTION_MASK_STRING / TARGET_STRING handling.
3474 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3475 (const rs6000_opt_masks) <"string">: Change mask to 0.
3476 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3477 (MASK_STRING): Delete.
3478 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3480 (load_multiple): Delete.
3487 (store_multiple): Delete.
3494 (movmemsi_8reg): Delete.
3495 (corresponding unnamed define_insn): Delete.
3496 (movmemsi_6reg): Delete.
3497 (corresponding unnamed define_insn): Delete.
3498 (movmemsi_4reg): Delete.
3499 (corresponding unnamed define_insn): Delete.
3500 (movmemsi_2reg): Delete.
3501 (corresponding unnamed define_insn): Delete.
3502 (movmemsi_1reg): Delete.
3503 (corresponding unnamed define_insn): Delete.
3504 * config/rs6000/rs6000.opt (mno-string): New.
3505 (mstring): Replace by deprecation warning stub.
3506 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3508 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3510 * regrename.c (regrename_do_replace): If replacing the same
3511 reg multiple times, try to reuse last created gen_raw_REG.
3514 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3515 main to workaround a bug in GDB.
3517 2018-01-12 Tom de Vries <tom@codesourcery.com>
3520 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3522 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3524 PR rtl-optimization/80481
3525 * ira-color.c (get_cap_member): New function.
3526 (allocnos_conflict_by_live_ranges_p): Use it.
3527 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3528 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3530 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3533 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3534 (*saddl_se_1): Ditto.
3536 (*ssubl_se_1): Ditto.
3538 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3540 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3541 rather than wi::to_widest for DR_INITs.
3542 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3543 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3544 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3546 (vect_analyze_group_access_1): Note that here.
3548 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3550 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3551 polynomial type sizes.
3553 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3555 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3556 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3557 (gimple_add_tmp_var): Likewise.
3559 2018-01-12 Martin Liska <mliska@suse.cz>
3561 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3562 (gimple_alloc_sizes): Likewise.
3563 (dump_gimple_statistics): Use PRIu64 in printf format.
3564 * gimple.h: Change uint64_t to int.
3566 2018-01-12 Martin Liska <mliska@suse.cz>
3568 * tree-core.h: Use uint64_t instead of int.
3569 * tree.c (tree_node_counts): Likewise.
3570 (tree_node_sizes): Likewise.
3571 (dump_tree_statistics): Use PRIu64 in printf format.
3573 2018-01-12 Martin Liska <mliska@suse.cz>
3575 * Makefile.in: As qsort_chk is implemented in vec.c, add
3576 vec.o to linkage of gencfn-macros.
3577 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3578 passing the info to record_node_allocation_statistics.
3579 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3581 * ggc-common.c (struct ggc_usage): Add operator== and use
3582 it in operator< and compare function.
3583 * mem-stats.h (struct mem_usage): Likewise.
3584 * vec.c (struct vec_usage): Remove operator< and compare
3585 function. Can be simply inherited.
3587 2018-01-12 Martin Jambor <mjambor@suse.cz>
3590 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3591 * tree-ssa-math-opts.c: Include domwalk.h.
3592 (convert_mult_to_fma_1): New function.
3593 (fma_transformation_info): New type.
3594 (fma_deferring_state): Likewise.
3595 (cancel_fma_deferring): New function.
3596 (result_of_phi): Likewise.
3597 (last_fma_candidate_feeds_initial_phi): Likewise.
3598 (convert_mult_to_fma): Added deferring logic, split actual
3599 transformation to convert_mult_to_fma_1.
3600 (math_opts_dom_walker): New type.
3601 (math_opts_dom_walker::after_dom_children): New method, body moved
3602 here from pass_optimize_widening_mul::execute, added deferring logic
3604 (pass_optimize_widening_mul::execute): Moved most of code to
3605 math_opts_dom_walker::after_dom_children.
3606 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3607 * config/i386/i386.c (ix86_option_override_internal): Added
3608 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3610 2018-01-12 Richard Biener <rguenther@suse.de>
3613 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3614 inline instance vars.
3616 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3619 * config/rx/rx.c (rx_is_restricted_memory_address):
3622 2018-01-12 Richard Biener <rguenther@suse.de>
3624 PR tree-optimization/80846
3625 * target.def (split_reduction): New target hook.
3626 * targhooks.c (default_split_reduction): New function.
3627 * targhooks.h (default_split_reduction): Declare.
3628 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3629 target requests first reduce vectors by combining low and high
3631 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3632 (get_vectype_for_scalar_type_and_size): Export.
3633 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3634 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3635 * doc/tm.texi: Regenerate.
3636 * config/i386/i386.c (ix86_split_reduction): Implement
3637 TARGET_VECTORIZE_SPLIT_REDUCTION.
3639 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3642 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3643 in PIC mode except for TARGET_VXWORKS_RTP.
3644 * config/sparc/sparc.c: Include cfgrtl.h.
3645 (TARGET_INIT_PIC_REG): Define.
3646 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3647 (sparc_pic_register_p): New predicate.
3648 (sparc_legitimate_address_p): Use it.
3649 (sparc_legitimize_pic_address): Likewise.
3650 (sparc_delegitimize_address): Likewise.
3651 (sparc_mode_dependent_address_p): Likewise.
3652 (gen_load_pcrel_sym): Remove 4th parameter.
3653 (load_got_register): Adjust call to above. Remove obsolete stuff.
3654 (sparc_expand_prologue): Do not call load_got_register here.
3655 (sparc_flat_expand_prologue): Likewise.
3656 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3657 (sparc_use_pseudo_pic_reg): New function.
3658 (sparc_init_pic_reg): Likewise.
3659 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3660 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3662 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3664 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3665 Add item for branch_cost.
3667 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3669 PR rtl-optimization/83565
3670 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3671 not extend the result to a larger mode for rotate operations.
3672 (num_sign_bit_copies1): Likewise.
3674 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3677 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3679 Use values-Xc.o for -pedantic.
3680 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3682 2018-01-12 Martin Liska <mliska@suse.cz>
3685 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3687 (possible_polymorphic_call_targets): Use it.
3688 (ipa_devirt): Likewise.
3690 2018-01-12 Martin Liska <mliska@suse.cz>
3692 * profile-count.h (enum profile_quality): Use 0 as invalid
3693 enum value of profile_quality.
3695 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3697 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3698 -mext-string options.
3700 2018-01-12 Richard Biener <rguenther@suse.de>
3702 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3703 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3704 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3706 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3708 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3710 * configure.ac (--with-long-double-format): Add support for the
3711 configuration option to change the default long double format on
3713 * config.gcc (powerpc*-linux*-*): Likewise.
3714 * configure: Regenerate.
3715 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3716 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3717 used without modification.
3719 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3721 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3722 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3723 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3724 MISC_BUILTIN_SPEC_BARRIER.
3725 (rs6000_init_builtins): Likewise.
3726 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3728 (speculation_barrier): New define_insn.
3729 * doc/extend.texi: Document __builtin_speculation_barrier.
3731 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3734 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3735 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3736 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3738 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3739 integral modes instead of "ss" and "sd".
3740 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3741 vectors with 32-bit and 64-bit elements.
3742 (vecdupssescalarmodesuffix): New mode attribute.
3743 (vec_dup<mode>): Use it.
3745 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3748 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3749 frame if argument is passed on stack.
3751 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3754 * ree.c (combine_reaching_defs): Optimize also
3755 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3756 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3758 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3761 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3763 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3766 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3767 after they are computed.
3769 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3771 PR tree-optimization/83695
3772 * gimple-loop-linterchange.cc
3773 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3774 reset cached scev information after interchange.
3775 (pass_linterchange::execute): Remove call to scev_reset_htab.
3777 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3779 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3780 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3781 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3782 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3783 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3784 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3785 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3786 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3787 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3788 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3789 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3790 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3791 (V_lane_reg): Likewise.
3792 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3794 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3795 (vfmal_lane_low<mode>_intrinsic,
3796 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3797 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3798 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3799 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3800 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3801 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3803 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3805 * config/arm/arm-cpus.in (fp16fml): New feature.
3806 (ALL_SIMD): Add fp16fml.
3807 (armv8.2-a): Add fp16fml as an option.
3808 (armv8.3-a): Likewise.
3809 (armv8.4-a): Add fp16fml as part of fp16.
3810 * config/arm/arm.h (TARGET_FP16FML): Define.
3811 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3813 * config/arm/arm-modes.def (V2HF): Define.
3814 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3815 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3816 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3817 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3818 vfmsl_low, vfmsl_high): New set of builtins.
3819 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3820 (vfml_op): New code attribute.
3821 (VFMLHALVES): New int iterator.
3822 (VFML, VFMLSEL): New mode attributes.
3823 (V_reg): Define mapping for V2HF.
3824 (V_hi, V_lo): New mode attributes.
3825 (VF_constraint): Likewise.
3826 (vfml_half, vfml_half_selector): New int attributes.
3827 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3829 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3830 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3832 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3833 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3834 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3835 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3837 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3838 Document new effective target and option set.
3840 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3842 * config/arm/arm-cpus.in (armv8_4): New feature.
3843 (ARMv8_4a): New fgroup.
3844 (armv8.4-a): New arch.
3845 * config/arm/arm-tables.opt: Regenerate.
3846 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3847 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3848 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3849 Add matching rules for -march=armv8.4-a and extensions.
3850 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3852 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3855 * config/rx/rx.md (BW): New mode attribute.
3856 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3858 2018-01-11 Richard Biener <rguenther@suse.de>
3860 PR tree-optimization/83435
3861 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3862 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3863 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3865 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3866 Alan Hayward <alan.hayward@arm.com>
3867 David Sherwood <david.sherwood@arm.com>
3869 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3871 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3872 (aarch64_print_address_internal): Use it to check for a zero offset.
3874 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3875 Alan Hayward <alan.hayward@arm.com>
3876 David Sherwood <david.sherwood@arm.com>
3878 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3879 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3880 Return a poly_int64 rather than a HOST_WIDE_INT.
3881 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3882 rather than a HOST_WIDE_INT.
3883 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3884 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3885 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3886 final_offset from HOST_WIDE_INT to poly_int64.
3887 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3888 to_constant when getting the number of units in an Advanced SIMD
3890 (aarch64_builtin_vectorized_function): Check for a constant number
3892 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3894 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3895 attribute instead of GET_MODE_NUNITS.
3896 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3897 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3898 GET_MODE_SIZE for fixed-size registers.
3899 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3900 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3901 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3902 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3903 (aarch64_print_operand, aarch64_print_address_internal)
3904 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3905 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3906 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3907 Handle polynomial GET_MODE_SIZE.
3908 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3909 wider than SImode without modification.
3910 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3911 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3912 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3913 passing and returning SVE modes.
3914 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3915 rather than GEN_INT.
3916 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3917 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3918 (aarch64_allocate_and_probe_stack_space): Likewise.
3919 (aarch64_layout_frame): Cope with polynomial offsets.
3920 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3921 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3923 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3924 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3925 poly_int64 rather than a HOST_WIDE_INT.
3926 (aarch64_get_separate_components, aarch64_process_components)
3927 (aarch64_expand_prologue, aarch64_expand_epilogue)
3928 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3929 (aarch64_anchor_offset): New function, split out from...
3930 (aarch64_legitimize_address): ...here.
3931 (aarch64_builtin_vectorization_cost): Handle polynomial
3932 TYPE_VECTOR_SUBPARTS.
3933 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3935 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3936 number of elements from the PARALLEL rather than the mode.
3937 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3938 rather than GET_MODE_BITSIZE.
3939 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3940 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3941 (aarch64_expand_vec_perm_const_1): Handle polynomial
3942 d->perm.length () and d->perm elements.
3943 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3944 Apply to_constant to d->perm elements.
3945 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3946 polynomial CONST_VECTOR_NUNITS.
3947 (aarch64_move_pointer): Take amount as a poly_int64 rather
3949 (aarch64_progress_pointer): Avoid temporary variable.
3950 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3951 the mode attribute instead of GET_MODE.
3953 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3954 Alan Hayward <alan.hayward@arm.com>
3955 David Sherwood <david.sherwood@arm.com>
3957 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3958 x exists before using it.
3959 (aarch64_add_constant_internal): Rename to...
3960 (aarch64_add_offset_1): ...this. Replace regnum with separate
3961 src and dest rtxes. Handle the case in which they're different,
3962 including when the offset is zero. Replace scratchreg with an rtx.
3963 Use 2 additions if there is no spare register into which we can
3964 move a 16-bit constant.
3965 (aarch64_add_constant): Delete.
3966 (aarch64_add_offset): Replace reg with separate src and dest
3967 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3968 Use aarch64_add_offset_1.
3969 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3970 an rtx rather than an int. Take the delta as a poly_int64
3971 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3972 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3973 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3974 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3975 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3977 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3978 aarch64_add_constant.
3980 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3982 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3983 Use scalar_float_mode.
3985 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3987 * config/aarch64/aarch64-simd.md
3988 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3989 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3990 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3991 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3992 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3993 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3994 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3995 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3996 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3997 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3999 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4002 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4003 targ_options->x_arm_arch_string is non NULL.
4005 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4007 * config/aarch64/aarch64.h
4008 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4010 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4013 * expmed.c (emit_store_flag_force): Swap if const op0
4014 and change VOIDmode to mode of op0.
4016 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4018 PR rtl-optimization/83761
4019 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4020 than bytes to mode_for_size.
4022 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4025 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4026 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4029 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4032 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4033 when in layout mode.
4034 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4035 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4038 2018-01-10 Michael Collison <michael.collison@arm.com>
4040 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4041 * config/aarch64/aarch64-option-extension.def: Add
4042 AARCH64_OPT_EXTENSION of 'fp16fml'.
4043 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4044 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4045 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4046 * config/aarch64/constraints.md (Ui7): New constraint.
4047 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4048 (VFMLA_SEL_W): Ditto.
4051 (VFMLA16_LOW): New int iterator.
4052 (VFMLA16_HIGH): Ditto.
4053 (UNSPEC_FMLAL): New unspec.
4054 (UNSPEC_FMLSL): Ditto.
4055 (UNSPEC_FMLAL2): Ditto.
4056 (UNSPEC_FMLSL2): Ditto.
4057 (f16mac): New code attribute.
4058 * config/aarch64/aarch64-simd-builtins.def
4059 (aarch64_fmlal_lowv2sf): Ditto.
4060 (aarch64_fmlsl_lowv2sf): Ditto.
4061 (aarch64_fmlalq_lowv4sf): Ditto.
4062 (aarch64_fmlslq_lowv4sf): Ditto.
4063 (aarch64_fmlal_highv2sf): Ditto.
4064 (aarch64_fmlsl_highv2sf): Ditto.
4065 (aarch64_fmlalq_highv4sf): Ditto.
4066 (aarch64_fmlslq_highv4sf): Ditto.
4067 (aarch64_fmlal_lane_lowv2sf): Ditto.
4068 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4069 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4070 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4071 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4072 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4073 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4074 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4075 (aarch64_fmlal_lane_highv2sf): Ditto.
4076 (aarch64_fmlsl_lane_highv2sf): Ditto.
4077 (aarch64_fmlal_laneq_highv2sf): Ditto.
4078 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4079 (aarch64_fmlalq_lane_highv4sf): Ditto.
4080 (aarch64_fmlsl_lane_highv4sf): Ditto.
4081 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4082 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4083 * config/aarch64/aarch64-simd.md:
4084 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4085 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4086 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4087 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4088 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4089 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4090 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4091 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4092 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4093 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4094 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4095 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4096 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4097 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4098 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4099 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4100 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4101 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4102 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4103 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4104 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4105 (vfmlsl_low_u32): Ditto.
4106 (vfmlalq_low_u32): Ditto.
4107 (vfmlslq_low_u32): Ditto.
4108 (vfmlal_high_u32): Ditto.
4109 (vfmlsl_high_u32): Ditto.
4110 (vfmlalq_high_u32): Ditto.
4111 (vfmlslq_high_u32): Ditto.
4112 (vfmlal_lane_low_u32): Ditto.
4113 (vfmlsl_lane_low_u32): Ditto.
4114 (vfmlal_laneq_low_u32): Ditto.
4115 (vfmlsl_laneq_low_u32): Ditto.
4116 (vfmlalq_lane_low_u32): Ditto.
4117 (vfmlslq_lane_low_u32): Ditto.
4118 (vfmlalq_laneq_low_u32): Ditto.
4119 (vfmlslq_laneq_low_u32): Ditto.
4120 (vfmlal_lane_high_u32): Ditto.
4121 (vfmlsl_lane_high_u32): Ditto.
4122 (vfmlal_laneq_high_u32): Ditto.
4123 (vfmlsl_laneq_high_u32): Ditto.
4124 (vfmlalq_lane_high_u32): Ditto.
4125 (vfmlslq_lane_high_u32): Ditto.
4126 (vfmlalq_laneq_high_u32): Ditto.
4127 (vfmlslq_laneq_high_u32): Ditto.
4128 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4129 (AARCH64_FL_FOR_ARCH8_4): New.
4130 (AARCH64_ISA_F16FML): New ISA flag.
4131 (TARGET_F16FML): New feature flag for fp16fml.
4132 (doc/invoke.texi): Document new fp16fml option.
4134 2018-01-10 Michael Collison <michael.collison@arm.com>
4136 * config/aarch64/aarch64-builtins.c:
4137 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4138 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4139 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4140 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4141 (AARCH64_ISA_SHA3): New ISA flag.
4142 (TARGET_SHA3): New feature flag for sha3.
4143 * config/aarch64/iterators.md (sha512_op): New int attribute.
4144 (CRYPTO_SHA512): New int iterator.
4145 (UNSPEC_SHA512H): New unspec.
4146 (UNSPEC_SHA512H2): Ditto.
4147 (UNSPEC_SHA512SU0): Ditto.
4148 (UNSPEC_SHA512SU1): Ditto.
4149 * config/aarch64/aarch64-simd-builtins.def
4150 (aarch64_crypto_sha512hqv2di): New builtin.
4151 (aarch64_crypto_sha512h2qv2di): Ditto.
4152 (aarch64_crypto_sha512su0qv2di): Ditto.
4153 (aarch64_crypto_sha512su1qv2di): Ditto.
4154 (aarch64_eor3qv8hi): Ditto.
4155 (aarch64_rax1qv2di): Ditto.
4156 (aarch64_xarqv2di): Ditto.
4157 (aarch64_bcaxqv8hi): Ditto.
4158 * config/aarch64/aarch64-simd.md:
4159 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4160 (aarch64_crypto_sha512su0qv2di): Ditto.
4161 (aarch64_crypto_sha512su1qv2di): Ditto.
4162 (aarch64_eor3qv8hi): Ditto.
4163 (aarch64_rax1qv2di): Ditto.
4164 (aarch64_xarqv2di): Ditto.
4165 (aarch64_bcaxqv8hi): Ditto.
4166 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4167 (vsha512h2q_u64): Ditto.
4168 (vsha512su0q_u64): Ditto.
4169 (vsha512su1q_u64): Ditto.
4170 (veor3q_u16): Ditto.
4171 (vrax1q_u64): Ditto.
4173 (vbcaxq_u16): Ditto.
4174 * config/arm/types.md (crypto_sha512): New type attribute.
4175 (crypto_sha3): Ditto.
4176 (doc/invoke.texi): Document new sha3 option.
4178 2018-01-10 Michael Collison <michael.collison@arm.com>
4180 * config/aarch64/aarch64-builtins.c:
4181 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4182 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4183 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4184 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4185 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4186 (AARCH64_ISA_SM4): New ISA flag.
4187 (TARGET_SM4): New feature flag for sm4.
4188 * config/aarch64/aarch64-simd-builtins.def
4189 (aarch64_sm3ss1qv4si): Ditto.
4190 (aarch64_sm3tt1aq4si): Ditto.
4191 (aarch64_sm3tt1bq4si): Ditto.
4192 (aarch64_sm3tt2aq4si): Ditto.
4193 (aarch64_sm3tt2bq4si): Ditto.
4194 (aarch64_sm3partw1qv4si): Ditto.
4195 (aarch64_sm3partw2qv4si): Ditto.
4196 (aarch64_sm4eqv4si): Ditto.
4197 (aarch64_sm4ekeyqv4si): Ditto.
4198 * config/aarch64/aarch64-simd.md:
4199 (aarch64_sm3ss1qv4si): Ditto.
4200 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4201 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4202 (aarch64_sm4eqv4si): Ditto.
4203 (aarch64_sm4ekeyqv4si): Ditto.
4204 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4205 (sm3part_op): Ditto.
4206 (CRYPTO_SM3TT): Ditto.
4207 (CRYPTO_SM3PART): Ditto.
4208 (UNSPEC_SM3SS1): New unspec.
4209 (UNSPEC_SM3TT1A): Ditto.
4210 (UNSPEC_SM3TT1B): Ditto.
4211 (UNSPEC_SM3TT2A): Ditto.
4212 (UNSPEC_SM3TT2B): Ditto.
4213 (UNSPEC_SM3PARTW1): Ditto.
4214 (UNSPEC_SM3PARTW2): Ditto.
4215 (UNSPEC_SM4E): Ditto.
4216 (UNSPEC_SM4EKEY): Ditto.
4217 * config/aarch64/constraints.md (Ui2): New constraint.
4218 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4219 * config/arm/types.md (crypto_sm3): New type attribute.
4220 (crypto_sm4): Ditto.
4221 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4222 (vsm3tt1aq_u32): Ditto.
4223 (vsm3tt1bq_u32): Ditto.
4224 (vsm3tt2aq_u32): Ditto.
4225 (vsm3tt2bq_u32): Ditto.
4226 (vsm3partw1q_u32): Ditto.
4227 (vsm3partw2q_u32): Ditto.
4228 (vsm4eq_u32): Ditto.
4229 (vsm4ekeyq_u32): Ditto.
4230 (doc/invoke.texi): Document new sm4 option.
4232 2018-01-10 Michael Collison <michael.collison@arm.com>
4234 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4235 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4236 (AARCH64_FL_FOR_ARCH8_4): New.
4237 (AARCH64_FL_V8_4): New flag.
4238 (doc/invoke.texi): Document new armv8.4-a option.
4240 2018-01-10 Michael Collison <michael.collison@arm.com>
4242 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4243 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4244 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4245 * config/aarch64/aarch64-option-extension.def: Add
4246 AARCH64_OPT_EXTENSION of 'sha2'.
4247 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4248 (crypto): Disable sha2 and aes if crypto disabled.
4249 (crypto): Enable aes and sha2 if enabled.
4250 (simd): Disable sha2 and aes if simd disabled.
4251 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4253 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4254 (TARGET_SHA2): New feature flag for sha2.
4255 (TARGET_AES): New feature flag for aes.
4256 * config/aarch64/aarch64-simd.md:
4257 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4258 conditional on TARGET_AES.
4259 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4260 (aarch64_crypto_sha1hsi): Make pattern conditional
4262 (aarch64_crypto_sha1hv4si): Ditto.
4263 (aarch64_be_crypto_sha1hv4si): Ditto.
4264 (aarch64_crypto_sha1su1v4si): Ditto.
4265 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4266 (aarch64_crypto_sha1su0v4si): Ditto.
4267 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4268 (aarch64_crypto_sha256su0v4si): Ditto.
4269 (aarch64_crypto_sha256su1v4si): Ditto.
4270 (doc/invoke.texi): Document new aes and sha2 options.
4272 2018-01-10 Martin Sebor <msebor@redhat.com>
4274 PR tree-optimization/83781
4275 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4278 2018-01-11 Martin Sebor <msebor@gmail.com>
4279 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4281 PR tree-optimization/83501
4282 PR tree-optimization/81703
4284 * tree-ssa-strlen.c (get_string_cst): Rename...
4285 (get_string_len): ...to this. Handle global constants.
4286 (handle_char_store): Adjust.
4288 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4289 Jim Wilson <jimw@sifive.com>
4291 * config/riscv/riscv-protos.h (riscv_output_return): New.
4292 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4293 (riscv_attribute_table, riscv_output_return),
4294 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4295 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4296 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4298 (riscv_expand_prologue): Add early return for naked function.
4299 (riscv_expand_epilogue): Likewise.
4300 (riscv_function_ok_for_sibcall): Return false for naked function.
4301 (riscv_set_current_function): New.
4302 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4303 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4304 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4305 * doc/extend.texi (RISC-V Function Attributes): New.
4307 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4309 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4310 check for 128-bit long double before checking TCmode.
4311 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4312 128-bit long doubles before checking TFmode or TCmode.
4313 (FLOAT128_IBM_P): Likewise.
4315 2018-01-10 Martin Sebor <msebor@redhat.com>
4317 PR tree-optimization/83671
4318 * builtins.c (c_strlen): Unconditionally return zero for the empty
4320 Use -Warray-bounds for warnings.
4321 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4322 for non-constant array indices with COMPONENT_REF, arrays of
4323 arrays, and pointers to arrays.
4324 (gimple_fold_builtin_strlen): Determine and set length range for
4325 non-constant character arrays.
4327 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4330 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4333 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4335 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4337 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4340 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4341 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4342 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4343 indexed_or_indirect_operand predicate.
4344 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4345 (*vsx_le_perm_load_v8hi): Likewise.
4346 (*vsx_le_perm_load_v16qi): Likewise.
4347 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4348 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4349 (*vsx_le_perm_store_v8hi): Likewise.
4350 (*vsx_le_perm_store_v16qi): Likewise.
4351 (eight unnamed splitters): Likewise.
4353 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4355 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4356 * config/rs6000/emmintrin.h: Likewise.
4357 * config/rs6000/mmintrin.h: Likewise.
4358 * config/rs6000/xmmintrin.h: Likewise.
4360 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4363 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4365 * tree.c (tree_nop_conversion): Return true for location wrapper
4367 (maybe_wrap_with_location): New function.
4368 (selftest::check_strip_nops): New function.
4369 (selftest::test_location_wrappers): New function.
4370 (selftest::tree_c_tests): Call it.
4371 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4372 (maybe_wrap_with_location): New decl.
4373 (EXPR_LOCATION_WRAPPER_P): New macro.
4374 (location_wrapper_p): New inline function.
4375 (tree_strip_any_location_wrapper): New inline function.
4377 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4380 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4381 stack_realign_offset for the largest alignment of stack slot
4383 (ix86_find_max_used_stack_alignment): New function.
4384 (ix86_finalize_stack_frame_flags): Use it. Set
4385 max_used_stack_alignment if we don't realign stack.
4386 * config/i386/i386.h (machine_function): Add
4387 max_used_stack_alignment.
4389 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4391 * config/arm/arm.opt (-mbranch-cost): New option.
4392 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4395 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4398 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4399 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4401 2018-01-10 Richard Biener <rguenther@suse.de>
4404 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4405 early out so it also covers the case where we have a non-NULL
4408 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4410 PR tree-optimization/83753
4411 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4412 for non-strided grouped accesses if the number of elements is 1.
4414 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4417 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4418 * i386.h (TARGET_USE_GATHER): Define.
4419 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4421 2018-01-10 Martin Liska <mliska@suse.cz>
4424 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4425 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4427 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4428 CLEANUP_NO_PARTITIONING is not set.
4430 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4432 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4433 for vectors, as a partial revert of r254296.
4434 * rtl.h (const_vec_p): Delete.
4435 (const_vec_duplicate_p): Don't test for vector CONSTs.
4436 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4437 * expmed.c (make_tree): Likewise.
4440 * common.md (E, F): Use CONSTANT_P instead of checking for
4442 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4443 checking for CONST_VECTOR.
4445 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4448 * predict.c (force_edge_cold): Handle in more sane way edges
4451 2018-01-09 Carl Love <cel@us.ibm.com>
4453 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4455 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4456 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4457 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4458 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4459 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4460 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4461 * config/rs6000/rs6000-protos.h: Add extern defition for
4462 rs6000_generate_float2_double_code.
4463 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4465 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4466 (float2_v2df): Add define_expand.
4468 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4471 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4472 op_mode in the force_to_mode call.
4474 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4476 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4477 instead of checking each element individually.
4478 (aarch64_evpc_uzp): Likewise.
4479 (aarch64_evpc_zip): Likewise.
4480 (aarch64_evpc_ext): Likewise.
4481 (aarch64_evpc_rev): Likewise.
4482 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4483 instead of checking each element individually. Return true without
4485 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4486 whether all selected elements come from the same input, instead of
4487 checking each element individually. Remove calls to gen_rtx_REG,
4488 start_sequence and end_sequence and instead assert that no rtl is
4491 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4493 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4494 order of HIGH and CONST checks.
4496 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4498 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4499 if the destination isn't an SSA_NAME.
4501 2018-01-09 Richard Biener <rguenther@suse.de>
4503 PR tree-optimization/83668
4504 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4506 (canonicalize_loop_form): ... here, renamed from ...
4507 (canonicalize_loop_closed_ssa_form): ... this and amended to
4508 swap successor edges for loop exit blocks to make us use
4509 the RPO order we need for initial schedule generation.
4511 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4513 PR tree-optimization/64811
4514 * match.pd: When optimizing comparisons with Inf, avoid
4515 introducing or losing exceptions from comparisons with NaN.
4517 2018-01-09 Martin Liska <mliska@suse.cz>
4520 * asan.c (shadow_mem_size): Add gcc_assert.
4522 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4524 Don't save registers in main().
4527 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4528 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4529 * config/avr/avr.c (avr_set_current_function): Don't error if
4530 naked, OS_task or OS_main are specified at the same time.
4531 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4533 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4535 * common/config/avr/avr-common.c (avr_option_optimization_table):
4536 Switch on -mmain-is-OS_task for optimizing compilations.
4538 2018-01-09 Richard Biener <rguenther@suse.de>
4540 PR tree-optimization/83572
4541 * graphite.c: Include cfganal.h.
4542 (graphite_transform_loops): Connect infinite loops to exit
4543 and remove fake edges at the end.
4545 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4547 * ipa-inline.c (edge_badness): Revert accidental checkin.
4549 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4552 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4553 symbols; not inline clones.
4555 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4558 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4559 hard registers. Formatting fixes.
4561 PR preprocessor/83722
4562 * gcc.c (try_generate_repro): Pass
4563 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4564 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4567 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4568 Kito Cheng <kito.cheng@gmail.com>
4570 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4571 (riscv_leaf_function_p): Delete.
4572 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4574 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4576 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4578 (do_ifelse): New function.
4579 (do_isel): New function.
4580 (do_sub3): New function.
4581 (do_add3): New function.
4582 (do_load_mask_compare): New function.
4583 (do_overlap_load_compare): New function.
4584 (expand_compare_loop): New function.
4585 (expand_block_compare): Call expand_compare_loop() when appropriate.
4586 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4588 (-mblock-compare-inline-loop-limit): New option.
4590 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4593 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4594 Reverse order of second and third operands in first alternative.
4595 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4596 of first and second elements in UNSPEC_VPERMR vector.
4597 (altivec_expand_vec_perm_le): Likewise.
4599 2018-01-08 Jeff Law <law@redhat.com>
4601 PR rtl-optimizatin/81308
4602 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4603 (process_switch): If group_case_labels makes a change, then set
4605 (pass_convert_switch::execute): If a switch is converted, then
4606 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4608 PR rtl-optimization/81308
4609 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4612 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4614 PR target/83663 - Revert r255946
4615 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4616 generation for cases where splatting a value is not useful.
4617 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4618 across a vec_duplicate and a paradoxical subreg forming a vector
4619 mode to a vec_concat.
4621 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4623 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4624 -march=armv8.3-a variants.
4625 * config/arm/t-multilib: Likewise.
4626 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4628 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4630 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4632 (cceq_ior_compare_complement): Give it a name so I can use it, and
4633 change boolean_or_operator predicate to boolean_operator so it can
4634 be used to generate a crand.
4635 (eqne): New code iterator.
4636 (bd/bd_neg): New code_attrs.
4637 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4638 a single define_insn.
4639 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4640 decrement (bdnzt/bdnzf/bdzt/bdzf).
4641 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4642 with the new names of the branch decrement patterns, and added the
4643 names of the branch decrement conditional patterns.
4645 2018-01-08 Richard Biener <rguenther@suse.de>
4647 PR tree-optimization/83563
4648 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4651 2018-01-08 Richard Biener <rguenther@suse.de>
4654 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4656 2018-01-08 Richard Biener <rguenther@suse.de>
4658 PR tree-optimization/83685
4659 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4660 references to abnormals.
4662 2018-01-08 Richard Biener <rguenther@suse.de>
4665 * dwarf2out.c (output_indirect_strings): Handle empty
4666 skeleton_debug_str_hash.
4667 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4669 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4671 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4672 (emit_store_direct): Likewise.
4673 (arc_trampoline_adjust_address): Likewise.
4674 (arc_asm_trampoline_template): New function.
4675 (arc_initialize_trampoline): Use asm_trampoline_template.
4676 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4677 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4678 * config/arc/arc.md (flush_icache): Delete pattern.
4680 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4682 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4683 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4686 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4689 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4690 by not USED_FOR_TARGET.
4691 (make_pass_resolve_sw_modes): Likewise.
4693 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4695 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4698 2018-01-08 Richard Biener <rguenther@suse.de>
4701 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4703 2018-01-08 Richard Biener <rguenther@suse.de>
4706 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4708 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4711 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4712 basic blocks with a small number of successors.
4713 (convert_control_dep_chain_into_preds): Improve handling of
4715 (dump_predicates): Split apart into...
4716 (dump_pred_chain): ...here...
4717 (dump_pred_info): ...and here.
4718 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4719 (can_chain_union_be_invalidated_p): Improve check for invalidation
4721 (uninit_uses_cannot_happen): Avoid unnecessary if
4722 convert_control_dep_chain_into_preds yielded nothing.
4724 2018-01-06 Martin Sebor <msebor@redhat.com>
4726 PR tree-optimization/83640
4727 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4728 subtracting negative offset from size.
4729 (builtin_access::overlap): Adjust offset bounds of the access to fall
4730 within the size of the object if possible.
4732 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4734 PR rtl-optimization/83699
4735 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4736 extract_bit_field_as_subreg to cases in which the extracted
4737 value is also a vector.
4739 * lra-constraints.c (process_alt_operands): Test for the equivalence
4740 substitutions when detecting a possible reload cycle.
4742 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4745 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4746 by default if flag_selective_schedling{,2}. Formatting fixes.
4748 PR rtl-optimization/83682
4749 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4750 if it has non-VECTOR_MODE element mode.
4751 (vec_duplicate_p): Likewise.
4754 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4755 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4757 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4760 * config/i386/i386-builtin.def
4761 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4762 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4763 Require also OPTION_MASK_ISA_AVX512F in addition to
4764 OPTION_MASK_ISA_GFNI.
4765 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4766 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4767 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4768 to OPTION_MASK_ISA_GFNI.
4769 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4770 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4771 OPTION_MASK_ISA_AVX512BW.
4772 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4773 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4774 addition to OPTION_MASK_ISA_GFNI.
4775 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4776 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4777 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4778 to OPTION_MASK_ISA_GFNI.
4779 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4780 a requirement for all ISAs rather than any of them with a few
4782 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4784 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4785 bitmasks to be enabled with 3 exceptions, instead of requiring any
4786 enabled ISA with lots of exceptions.
4787 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4788 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4789 Change avx512bw in isa attribute to avx512f.
4790 * config/i386/sgxintrin.h: Add license boilerplate.
4791 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4792 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4793 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4794 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4796 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4797 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4798 temporarily sse2 rather than sse if not enabled already.
4801 * config/i386/sse.md (VI248_VLBW): Rename to ...
4802 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4803 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4804 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4805 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4806 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4807 mode iterator instead of VI248_VLBW.
4809 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4811 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4812 (record_modified): Skip clobbers; add debug output.
4813 (param_change_prob): Use sreal frequencies.
4815 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4817 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4818 punt for user-aligned variables.
4820 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4822 * tree-chrec.c (chrec_contains_symbols): Return true for
4825 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4828 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4829 of (x|y) == x for BICS pattern.
4831 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4833 PR tree-optimization/83605
4834 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4835 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4838 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4840 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4841 * config/epiphany/rtems.h: New file.
4843 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4844 Uros Bizjak <ubizjak@gmail.com>
4847 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4848 QIreg_operand instead of register_operand predicate.
4849 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4850 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4851 comments instead of -fmitigate[-_]rop.
4853 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4856 * cgraphunit.c (symbol_table::compile): Switch to text_section
4857 before calling assembly_start debug hook.
4858 * run-rtl-passes.c (run_rtl_passes): Likewise.
4861 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4863 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4864 range_int_cst_p rather than !symbolic_range_p before calling
4865 extract_range_from_multiplicative_op_1.
4867 2018-01-04 Jeff Law <law@redhat.com>
4869 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4870 redundant test in assertion.
4872 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4874 * doc/rtl.texi: Document machine_mode wrapper classes.
4876 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4878 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4881 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4883 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4884 the VEC_PERM_EXPR fold to fail.
4886 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4889 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4890 to switched_sections.
4892 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4895 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4898 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4901 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4902 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4904 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4907 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4908 is BLKmode and bitpos not zero or mode change is needed.
4910 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4913 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4916 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4919 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4920 instead of MULT rtx. Update all corresponding splitters.
4922 (*ssub<modesuffix>): Ditto.
4924 (*cmp_sadd_di): Update split patterns.
4925 (*cmp_sadd_si): Ditto.
4926 (*cmp_sadd_sidi): Ditto.
4927 (*cmp_ssub_di): Ditto.
4928 (*cmp_ssub_si): Ditto.
4929 (*cmp_ssub_sidi): Ditto.
4930 * config/alpha/predicates.md (const23_operand): New predicate.
4931 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4932 Look for ASHIFT, not MULT inner operand.
4933 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4935 2018-01-04 Martin Liska <mliska@suse.cz>
4937 PR gcov-profile/83669
4938 * gcov.c (output_intermediate_file): Add version to intermediate
4940 * doc/gcov.texi: Document new field 'version' in intermediate
4941 file format. Fix location of '-k' option of gcov command.
4943 2018-01-04 Martin Liska <mliska@suse.cz>
4946 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4948 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4950 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4952 2018-01-03 Martin Sebor <msebor@redhat.com>
4954 PR tree-optimization/83655
4955 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4956 checking calls with invalid arguments.
4958 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4960 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4961 (vectorizable_mask_load_store): Delete.
4962 (vectorizable_call): Return false for masked loads and stores.
4963 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4964 instead of gimple_assign_rhs1.
4965 (vectorizable_load): Handle IFN_MASK_LOAD.
4966 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4970 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4972 (vectorizable_mask_load_store): ...here.
4973 (vectorizable_load): ...and here.
4975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4977 * tree-vect-stmts.c (vect_build_all_ones_mask)
4978 (vect_build_zero_merge_argument): New functions, split out from...
4979 (vectorizable_load): ...here.
4981 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4983 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4985 (vectorizable_mask_load_store): ...here.
4986 (vectorizable_store): ...and here.
4988 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4990 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4992 (vectorizable_mask_load_store): ...here.
4994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4996 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4997 (vect_model_store_cost): Take a vec_load_store_type instead of a
4999 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5000 (vect_model_store_cost): Take a vec_load_store_type instead of a
5002 (vectorizable_mask_load_store): Update accordingly.
5003 (vectorizable_store): Likewise.
5004 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5006 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5008 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5009 IFN_MASK_LOAD calls here rather than...
5010 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5012 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5013 Alan Hayward <alan.hayward@arm.com>
5014 David Sherwood <david.sherwood@arm.com>
5016 * expmed.c (extract_bit_field_1): For vector extracts,
5017 fall back to extract_bit_field_as_subreg if vec_extract
5020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5021 Alan Hayward <alan.hayward@arm.com>
5022 David Sherwood <david.sherwood@arm.com>
5024 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5025 they are variable or constant sized.
5026 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5027 slots for constant-sized data.
5029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5030 Alan Hayward <alan.hayward@arm.com>
5031 David Sherwood <david.sherwood@arm.com>
5033 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5034 handling COND_EXPRs with boolean comparisons, try to find a better
5035 basis for the mask type than the boolean itself.
5037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5039 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5040 is calculated and how it can be overridden.
5041 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5042 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5044 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5048 Alan Hayward <alan.hayward@arm.com>
5049 David Sherwood <david.sherwood@arm.com>
5051 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5052 Remove the mode argument.
5053 (aarch64_simd_valid_immediate): Remove the mode and inverse
5055 * config/aarch64/iterators.md (bitsize): New iterator.
5056 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5057 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5058 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5059 aarch64_simd_valid_immediate.
5060 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5061 (aarch64_reg_or_bic_imm): Likewise.
5062 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5063 with an insn_type enum and msl with a modifier_type enum.
5064 Replace element_width with a scalar_mode. Change the shift
5065 to unsigned int. Add constructors for scalar_float_mode and
5066 scalar_int_mode elements.
5067 (aarch64_vect_float_const_representable_p): Delete.
5068 (aarch64_can_const_movi_rtx_p)
5069 (aarch64_simd_scalar_immediate_valid_for_move)
5070 (aarch64_simd_make_constant): Update call to
5071 aarch64_simd_valid_immediate.
5072 (aarch64_advsimd_valid_immediate_hs): New function.
5073 (aarch64_advsimd_valid_immediate): Likewise.
5074 (aarch64_simd_valid_immediate): Remove mode and inverse
5075 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5076 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5077 and aarch64_float_const_representable_p on the result.
5078 (aarch64_output_simd_mov_immediate): Remove mode argument.
5079 Update call to aarch64_simd_valid_immediate and use of
5080 simd_immediate_info.
5081 (aarch64_output_scalar_simd_mov_immediate): Update call
5084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5085 Alan Hayward <alan.hayward@arm.com>
5086 David Sherwood <david.sherwood@arm.com>
5088 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5089 (mode_nunits): Likewise CONST_MODE_NUNITS.
5090 * machmode.def (ADJUST_NUNITS): Document.
5091 * genmodes.c (mode_data::need_nunits_adj): New field.
5092 (blank_mode): Update accordingly.
5093 (adj_nunits): New variable.
5094 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5096 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5097 listed in adj_nunits.
5098 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5099 listed in adj_nunits. Don't emit case statements for such modes.
5100 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5101 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5102 nothing if adj_nunits is nonnull.
5103 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5104 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5105 (emit_mode_fbit): Update use of print_maybe_const_decl.
5106 (emit_move_size): Likewise. Treat the array as non-const
5108 (emit_mode_adjustments): Handle adj_nunits.
5110 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5112 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5113 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5114 (VECTOR_MODES): Use it.
5115 (make_vector_modes): Take the prefix as an argument.
5117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5118 Alan Hayward <alan.hayward@arm.com>
5119 David Sherwood <david.sherwood@arm.com>
5121 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5122 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5123 for MODE_VECTOR_BOOL.
5124 * machmode.def (VECTOR_BOOL_MODE): Document.
5125 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5126 (make_vector_bool_mode): New function.
5127 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5129 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5130 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5132 * stor-layout.c (int_mode_for_mode): Likewise.
5133 * tree.c (build_vector_type_for_mode): Likewise.
5134 * varasm.c (output_constant_pool_2): Likewise.
5135 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5136 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5137 for MODE_VECTOR_BOOL.
5138 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5139 of mode class checks.
5140 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5141 instead of a list of mode class checks.
5142 (expand_vector_scalar_condition): Likewise.
5143 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5146 Alan Hayward <alan.hayward@arm.com>
5147 David Sherwood <david.sherwood@arm.com>
5149 * machmode.h (mode_size): Change from unsigned short to
5151 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5152 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5153 or if measurement_type is not polynomial.
5154 (fixed_size_mode::includes_p): Check for constant-sized modes.
5155 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5156 return a poly_uint16 rather than an unsigned short.
5157 (emit_mode_size): Change the type of mode_size from unsigned short
5158 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5159 (emit_mode_adjustments): Cope with polynomial vector sizes.
5160 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5162 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5164 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5165 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5166 * caller-save.c (setup_save_areas): Likewise.
5167 (replace_reg_with_saved_mem): Likewise.
5168 * calls.c (emit_library_call_value_1): Likewise.
5169 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5170 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5171 (gen_lowpart_for_combine): Likewise.
5172 * convert.c (convert_to_integer_1): Likewise.
5173 * cse.c (equiv_constant, cse_insn): Likewise.
5174 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5175 (cselib_subst_to_values): Likewise.
5176 * dce.c (word_dce_process_block): Likewise.
5177 * df-problems.c (df_word_lr_mark_ref): Likewise.
5178 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5179 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5180 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5181 (rtl_for_decl_location): Likewise.
5182 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5183 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5184 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5185 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5186 (expand_expr_real_1): Likewise.
5187 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5188 (pad_below): Likewise.
5189 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5190 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5191 * ira.c (get_subreg_tracking_sizes): Likewise.
5192 * ira-build.c (ira_create_allocno_objects): Likewise.
5193 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5194 (ira_sort_regnos_for_alter_reg): Likewise.
5195 * ira-costs.c (record_operand_costs): Likewise.
5196 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5197 (resolve_simple_move): Likewise.
5198 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5199 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5200 (lra_constraints): Likewise.
5201 (CONST_POOL_OK_P): Reject variable-sized modes.
5202 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5203 (add_pseudo_to_slot, lra_spill): Likewise.
5204 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5205 * optabs-query.c (get_best_extraction_insn): Likewise.
5206 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5207 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5208 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5209 * recog.c (offsettable_address_addr_space_p): Likewise.
5210 * regcprop.c (maybe_mode_change): Likewise.
5211 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5212 * regrename.c (build_def_use): Likewise.
5213 * regstat.c (dump_reg_info): Likewise.
5214 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5215 (find_reloads, find_reloads_subreg_address): Likewise.
5216 * reload1.c (eliminate_regs_1): Likewise.
5217 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5218 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5219 (simplify_binary_operation_1, simplify_subreg): Likewise.
5220 * targhooks.c (default_function_arg_padding): Likewise.
5221 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5222 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5223 (verify_gimple_assign_ternary): Likewise.
5224 * tree-inline.c (estimate_move_cost): Likewise.
5225 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5226 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5227 (get_address_cost_ainc): Likewise.
5228 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5229 (vect_supportable_dr_alignment): Likewise.
5230 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5231 (vectorizable_reduction): Likewise.
5232 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5233 (vectorizable_operation, vectorizable_load): Likewise.
5234 * tree.c (build_same_sized_truth_vector_type): Likewise.
5235 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5236 * var-tracking.c (emit_note_insn_var_location): Likewise.
5237 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5238 (ADDR_VEC_ALIGN): Likewise.
5240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5241 Alan Hayward <alan.hayward@arm.com>
5242 David Sherwood <david.sherwood@arm.com>
5244 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5246 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5247 or if measurement_type is polynomial.
5248 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5249 * combine.c (make_extraction): Likewise.
5250 * dse.c (find_shift_sequence): Likewise.
5251 * dwarf2out.c (mem_loc_descriptor): Likewise.
5252 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5253 (extract_bit_field, extract_low_bits): Likewise.
5254 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5255 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5256 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5257 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5258 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5259 * reload.c (find_reloads): Likewise.
5260 * reload1.c (alter_reg): Likewise.
5261 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5262 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5263 * tree-if-conv.c (predicate_mem_writes): Likewise.
5264 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5265 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5266 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5267 * valtrack.c (dead_debug_insert_temp): Likewise.
5268 * varasm.c (mergeable_constant_section): Likewise.
5269 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5271 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5272 Alan Hayward <alan.hayward@arm.com>
5273 David Sherwood <david.sherwood@arm.com>
5275 * expr.c (expand_assignment): Cope with polynomial mode sizes
5276 when assigning to a CONCAT.
5278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5279 Alan Hayward <alan.hayward@arm.com>
5280 David Sherwood <david.sherwood@arm.com>
5282 * machmode.h (mode_precision): Change from unsigned short to
5284 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5286 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5287 or if measurement_type is not polynomial.
5288 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5289 in which the mode is already known to be a scalar_int_mode.
5290 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5291 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5293 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5294 for GET_MODE_PRECISION.
5295 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5296 for GET_MODE_PRECISION.
5297 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5299 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5300 (expand_field_assignment, make_extraction): Likewise.
5301 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5302 (get_last_value): Likewise.
5303 * convert.c (convert_to_integer_1): Likewise.
5304 * cse.c (cse_insn): Likewise.
5305 * expr.c (expand_expr_real_1): Likewise.
5306 * lra-constraints.c (simplify_operand_subreg): Likewise.
5307 * optabs-query.c (can_atomic_load_p): Likewise.
5308 * optabs.c (expand_atomic_load): Likewise.
5309 (expand_atomic_store): Likewise.
5310 * ree.c (combine_reaching_defs): Likewise.
5311 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5312 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5313 * tree.h (type_has_mode_precision_p): Likewise.
5314 * ubsan.c (instrument_si_overflow): Likewise.
5316 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5317 Alan Hayward <alan.hayward@arm.com>
5318 David Sherwood <david.sherwood@arm.com>
5320 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5321 polynomial numbers of units.
5322 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5323 (valid_vector_subparts_p): New function.
5324 (build_vector_type): Remove temporary shim and take the number
5325 of units as a poly_uint64 rather than an int.
5326 (build_opaque_vector_type): Take the number of units as a
5327 poly_uint64 rather than an int.
5328 * tree.c (build_vector_from_ctor): Handle polynomial
5329 TYPE_VECTOR_SUBPARTS.
5330 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5331 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5332 (build_vector_from_val): If the number of units is variable,
5333 use build_vec_duplicate_cst for constant operands and
5334 VEC_DUPLICATE_EXPR otherwise.
5335 (make_vector_type): Remove temporary is_constant ().
5336 (build_vector_type, build_opaque_vector_type): Take the number of
5337 units as a poly_uint64 rather than an int.
5338 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5340 * cfgexpand.c (expand_debug_expr): Likewise.
5341 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5342 (store_constructor, expand_expr_real_1): Likewise.
5343 (const_scalar_mask_from_tree): Likewise.
5344 * fold-const-call.c (fold_const_reduction): Likewise.
5345 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5346 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5347 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5348 (fold_relational_const): Likewise.
5349 (native_interpret_vector): Likewise. Change the size from an
5350 int to an unsigned int.
5351 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5352 TYPE_VECTOR_SUBPARTS.
5353 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5354 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5355 duplicating a non-constant operand into a variable-length vector.
5356 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5357 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5358 * ipa-icf.c (sem_variable::equals): Likewise.
5359 * match.pd: Likewise.
5360 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5361 * print-tree.c (print_node): Likewise.
5362 * stor-layout.c (layout_type): Likewise.
5363 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5364 * tree-cfg.c (verify_gimple_comparison): Likewise.
5365 (verify_gimple_assign_binary): Likewise.
5366 (verify_gimple_assign_ternary): Likewise.
5367 (verify_gimple_assign_single): Likewise.
5368 * tree-pretty-print.c (dump_generic_node): Likewise.
5369 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5370 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5371 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5372 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5373 (vect_shift_permute_load_chain): Likewise.
5374 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5375 (expand_vector_condition, optimize_vector_constructor): Likewise.
5376 (lower_vec_perm, get_compute_type): Likewise.
5377 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5378 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5379 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5380 (vect_recog_mask_conversion_pattern): Likewise.
5381 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5382 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5383 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5384 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5385 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5386 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5387 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5388 (supportable_widening_operation): Likewise.
5389 (supportable_narrowing_operation): Likewise.
5390 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5392 * varasm.c (output_constant): Likewise.
5394 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5395 Alan Hayward <alan.hayward@arm.com>
5396 David Sherwood <david.sherwood@arm.com>
5398 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5399 so that both the length == 3 and length != 3 cases set up their
5400 own permute vectors. Add comments explaining why we know the
5401 number of elements is constant.
5402 (vect_permute_load_chain): Likewise.
5404 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5405 Alan Hayward <alan.hayward@arm.com>
5406 David Sherwood <david.sherwood@arm.com>
5408 * machmode.h (mode_nunits): Change from unsigned char to
5410 (ONLY_FIXED_SIZE_MODES): New macro.
5411 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5412 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5413 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5415 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5416 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5417 or if measurement_type is not polynomial.
5418 * genmodes.c (ZERO_COEFFS): New macro.
5419 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5421 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5422 Use ZERO_COEFFS when emitting initializers.
5423 * data-streamer.h (bp_pack_poly_value): New function.
5424 (bp_unpack_poly_value): Likewise.
5425 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5426 for GET_MODE_NUNITS.
5427 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5428 for GET_MODE_NUNITS.
5429 * tree.c (make_vector_type): Remove temporary shim and make
5430 the real function take the number of units as a poly_uint64
5432 (build_vector_type_for_mode): Handle polynomial nunits.
5433 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5434 * emit-rtl.c (const_vec_series_p_1): Likewise.
5435 (gen_rtx_CONST_VECTOR): Likewise.
5436 * fold-const.c (test_vec_duplicate_folding): Likewise.
5437 * genrecog.c (validate_pattern): Likewise.
5438 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5439 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5440 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5441 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5442 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5443 * rtlanal.c (subreg_get_info): Likewise.
5444 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5445 (vect_grouped_load_supported): Likewise.
5446 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5447 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5448 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5449 (simplify_const_unary_operation, simplify_binary_operation_1)
5450 (simplify_const_binary_operation, simplify_ternary_operation)
5451 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5452 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5453 instead of CONST_VECTOR_NUNITS.
5454 * varasm.c (output_constant_pool_2): Likewise.
5455 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5456 explicit-encoded elements in the XVEC for variable-length vectors.
5458 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5460 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5463 Alan Hayward <alan.hayward@arm.com>
5464 David Sherwood <david.sherwood@arm.com>
5466 * coretypes.h (fixed_size_mode): Declare.
5467 (fixed_size_mode_pod): New typedef.
5468 * builtins.h (target_builtins::x_apply_args_mode)
5469 (target_builtins::x_apply_result_mode): Change type to
5470 fixed_size_mode_pod.
5471 * builtins.c (apply_args_size, apply_result_size, result_vector)
5472 (expand_builtin_apply_args_1, expand_builtin_apply)
5473 (expand_builtin_return): Update accordingly.
5475 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5477 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5478 * cselib.c (cselib_hash_rtx): Likewise.
5479 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5480 CONST_VECTOR encoding.
5482 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5483 Jeff Law <law@redhat.com>
5486 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5487 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5488 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5489 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5492 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5493 explicitly probe *sp in a noreturn function if there were any callee
5494 register saves or frame pointer is needed.
5496 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5499 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5500 BLKmode for ternary, binary or unary expressions.
5503 * var-tracking.c (delete_vta_debug_insn): New inline function.
5504 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5505 insns from get_insns () to NULL instead of each bb separately.
5506 Use delete_vta_debug_insn. No longer static.
5507 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5508 delete_vta_debug_insns callers.
5509 * rtl.h (delete_vta_debug_insns): Declare.
5510 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5511 instead of variable_tracking_main.
5513 2018-01-03 Martin Sebor <msebor@redhat.com>
5515 PR tree-optimization/83603
5516 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5517 arguments past the endof the argument list in functions declared
5518 without a prototype.
5519 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5520 Avoid checking when arguments are null.
5522 2018-01-03 Martin Sebor <msebor@redhat.com>
5525 * doc/extend.texi (attribute const): Fix a typo.
5526 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5527 issuing -Wsuggest-attribute for void functions.
5529 2018-01-03 Martin Sebor <msebor@redhat.com>
5531 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5532 offset_int::from instead of wide_int::to_shwi.
5533 (maybe_diag_overlap): Remove assertion.
5534 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5535 * gimple-ssa-sprintf.c (format_directive): Same.
5536 (parse_directive): Same.
5537 (sprintf_dom_walker::compute_format_length): Same.
5538 (try_substitute_return_value): Same.
5540 2018-01-03 Jeff Law <law@redhat.com>
5543 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5544 non-constant residual for zero at runtime and avoid probing in
5545 that case. Reorganize code for trailing problem to mirror handling
5548 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5550 PR tree-optimization/83501
5551 * tree-ssa-strlen.c (get_string_cst): New.
5552 (handle_char_store): Call get_string_cst.
5554 2018-01-03 Martin Liska <mliska@suse.cz>
5556 PR tree-optimization/83593
5557 * tree-ssa-strlen.c: Include tree-cfg.h.
5558 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5559 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5560 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5562 (strlen_dom_walker::before_dom_children): Call
5563 gimple_purge_dead_eh_edges. Dump tranformation with details
5565 (strlen_dom_walker::before_dom_children): Update call by adding
5566 new argument cleanup_eh.
5567 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5569 2018-01-03 Martin Liska <mliska@suse.cz>
5572 * cif-code.def (VARIADIC_THUNK): New enum value.
5573 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5576 2018-01-03 Jan Beulich <jbeulich@suse.com>
5578 * sse.md (mov<mode>_internal): Tighten condition for when to use
5579 vmovdqu<ssescalarsize> for TI and OI modes.
5581 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5583 Update copyright years.
5585 2018-01-03 Martin Liska <mliska@suse.cz>
5588 * ipa-visibility.c (function_and_variable_visibility): Skip
5589 functions with noipa attribure.
5591 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5593 * gcc.c (process_command): Update copyright notice dates.
5594 * gcov-dump.c (print_version): Ditto.
5595 * gcov.c (print_version): Ditto.
5596 * gcov-tool.c (print_version): Ditto.
5597 * gengtype.c (create_file): Ditto.
5598 * doc/cpp.texi: Bump @copying's copyright year.
5599 * doc/cppinternals.texi: Ditto.
5600 * doc/gcc.texi: Ditto.
5601 * doc/gccint.texi: Ditto.
5602 * doc/gcov.texi: Ditto.
5603 * doc/install.texi: Ditto.
5604 * doc/invoke.texi: Ditto.
5606 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5608 * vector-builder.h (vector_builder::m_full_nelts): Change from
5609 unsigned int to poly_uint64.
5610 (vector_builder::full_nelts): Update prototype accordingly.
5611 (vector_builder::new_vector): Likewise.
5612 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5613 (vector_builder::operator ==): Likewise.
5614 (vector_builder::finalize): Likewise.
5615 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5616 Take the number of elements as a poly_uint64 rather than an
5618 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5619 from unsigned int to poly_uint64.
5620 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5621 (vec_perm_indices::new_vector): Likewise.
5622 (vec_perm_indices::length): Likewise.
5623 (vec_perm_indices::nelts_per_input): Likewise.
5624 (vec_perm_indices::input_nelts): Likewise.
5625 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5626 number of elements per input as a poly_uint64 rather than an
5627 unsigned int. Use the original encoding for variable-length
5628 vectors, rather than clamping each individual element.
5629 For the second and subsequent elements in each pattern,
5630 clamp the step and base before clamping their sum.
5631 (vec_perm_indices::series_p): Handle polynomial element counts.
5632 (vec_perm_indices::all_in_range_p): Likewise.
5633 (vec_perm_indices_to_tree): Likewise.
5634 (vec_perm_indices_to_rtx): Likewise.
5635 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5636 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5637 (tree_vector_builder::new_binary_operation): Handle polynomial
5638 element counts. Return false if we need to know the number
5639 of elements at compile time.
5640 * fold-const.c (fold_vec_perm): Punt if the number of elements
5641 isn't known at compile time.
5643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5645 * vec-perm-indices.h (vec_perm_builder): Change element type
5646 from HOST_WIDE_INT to poly_int64.
5647 (vec_perm_indices::element_type): Update accordingly.
5648 (vec_perm_indices::clamp): Handle polynomial element_types.
5649 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5650 (vec_perm_indices::all_in_range_p): Likewise.
5651 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5653 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5654 polynomial vec_perm_indices element types.
5655 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5656 * fold-const.c (fold_vec_perm): Likewise.
5657 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5658 * tree-vect-generic.c (lower_vec_perm): Likewise.
5659 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5660 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5661 element type to HOST_WIDE_INT.
5663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5664 Alan Hayward <alan.hayward@arm.com>
5665 David Sherwood <david.sherwood@arm.com>
5667 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5668 rather than an int. Use plus_constant.
5669 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5670 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
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 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5677 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
5684 mode sizes. Require a constant size for BLKmode parameters
5685 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5686 forces a parameter to be padded at the lsb end in order to
5687 fill a complete number of words, require the parameter size
5688 to be ordered wrt UNITS_PER_WORD.
5690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5691 Alan Hayward <alan.hayward@arm.com>
5692 David Sherwood <david.sherwood@arm.com>
5694 * reload1.c (spill_stack_slot_width): Change element type
5695 from unsigned int to poly_uint64_pod.
5696 (alter_reg): Treat mode sizes as polynomial.
5698 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5699 Alan Hayward <alan.hayward@arm.com>
5700 David Sherwood <david.sherwood@arm.com>
5702 * reload.c (complex_word_subreg_p): New function.
5703 (reload_inner_reg_of_subreg, push_reload): Use it.
5705 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5706 Alan Hayward <alan.hayward@arm.com>
5707 David Sherwood <david.sherwood@arm.com>
5709 * lra-constraints.c (process_alt_operands): Reject matched
5710 operands whose sizes aren't ordered.
5711 (match_reload): Refer to this check here.
5713 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5714 Alan Hayward <alan.hayward@arm.com>
5715 David Sherwood <david.sherwood@arm.com>
5717 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5718 that the mode size is in the set {1, 2, 4, 8, 16}.
5720 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5721 Alan Hayward <alan.hayward@arm.com>
5722 David Sherwood <david.sherwood@arm.com>
5724 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5725 Use plus_constant instead of gen_rtx_PLUS.
5727 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5728 Alan Hayward <alan.hayward@arm.com>
5729 David Sherwood <david.sherwood@arm.com>
5731 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5732 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5733 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5734 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5735 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5736 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5737 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5738 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5739 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5740 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5742 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5743 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5744 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5745 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5746 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5747 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5748 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5749 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5750 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5751 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5753 * expr.c (emit_move_resolve_push): Treat the input and result
5754 of PUSH_ROUNDING as a poly_int64.
5755 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5756 (emit_push_insn): Likewise.
5757 * lra-eliminations.c (mark_not_eliminable): Likewise.
5758 * recog.c (push_operand): Likewise.
5759 * reload1.c (elimination_effects): Likewise.
5760 * rtlanal.c (nonzero_bits1): Likewise.
5761 * calls.c (store_one_arg): Likewise. Require the padding to be
5762 known at compile time.
5764 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5765 Alan Hayward <alan.hayward@arm.com>
5766 David Sherwood <david.sherwood@arm.com>
5768 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5769 Use plus_constant instead of gen_rtx_PLUS.
5771 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5772 Alan Hayward <alan.hayward@arm.com>
5773 David Sherwood <david.sherwood@arm.com>
5775 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5778 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5779 Alan Hayward <alan.hayward@arm.com>
5780 David Sherwood <david.sherwood@arm.com>
5782 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5783 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5784 via stack temporaries. Treat the mode size as polynomial too.
5786 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5787 Alan Hayward <alan.hayward@arm.com>
5788 David Sherwood <david.sherwood@arm.com>
5790 * expr.c (expand_expr_real_2): When handling conversions involving
5791 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5792 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5793 as a poly_uint64 too.
5795 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5796 Alan Hayward <alan.hayward@arm.com>
5797 David Sherwood <david.sherwood@arm.com>
5799 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5801 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5802 Alan Hayward <alan.hayward@arm.com>
5803 David Sherwood <david.sherwood@arm.com>
5805 * combine.c (can_change_dest_mode): Handle polynomial
5806 REGMODE_NATURAL_SIZE.
5807 * expmed.c (store_bit_field_1): Likewise.
5808 * expr.c (store_constructor): Likewise.
5809 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5810 and polynomial REGMODE_NATURAL_SIZE.
5811 (gen_lowpart_common): Likewise.
5812 * reginfo.c (record_subregs_of_mode): Likewise.
5813 * rtlanal.c (read_modify_subreg_p): Likewise.
5815 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5816 Alan Hayward <alan.hayward@arm.com>
5817 David Sherwood <david.sherwood@arm.com>
5819 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5820 numbers of elements.
5822 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5823 Alan Hayward <alan.hayward@arm.com>
5824 David Sherwood <david.sherwood@arm.com>
5826 * match.pd: Cope with polynomial numbers of vector elements.
5828 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5829 Alan Hayward <alan.hayward@arm.com>
5830 David Sherwood <david.sherwood@arm.com>
5832 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5833 in a POINTER_PLUS_EXPR.
5835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5836 Alan Hayward <alan.hayward@arm.com>
5837 David Sherwood <david.sherwood@arm.com>
5839 * omp-simd-clone.c (simd_clone_subparts): New function.
5840 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5841 (ipa_simd_modify_function_body): Likewise.
5843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5844 Alan Hayward <alan.hayward@arm.com>
5845 David Sherwood <david.sherwood@arm.com>
5847 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5848 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5849 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5850 (expand_vector_condition, vector_element): Likewise.
5851 (subparts_gt): New function.
5852 (get_compute_type): Use subparts_gt.
5853 (count_type_subparts): Delete.
5854 (expand_vector_operations_1): Use subparts_gt instead of
5855 count_type_subparts.
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-data-refs.c (vect_no_alias_p): Replace with...
5862 (vect_compile_time_alias): ...this new function. Do the calculation
5863 on poly_ints rather than trees.
5864 (vect_prune_runtime_alias_test_list): Update call accordingly.
5866 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5867 Alan Hayward <alan.hayward@arm.com>
5868 David Sherwood <david.sherwood@arm.com>
5870 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5872 (vect_schedule_slp_instance): Likewise.
5874 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5875 Alan Hayward <alan.hayward@arm.com>
5876 David Sherwood <david.sherwood@arm.com>
5878 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5879 constant and extern definitions for variable-length vectors.
5880 (vect_get_constant_vectors): Note that the number of units
5881 is known to be constant.
5883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5884 Alan Hayward <alan.hayward@arm.com>
5885 David Sherwood <david.sherwood@arm.com>
5887 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5888 of units as polynomial. Choose between WIDE and NARROW based
5891 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5892 Alan Hayward <alan.hayward@arm.com>
5893 David Sherwood <david.sherwood@arm.com>
5895 * tree-vect-stmts.c (simd_clone_subparts): New function.
5896 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5899 Alan Hayward <alan.hayward@arm.com>
5900 David Sherwood <david.sherwood@arm.com>
5902 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5903 vectors as polynomial. Use build_index_vector for
5906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5907 Alan Hayward <alan.hayward@arm.com>
5908 David Sherwood <david.sherwood@arm.com>
5910 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5911 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5912 for variable-length vectors.
5913 (vectorizable_mask_load_store): Treat the number of units as
5914 polynomial, asserting that it is constant if the condition has
5915 already been enforced.
5916 (vectorizable_store, vectorizable_load): Likewise.
5918 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5919 Alan Hayward <alan.hayward@arm.com>
5920 David Sherwood <david.sherwood@arm.com>
5922 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5923 of units as polynomial. Punt if we can't tell at compile time
5924 which vector contains the final result.
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-loop.c (vectorizable_induction): Treat the number
5931 of units as polynomial. Punt on SLP inductions. Use an integer
5932 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5933 cast of such a series for variable-length floating-point
5936 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5937 Alan Hayward <alan.hayward@arm.com>
5938 David Sherwood <david.sherwood@arm.com>
5940 * tree.h (build_index_vector): Declare.
5941 * tree.c (build_index_vector): New function.
5942 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5943 of units as polynomial, forcibly converting it to a constant if
5944 vectorizable_reduction has already enforced the condition.
5945 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5946 to create a {1,2,3,...} vector.
5947 (vectorizable_reduction): Treat the number of units as polynomial.
5948 Choose vectype_in based on the largest scalar element size rather
5949 than the smallest number of units. Enforce the restrictions
5952 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5953 Alan Hayward <alan.hayward@arm.com>
5954 David Sherwood <david.sherwood@arm.com>
5956 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5957 number of units as polynomial.
5959 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5960 Alan Hayward <alan.hayward@arm.com>
5961 David Sherwood <david.sherwood@arm.com>
5963 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5964 * target.def (autovectorize_vector_sizes): Return the vector sizes
5965 by pointer, using vector_sizes rather than a bitmask.
5966 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5967 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5968 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5970 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5971 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5972 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5973 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5974 * omp-general.c (omp_max_vf): Likewise.
5975 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5976 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5977 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5978 * tree-vect-slp.c (vect_slp_bb): Likewise.
5979 * doc/tm.texi: Regenerate.
5980 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5982 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5983 the vector size as a poly_uint64 rather than an unsigned int.
5984 (current_vector_size): Change from an unsigned int to a poly_uint64.
5985 (get_vectype_for_scalar_type): Update accordingly.
5986 * tree.h (build_truth_vector_type): Take the size and number of
5987 units as a poly_uint64 rather than an unsigned int.
5988 (build_vector_type): Add a temporary overload that takes
5989 the number of units as a poly_uint64 rather than an unsigned int.
5990 * tree.c (make_vector_type): Likewise.
5991 (build_truth_vector_type): Take the number of units as a poly_uint64
5992 rather than an unsigned int.
5994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5995 Alan Hayward <alan.hayward@arm.com>
5996 David Sherwood <david.sherwood@arm.com>
5998 * target.def (get_mask_mode): Take the number of units and length
5999 as poly_uint64s rather than unsigned ints.
6000 * targhooks.h (default_get_mask_mode): Update accordingly.
6001 * targhooks.c (default_get_mask_mode): Likewise.
6002 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6003 * doc/tm.texi: Regenerate.
6005 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6006 Alan Hayward <alan.hayward@arm.com>
6007 David Sherwood <david.sherwood@arm.com>
6009 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6010 * omp-general.c (omp_max_vf): Likewise.
6011 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6012 (expand_omp_simd): Handle polynomial safelen.
6013 * omp-low.c (omplow_simd_context): Add a default constructor.
6014 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6015 (lower_rec_simd_input_clauses): Update accordingly.
6016 (lower_rec_input_clauses): Likewise.
6018 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6019 Alan Hayward <alan.hayward@arm.com>
6020 David Sherwood <david.sherwood@arm.com>
6022 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6023 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6024 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6025 (vect_analyze_slp_cost): Likewise.
6026 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6027 (vect_model_load_cost): Likewise.
6029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6030 Alan Hayward <alan.hayward@arm.com>
6031 David Sherwood <david.sherwood@arm.com>
6033 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6034 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6035 from an unsigned int * to a poly_uint64_pod *.
6036 (calculate_unrolling_factor): New function.
6037 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6040 Alan Hayward <alan.hayward@arm.com>
6041 David Sherwood <david.sherwood@arm.com>
6043 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6044 from an unsigned int to a poly_uint64.
6045 (_loop_vec_info::slp_unrolling_factor): Likewise.
6046 (_loop_vec_info::vectorization_factor): Change from an int
6048 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6049 (vect_get_num_vectors): New function.
6050 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6051 (vect_get_num_copies): Use vect_get_num_vectors.
6052 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6053 to an unsigned int *.
6054 (vect_analyze_data_refs): Change min_vf from an int * to a
6056 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6057 than an unsigned HOST_WIDE_INT.
6058 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6059 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6060 to an unsigned int *.
6061 (vect_analyze_data_ref_dependences): Likewise.
6062 (vect_compute_data_ref_alignment): Handle polynomial vf.
6063 (vect_enhance_data_refs_alignment): Likewise.
6064 (vect_prune_runtime_alias_test_list): Likewise.
6065 (vect_shift_permute_load_chain): Likewise.
6066 (vect_supportable_dr_alignment): Likewise.
6067 (dependence_distance_ge_vf): Take the vectorization factor as a
6068 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6069 (vect_analyze_data_refs): Change min_vf from an int * to a
6071 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6072 vfm1 as a poly_uint64 rather than an int. Make the same change
6073 for the returned bound_scalar.
6074 (vect_gen_vector_loop_niters): Handle polynomial vf.
6075 (vect_do_peeling): Likewise. Update call to
6076 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6077 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6079 * tree-vect-loop.c (vect_determine_vectorization_factor)
6080 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6081 (vect_get_known_peeling_cost): Likewise.
6082 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6083 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6084 (vect_transform_loop): Likewise. Use the lowest possible VF when
6085 updating the upper bounds of the loop.
6086 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6088 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6089 polynomial unroll factors.
6090 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6091 (vect_make_slp_decision): Likewise.
6092 (vect_supported_load_permutation_p): Likewise, and polynomial
6094 (vect_analyze_slp_cost): Handle polynomial vf.
6095 (vect_slp_analyze_node_operations): Likewise.
6096 (vect_slp_analyze_bb_1): Likewise.
6097 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6098 than an unsigned HOST_WIDE_INT.
6099 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6100 (vectorizable_load): Handle polynomial vf.
6101 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6103 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6106 Alan Hayward <alan.hayward@arm.com>
6107 David Sherwood <david.sherwood@arm.com>
6109 * match.pd: Handle bit operations involving three constants
6110 and try to fold one pair.
6112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6114 * tree-vect-loop-manip.c: Include gimple-fold.h.
6115 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6116 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6117 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6118 Add a path that uses a step of VF instead of 1, but disable it
6120 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6121 and niters_no_overflow parameters. Update calls to
6122 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6123 Create a new SSA name if the latter choses to use a ste other
6124 than zero, and return it via niters_vector_mult_vf_var.
6125 * tree-vect-loop.c (vect_transform_loop): Update calls to
6126 vect_do_peeling, vect_gen_vector_loop_niters and
6127 slpeel_make_loop_iterate_ntimes.
6128 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6129 (vect_gen_vector_loop_niters): Update declarations after above changes.
6131 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6133 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6134 128-bit round to integer instructions.
6135 (ceil<mode>2): Likewise.
6136 (btrunc<mode>2): Likewise.
6137 (round<mode>2): Likewise.
6139 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6141 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6142 unaligned VSX load/store on P8/P9.
6143 (expand_block_clear): Allow the use of unaligned VSX
6144 load/store on P8/P9.
6146 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6148 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6150 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6151 swap associated with both a load and a store.
6153 2018-01-02 Andrew Waterman <andrew@sifive.com>
6155 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6156 * config/riscv/riscv.md (clear_cache): Use it.
6158 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6160 * web.c: Remove out-of-date comment.
6162 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6164 * expr.c (fixup_args_size_notes): Check that any existing
6165 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6166 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6167 (emit_single_push_insn): ...here.
6169 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6171 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6172 (const_vector_encoded_nelts): New function.
6173 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6174 (const_vector_int_elt, const_vector_elt): Declare.
6175 * emit-rtl.c (const_vector_int_elt_1): New function.
6176 (const_vector_elt): Likewise.
6177 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6178 of CONST_VECTOR_ELT.
6180 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6182 * expr.c: Include rtx-vector-builder.h.
6183 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6184 directly on the tree encoding.
6185 (const_vector_from_tree): Likewise.
6186 * optabs.c: Include rtx-vector-builder.h.
6187 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6188 sequence of "u" values.
6189 * vec-perm-indices.c: Include rtx-vector-builder.h.
6190 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6191 directly on the vec_perm_indices encoding.
6193 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6195 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6196 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6197 * rtx-vector-builder.h: New file.
6198 * rtx-vector-builder.c: Likewise.
6199 * rtl.h (rtx_def::u2): Add a const_vector field.
6200 (CONST_VECTOR_NPATTERNS): New macro.
6201 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6202 (CONST_VECTOR_DUPLICATE_P): Likewise.
6203 (CONST_VECTOR_STEPPED_P): Likewise.
6204 (CONST_VECTOR_ENCODED_ELT): Likewise.
6205 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6206 (unwrap_const_vec_duplicate): Likewise.
6207 (const_vec_series_p): Check for a non-duplicated vector encoding.
6208 Say that the function only returns true for integer vectors.
6209 * emit-rtl.c: Include rtx-vector-builder.h.
6210 (gen_const_vec_duplicate_1): Delete.
6211 (gen_const_vector): Call gen_const_vec_duplicate instead of
6212 gen_const_vec_duplicate_1.
6213 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6214 (gen_const_vec_duplicate): Use rtx_vector_builder.
6215 (gen_const_vec_series): Likewise.
6216 (gen_rtx_CONST_VECTOR): Likewise.
6217 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6218 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6219 Build a new vector rather than modifying a CONST_VECTOR in-place.
6220 (handle_special_swappables): Update call accordingly.
6221 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6222 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6223 Build a new vector rather than modifying a CONST_VECTOR in-place.
6224 (handle_special_swappables): Update call accordingly.
6226 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6228 * simplify-rtx.c (simplify_const_binary_operation): Use
6229 CONST_VECTOR_ELT instead of XVECEXP.
6231 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6233 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6234 the selector elements to be different from the data elements
6235 if the selector is a VECTOR_CST.
6236 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6237 ssizetype for the selector.
6239 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6241 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6242 before testing each element individually.
6243 * tree-vect-generic.c (lower_vec_perm): Likewise.
6245 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6247 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6248 * selftest-run-tests.c (selftest::run_tests): Call it.
6249 * vector-builder.h (vector_builder::operator ==): New function.
6250 (vector_builder::operator !=): Likewise.
6251 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6252 (vec_perm_indices::all_from_input_p): New function.
6253 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6254 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6255 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6256 instead of reading the VECTOR_CST directly. Detect whether both
6257 vector inputs are the same before constructing the vec_perm_indices,
6258 and update the number of inputs argument accordingly. Use the
6259 utility functions added above. Only construct sel2 if we need to.
6261 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6263 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6264 the broadcast of the low byte.
6265 (expand_mult_highpart): Use an explicit encoding for the permutes.
6266 * optabs-query.c (can_mult_highpart_p): Likewise.
6267 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6268 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6269 (vectorizable_bswap): Likewise.
6270 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6271 explicit encoding for the power-of-2 permutes.
6272 (vect_permute_store_chain): Likewise.
6273 (vect_grouped_load_supported): Likewise.
6274 (vect_permute_load_chain): Likewise.
6276 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6278 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6279 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6280 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6281 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6282 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6283 (vect_gen_perm_mask_any): Likewise.
6285 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6287 * int-vector-builder.h: New file.
6288 * vec-perm-indices.h: Include int-vector-builder.h.
6289 (vec_perm_indices): Redefine as an int_vector_builder.
6290 (auto_vec_perm_indices): Delete.
6291 (vec_perm_builder): Redefine as a stand-alone class.
6292 (vec_perm_indices::vec_perm_indices): New function.
6293 (vec_perm_indices::clamp): Likewise.
6294 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6295 (vec_perm_indices::new_vector): New function.
6296 (vec_perm_indices::new_expanded_vector): Update for new
6297 vec_perm_indices class.
6298 (vec_perm_indices::rotate_inputs): New function.
6299 (vec_perm_indices::all_in_range_p): Operate directly on the
6300 encoded form, without computing elided elements.
6301 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6302 encoding. Update for new vec_perm_indices class.
6303 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6304 the given vec_perm_builder.
6305 (expand_vec_perm_var): Update vec_perm_builder constructor.
6306 (expand_mult_highpart): Use vec_perm_builder instead of
6307 auto_vec_perm_indices.
6308 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6309 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6310 or double series encoding as appropriate.
6311 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6312 vec_perm_indices instead of auto_vec_perm_indices.
6313 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6314 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6315 (vect_permute_store_chain): Likewise.
6316 (vect_grouped_load_supported): Likewise.
6317 (vect_permute_load_chain): Likewise.
6318 (vect_shift_permute_load_chain): Likewise.
6319 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6320 (vect_transform_slp_perm_load): Likewise.
6321 (vect_schedule_slp_instance): Likewise.
6322 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6323 (vectorizable_mask_load_store): Likewise.
6324 (vectorizable_bswap): Likewise.
6325 (vectorizable_store): Likewise.
6326 (vectorizable_load): Likewise.
6327 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6328 vec_perm_indices instead of auto_vec_perm_indices. Use
6329 tree_to_vec_perm_builder to read the vector from a tree.
6330 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6331 vec_perm_builder instead of a vec_perm_indices.
6332 (have_whole_vector_shift): Use vec_perm_builder and
6333 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6334 truncation to calc_vec_perm_mask_for_shift.
6335 (vect_create_epilog_for_reduction): Likewise.
6336 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6337 from auto_vec_perm_indices to vec_perm_indices.
6338 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6339 instead of changing individual elements.
6340 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6341 the vector in d.perm.
6342 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6343 from auto_vec_perm_indices to vec_perm_indices.
6344 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6345 instead of changing individual elements.
6346 (arm_vectorize_vec_perm_const): Use new_vector to install
6347 the vector in d.perm.
6348 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6349 Update vec_perm_builder constructor.
6350 (rs6000_expand_interleave): Likewise.
6351 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6352 (rs6000_expand_interleave): Likewise.
6354 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6356 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6357 to qimode could truncate the indices.
6358 * optabs.c (expand_vec_perm_var): Likewise.
6360 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6362 * Makefile.in (OBJS): Add vec-perm-indices.o.
6363 * vec-perm-indices.h: New file.
6364 * vec-perm-indices.c: Likewise.
6365 * target.h (vec_perm_indices): Replace with a forward class
6367 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6368 * optabs.h: Include vec-perm-indices.h.
6369 (expand_vec_perm): Delete.
6370 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6371 (expand_vec_perm_const): Declare.
6372 * target.def (vec_perm_const_ok): Replace with...
6373 (vec_perm_const): ...this new hook.
6374 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6375 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6376 * doc/tm.texi: Regenerate.
6377 * optabs.def (vec_perm_const): Delete.
6378 * doc/md.texi (vec_perm_const): Likewise.
6379 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6380 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6381 expand_vec_perm for constant permutation vectors. Assert that
6382 the mode of variable permutation vectors is the integer equivalent
6383 of the mode that is being permuted.
6384 * optabs-query.h (selector_fits_mode_p): Declare.
6385 * optabs-query.c: Include vec-perm-indices.h.
6386 (selector_fits_mode_p): New function.
6387 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6388 is defined, instead of checking whether the vec_perm_const_optab
6389 exists. Use targetm.vectorize.vec_perm_const instead of
6390 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6391 fit in the vector mode before using a variable permute.
6392 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6393 vec_perm_indices instead of an rtx.
6394 (expand_vec_perm): Replace with...
6395 (expand_vec_perm_const): ...this new function. Take the selector
6396 as a vec_perm_indices rather than an rtx. Also take the mode of
6397 the selector. Update call to shift_amt_for_vec_perm_mask.
6398 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6399 Use vec_perm_indices::new_expanded_vector to expand the original
6400 selector into bytes. Check whether the indices fit in the vector
6401 mode before using a variable permute.
6402 (expand_vec_perm_var): Make global.
6403 (expand_mult_highpart): Use expand_vec_perm_const.
6404 * fold-const.c: Includes vec-perm-indices.h.
6405 * tree-ssa-forwprop.c: Likewise.
6406 * tree-vect-data-refs.c: Likewise.
6407 * tree-vect-generic.c: Likewise.
6408 * tree-vect-loop.c: Likewise.
6409 * tree-vect-slp.c: Likewise.
6410 * tree-vect-stmts.c: Likewise.
6411 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6413 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6414 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6415 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6416 (aarch64_vectorize_vec_perm_const): ...this new function.
6417 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6418 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6419 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6420 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6421 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6422 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6423 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6425 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6426 check for NEON modes.
6427 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6428 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6429 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6430 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6432 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6433 the old VEC_PERM_CONST conditions.
6434 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6435 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6436 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6437 (ia64_vectorize_vec_perm_const_ok): Merge into...
6438 (ia64_vectorize_vec_perm_const): ...this new function.
6439 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6440 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6441 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6442 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6443 * config/mips/mips.c (mips_expand_vec_perm_const)
6444 (mips_vectorize_vec_perm_const_ok): Merge into...
6445 (mips_vectorize_vec_perm_const): ...this new function.
6446 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6447 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6448 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6449 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6450 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6451 (rs6000_expand_vec_perm_const): Delete.
6452 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6454 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6455 (altivec_expand_vec_perm_const_le): Take each operand individually.
6456 Operate on constant selectors rather than rtxes.
6457 (altivec_expand_vec_perm_const): Likewise. Update call to
6458 altivec_expand_vec_perm_const_le.
6459 (rs6000_expand_vec_perm_const): Delete.
6460 (rs6000_vectorize_vec_perm_const_ok): Delete.
6461 (rs6000_vectorize_vec_perm_const): New function.
6462 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6463 an element count and rtx array.
6464 (rs6000_expand_extract_even): Update call accordingly.
6465 (rs6000_expand_interleave): Likewise.
6466 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6467 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6468 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6469 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6470 (rs6000_expand_vec_perm_const): Delete.
6471 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6472 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6473 (altivec_expand_vec_perm_const_le): Take each operand individually.
6474 Operate on constant selectors rather than rtxes.
6475 (altivec_expand_vec_perm_const): Likewise. Update call to
6476 altivec_expand_vec_perm_const_le.
6477 (rs6000_expand_vec_perm_const): Delete.
6478 (rs6000_vectorize_vec_perm_const_ok): Delete.
6479 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6480 reference to the SPE evmerge intructions.
6481 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6482 an element count and rtx array.
6483 (rs6000_expand_extract_even): Update call accordingly.
6484 (rs6000_expand_interleave): Likewise.
6485 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6486 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6488 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6490 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6492 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6493 vector mode and that that mode matches the mode of the data
6495 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6496 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6497 directly using expand_vec_perm_1 when forcing selectors into
6499 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6501 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6503 * optabs-query.h (can_vec_perm_p): Delete.
6504 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6505 * optabs-query.c (can_vec_perm_p): Split into...
6506 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6507 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6508 particular selector is valid.
6509 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6510 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6511 (vect_grouped_load_supported): Likewise.
6512 (vect_shift_permute_load_chain): Likewise.
6513 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6514 (vect_transform_slp_perm_load): Likewise.
6515 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6516 (vectorizable_bswap): Likewise.
6517 (vect_gen_perm_mask_checked): Likewise.
6518 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6519 implementations of variable permutation vectors into account
6520 when deciding which selector to use.
6521 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6522 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6523 with a false third argument.
6524 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6525 to test whether the constant selector is valid and can_vec_perm_var_p
6526 to test whether a variable selector is valid.
6528 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6530 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6531 * optabs-query.c (can_vec_perm_p): Likewise.
6532 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6533 instead of vec_perm_indices.
6534 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6535 (vect_gen_perm_mask_checked): Likewise,
6536 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6537 (vect_gen_perm_mask_checked): Likewise,
6539 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6541 * optabs-query.h (qimode_for_vec_perm): Declare.
6542 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6543 (qimode_for_vec_perm): ...this new function.
6544 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6546 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6548 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6549 does not have a conditional at the top.
6551 2018-01-02 Richard Biener <rguenther@suse.de>
6553 * ipa-inline.c (big_speedup_p): Fix expression.
6555 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6558 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6561 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6565 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6566 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6567 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6568 cond_taken_branch_cost 3->4.
6570 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6572 PR tree-optimization/83581
6573 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6574 TODO_cleanup_cfg if any changes have been made.
6577 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6578 convert_modes if target mode has the right side, but different mode
6582 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6583 last argument when extracting from CONCAT. If either from_real or
6584 from_imag is NULL, use expansion through memory. If result is not
6585 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6586 the parts directly to inner mode, if even that fails, use expansion
6590 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6591 check for bswap in mode rather than HImode and use that in expand_unop
6594 Copyright (C) 2018 Free Software Foundation, Inc.
6596 Copying and distribution of this file, with or without modification,
6597 are permitted in any medium without royalty provided the copyright
6598 notice and this notice are preserved.