1 2018-02-09 Jakub Jelinek <jakub@redhat.com>
4 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
5 constraint from =wa to wa. Avoid a subreg on the output operand,
6 instead use a pseudo and subreg it in a move.
7 (p9_xxbrd_<mode>): Changed to ...
8 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
9 (p9_xxbrd_v2df): New expander.
10 (p9_xxbrw_<mode>): Changed to ...
11 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
12 (p9_xxbrw_v4sf): New expander.
14 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
16 * config/rx.md: updated "movsicc" expand to be matched by GCC
17 * testsuite/gcc.target/rx/movsicc.c: new test case
19 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
22 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
23 multiply in 32-bit mode.
24 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
25 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
28 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
30 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
31 to allow or block "symbol_ref" depending on value of TARGET_JSR
32 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
33 call_value_internal insns
35 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
38 * dwarf2out.c (is_trivial_indirect_ref): New function.
39 (dwarf2out_late_global_decl): Do not generate a location
40 attribute for variables that have a non-trivial DECL_VALUE_EXPR
41 and that are not defined in the current unit.
43 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
45 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
46 instead of a libcall for UNORDERED.
48 2018-02-09 Tamar Christina <tamar.christina@arm.com>
51 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
52 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
54 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
57 * config/s390/s390.c (s390_set_current_function): Invoke
58 s390_indirect_branch_settings also if fndecl didn't change.
60 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
62 * config/rs6000/rs6000.md (blockage): Set length to zero.
64 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
66 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
68 2018-02-09 Jakub Jelinek <jakub@redhat.com>
71 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
72 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
76 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
77 PARALLEL incoming that failed vt_get_decl_and_offset check.
80 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
81 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
83 (get_variable_section): For decls in named .bss* sections pass true as
84 second argument to bss_initializer_p.
86 2018-02-09 Marek Polacek <polacek@redhat.com>
87 Jakub Jelinek <jakub@redhat.com>
90 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
91 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
92 Sync some changes from cxx_fold_indirect_ref.
94 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
96 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
98 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
99 (BLOCK_INLINE_ENTRY_LABEL): New.
100 (dwarf2out_var_location): Disregard inline entry markers.
101 (inline_entry_data): New struct.
102 (inline_entry_data_hasher): New hashtable type.
103 (inline_entry_data_hasher::hash): New.
104 (inline_entry_data_hasher::equal): New.
105 (inline_entry_data_table): New variable.
106 (add_high_low_attributes): Add DW_AT_entry_pc and
107 DW_AT_GNU_entry_view attributes if a pending entry is found
108 in inline_entry_data_table. Add old entry_pc attribute only
109 if debug nonbinding markers are disabled.
110 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
112 (block_within_block_p, dwarf2out_inline_entry): New.
113 (dwarf2out_finish): Check that no entries remained in
114 inline_entry_data_table.
115 * final.c (reemit_insn_block_notes): Handle inline entry notes.
116 (final_scan_insn, notice_source_line): Likewise.
117 (rest_of_clean_state): Skip inline entry markers.
118 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
120 * gimple.c (gimple_build_debug_inline_entry): New.
121 * gimple.h (enum gimple_debug_subcode): Add
122 GIMPLE_DEBUG_INLINE_ENTRY.
123 (gimple_build_debug_inline_entry): Declare.
124 (gimple_debug_inline_entry_p): New.
125 (gimple_debug_nonbind_marker_p): Adjust.
126 * insn-notes.def (INLINE_ENTRY): New.
127 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
128 inline entry marker notes.
129 (print_insn): Likewise.
130 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
131 (INSN_DEBUG_MARKER_KIND): Likewise.
132 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
133 * tree-inline.c (expand_call_inline): Build and insert
134 debug_inline_entry stmt.
135 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
136 inline entry blocks early, if nonbind markers are enabled.
137 (dump_scope_block): Dump fragment info.
138 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
139 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
140 (gimple_build_debug_inline_entry): New.
141 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
142 Enable/disable inline entry points too.
143 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
144 (DEBUG_INSN): Describe inline entry markers.
146 * common.opt (gvariable-location-views): New.
147 (gvariable-location-views=incompat5): New.
148 * config.in: Rebuilt.
149 * configure: Rebuilt.
150 * configure.ac: Test assembler for view support.
151 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
152 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
153 * dwarf2out.c (var_loc_view): New typedef.
154 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
155 (dwarf2out_locviews_in_attribute): New.
156 (dwarf2out_locviews_in_loclist): New.
157 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
158 (enum dw_line_info_opcode): Add LI_adv_address.
159 (struct dw_line_info_table): Add view.
160 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
161 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
162 (zero_view_p): New variable.
163 (ZERO_VIEW_P): New macro.
164 (output_asm_line_debug_info): New.
165 (struct var_loc_node): Add view.
166 (add_AT_view_list, AT_loc_list): New.
167 (add_var_loc_to_decl): Add view param. Test it against last.
168 (new_loc_list): Add view params. Record them.
169 (AT_loc_list_ptr): Handle loc and view lists.
170 (view_list_to_loc_list_val_node): New.
171 (print_dw_val): Handle dw_val_class_view_list.
172 (size_of_die): Likewise.
173 (value_format): Likewise.
174 (loc_list_has_views): New.
175 (gen_llsym): Set vl_symbol too.
176 (maybe_gen_llsym, skip_loc_list_entry): New.
177 (dwarf2out_maybe_output_loclist_view_pair): New.
178 (output_loc_list): Output view list or entries too.
179 (output_view_list_offset): New.
180 (output_die): Handle dw_val_class_view_list.
181 (output_dwarf_version): New.
182 (output_compilation_unit_header): Use it.
183 (output_skeleton_debug_sections): Likewise.
184 (output_rnglists, output_line_info): Likewise.
185 (output_pubnames, output_aranges): Update version comments.
186 (output_one_line_info_table): Output view numbers in asm comments.
187 (dw_loc_list): Determine current endview, pass it to new_loc_list.
188 Call maybe_gen_llsym.
189 (loc_list_from_tree_1): Adjust.
190 (add_AT_location_description): Create view list attribute if
191 needed, check it's absent otherwise.
192 (convert_cfa_to_fb_loc_list): Adjust.
193 (maybe_emit_file): Call output_asm_line_debug_info for test.
194 (dwarf2out_var_location): Reset views as needed. Precompute
195 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
197 (new_line_info_table): Reset next view.
198 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
199 (dwarf2out_source_line): Likewise. Output view resets and labels to
200 the assembler, or select appropriate line info opcodes.
201 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
202 (optimize_string_length): Catch it. Adjust.
203 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
204 dw_val_class_view_list, and remove it if no longer needed.
205 (hash_loc_list): Hash view numbers.
206 (loc_list_hasher::equal): Compare them.
207 (optimize_location_lists): Check whether a view list symbol is
208 needed, and whether the locview attribute is present, and
209 whether they match. Remove the locview attribute if no longer
211 (index_location_lists): Call skip_loc_list_entry for test.
212 (dwarf2out_finish): Call output_asm_line_debug_info for test.
213 Use output_dwarf_version.
214 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
215 (struct dw_val_node): Add val_view_list.
216 * final.c (SEEN_NEXT_VIEW): New.
217 (set_next_view_needed): New.
218 (clear_next_view_needed): New.
219 (maybe_output_next_view): New.
220 (final_start_function): Rename to...
221 (final_start_function_1): ... this. Take pointer to FIRST,
222 add SEEN parameter. Emit param bindings in the initial view.
223 (final_start_function): Reintroduce SEEN-less interface.
224 (final): Rename to...
225 (final_1): ... this. Take SEEN parameter. Output final pending
226 next view at the end.
227 (final): Reintroduce seen-less interface.
228 (final_scan_insn): Output pending next view before switching
229 sections or ending a block. Mark the next view as needed when
230 outputting variable locations. Notify debug backend of section
231 changes, and of location view changes.
232 (rest_of_handle_final): Adjust.
233 * toplev.c (process_options): Autodetect value for debug variable
234 location views option. Warn on incompat5 without -gdwarf-5.
235 * doc/invoke.texi (gvariable-location-views): New.
236 (gvariable-location-views=incompat5): New.
237 (gno-variable-location-views): New.
239 2018-02-08 David Malcolm <dmalcolm@redhat.com>
241 PR tree-optimization/84136
242 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
243 that the result of find_edge is non-NULL.
245 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
248 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
249 storing integer register in SImode. Fix cost of 256 and 512
250 byte aligned SSE register store.
252 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
254 * config/i386/i386.c (ix86_multiplication_cost): Fix
255 multiplication cost for TARGET_AVX512DQ.
257 2018-02-08 Marek Polacek <polacek@redhat.com>
259 PR tree-optimization/84238
260 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
263 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
265 PR tree-optimization/84265
266 * tree-vect-stmts.c (vectorizable_store): Don't treat
267 VMAT_CONTIGUOUS accesses as grouped.
268 (vectorizable_load): Likewise.
270 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
272 PR tree-optimization/81635
273 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
274 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
275 (test_round_for_mask): New functions.
276 (wide_int_cc_tests): Call test_round_for_mask.
277 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
278 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
279 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
280 range returned by get_range_info.
282 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
285 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
286 * symtab.c: Include builtins.h
287 (symtab_node::output_to_lto_symbol_table_p): Move here
288 from lto-streamer-out.c:output_symbol_p.
289 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
290 (output_symbol_p): Move all logic to symtab.c
291 (produce_symtab): Update.
293 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
295 * config/s390/s390-opts.h (enum indirect_branch): Define.
296 * config/s390/s390-protos.h (s390_return_addr_from_memory)
297 (s390_indirect_branch_via_thunk)
298 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
299 (enum s390_indirect_branch_type): Define.
300 * config/s390/s390.c (struct s390_frame_layout, struct
301 machine_function): Remove.
302 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
303 (indirect_branch_table_label_no, indirect_branch_table_name):
305 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
306 (enum s390_indirect_branch_option): Define.
307 (s390_return_addr_from_memory): New function.
308 (s390_handle_string_attribute): New function.
309 (s390_attribute_table): Add new attribute handler.
310 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
311 (s390_indirect_branch_via_thunk): New function.
312 (s390_indirect_branch_via_inline_thunk): New function.
313 (s390_function_ok_for_sibcall): When jumping via thunk disallow
314 sibling call optimization for non z10 compiles.
315 (s390_emit_call): Force indirect branch target to be a single
316 register. Add r1 clobber for non-z10 compiles.
317 (s390_emit_epilogue): Emit return jump via return_use expander.
318 (s390_reorg): Handle JUMP_INSNs as execute targets.
319 (s390_option_override_internal): Perform validity checks for the
320 new command line options.
321 (s390_indirect_branch_attrvalue): New function.
322 (s390_indirect_branch_settings): New function.
323 (s390_set_current_function): Invoke s390_indirect_branch_settings.
324 (s390_output_indirect_thunk_function): New function.
325 (s390_code_end): Implement target hook.
326 (s390_case_values_threshold): Implement target hook.
327 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
329 * config/s390/s390.h (struct s390_frame_layout)
330 (struct machine_function): Move here from s390.c.
331 (TARGET_INDIRECT_BRANCH_NOBP_RET)
332 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
333 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
334 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
335 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
336 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
337 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
338 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
339 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
340 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
341 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
342 (mnemonic attribute): Add values which aren't recognized
344 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
345 pattern for branch conversion. Fix mnemonic attribute.
346 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
347 indirect branch via thunk if requested.
348 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
349 ("*indirect_jump"): Disable for branch conversion using out of
351 ("indirect_jump_via_thunk<mode>_z10")
352 ("indirect_jump_via_thunk<mode>")
353 ("indirect_jump_via_inlinethunk<mode>_z10")
354 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
355 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
356 ("casesi_jump_via_inlinethunk<mode>_z10")
357 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
358 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
359 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
360 ("*indirect2_jump"): Disable for branch conversion.
361 ("casesi_jump"): Turn into expander and expand patterns for branch
363 ("return_use"): New expander.
364 ("*return"): Emit return via thunk and rename it to ...
365 ("*return<mode>"): ... this one.
366 * config/s390/s390.opt: Add new options and and enum for the
369 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
371 * lra-constraints.c (match_reload): Unconditionally use
372 gen_lowpart_SUBREG, rather than selecting between that
373 and equivalent gen_rtx_SUBREG code.
375 2018-02-08 Richard Biener <rguenther@suse.de>
377 PR tree-optimization/84233
378 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
379 changed flag instead of boguously re-using phi_inserted.
381 2018-02-08 Martin Jambor <mjambor@suse.cz>
383 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
384 static local variables.
386 2018-02-08 Richard Biener <rguenther@suse.de>
388 PR tree-optimization/84278
389 * tree-vect-stmts.c (vectorizable_store): When looking for
390 smaller vector types to perform grouped strided loads/stores
391 make sure the mode is supported by the target.
392 (vectorizable_load): Likewise.
394 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
396 * config/aarch64/aarch64.c (aarch64_components_for_bb):
397 Increase LDP/STP opportunities by adding adjacent callee-saves.
399 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
401 PR rtl-optimization/84068
402 PR rtl-optimization/83459
403 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
405 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
407 PR tree-optimization/84224
408 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
409 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
412 2018-02-07 Iain Sandoe <iain@codesourcery.com>
415 * config/rs6000/altivec.md (*restore_world): Remove LR use.
416 * config/rs6000/predicates.md (restore_world_operation): Adjust op
417 count, remove one USE.
419 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
421 * doc/install.texi (Configuration): Document the
422 --with-long-double-format={ibm,ieee} PowerPC configuration
426 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
427 Convert from define_expand to be define_insn_and_split. Rework
428 float/double/_Float128 conversions to QI/HI/SImode to work with
429 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
430 conversions to QI/HImode types did a store and then a load to
431 truncate the value. For conversions to VSX registers, don't split
432 the insn, instead emit the code directly. Use the code iterator
433 any_fix to combine signed and unsigned conversions.
434 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
435 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
436 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
437 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
438 (fix_<mode>di2_hw): Likewise.
439 (fixuns_<mode>di2_hw): Likewise.
440 (fix_<mode>si2_hw): Likewise.
441 (fixuns_<mode>si2_hw): Likewise.
442 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
443 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
444 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
445 fix<uns>_trunc<SFDF:mode>si2_p8.
446 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
448 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
449 (fix<uns>_<mode>_mem): Likewise.
450 (fctiw<u>z_<mode>_mem): Likewise.
451 (fix<uns>_<mode>_mem): Likewise.
452 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
453 the register allocator from doing a direct move to the GPRs to do
454 a store, and instead use the ISA 3.0 store byte/half-word from
455 vector register instruction. For IEEE 128-bit floating point,
456 also optimize stores of 32-bit ints.
457 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
459 2018-02-07 Alan Hayward <alan.hayward@arm.com>
461 * genextract.c (push_pathstr_operand): New function to support
463 (walk_rtx): Call push_pathstr_operand.
464 (print_path): Support [a-zA-Z].
466 2018-02-07 Richard Biener <rguenther@suse.de>
468 PR tree-optimization/84037
469 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
470 (cse_and_gimplify_to_preheader): Declare.
471 (vect_get_place_in_interleaving_chain): Likewise.
472 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
474 (_loop_vec_info::~_loop_vec_info): Delete it.
475 (cse_and_gimplify_to_preheader): New function.
476 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
477 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
478 (vectorizable_load): Likewise. For grouped stores always base
479 the IV on the first element.
480 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
481 condition before gimplifying.
483 2018-02-07 Jakub Jelinek <jakub@redhat.com>
485 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
486 *DIV_EXPR and *MOD_EXPR.
488 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
491 * config/i386/i386.c (ix86_option_override_internal): Mask out
492 the CF_SET bit when checking -fcf-protection.
494 2018-02-07 Tom de Vries <tom@codesourcery.com>
497 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
500 2018-02-07 Richard Biener <rguenther@suse.de>
502 PR tree-optimization/84204
503 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
506 PR tree-optimization/84205
507 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
508 special-case isl_ast_op_zdiv_r.
510 PR tree-optimization/84223
511 * graphite-scop-detection.c (gather_bbs::before_dom_children):
512 Only add conditions from within the region.
513 (gather_bbs::after_dom_children): Adjust.
515 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
518 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
519 * config/avr/avr.md: Only post-reload split REG-REG moves if
520 either register is REGERAL_REG_P.
522 2018-02-07 Jakub Jelinek <jakub@redhat.com>
524 PR tree-optimization/84235
525 * tree-ssa-scopedtables.c
526 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
527 if the subtraction is performed in floating point type where NaNs are
528 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
529 build 1. Formatting fix.
531 2018-02-06 Jakub Jelinek <jakub@redhat.com>
534 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
535 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
536 and skip it regardless of bb boundaries. Use CALL_P macro,
537 don't test INSN_P (insn) together with CALL_P or JUMP_P check
538 unnecessarily, formatting fix.
540 2018-02-06 Michael Collison <michael.collison@arm.com>
542 * config/arm/thumb2.md:
543 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
544 (*thumb_mov_notscc): Ditto.
546 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
549 * config/rs6000/rs6000.md (su code attribute): Use "u" for
550 unsigned_fix, not "s".
552 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
554 * configure.ac (gcc_fn_eh_frame_ro): New function.
555 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
556 correct .eh_frame permissions.
557 * configure: Regenerate.
559 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
561 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
564 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
566 * config/rs6000/rs6000.c (rs6000_option_override_internal):
567 Display warning message for -mno-speculate-indirect-jumps.
569 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
571 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
573 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
575 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
577 PR tree-optimization/84225
578 * tree-eh.c (find_trapping_overflow): Only call
579 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
581 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
584 * config/i386/i386.c: Reimplement the check of possible options
585 -mibt/-mshstk conbination. Change error messages.
586 * doc/invoke.texi: Fix a typo: remove extra '='.
588 2018-02-06 Marek Polacek <polacek@redhat.com>
590 PR tree-optimization/84228
591 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
593 2018-02-06 Tamar Christina <tamar.christina@arm.com>
596 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
597 emitted arch directives.
598 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
599 __ARM_FEATURE_COPROC before changing architectures.
601 2018-02-06 Richard Biener <rguenther@suse.de>
603 * config/i386/i386.c (print_reg): Fix typo.
604 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
606 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
608 * configure: Regenerate.
610 2018-02-05 Martin Sebor <msebor@redhat.com>
612 PR tree-optimization/83369
613 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
616 2018-02-05 Martin Liska <mliska@suse.cz>
618 * doc/invoke.texi: Cherry-pick upstream r323995.
620 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
622 * ira.c (ira_init_register_move_cost): Adjust comment.
624 2018-02-05 Martin Liska <mliska@suse.cz>
626 PR gcov-profile/84137
627 * doc/gcov.texi: Fix typo in documentation.
629 2018-02-05 Martin Liska <mliska@suse.cz>
631 PR gcov-profile/83879
632 * doc/gcov.texi: Document necessity of --dynamic-list-data when
633 using dlopen functionality.
635 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
637 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
638 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
639 _mm_maskz_range_ss, _mm_mask_range_round_ss,
640 _mm_maskz_range_round_ss): New intrinsics.
641 (__builtin_ia32_rangesd128_round)
642 (__builtin_ia32_rangess128_round): Remove.
643 (__builtin_ia32_rangesd128_mask_round,
644 __builtin_ia32_rangess128_mask_round): New builtins.
645 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
646 __builtin_ia32_rangess128_round): Remove.
647 (__builtin_ia32_rangesd128_mask_round,
648 __builtin_ia32_rangess128_mask_round): New builtins.
649 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
650 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
651 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
652 "<round_saeonly_constraint>")): Changed to ...
653 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
654 "<round_saeonly_scalar_constraint>")): ... this.
655 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
656 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
657 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
658 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
659 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
661 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
663 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
665 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
666 Remove all values except native, 8540 and 8548.
668 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
670 * config/i386/i386.c (ix86_output_function_return): Pass
671 INVALID_REGNUM, instead of -1, as invalid register number to
672 indirect_thunk_name and output_indirect_thunk.
674 2018-02-02 Julia Koval <julia.koval@intel.com>
676 * config.gcc: Add -march=icelake.
677 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
678 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
679 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
680 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
681 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
682 (processor_target_table): Add icelake.
683 (ix86_option_override_internal): Handle new PTAs.
684 (get_builtin_code_for_version): Handle icelake.
685 (M_INTEL_COREI7_ICELAKE): New.
686 (fold_builtin_cpu): Handle icelake.
687 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
688 * doc/invoke.texi: Add -march=icelake.
690 2018-02-02 Julia Koval <julia.koval@intel.com>
692 * config/i386/i386.c (ix86_option_override_internal): Change flags type
694 * wide-int-bitmask.h: New.
696 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
699 * config/i386/i386.md: Replace Pmode with word_mode in
700 builtin_setjmp_setup and builtin_longjmp to support x32.
702 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
706 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
708 (rs6000_supported_cpu_names): New static variable.
709 (linux_cpu_translation_table): Likewise.
710 (elf_platform) <cpu>: Define new static variable and use it.
711 Translate kernel AT_PLATFORM name to canonical name if needed.
712 Error if platform name is unknown.
714 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
717 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
719 2018-02-01 Jeff Law <law@redhat.com>
722 * config/i386/i386.c (release_scratch_register_on_entry): Add new
723 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
724 the scratch if RELEASE_VIA_POP is false.
725 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
726 If we have to save a temporary register, decrement SIZE appropriately.
727 Pass new arguments to release_scratch_register_on_entry.
728 (ix86_adjust_stack_and_probe): Likewise.
729 (ix86_emit_probe_stack_range): Pass new arguments to
730 release_scratch_register_on_entry.
732 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
734 PR rtl-optimization/84157
735 * combine.c (change_zero_ext): Use REG_P predicate in
736 front of HARD_REGISTER_P predicate.
738 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
740 * config/avr/avr.c (avr_option_override): Move disabling of
741 -fdelete-null-pointer-checks to...
742 * common/config/avr/avr-common.c (avr_option_optimization_table):
745 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
747 PR tree-optimization/81635
748 * tree-data-ref.c (split_constant_offset_1): For types that
749 wrap on overflow, try to use range info to prove that wrapping
752 2018-02-01 Renlin Li <renlin.li@arm.com>
755 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
757 (aarch64_register_move_cost): Likewise.
758 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
760 (REG_CLASS_NAMES): Likewise.
761 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
762 TAILCALL_ADDR_REGS. Remove IP registers.
763 * config/aarch64/aarch64.md (Ucs): Update register constraint.
765 2018-02-01 Richard Biener <rguenther@suse.de>
767 * domwalk.h (dom_walker::dom_walker): Add additional constructor
768 for specifying RPO order and allow NULL for that.
769 * domwalk.c (dom_walker::dom_walker): Likewise.
770 (dom_walker::walk): Handle NULL RPO order.
771 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
773 (rewrite_update_dom_walker): Likewise.
774 (mark_def_dom_walker): Likewise.
776 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
778 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
779 (aarch64_maybe_expand_sve_subreg_move): Declare.
780 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
781 * config/aarch64/predicates.md (aarch64_any_register_operand): New
783 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
784 that are semantically a reverse operation.
785 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
786 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
787 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
789 (aarch64_can_change_mode_class): For big-endian, forbid changes
790 between two SVE modes if they have different element sizes.
792 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
794 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
795 the TImode handling for big-endian targets.
797 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
799 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
800 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
802 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
803 Remove BSWAP handing for big-endian targets and use the form of
804 LD1RQ appropariate for the mode.
806 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
808 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
809 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
812 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
815 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
816 check for operands that need to go through aarch64_sve_reload_be.
818 2018-02-01 Jakub Jelinek <jakub@redhat.com>
820 PR tree-optimization/81661
821 PR tree-optimization/84117
822 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
823 * tree-eh.c: Include gimplify.h.
824 (find_trapping_overflow, replace_trapping_overflow,
825 rewrite_to_non_trapping_overflow): New functions.
826 * tree-vect-loop.c: Include tree-eh.h.
827 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
828 * tree-data-ref.c: Include tree-eh.h.
829 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
831 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
833 PR rtl-optimization/84123
834 * combine.c (change_zero_ext): Check if hard register satisfies
835 can_change_dest_mode before calling gen_lowpart_SUBREG.
837 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
840 * ira.c (ira_init_register_move_cost): Remove assert.
842 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
844 PR rtl-optimization/84071
845 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
846 * doc/tm.texi: Regenerate.
848 2018-01-31 Richard Biener <rguenther@suse.de>
850 PR tree-optimization/84132
851 * tree-data-ref.c (analyze_miv_subscript): Properly
852 check whether evolution_function_is_affine_multivariate_p
853 before calling gcd_of_steps_may_divide_p.
855 2018-01-31 Julia Koval <julia.koval@intel.com>
858 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
859 * config/i386/i386.md (rdpid_rex64) New.
860 (rdpid): Make 32bit only.
862 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
865 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
866 an IDENTIFIER_NODE for FUNCTION_TYPE's.
868 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
871 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
873 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
875 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
877 PR rtl-optimization/84071
878 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
879 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
881 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
883 * config/arc/arc.c (arc_handle_aux_attribute): New function.
884 (arc_attribute_table): Add 'aux' attribute.
885 (arc_in_small_data_p): Consider aux like variables.
886 (arc_is_aux_reg_p): New function.
887 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
888 (arc_get_aux_arg): New function.
889 (prepare_move_operands): Handle aux-register access.
890 (arc_handle_aux_attribute): New function.
891 * doc/extend.texi (ARC Variable attributes): Add subsection.
893 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
895 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
896 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
897 (arc_attribute_table): Add 'uncached' attribute.
898 (arc_print_operand): Print '.di' flag for uncached memory
900 (arc_in_small_data_p): Do not consider for small data the uncached
902 (arc_is_uncached_mem_p): New function.
903 * config/arc/predicates.md (compact_store_memory_operand): Check
904 for uncached memory accesses.
905 (nonvol_nonimm_operand): Likewise.
906 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
908 2018-01-31 Jakub Jelinek <jakub@redhat.com>
911 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
912 falign-loops=): Add Optimization flag.
914 2018-01-30 Jeff Law <law@redhat.com>
917 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
918 INT_REGISTERS_SAVED. Check it prior to calling
919 get_scratch_register_on_entry.
920 (ix86_adjust_stack_and_probe): Similarly.
921 (ix86_emit_probe_stack_range): Similarly.
922 (ix86_expand_prologue): Corresponding changes.
924 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
927 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
928 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
930 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
933 * lra-constraints.c (curr_insn_transform): Process AND in the
936 2018-01-30 Jakub Jelinek <jakub@redhat.com>
938 PR rtl-optimization/83986
939 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
940 dependence against last_pending_memory_flush in addition to
943 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
945 PR tree-optimization/81611
946 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
949 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
951 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
954 2018-01-30 Richard Biener <rguenther@suse.de>
955 Jakub Jelinek <jakub@redhat.com>
957 PR tree-optimization/84111
958 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
959 inner loops added during recursion, as they don't have up-to-date
962 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
965 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
966 (can_inline_edge_by_limits_p): ... here.
967 (can_early_inline_edge_p, check_callers,
968 update_caller_keys, update_callee_keys, recursive_inlining,
969 add_new_edges_to_heap, speculation_useful_p,
970 inline_small_functions,
971 inline_small_functions, flatten_function,
972 inline_to_all_callers_1): Update.
974 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
976 * profile-count.c (profile_count::combine_with_ipa_count): Handle
979 2018-01-30 Richard Biener <rguenther@suse.de>
981 PR tree-optimization/83008
982 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
983 invariant and constant vector uses in stmts when they need
986 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
989 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
990 * configure: Regenerate.
992 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
994 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
996 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
997 Use gen_rtx_REG rather than gen_lowpart.
999 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1001 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1002 rather than 0 when creating partial subregs.
1004 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1006 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1009 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1012 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1013 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1014 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1015 flags. This restores the settings used before the 2017-07-24.
1016 Turning off pre increment/decrement/modify allows IVOPTS to
1017 optimize DF/SF loops where the index is an int.
1019 2018-01-29 Richard Biener <rguenther@suse.de>
1020 Kelvin Nilsen <kelvin@gcc.gnu.org>
1023 * tree-vect-stmts.c (vectorizable_call): Don't call
1024 targetm.vectorize_builtin_md_vectorized_function if callee is
1027 2018-01-22 Carl Love <cel@us.ibm.com>
1029 * doc/extend.tex: Fix typo in second arg in
1030 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1032 2018-01-29 Richard Biener <rguenther@suse.de>
1034 PR tree-optimization/84086
1035 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1036 (flush_ssaname_freelist): When SSA names were released reset
1037 the SCEV hash table.
1039 2018-01-29 Richard Biener <rguenther@suse.de>
1041 PR tree-optimization/84057
1042 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1043 removed paths when removing edges.
1045 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1047 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1048 -mfunction-return=@var{choice}.
1050 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1053 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1055 (layout::print_source_line): Likewise.
1056 (test_get_line_width_without_trailing_whitespace): Add test cases.
1058 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1061 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1064 2018-01-26 Jim Wilson <jimw@sifive.com>
1066 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1068 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1071 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1073 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1074 and CMP + SUB-immediate -> SUBS.
1076 2018-01-26 Martin Sebor <msebor@redhat.com>
1078 PR tree-optimization/83896
1079 * tree-ssa-strlen.c (get_string_len): Rename...
1080 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1081 Avoid assuming length is constant.
1082 (handle_char_store): Use HOST_WIDE_INT for string length.
1084 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1087 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1088 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1090 2018-01-26 Richard Biener <rguenther@suse.de>
1092 PR rtl-optimization/84003
1093 * dse.c (record_store): Only record redundant stores when
1094 the earlier store aliases at least all accesses the later one does.
1096 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1098 PR rtl-optimization/83985
1099 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1100 REG_CFA_RESTORE insns.
1101 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1102 REG_CFA_RESTORE insns here.
1105 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1106 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1108 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1110 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1111 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1112 (arc_init): Likewise.
1113 (arc_override_options): Likewise.
1114 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1116 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1118 * config/arc/arc.h (TARGET_DBNZ): Define.
1119 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1120 properly set the tune attribute.
1121 (dbnz): Use TARGET_DBNZ guard.
1122 * config/arc/arc.opt (mtune): Add core3 option.
1124 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1126 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1127 recognize new pic like addresses.
1128 (arc_delegitimize_address): Clean up.
1130 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1132 * config/arc/arc-arches.def: Option mrf16 valid for all
1134 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1135 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1136 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1137 * config/arc/arc-tables.opt: Regenerate.
1138 * config/arc/arc.c (arc_conditional_register_usage): Handle
1139 reduced register file case.
1140 (arc_file_start): Set must have build attributes.
1141 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1143 * config/arc/arc.opt (mrf16): Add new option.
1144 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1145 * config/arc/genmultilib.awk: Handle new mrf16 option.
1146 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1147 * config/arc/t-multilib: Regenerate.
1148 * doc/invoke.texi (ARC Options): Document mrf16 option.
1150 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1152 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1153 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1154 (arc_attribute_table): Add 'secure_call' attribute.
1155 (arc_print_operand): Print secure call operand.
1156 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1158 (arc_is_secure_call_p): New function. * config/arc/arc.md
1159 (call_i): Add support for sjli instruction.
1160 (call_value_i): Likewise.
1161 * config/arc/constraints.md (Csc): New constraint.
1163 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1164 John Eric Martin <John.Martin@emmicro-us.com>
1166 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1167 * config/arc/arc.c (_arc_jli_section): New struct.
1168 (arc_jli_section): New type.
1169 (rc_jli_sections): New static variable.
1170 (arc_handle_jli_attribute): New function.
1171 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1172 (arc_file_end): New function.
1173 (TARGET_ASM_FILE_END): Define.
1174 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1175 (arc_add_jli_section): New function.
1176 (jli_call_scan): Likewise.
1177 (arc_reorg): Call jli_call_scan.
1178 (arc_output_addsi): Remove 'S' from printing asm operand.
1179 (arc_is_jli_call_p): New function.
1180 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1182 (movhi_insn): Likewise.
1183 (movsi_insn): Likewise.
1184 (movsi_set_cc_insn): Likewise.
1185 (loadqi_update): Likewise.
1186 (load_zeroextendqisi_update): Likewise.
1187 (load_signextendqisi_update): Likewise.
1188 (loadhi_update): Likewise.
1189 (load_zeroextendhisi_update): Likewise.
1190 (load_signextendhisi_update): Likewise.
1191 (loadsi_update): Likewise.
1192 (loadsf_update): Likewise.
1193 (movsicc_insn): Likewise.
1194 (bset_insn): Likewise.
1195 (bxor_insn): Likewise.
1196 (bclr_insn): Likewise.
1197 (bmsk_insn): Likewise.
1198 (bicsi3_insn): Likewise.
1199 (cmpsi_cc_c_insn): Likewise.
1200 (movsi_ne): Likewise.
1201 (movsi_cond_exec): Likewise.
1202 (clrsbsi2): Likewise.
1213 (call_i): Remove 'S' asm letter, add jli instruction.
1214 (call_value_i): Likewise.
1215 * config/arc/arc.op (mjli-always): New option.
1216 * config/arc/constraints.md (Cji): New constraint.
1217 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1219 (subsf3_fpx): Likewise.
1220 (mulsf3_fpx): Likewise.
1221 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1223 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1225 * doc/invoke.texi (ARC): Document mjli-always option.
1227 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1229 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1230 and use incw and decw where possible
1231 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1233 2018-01-26 Richard Biener <rguenther@suse.de>
1235 PR tree-optimization/81082
1236 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1237 association if it requires casting to unsigned.
1238 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1239 from fold_plusminus_mult_expr to catch important cases late when
1240 range info is available.
1242 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1244 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1245 * configure.ac (hidden_linkonce): New test.
1246 * configure: Regenerate.
1247 * config.in: Regenerate.
1249 2018-01-26 Julia Koval <julia.koval@intel.com>
1251 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1252 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1253 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1254 _mm_mask_bitshuffle_epi64_mask): Fix type.
1255 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1256 USI_FTYPE_V4DI_V4DI_USI): Remove.
1257 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1258 __builtin_ia32_vpshufbitqmb256_mask,
1259 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1260 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1261 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1263 2018-01-26 Alan Modra <amodra@gmail.com>
1266 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1267 UNSPEC_VBPERMQ. Sort other unspecs.
1269 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1271 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1273 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1276 * predict.c (drop_profile): Do not push/pop cfun; update also
1278 (handle_missing_profiles): Fix logic looking for zero profiles.
1280 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1283 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1284 on functions with #pragma omp declare simd or functions with simd
1286 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1287 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1288 Remove trailing \n from warning_at calls.
1290 2018-01-25 Tom de Vries <tom@codesourcery.com>
1293 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1294 for neutered workers.
1296 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1299 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1300 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1302 2018-01-24 Jeff Law <law@redhat.com>
1305 * i386.c (get_probe_interval): Move to earlier point.
1306 (ix86_compute_frame_layout): If -fstack-clash-protection and
1307 the frame is larger than the probe interval, then use pushes
1308 to save registers rather than reg->mem moves.
1309 (ix86_expand_prologue): Remove conditional for int_registers_saved
1312 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1315 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1316 min/max for never referenced object.
1318 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1321 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1323 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1324 attributes from DECL_ATTRIBUTES (decl) without affecting
1325 DECL_ATTRIBUTES (current_function_decl).
1326 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1327 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1329 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1331 PR tree-optimization/83979
1332 * fold-const.c (fold_comparison): Use constant_boolean_node
1333 instead of boolean_{true,false}_node.
1335 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1337 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1340 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1342 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1343 Simplify the clause that sets the length attribute.
1344 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1345 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1346 clause that sets the length attribute.
1347 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1349 2018-01-24 Tom de Vries <tom@codesourcery.com>
1352 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1353 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1354 Add strict parameter.
1355 (prevent_branch_around_nothing): Insert dummy insn between branch to
1356 label and label with no ptx insn inbetween.
1357 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1359 2018-01-24 Tom de Vries <tom@codesourcery.com>
1362 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1363 for neutered threads in warp.
1364 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1366 2018-01-24 Richard Biener <rguenther@suse.de>
1368 PR tree-optimization/83176
1369 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1372 2018-01-24 Richard Biener <rguenther@suse.de>
1374 PR tree-optimization/82819
1375 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1376 code generating pluses that are no-ops in the target precision.
1378 2018-01-24 Richard Biener <rguenther@suse.de>
1381 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1383 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1385 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1386 to merge probabilities.
1387 * predict.c (probably_never_executed): Also mark as cold functions
1388 with global 0 profile and guessed local profile.
1389 * profile-count.c (profile_probability::combine_with_count): New
1391 * profile-count.h (profile_probability::operator*,
1392 profile_probability::operator*=, profile_probability::operator/,
1393 profile_probability::operator/=): Reduce precision to adjusted
1394 and set value to guessed on contradictory divisions.
1395 (profile_probability::combine_with_freq): Remove.
1396 (profile_probability::combine_wiht_count): Declare.
1397 (profile_count::force_nonzero):: Set to adjusted.
1398 (profile_count::probability_in):: Set quality to adjusted.
1399 * tree-ssa-tail-merge.c (replace_block_by): Use
1402 2018-01-23 Andrew Waterman <andrew@sifive.com>
1403 Jim Wilson <jimw@sifive.com>
1405 * config/riscv/riscv.c (riscv_stack_boundary): New.
1406 (riscv_option_override): Set riscv_stack_boundary. Handle
1407 riscv_preferred_stack_boundary_arg.
1408 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1409 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1410 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1411 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1412 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1413 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1415 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1418 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1419 of struct ix86_frame.
1420 (ix86_expand_epilogue): Likewise. Add a local variable for
1421 the reg_save_offset field in struct ix86_frame.
1423 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1425 PR tree-optimization/82604
1426 * tree-loop-distribution.c (enum partition_kind): New enum item
1427 PKIND_PARTIAL_MEMSET.
1428 (partition_builtin_p): Support above new enum item.
1429 (generate_code_for_partition): Ditto.
1430 (compute_access_range): Differentiate cases that equality can be
1431 proven at all loops, the innermost loops or no loops.
1432 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1433 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1434 (finalize_partitions, distribute_loop): Don't fuse partition of
1435 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1436 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1439 2018-01-23 Martin Liska <mliska@suse.cz>
1441 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1442 order to ignore the predictor.
1443 (PRED_POLYMORPHIC_CALL): Likewise.
1444 (PRED_RECURSIVE_CALL): Likewise.
1446 2018-01-23 Martin Liska <mliska@suse.cz>
1448 * tree-profile.c (tree_profiling): Print function header to
1449 aware reader which function we are working on.
1450 * value-prof.c (gimple_find_values_to_profile): Do not print
1451 not interesting value histograms.
1453 2018-01-23 Martin Liska <mliska@suse.cz>
1455 * profile-count.h (enum profile_quality): Add
1456 profile_uninitialized as the first value. Do not number values
1457 as they are zero based.
1458 (profile_count::verify): Update sanity check.
1459 (profile_probability::verify): Likewise.
1461 2018-01-23 Nathan Sidwell <nathan@acm.org>
1463 * doc/invoke.texi (ffor-scope): Deprecate.
1465 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1467 PR tree-optimization/83510
1468 * domwalk.c (set_all_edges_as_executable): New function.
1469 (dom_walker::dom_walker): Convert bool param
1470 "skip_unreachable_blocks" to enum reachability. Move setup of
1471 edge flags to set_all_edges_as_executable and only do it when
1472 reachability is REACHABLE_BLOCKS.
1473 * domwalk.h (enum dom_walker::reachability): New enum.
1474 (dom_walker::dom_walker): Convert bool param
1475 "skip_unreachable_blocks" to enum reachability.
1476 (set_all_edges_as_executable): New decl.
1477 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1478 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1480 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1481 but converting true to REACHABLE_BLOCKS.
1482 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1484 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1485 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1486 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1488 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1489 if check_all_array_refs will be called.
1491 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1493 * tree.c (selftest::test_location_wrappers): Add more test
1496 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1498 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1499 (selftest::test_bit_in_range): Likewise.
1501 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1504 * doc/sourcebuild.texi (vect_float): Say that the selector
1505 only describes the situation when -funsafe-math-optimizations is on.
1506 (vect_float_strict): Document.
1508 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1510 PR tree-optimization/83965
1511 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1512 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1513 instead of checking only for a reduction.
1514 (vect_recog_widen_sum_pattern): Likewise.
1516 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1518 * predict.c (probably_never_executed): Only use precise profile info.
1519 (compute_function_frequency): Skip after inlining hack since we now
1520 have quality checking.
1522 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1524 * profile-count.h (profile_probability::very_unlikely,
1525 profile_probability::unlikely, profile_probability::even): Set
1526 precision to guessed.
1528 2018-01-23 Richard Biener <rguenther@suse.de>
1530 PR tree-optimization/83963
1531 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1532 Properly terminate dominator walk when crossing the exit edge not
1533 when visiting its source block.
1535 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1538 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1539 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1541 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1543 PR tree-optimization/83957
1544 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1545 semicolon after for body surrounded by braces.
1547 PR tree-optimization/83081
1548 * profile-count.h (profile_probability::split): New method.
1549 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1550 Use profile_probability::split.
1551 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1552 when splitting a single conditional jump into 2.
1554 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1556 PR tree-optimization/69452
1557 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1560 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1562 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1563 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1564 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1566 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1568 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1569 * config/rl78/rl78.md: New define_expand "movdi"
1570 * config/rl78/rl78.c: New function definition rl78_split_movdi
1572 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1575 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1577 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1578 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1579 128-bit to produce an UNSPEC move to get the double word with the
1580 signbit and then a shift directly to do signbit.
1581 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1582 implementation with a new version that just does either a direct
1583 move or a regular move. Move memory interface to separate insns.
1584 Move insns so they are next to the expander.
1585 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1586 with signbit move. Split big and little endian case.
1587 (signbit<mode>2_dm_mem_le): Likewise.
1588 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1589 (signbit<mode>2_dm2): Likewise.
1591 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1593 * config/rl78/rl78.md: New define_expand "anddi3".
1595 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1597 * config/rl78/rl78.md: New define_expand "umindi3".
1599 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1601 * config/rl78/rl78.md: New define_expand "smindi3".
1603 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1605 * config/rl78/rl78.md: New define_expand "smaxdi3".
1607 2018-01-22 Carl Love <cel@us.ibm.com>
1609 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1610 LVX_V1TI): Add macro expansion.
1611 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1612 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1613 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1614 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1615 Change check to determine if the instruction is a byte reversing
1616 entry. Fix typo in comment.
1617 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1618 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1619 Add def_builtin calls for new builtins.
1620 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1621 Add define_insn expansion.
1623 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1625 * config/rl78/rl78.md: New define_expand "umaxdi3".
1627 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1629 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1630 for non-QImode registers
1632 2018-01-22 Richard Biener <rguenther@suse.de>
1634 PR tree-optimization/83963
1635 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1636 including the loop exit block.
1637 (scop_detection::merge_sese): Likewise.
1638 (scop_detection::add_scop): Do it here instead.
1640 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1642 * doc/sourcebuild.texi (arm_softfloat): Document.
1644 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1647 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1648 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1649 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1651 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1652 David Edelsohn <dje.gcc@gmail.com>
1655 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1656 Change "crset eq" to "crset 2".
1657 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1658 (*call_indirect_aix<mode>_nospec): Likewise.
1659 (*call_value_indirect_aix<mode>_nospec): Likewise.
1660 (*call_indirect_elfv2<mode>_nospec): Likewise.
1661 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1662 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1663 change assembly output from . to $.
1664 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1665 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1666 (*tablejump<mode>_internal1_nospec): Likewise.
1668 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1671 * config/sh/sh_optimize_sett_clrt.cc:
1672 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1674 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1676 PR tree-optimization/83940
1677 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1678 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1679 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1680 use it to pass back the definition type.
1681 (vect_check_store_rhs): Likewise rhs_dt_out.
1682 (vect_build_gather_load_calls): Add a mask_dt argument and use
1683 it instead of a call to vect_is_simple_use.
1684 (vectorizable_store): Update calls to vect_check_load_store_mask
1685 and vect_check_store_rhs. Use the dt returned by the latter instead
1686 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1687 instead of calls to vect_is_simple_use. Pass the scalar rather
1688 than the vector operand to vect_is_simple_use when handling
1689 second and subsequent copies of an rhs value.
1690 (vectorizable_load): Update calls to vect_check_load_store_mask
1691 and vect_build_gather_load_calls. Use the cached mask_dt and
1692 gs_info.offset_dt instead of calls to vect_is_simple_use.
1694 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1697 * tree-emutls.c: Include gimplify.h.
1698 (lower_emutls_2): New function.
1699 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1700 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1701 it before further processing.
1704 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1705 UINTVAL (trueop1) instead of INTVAL (op1).
1707 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1711 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1712 INCOMING_FRAME_SP_OFFSET if not defined.
1713 (scan_trace): Add ENTRY argument. If true and
1714 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1715 emit a note to adjust the CFA offset.
1716 (create_cfi_notes): Adjust scan_trace callers.
1717 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1718 INCOMING_FRAME_SP_OFFSET in the CIE.
1719 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1720 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1722 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1723 * doc/tm.texi: Regenerated.
1725 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1727 PR rtl-optimization/83147
1728 * lra-constraints.c (remove_inheritance_pseudos): Use
1729 lra_substitute_pseudo_within_insn.
1731 2018-01-19 Tom de Vries <tom@codesourcery.com>
1732 Cesar Philippidis <cesar@codesourcery.com>
1735 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1737 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1740 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1741 spaces for function labels.
1743 2018-01-19 Martin Liska <mliska@suse.cz>
1745 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1746 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1747 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1748 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1749 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1750 (PRED_CONST_RETURN): Change from 69 to 65.
1751 (PRED_NULL_RETURN): Change from 91 to 71.
1752 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1753 (PRED_LOOP_GUARD): Change from 66 to 73.
1755 2018-01-19 Martin Liska <mliska@suse.cz>
1757 * predict.c (predict_insn_def): Add new assert.
1758 (struct branch_predictor): Change type to signed integer.
1759 (test_prediction_value_range): Amend test to cover
1761 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1762 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1763 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1764 (PRED_LOOP_IV_COMPARE): Likewise.
1765 * predict.h (PROB_UNINITIALIZED): Define new constant.
1767 2018-01-19 Martin Liska <mliska@suse.cz>
1769 * predict.c (dump_prediction): Add new format for
1770 analyze_brprob.py script which is enabled with -details
1772 * profile-count.h (precise_p): New function.
1774 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1776 PR tree-optimization/83922
1777 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1778 there are no statements that need masking.
1779 (vect_active_double_reduction_p): New function.
1780 (vect_analyze_loop_operations): Use it when handling phis that
1781 are not in the loop header.
1783 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1785 PR tree-optimization/83914
1786 * tree-vect-loop.c (vectorizable_induction): Don't convert
1787 init_expr or apply the peeling adjustment for inductions
1788 that are nested within the vectorized loop.
1790 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1792 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1795 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1799 * function.h (gimplify_parameters): Add gimple_seq * argument.
1800 * function.c: Include gimple.h and options.h.
1801 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1802 for the added local temporaries if needed.
1803 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1804 if there are any parameter cleanups, wrap whole body into a
1805 try/finally with the cleanups.
1807 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1810 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1811 Use GET_MODE_CLASS for scalar floating point.
1813 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1817 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1818 Fix call of call_cgraph_insertion_hooks.
1820 2018-01-18 Martin Sebor <msebor@redhat.com>
1822 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1824 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1827 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1830 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1833 * common.opt: (-ffile-prefix-map): New option.
1834 * opts.c (common_handle_option): Defer it.
1835 * opts-global.c (handle_common_deferred_options): Handle it.
1836 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1837 * file-prefix-map.h: New file.
1838 (remap_debug_filename, add_debug_prefix_map): ...here.
1839 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1840 * final.c (debug_prefix_map, add_debug_prefix_map
1841 remap_debug_filename): Move to...
1842 * file-prefix-map.c: New file.
1843 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1844 generalize, get rid of alloca(), use strrchr() instead of strchr().
1845 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1846 Implement in terms of add_prefix_map().
1847 (remap_macro_filename, remap_debug_filename): Implement in term of
1849 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1850 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1851 * dbxout.c: Include file-prefix-map.h.
1852 * varasm.c: Likewise.
1853 * vmsdbgout.c: Likewise.
1854 * xcoffout.c: Likewise.
1855 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1856 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1857 * doc/invoke.texi (-ffile-prefix-map): Document.
1858 (-fdebug-prefix-map): Update description.
1860 2018-01-18 Martin Liska <mliska@suse.cz>
1862 * config/i386/i386.c (indirect_thunk_name): Document that also
1864 (output_indirect_thunk): Document why both instructions
1865 (pause and lfence) are generated.
1867 2018-01-18 Richard Biener <rguenther@suse.de>
1869 PR tree-optimization/83887
1870 * graphite-scop-detection.c
1871 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1872 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1873 (scop_detection::merge_sese): Re-implement with a flood-fill
1874 algorithm that properly finds a SESE region if it exists.
1876 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1879 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1880 pointer_diff optimizations use view_convert instead of convert.
1882 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1884 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1885 Generate different code for -mno-speculate-indirect-jumps.
1886 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1887 (*call_indirect_aix<mode>): Disable for
1888 -mno-speculate-indirect-jumps.
1889 (*call_indirect_aix<mode>_nospec): New define_insn.
1890 (*call_value_indirect_aix<mode>): Disable for
1891 -mno-speculate-indirect-jumps.
1892 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1893 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1894 -mno-speculate-indirect-jumps.
1895 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1897 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1899 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1900 long double type, set the flags for noting the default long double
1901 type, even if we don't pass or return a long double type.
1903 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1906 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1909 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1911 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1912 support for merge[hl].
1913 (fold_mergehl_helper): New helper function.
1914 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1915 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1916 (altivec_vmrglw_direct): Add xxmrglw insn.
1918 2018-01-17 Andrew Waterman <andrew@sifive.com>
1920 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1921 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1923 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1926 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1927 call the lto_location_cache before reading the
1928 DECL_SOURCE_LOCATION of the types.
1930 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1931 Richard Sandiford <richard.sandiford@linaro.org>
1933 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1934 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1935 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1936 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1937 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1939 * config/aarch64/constraints.md (aarch64_movti_operand):
1941 * config/aarch64/predicates.md (Uti): Add new constraint.
1943 2018-01-17 Carl Love <cel@us.ibm.com>
1944 * config/rs6000/vsx.md (define_expand xl_len_r,
1945 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1946 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1948 (define_expand, define_insn): Move the shift left from the
1949 define_insn to the define_expand for lxvl and stxvl instructions.
1950 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1951 and XL_LEN_R definitions to PURE.
1953 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1955 * config/i386/i386.c (indirect_thunk_name): Declare regno
1956 as unsigned int. Compare regno with INVALID_REGNUM.
1957 (output_indirect_thunk): Ditto.
1958 (output_indirect_thunk_function): Ditto.
1959 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1960 in the call to output_indirect_thunk_function.
1962 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1965 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1966 rather than the size of inner_type to determine the stack slot size
1967 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1969 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1972 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1975 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1977 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1978 endian Linux systems to optionally enable multilibs for selecting
1979 the long double type if the user configured an explicit type.
1980 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1981 have no long double multilibs if not defined.
1982 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1983 warn if the user used -mabi={ieee,ibm}longdouble and we built
1984 multilibs for long double.
1985 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1986 appropriate multilib option.
1987 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1989 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1990 for building long double multilibs.
1991 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1993 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1995 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1998 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2000 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2003 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2006 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2009 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2011 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2012 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2014 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2016 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2017 different rtl trees depending on TARGET_64BIT.
2018 (rs6000_gen_lvx): Likewise.
2020 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2022 * config/visium/visium.md (nop): Tweak comment.
2023 (hazard_nop): Likewise.
2025 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2027 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2028 -mspeculate-indirect-jumps.
2029 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2030 for -mno-speculate-indirect-jumps.
2031 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2032 (*call_value_indirect_elfv2<mode>): Disable for
2033 -mno-speculate-indirect-jumps.
2034 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2035 (indirect_jump): Emit different RTL for
2036 -mno-speculate-indirect-jumps.
2037 (*indirect_jump<mode>): Disable for
2038 -mno-speculate-indirect-jumps.
2039 (*indirect_jump<mode>_nospec): New define_insn.
2040 (tablejump): Emit different RTL for
2041 -mno-speculate-indirect-jumps.
2042 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2043 (tablejumpsi_nospec): New define_expand.
2044 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2045 (tablejumpdi_nospec): New define_expand.
2046 (*tablejump<mode>_internal1): Disable for
2047 -mno-speculate-indirect-jumps.
2048 (*tablejump<mode>_internal1_nospec): New define_insn.
2049 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2052 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2054 * caller-save.c (insert_save): Drop unnecessary parameter. All
2057 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2058 Richard Biener <rguenth@suse.de>
2061 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2062 return early, inline manually is_gimple_sizepos. Make sure if we
2063 call gimplify_expr we don't end up with a gimple constant.
2064 * tree.c (variably_modified_type_p): Don't return true for
2065 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2066 * gimplify.h (is_gimple_sizepos): Remove.
2068 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2070 PR tree-optimization/83857
2071 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2072 vectorizable_live_operation for pure SLP statements.
2073 (vectorizable_live_operation): Handle PHIs.
2075 2018-01-16 Richard Biener <rguenther@suse.de>
2077 PR tree-optimization/83867
2078 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2079 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2081 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2084 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2085 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2086 If off is not INTEGER_CST, issue a may not be aligned warning
2087 rather than isn't aligned. Use isn%'t rather than isn't.
2088 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2090 <case MULT_EXPR>: Improve the case when bottom and one of the
2091 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2092 operand, in that case check if the other operand is multiple of
2093 bottom divided by the INTEGER_CST operand.
2095 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2098 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2099 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2100 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2101 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2102 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2103 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2104 (pa_function_arg_size): New function.
2106 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2108 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2109 in a separate statement.
2111 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2113 PR tree-optimization/83847
2114 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2115 group gathers and scatters.
2117 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2119 PR rtl-optimization/86620
2120 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2122 PR rtl-optimization/83213
2123 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2124 to last if both are JUMP_INSNs.
2126 PR tree-optimization/83843
2127 * gimple-ssa-store-merging.c
2128 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2129 store_immediate_info for bswap/nop orig_stores.
2131 2018-01-15 Andrew Waterman <andrew@sifive.com>
2133 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2135 <UDIV>: Increase cost if !TARGET_DIV.
2137 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2139 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2140 (define_attr "cr_logical_3op"): New.
2141 (cceq_ior_compare): Adjust.
2142 (cceq_ior_compare_complement): Adjust.
2143 (*cceq_rev_compare): Adjust.
2144 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2145 (is_cracked_insn): Adjust.
2146 (insn_must_be_first_in_group): Adjust.
2147 * config/rs6000/40x.md: Adjust.
2148 * config/rs6000/440.md: Adjust.
2149 * config/rs6000/476.md: Adjust.
2150 * config/rs6000/601.md: Adjust.
2151 * config/rs6000/603.md: Adjust.
2152 * config/rs6000/6xx.md: Adjust.
2153 * config/rs6000/7450.md: Adjust.
2154 * config/rs6000/7xx.md: Adjust.
2155 * config/rs6000/8540.md: Adjust.
2156 * config/rs6000/cell.md: Adjust.
2157 * config/rs6000/e300c2c3.md: Adjust.
2158 * config/rs6000/e500mc.md: Adjust.
2159 * config/rs6000/e500mc64.md: Adjust.
2160 * config/rs6000/e5500.md: Adjust.
2161 * config/rs6000/e6500.md: Adjust.
2162 * config/rs6000/mpc.md: Adjust.
2163 * config/rs6000/power4.md: Adjust.
2164 * config/rs6000/power5.md: Adjust.
2165 * config/rs6000/power6.md: Adjust.
2166 * config/rs6000/power7.md: Adjust.
2167 * config/rs6000/power8.md: Adjust.
2168 * config/rs6000/power9.md: Adjust.
2169 * config/rs6000/rs64.md: Adjust.
2170 * config/rs6000/titan.md: Adjust.
2172 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2174 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2175 ix86_indirect_branch_register logic.
2177 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2179 * config/i386/constraints.md (Bs): Update
2180 ix86_indirect_branch_register check. Don't check
2181 ix86_indirect_branch_register with GOT_memory_operand.
2183 * config/i386/predicates.md (GOT_memory_operand): Don't check
2184 ix86_indirect_branch_register here.
2185 (GOT32_symbol_operand): Likewise.
2187 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2189 * config/i386/predicates.md (constant_call_address_operand):
2190 Rewrite ix86_indirect_branch_register logic.
2191 (sibcall_insn_operand): Likewise.
2193 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2195 * config/i386/constraints.md (Bs): Replace
2196 ix86_indirect_branch_thunk_register with
2197 ix86_indirect_branch_register.
2199 * config/i386/i386.md (indirect_jump): Likewise.
2200 (tablejump): Likewise.
2201 (*sibcall_memory): Likewise.
2202 (*sibcall_value_memory): Likewise.
2203 Peepholes of indirect call and jump via memory: Likewise.
2204 * config/i386/i386.opt: Likewise.
2205 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2206 (GOT_memory_operand): Likewise.
2207 (call_insn_operand): Likewise.
2208 (sibcall_insn_operand): Likewise.
2209 (GOT32_symbol_operand): Likewise.
2211 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2214 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2215 type rather than type addr's type points to.
2216 (expand_omp_atomic_mutex): Likewise.
2217 (expand_omp_atomic): Likewise.
2219 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2222 * config/i386/i386.c (output_indirect_thunk_function): Use
2223 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2224 for __x86_return_thunk.
2226 2018-01-15 Richard Biener <rguenther@suse.de>
2229 * expmed.c (extract_bit_field_1): Fix typo.
2231 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2234 * config/arm/iterators.md (VF): New mode iterator.
2235 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2236 Remove integer-related logic from pattern.
2237 (neon_vabd<mode>_3): Likewise.
2239 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2242 * common.opt (fstrict-overflow): No longer an alias.
2243 (fwrapv-pointer): New option.
2244 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2245 also for pointer types based on flag_wrapv_pointer.
2246 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2247 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2248 opts->x_flag_wrapv got set.
2249 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2250 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2251 POINTER_TYPE_OVERFLOW_UNDEFINED.
2252 * match.pd: Likewise in address comparison pattern.
2253 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2255 2018-01-15 Richard Biener <rguenther@suse.de>
2258 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2259 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2260 Reset type names to their identifier if their TYPE_DECL doesn't
2261 have linkage (and thus is used for ODR and devirt).
2262 (save_debug_info_for_decl): Remove.
2263 (save_debug_info_for_type): Likewise.
2264 (add_tree_to_fld_list): Adjust.
2265 * tree-pretty-print.c (dump_generic_node): Make dumping of
2266 type names more robust.
2268 2018-01-15 Richard Biener <rguenther@suse.de>
2270 * BASE-VER: Bump to 8.0.1.
2272 2018-01-14 Martin Sebor <msebor@redhat.com>
2275 * builtins.c (check_access): Avoid warning when the no-warning bit
2278 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2280 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2281 * ira-color (allocno_hard_regs_compare): Likewise.
2283 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2286 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2287 Use .pushsection/.popsection.
2289 2018-01-14 Martin Sebor <msebor@redhat.com>
2292 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2294 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2296 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2297 entry from extra_headers.
2298 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2299 extra_headers, make the list bitwise identical to the i?86-*-* one.
2301 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2303 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2304 -mcmodel=large with -mindirect-branch=thunk,
2305 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2306 -mfunction-return=thunk-extern.
2307 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2308 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2309 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2311 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2313 * config/i386/i386.c (print_reg): Print the name of the full
2314 integer register without '%'.
2315 (ix86_print_operand): Handle 'V'.
2316 * doc/extend.texi: Document 'V' modifier.
2318 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2320 * config/i386/constraints.md (Bs): Disallow memory operand for
2321 -mindirect-branch-register.
2323 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2324 (GOT_memory_operand): Likewise.
2325 (call_insn_operand): Likewise.
2326 (sibcall_insn_operand): Likewise.
2327 (GOT32_symbol_operand): Likewise.
2328 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2329 for -mindirect-branch-register.
2330 (tablejump): Likewise.
2331 (*sibcall_memory): Likewise.
2332 (*sibcall_value_memory): Likewise.
2333 Disallow peepholes of indirect call and jump via memory for
2334 -mindirect-branch-register.
2335 (*call_pop): Replace m with Bw.
2336 (*call_value_pop): Likewise.
2337 (*sibcall_pop_memory): Replace m with Bs.
2338 * config/i386/i386.opt (mindirect-branch-register): New option.
2339 * doc/invoke.texi: Document -mindirect-branch-register option.
2341 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2343 * config/i386/i386-protos.h (ix86_output_function_return): New.
2344 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2345 set function_return_type.
2346 (indirect_thunk_name): Add ret_p to indicate thunk for function
2348 (output_indirect_thunk_function): Pass false to
2349 indirect_thunk_name.
2350 (ix86_output_indirect_branch_via_reg): Likewise.
2351 (ix86_output_indirect_branch_via_push): Likewise.
2352 (output_indirect_thunk_function): Create alias for function
2353 return thunk if regno < 0.
2354 (ix86_output_function_return): New function.
2355 (ix86_handle_fndecl_attribute): Handle function_return.
2356 (ix86_attribute_table): Add function_return.
2357 * config/i386/i386.h (machine_function): Add
2358 function_return_type.
2359 * config/i386/i386.md (simple_return_internal): Use
2360 ix86_output_function_return.
2361 (simple_return_internal_long): Likewise.
2362 * config/i386/i386.opt (mfunction-return=): New option.
2363 (indirect_branch): Mention -mfunction-return=.
2364 * doc/extend.texi: Document function_return function attribute.
2365 * doc/invoke.texi: Document -mfunction-return= option.
2367 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2369 * config/i386/i386-opts.h (indirect_branch): New.
2370 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2371 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2372 with local indirect jump when converting indirect call and jump.
2373 (ix86_set_indirect_branch_type): New.
2374 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2375 (indirectlabelno): New.
2376 (indirect_thunk_needed): Likewise.
2377 (indirect_thunk_bnd_needed): Likewise.
2378 (indirect_thunks_used): Likewise.
2379 (indirect_thunks_bnd_used): Likewise.
2380 (INDIRECT_LABEL): Likewise.
2381 (indirect_thunk_name): Likewise.
2382 (output_indirect_thunk): Likewise.
2383 (output_indirect_thunk_function): Likewise.
2384 (ix86_output_indirect_branch_via_reg): Likewise.
2385 (ix86_output_indirect_branch_via_push): Likewise.
2386 (ix86_output_indirect_branch): Likewise.
2387 (ix86_output_indirect_jmp): Likewise.
2388 (ix86_code_end): Call output_indirect_thunk_function if needed.
2389 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2391 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2392 (ix86_attribute_table): Add indirect_branch.
2393 * config/i386/i386.h (machine_function): Add indirect_branch_type
2394 and has_local_indirect_jump.
2395 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2397 (tablejump): Likewise.
2398 (*indirect_jump): Use ix86_output_indirect_jmp.
2399 (*tablejump_1): Likewise.
2400 (simple_return_indirect_internal): Likewise.
2401 * config/i386/i386.opt (mindirect-branch=): New option.
2402 (indirect_branch): New.
2405 (thunk-inline): Likewise.
2406 (thunk-extern): Likewise.
2407 * doc/extend.texi: Document indirect_branch function attribute.
2408 * doc/invoke.texi: Document -mindirect-branch= option.
2410 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2413 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2415 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2417 * ipa-inline.c (want_inline_small_function_p): Return false if
2418 inlining has already failed with CIF_FINAL_ERROR.
2419 (update_caller_keys): Call want_inline_small_function_p before
2421 (update_callee_keys): Likewise.
2423 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2425 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2427 (rs6000_quadword_masked_address_p): Likewise.
2428 (quad_aligned_load_p): Likewise.
2429 (quad_aligned_store_p): Likewise.
2430 (const_load_sequence_p): Add comment to describe the outer-most loop.
2431 (mimic_memory_attributes_and_flags): New function.
2432 (rs6000_gen_stvx): Likewise.
2433 (replace_swapped_aligned_store): Likewise.
2434 (rs6000_gen_lvx): Likewise.
2435 (replace_swapped_aligned_load): Likewise.
2436 (replace_swapped_load_constant): Capitalize argument name in
2437 comment describing this function.
2438 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2439 and stores that access quad-word aligned addresses and replace
2440 with stvx or lvx instructions when appropriate.
2441 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2442 New function prototype.
2443 (rs6000_quadword_masked_address_p): Likewise.
2444 (rs6000_gen_lvx): Likewise.
2445 (rs6000_gen_stvx): Likewise.
2446 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2447 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2448 when memory address is aligned.
2449 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2450 this split to select lvx instruction when memory address is aligned.
2451 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2452 instruction when memory address is aligned.
2453 (*vsx_le_perm_load_v16qi): Likewise.
2454 (four unnamed splitters): Modify to select the stvx instruction
2455 when memory is aligned.
2457 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2459 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2460 which appears in the queue multiple times.
2462 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2463 Alan Hayward <alan.hayward@arm.com>
2464 David Sherwood <david.sherwood@arm.com>
2466 * tree-vectorizer.h (vec_lower_bound): New structure.
2467 (_loop_vec_info): Add check_nonzero and lower_bounds.
2468 (LOOP_VINFO_CHECK_NONZERO): New macro.
2469 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2470 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2471 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2472 fields. Make seg_len the distance travelled, not including the
2474 (dr_direction_indicator): Declare.
2475 (dr_zero_step_indicator): Likewise.
2476 (dr_known_forward_stride_p): Likewise.
2477 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2479 (runtime_alias_check_p): Allow runtime alias checks with
2481 (operator ==): Compare access_size and align.
2482 (prune_runtime_alias_test_list): Rework for new distinction between
2483 the access_size and seg_len.
2484 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2486 (get_segment_min_max): New function.
2487 (create_intersect_range_checks): Use it.
2488 (dr_step_indicator): New function.
2489 (dr_direction_indicator): Likewise.
2490 (dr_zero_step_indicator): Likewise.
2491 (dr_known_forward_stride_p): Likewise.
2492 * tree-loop-distribution.c (data_ref_segment_size): Return
2493 DR_STEP * (niters - 1).
2494 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2496 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2497 (vect_preserves_scalar_order_p): New function, split out from...
2498 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2499 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2500 (vect_vfa_access_size): New function.
2501 (vect_vfa_align): Likewise.
2502 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2503 (dump_lower_bound): New function.
2504 (vect_check_lower_bound): Likewise.
2505 (vect_small_gap_p): Likewise.
2506 (vectorizable_with_step_bound_p): Likewise.
2507 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2508 depencies if the vectorization factor is 1. Convert the checks
2509 for nonzero steps into checks on the bounds of DR_STEP. Try using
2510 a bunds check for variable steps if the minimum required step is
2511 relatively small. Update calls to the dr_with_seg_len
2512 constructor and to vect_compile_time_alias.
2513 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2515 (vect_loop_versioning): Call it.
2516 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2518 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2520 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2521 Alan Hayward <alan.hayward@arm.com>
2522 David Sherwood <david.sherwood@arm.com>
2524 * doc/sourcebuild.texi (vect_scatter_store): Document.
2525 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2527 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2529 * genopinit.c (main): Add supports_vec_scatter_store and
2530 supports_vec_scatter_store_cached to target_optabs.
2531 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2532 IFN_MASK_SCATTER_STORE.
2533 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2535 * internal-fn.h (internal_store_fn_p): Declare.
2536 (internal_fn_stored_value_index): Likewise.
2537 * internal-fn.c (scatter_store_direct): New macro.
2538 (expand_scatter_store_optab_fn): New function.
2539 (direct_scatter_store_optab_supported_p): New macro.
2540 (internal_store_fn_p): New function.
2541 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2542 IFN_MASK_SCATTER_STORE.
2543 (internal_fn_mask_index): Likewise.
2544 (internal_fn_stored_value_index): New function.
2545 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2547 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2548 * optabs-query.c (supports_vec_scatter_store_p): New function.
2549 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2550 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2551 true for scatter stores.
2552 (vect_gather_scatter_fn_p): Handle scatter stores too.
2553 (vect_check_gather_scatter): Consider using scatter stores if
2554 supports_vec_scatter_store_p.
2555 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2557 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2558 internal_fn_stored_value_index.
2559 (check_load_store_masking): Handle scatter stores too.
2560 (vect_get_store_rhs): Make public.
2561 (vectorizable_call): Use internal_store_fn_p.
2562 (vectorizable_store): Handle scatter store internal functions.
2563 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2564 when deciding whether the end of the group has been reached.
2565 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2566 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2567 (mask_scatter_store<mode>): New insns.
2569 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2570 Alan Hayward <alan.hayward@arm.com>
2571 David Sherwood <david.sherwood@arm.com>
2573 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2574 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2575 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2577 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2578 Use vect_truncate_gather_scatter_offset if we can't treat the
2579 operation as a normal gather load or scatter store.
2580 (get_group_load_store_type): Take the gather_scatter_info
2581 as argument. Try using a gather load or scatter store for
2582 single-element groups.
2583 (get_load_store_type): Update calls to get_group_load_store_type
2584 and vect_use_strided_gather_scatters_p.
2586 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2587 Alan Hayward <alan.hayward@arm.com>
2588 David Sherwood <david.sherwood@arm.com>
2590 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2591 optional tree argument.
2592 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2594 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2595 but continue to use the current value as a fallback.
2596 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2597 to compare the updates.
2598 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2599 (get_load_store_type): Use it when handling a strided access.
2600 (vect_get_strided_load_store_ops): New function.
2601 (vect_get_data_ptr_increment): Likewise.
2602 (vectorizable_load): Handle strided gather loads. Always pass
2603 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2605 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2606 Alan Hayward <alan.hayward@arm.com>
2607 David Sherwood <david.sherwood@arm.com>
2609 * doc/md.texi (gather_load@var{m}): Document.
2610 (mask_gather_load@var{m}): Likewise.
2611 * genopinit.c (main): Add supports_vec_gather_load and
2612 supports_vec_gather_load_cached to target_optabs.
2613 * optabs-tree.c (init_tree_optimization_optabs): Use
2614 ggc_cleared_alloc to allocate target_optabs.
2615 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2616 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2618 * internal-fn.h (internal_load_fn_p): Declare.
2619 (internal_gather_scatter_fn_p): Likewise.
2620 (internal_fn_mask_index): Likewise.
2621 (internal_gather_scatter_fn_supported_p): Likewise.
2622 * internal-fn.c (gather_load_direct): New macro.
2623 (expand_gather_load_optab_fn): New function.
2624 (direct_gather_load_optab_supported_p): New macro.
2625 (direct_internal_fn_optab): New function.
2626 (internal_load_fn_p): Likewise.
2627 (internal_gather_scatter_fn_p): Likewise.
2628 (internal_fn_mask_index): Likewise.
2629 (internal_gather_scatter_fn_supported_p): Likewise.
2630 * optabs-query.c (supports_at_least_one_mode_p): New function.
2631 (supports_vec_gather_load_p): Likewise.
2632 * optabs-query.h (supports_vec_gather_load_p): Declare.
2633 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2634 and memory_type field.
2635 (NUM_PATTERNS): Bump to 15.
2636 * tree-vect-data-refs.c: Include internal-fn.h.
2637 (vect_gather_scatter_fn_p): New function.
2638 (vect_describe_gather_scatter_call): Likewise.
2639 (vect_check_gather_scatter): Try using internal functions for
2640 gather loads. Recognize existing calls to a gather load function.
2641 (vect_analyze_data_refs): Consider using gather loads if
2642 supports_vec_gather_load_p.
2643 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2644 (vect_get_gather_scatter_offset_type): Likewise.
2645 (vect_convert_mask_for_vectype): Likewise.
2646 (vect_add_conversion_to_patterm): Likewise.
2647 (vect_try_gather_scatter_pattern): Likewise.
2648 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2649 (vect_vect_recog_func_ptrs): Add it.
2650 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2651 internal_fn_mask_index and internal_gather_scatter_fn_p.
2652 (check_load_store_masking): Take the gather_scatter_info as an
2653 argument and handle gather loads.
2654 (vect_get_gather_scatter_ops): New function.
2655 (vectorizable_call): Check internal_load_fn_p.
2656 (vectorizable_load): Likewise. Handle gather load internal
2658 (vectorizable_store): Update call to check_load_store_masking.
2659 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2660 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2661 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2662 (aarch64_gather_scale_operand_d): New predicates.
2663 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2664 (mask_gather_load<mode>): New insns.
2666 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2667 Alan Hayward <alan.hayward@arm.com>
2668 David Sherwood <david.sherwood@arm.com>
2670 * optabs.def (fold_left_plus_optab): New optab.
2671 * doc/md.texi (fold_left_plus_@var{m}): Document.
2672 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2673 * internal-fn.c (fold_left_direct): Define.
2674 (expand_fold_left_optab_fn): Likewise.
2675 (direct_fold_left_optab_supported_p): Likewise.
2676 * fold-const-call.c (fold_const_fold_left): New function.
2677 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2678 * tree-parloops.c (valid_reduction_p): New function.
2679 (gather_scalar_reductions): Use it.
2680 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2681 (vect_finish_replace_stmt): Declare.
2682 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2683 (needs_fold_left_reduction_p): New function, split out from...
2684 (vect_is_simple_reduction): ...here. Accept reductions that
2685 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2686 (vect_force_simple_reduction): Also store the reduction type in
2687 the assignment's STMT_VINFO_REDUC_TYPE.
2688 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2689 (merge_with_identity): New function.
2690 (vect_expand_fold_left): Likewise.
2691 (vectorize_fold_left_reduction): Likewise.
2692 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2693 scalar phi in place for it. Check for target support and reject
2694 cases that would reassociate the operation. Defer the transform
2695 phase to vectorize_fold_left_reduction.
2696 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2697 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2698 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2700 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2702 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2703 call to ifc_temp_var.
2705 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2706 Alan Hayward <alan.hayward@arm.com>
2707 David Sherwood <david.sherwood@arm.com>
2709 * target.def (legitimize_address_displacement): Take the original
2710 offset as a poly_int.
2711 * targhooks.h (default_legitimize_address_displacement): Update
2713 * targhooks.c (default_legitimize_address_displacement): Likewise.
2714 * doc/tm.texi: Regenerate.
2715 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2716 as an argument, moving assert of ad->disp == ad->disp_term to...
2717 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2718 Try calling targetm.legitimize_address_displacement before expanding
2719 the address rather than afterwards, and adjust for the new interface.
2720 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2721 Match the new hook interface. Handle SVE addresses.
2722 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2725 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2727 * Makefile.in (OBJS): Add early-remat.o.
2728 * target.def (select_early_remat_modes): New hook.
2729 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2730 * doc/tm.texi: Regenerate.
2731 * targhooks.h (default_select_early_remat_modes): Declare.
2732 * targhooks.c (default_select_early_remat_modes): New function.
2733 * timevar.def (TV_EARLY_REMAT): New timevar.
2734 * passes.def (pass_early_remat): New pass.
2735 * tree-pass.h (make_pass_early_remat): Declare.
2736 * early-remat.c: New file.
2737 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2739 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2741 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2742 Alan Hayward <alan.hayward@arm.com>
2743 David Sherwood <david.sherwood@arm.com>
2745 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2746 vfm1 with a bound_epilog parameter.
2747 (vect_do_peeling): Update calls accordingly, and move the prologue
2748 call earlier in the function. Treat the base bound_epilog as 0 for
2749 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2750 this base when peeling for gaps.
2751 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2752 with fully-masked loops.
2753 (vect_estimate_min_profitable_iters): Handle the single peeled
2754 iteration in that case.
2756 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2757 Alan Hayward <alan.hayward@arm.com>
2758 David Sherwood <david.sherwood@arm.com>
2760 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2761 single-element interleaving even if the size is not a power of 2.
2762 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2763 accesses for single-element interleaving if the group size is
2766 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2767 Alan Hayward <alan.hayward@arm.com>
2768 David Sherwood <david.sherwood@arm.com>
2770 * doc/md.texi (fold_extract_last_@var{m}): Document.
2771 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2772 * optabs.def (fold_extract_last_optab): New optab.
2773 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2774 * internal-fn.c (fold_extract_direct): New macro.
2775 (expand_fold_extract_optab_fn): Likewise.
2776 (direct_fold_extract_optab_supported_p): Likewise.
2777 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2778 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2779 EXTRACT_LAST_REDUCTION.
2780 (get_initial_def_for_reduction): Do not create an initial vector
2781 for EXTRACT_LAST_REDUCTION reductions.
2782 (vectorizable_reduction): Leave the scalar phi in place for
2783 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2784 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2785 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2786 transform phase to vectorizable_condition.
2787 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2789 (vect_finish_stmt_generation): ...here.
2790 (vect_finish_replace_stmt): New function.
2791 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2792 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2794 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2796 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2797 Alan Hayward <alan.hayward@arm.com>
2798 David Sherwood <david.sherwood@arm.com>
2800 * doc/md.texi (extract_last_@var{m}): Document.
2801 * optabs.def (extract_last_optab): New optab.
2802 * internal-fn.def (EXTRACT_LAST): New internal function.
2803 * internal-fn.c (cond_unary_direct): New macro.
2804 (expand_cond_unary_optab_fn): Likewise.
2805 (direct_cond_unary_optab_supported_p): Likewise.
2806 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2807 loops using EXTRACT_LAST.
2808 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2809 (extract_last_<mode>): ...this optab.
2810 (vec_extract<mode><Vel>): Update accordingly.
2812 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2813 Alan Hayward <alan.hayward@arm.com>
2814 David Sherwood <david.sherwood@arm.com>
2816 * target.def (empty_mask_is_expensive): New hook.
2817 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2818 * doc/tm.texi: Regenerate.
2819 * targhooks.h (default_empty_mask_is_expensive): Declare.
2820 * targhooks.c (default_empty_mask_is_expensive): New function.
2821 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2822 if the target says that empty masks are expensive.
2823 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2825 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2827 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2828 Alan Hayward <alan.hayward@arm.com>
2829 David Sherwood <david.sherwood@arm.com>
2831 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2832 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2833 (vect_use_loop_mask_for_alignment_p): New function.
2834 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2835 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2836 niters_skip argument. Make sure that the first niters_skip elements
2837 of the first iteration are inactive.
2838 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2839 Update call to vect_set_loop_masks_directly.
2840 (get_misalign_in_elems): New function, split out from...
2841 (vect_gen_prolog_loop_niters): ...here.
2842 (vect_update_init_of_dr): Take a code argument that specifies whether
2843 the adjustment should be added or subtracted.
2844 (vect_update_init_of_drs): Likewise.
2845 (vect_prepare_for_masked_peels): New function.
2846 (vect_do_peeling): Skip prologue peeling if we're using a mask
2847 instead. Update call to vect_update_inits_of_drs.
2848 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2850 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2851 alignment. Do not include the number of peeled iterations in
2852 the minimum threshold in that case.
2853 (vectorizable_induction): Adjust the start value down by
2854 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2855 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2856 Take the number of skipped iterations into account when calculating
2858 * tree-vect-stmts.c (vect_gen_while_not): New function.
2860 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2861 Alan Hayward <alan.hayward@arm.com>
2862 David Sherwood <david.sherwood@arm.com>
2864 * doc/sourcebuild.texi (vect_fully_masked): Document.
2865 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2867 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2869 (vect_analyze_loop_2): ...here. Don't check the vectorization
2870 factor against the number of loop iterations if the loop is
2873 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2874 Alan Hayward <alan.hayward@arm.com>
2875 David Sherwood <david.sherwood@arm.com>
2877 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2878 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2879 (dump_groups): Update accordingly.
2880 (iv_use::mem_type): New member variable.
2881 (address_p): New function.
2882 (record_use): Add a mem_type argument and initialize the new
2884 (record_group_use): Add a mem_type argument. Use address_p.
2885 Remove obsolete null checks of base_object. Update call to record_use.
2886 (find_interesting_uses_op): Update call to record_group_use.
2887 (find_interesting_uses_cond): Likewise.
2888 (find_interesting_uses_address): Likewise.
2889 (get_mem_type_for_internal_fn): New function.
2890 (find_address_like_use): Likewise.
2891 (find_interesting_uses_stmt): Try find_address_like_use before
2892 calling find_interesting_uses_op.
2893 (addr_offset_valid_p): Use the iv mem_type field as the type
2894 of the addressed memory.
2895 (add_autoinc_candidates): Likewise.
2896 (get_address_cost): Likewise.
2897 (split_small_address_groups_p): Use address_p.
2898 (split_address_groups): Likewise.
2899 (add_iv_candidate_for_use): Likewise.
2900 (autoinc_possible_for_pair): Likewise.
2901 (rewrite_groups): Likewise.
2902 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2903 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2904 (get_alias_ptr_type_for_ptr_address): New function.
2905 (rewrite_use_address): Rewrite address uses in calls that were
2906 identified by find_address_like_use.
2908 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2909 Alan Hayward <alan.hayward@arm.com>
2910 David Sherwood <david.sherwood@arm.com>
2912 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2914 * gimple-expr.h (is_gimple_addressable: Likewise.
2915 * gimple-expr.c (is_gimple_address): Likewise.
2916 * internal-fn.c (expand_call_mem_ref): New function.
2917 (expand_mask_load_optab_fn): Use it.
2918 (expand_mask_store_optab_fn): Likewise.
2920 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2921 Alan Hayward <alan.hayward@arm.com>
2922 David Sherwood <david.sherwood@arm.com>
2924 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2925 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2926 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2927 (cond_umax@var{mode}): Document.
2928 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2929 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2930 (cond_umin_optab, cond_umax_optab): New optabs.
2931 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2932 (COND_IOR, COND_XOR): New internal functions.
2933 * internal-fn.h (get_conditional_internal_fn): Declare.
2934 * internal-fn.c (cond_binary_direct): New macro.
2935 (expand_cond_binary_optab_fn): Likewise.
2936 (direct_cond_binary_optab_supported_p): Likewise.
2937 (get_conditional_internal_fn): New function.
2938 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2939 Cope with reduction statements that are vectorized as calls rather
2941 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2942 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2943 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2944 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2945 (UNSPEC_COND_EOR): New unspecs.
2946 (optab): Add mappings for them.
2947 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2948 (sve_int_op, sve_fp_op): New int attributes.
2950 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2951 Alan Hayward <alan.hayward@arm.com>
2952 David Sherwood <david.sherwood@arm.com>
2954 * optabs.def (while_ult_optab): New optab.
2955 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2956 * internal-fn.def (WHILE_ULT): New internal function.
2957 * internal-fn.h (direct_internal_fn_supported_p): New override
2958 that takes two types as argument.
2959 * internal-fn.c (while_direct): New macro.
2960 (expand_while_optab_fn): New function.
2961 (convert_optab_supported_p): Likewise.
2962 (direct_while_optab_supported_p): New macro.
2963 * wide-int.h (wi::udiv_ceil): New function.
2964 * tree-vectorizer.h (rgroup_masks): New structure.
2965 (vec_loop_masks): New typedef.
2966 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2968 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2969 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2970 (vect_max_vf): New function.
2971 (slpeel_make_loop_iterate_ntimes): Delete.
2972 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2973 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2974 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2975 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2976 internal-fn.h, stor-layout.h and optabs-query.h.
2977 (vect_set_loop_mask): New function.
2978 (add_preheader_seq): Likewise.
2979 (add_header_seq): Likewise.
2980 (interleave_supported_p): Likewise.
2981 (vect_maybe_permute_loop_masks): Likewise.
2982 (vect_set_loop_masks_directly): Likewise.
2983 (vect_set_loop_condition_masked): Likewise.
2984 (vect_set_loop_condition_unmasked): New function, split out from
2985 slpeel_make_loop_iterate_ntimes.
2986 (slpeel_make_loop_iterate_ntimes): Rename to..
2987 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2988 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2989 (vect_do_peeling): Update call accordingly.
2990 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2992 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2993 mask_compare_type, can_fully_mask_p and fully_masked_p.
2994 (release_vec_loop_masks): New function.
2995 (_loop_vec_info): Use it to free the loop masks.
2996 (can_produce_all_loop_masks_p): New function.
2997 (vect_get_max_nscalars_per_iter): Likewise.
2998 (vect_verify_full_masking): Likewise.
2999 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3000 retries, and free the mask rgroups before retrying. Check loop-wide
3001 reasons for disallowing fully-masked loops. Make the final decision
3002 about whether use a fully-masked loop or not.
3003 (vect_estimate_min_profitable_iters): Do not assume that peeling
3004 for the number of iterations will be needed for fully-masked loops.
3005 (vectorizable_reduction): Disable fully-masked loops.
3006 (vectorizable_live_operation): Likewise.
3007 (vect_halve_mask_nunits): New function.
3008 (vect_double_mask_nunits): Likewise.
3009 (vect_record_loop_mask): Likewise.
3010 (vect_get_loop_mask): Likewise.
3011 (vect_transform_loop): Handle the case in which the final loop
3012 iteration might handle a partial vector. Call vect_set_loop_condition
3013 instead of slpeel_make_loop_iterate_ntimes.
3014 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3015 (check_load_store_masking): New function.
3016 (prepare_load_store_mask): Likewise.
3017 (vectorizable_store): Handle fully-masked loops.
3018 (vectorizable_load): Likewise.
3019 (supportable_widening_operation): Use vect_halve_mask_nunits for
3021 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3022 (vect_gen_while): New function.
3023 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3024 (aarch64_uqdec<mode>): New insn.
3026 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3027 Alan Hayward <alan.hayward@arm.com>
3028 David Sherwood <david.sherwood@arm.com>
3030 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3031 (reduc_xor_scal_optab): New optabs.
3032 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3033 (reduc_xor_scal_@var{m}): Document.
3034 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3035 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3037 * fold-const-call.c (fold_const_call): Handle them.
3038 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3039 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3040 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3041 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3042 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3043 (UNSPEC_XORV): New unspecs.
3044 (optab): Add entries for them.
3045 (BITWISEV): New int iterator.
3046 (bit_reduc_op): New int attributes.
3048 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3049 Alan Hayward <alan.hayward@arm.com>
3050 David Sherwood <david.sherwood@arm.com>
3052 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3053 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3054 * optabs.def (vec_shl_insert_optab): New optab.
3055 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3056 (duplicate_and_interleave): Likewise.
3057 * tree-vect-loop.c: Include internal-fn.h.
3058 (neutral_op_for_slp_reduction): New function, split out from
3059 get_initial_defs_for_reduction.
3060 (get_initial_def_for_reduction): Handle option 2 for variable-length
3061 vectors by loading the neutral value into a vector and then shifting
3062 the initial value into element 0.
3063 (get_initial_defs_for_reduction): Replace the code argument with
3064 the neutral value calculated by neutral_op_for_slp_reduction.
3065 Use gimple_build_vector for constant-length vectors.
3066 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3067 but the first group_size elements have a neutral value.
3068 Use duplicate_and_interleave otherwise.
3069 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3070 Update call to get_initial_defs_for_reduction. Handle SLP
3071 reductions for variable-length vectors by creating one vector
3072 result for each scalar result, with the elements associated
3073 with other scalar results stubbed out with the neutral value.
3074 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3075 Require IFN_VEC_SHL_INSERT for double reductions on
3076 variable-length vectors, or SLP reductions that have
3077 a neutral value. Require can_duplicate_and_interleave_p
3078 support for variable-length unchained SLP reductions if there
3079 is no neutral value, such as for MIN/MAX reductions. Also require
3080 the number of vector elements to be a multiple of the number of
3081 SLP statements when doing variable-length unchained SLP reductions.
3082 Update call to vect_create_epilog_for_reduction.
3083 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3084 and remove initial values.
3085 (duplicate_and_interleave): Make public.
3086 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3087 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3089 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3090 Alan Hayward <alan.hayward@arm.com>
3091 David Sherwood <david.sherwood@arm.com>
3093 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3094 (can_duplicate_and_interleave_p): New function.
3095 (vect_get_and_check_slp_defs): Take the vector of statements
3096 rather than just the current one. Remove excess parentheses.
3097 Restriction rejectinon of vect_constant_def and vect_external_def
3098 for variable-length vectors to boolean types, or types for which
3099 can_duplicate_and_interleave_p is false.
3100 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3101 (duplicate_and_interleave): New function.
3102 (vect_get_constant_vectors): Use gimple_build_vector for
3103 constant-length vectors and suitable variable-length constant
3104 vectors. Use duplicate_and_interleave for other variable-length
3105 vectors. Don't defer the update when inserting new statements.
3107 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3108 Alan Hayward <alan.hayward@arm.com>
3109 David Sherwood <david.sherwood@arm.com>
3111 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3112 min_profitable_iters doesn't go negative.
3114 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3115 Alan Hayward <alan.hayward@arm.com>
3116 David Sherwood <david.sherwood@arm.com>
3118 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3119 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3120 * optabs.def (vec_mask_load_lanes_optab): New optab.
3121 (vec_mask_store_lanes_optab): Likewise.
3122 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3123 (MASK_STORE_LANES): Likewise.
3124 * internal-fn.c (mask_load_lanes_direct): New macro.
3125 (mask_store_lanes_direct): Likewise.
3126 (expand_mask_load_optab_fn): Handle masked operations.
3127 (expand_mask_load_lanes_optab_fn): New macro.
3128 (expand_mask_store_optab_fn): Handle masked operations.
3129 (expand_mask_store_lanes_optab_fn): New macro.
3130 (direct_mask_load_lanes_optab_supported_p): Likewise.
3131 (direct_mask_store_lanes_optab_supported_p): Likewise.
3132 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3134 (vect_load_lanes_supported): Likewise.
3135 * tree-vect-data-refs.c (strip_conversion): New function.
3136 (can_group_stmts_p): Likewise.
3137 (vect_analyze_data_ref_accesses): Use it instead of checking
3138 for a pair of assignments.
3139 (vect_store_lanes_supported): Take a masked_p parameter.
3140 (vect_load_lanes_supported): Likewise.
3141 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3142 vect_store_lanes_supported and vect_load_lanes_supported.
3143 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3144 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3145 parameter. Don't allow gaps for masked accesses.
3146 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3147 and vect_load_lanes_supported.
3148 (get_load_store_type): Take a masked_p parameter and update
3149 call to get_group_load_store_type.
3150 (vectorizable_store): Update call to get_load_store_type.
3151 Handle IFN_MASK_STORE_LANES.
3152 (vectorizable_load): Update call to get_load_store_type.
3153 Handle IFN_MASK_LOAD_LANES.
3155 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3156 Alan Hayward <alan.hayward@arm.com>
3157 David Sherwood <david.sherwood@arm.com>
3159 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3161 * config/aarch64/aarch64-protos.h
3162 (aarch64_sve_struct_memory_operand_p): Declare.
3163 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3164 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3165 (VPRED, vpred): Handle SVE structure modes.
3166 * config/aarch64/constraints.md (Utx): New constraint.
3167 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3168 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3169 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3170 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3171 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3172 structure modes. Split into pieces after RA.
3173 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3174 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3176 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3177 SVE structure modes.
3178 (aarch64_classify_address): Likewise.
3179 (sizetochar): Move earlier in file.
3180 (aarch64_print_operand): Handle SVE register lists.
3181 (aarch64_array_mode): New function.
3182 (aarch64_sve_struct_memory_operand_p): Likewise.
3183 (TARGET_ARRAY_MODE): Redefine.
3185 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3186 Alan Hayward <alan.hayward@arm.com>
3187 David Sherwood <david.sherwood@arm.com>
3189 * target.def (array_mode): New target hook.
3190 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3191 * doc/tm.texi: Regenerate.
3192 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3193 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3194 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3196 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3199 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3200 Alan Hayward <alan.hayward@arm.com>
3201 David Sherwood <david.sherwood@arm.com>
3203 * fold-const.c (fold_binary_loc): Check the argument types
3204 rather than the result type when testing for a vector operation.
3206 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3208 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3209 * doc/tm.texi: Regenerate.
3211 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3212 Alan Hayward <alan.hayward@arm.com>
3213 David Sherwood <david.sherwood@arm.com>
3215 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3216 (sve): Document new AArch64 extension.
3217 * doc/md.texi (w): Extend the description of the AArch64
3218 constraint to include SVE vectors.
3219 (Upl, Upa): Document new AArch64 predicate constraints.
3220 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3222 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3223 (msve-vector-bits=): New option.
3224 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3225 SVE when these are disabled.
3226 (sve): New extension.
3227 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3228 modes. Adjust their number of units based on aarch64_sve_vg.
3229 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3230 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3231 aarch64_addr_query_type.
3232 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3233 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3234 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3235 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3236 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3237 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3238 (aarch64_simd_imm_zero_p): Delete.
3239 (aarch64_check_zero_based_sve_index_immediate): Declare.
3240 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3241 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3242 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3243 (aarch64_sve_float_mul_immediate_p): Likewise.
3244 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3246 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3247 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3248 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3249 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3250 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3251 (aarch64_regmode_natural_size): Likewise.
3252 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3253 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3255 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3256 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3257 for VG and the SVE predicate registers.
3258 (V_ALIASES): Add a "z"-prefixed alias.
3259 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3260 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3261 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3262 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3263 (REG_CLASS_NAMES): Add entries for them.
3264 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3265 and the predicate registers.
3266 (aarch64_sve_vg): Declare.
3267 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3268 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3269 (REGMODE_NATURAL_SIZE): Define.
3270 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3272 * config/aarch64/aarch64.c: Include cfgrtl.h.
3273 (simd_immediate_info): Add a constructor for series vectors,
3274 and an associated step field.
3275 (aarch64_sve_vg): New variable.
3276 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3277 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3278 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3279 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3280 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3281 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3282 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3283 (aarch64_get_mask_mode): New functions.
3284 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3285 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3286 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3287 predicate modes and predicate registers. Explicitly restrict
3288 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3289 to store a vector mode if it is recognized by
3290 aarch64_classify_vector_mode.
3291 (aarch64_regmode_natural_size): New function.
3292 (aarch64_hard_regno_caller_save_mode): Return the original mode
3294 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3295 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3296 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3297 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3299 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3300 does not overlap dest if the function is frame-related. Handle
3302 (aarch64_split_add_offset): New function.
3303 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3304 them aarch64_add_offset.
3305 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3306 and update call to aarch64_sub_sp.
3307 (aarch64_add_cfa_expression): New function.
3308 (aarch64_expand_prologue): Pass extra temporary registers to the
3309 functions above. Handle the case in which we need to emit new
3310 DW_CFA_expressions for registers that were originally saved
3311 relative to the stack pointer, but now have to be expressed
3312 relative to the frame pointer.
3313 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3315 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3316 IP0 and IP1 values for SVE frames.
3317 (aarch64_expand_vec_series): New function.
3318 (aarch64_expand_sve_widened_duplicate): Likewise.
3319 (aarch64_expand_sve_const_vector): Likewise.
3320 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3321 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3322 into the register, rather than emitting a SET directly.
3323 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3324 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3325 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3326 (offset_9bit_signed_scaled_p): New functions.
3327 (aarch64_replicate_bitmask_imm): New function.
3328 (aarch64_bitmask_imm): Use it.
3329 (aarch64_cannot_force_const_mem): Reject expressions involving
3330 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3331 (aarch64_classify_index): Handle SVE indices, by requiring
3332 a plain register index with a scale that matches the element size.
3333 (aarch64_classify_address): Handle SVE addresses. Assert that
3334 the mode of the address is VOIDmode or an integer mode.
3335 Update call to aarch64_classify_symbol.
3336 (aarch64_classify_symbolic_expression): Update call to
3337 aarch64_classify_symbol.
3338 (aarch64_const_vec_all_in_range_p): New function.
3339 (aarch64_print_vector_float_operand): Likewise.
3340 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3341 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3342 and the FP immediates 1.0 and 0.5.
3343 (aarch64_print_address_internal): Handle SVE addresses.
3344 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3345 (aarch64_regno_regclass): Handle predicate registers.
3346 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3348 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3349 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3350 (aarch64_convert_sve_vector_bits): New function.
3351 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3352 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3354 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3355 Handle SVE vector and predicate modes. Accept VL-based constants
3356 that need only one temporary register, and VL offsets that require
3357 no temporary registers.
3358 (aarch64_conditional_register_usage): Mark the predicate registers
3359 as fixed if SVE isn't available.
3360 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3361 Return true for SVE vector and predicate modes.
3362 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3363 rather than an unsigned int. Handle SVE modes.
3364 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3366 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3368 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3369 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3370 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3371 (aarch64_sve_float_mul_immediate_p): New functions.
3372 (aarch64_sve_valid_immediate): New function.
3373 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3374 Explicitly reject structure modes. Check for INDEX constants.
3375 Handle PTRUE and PFALSE constants.
3376 (aarch64_check_zero_based_sve_index_immediate): New function.
3377 (aarch64_simd_imm_zero_p): Delete.
3378 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3379 vector modes. Accept constants in the range of CNT[BHWD].
3380 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3381 ask for an Advanced SIMD mode.
3382 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3383 (aarch64_simd_vector_alignment): Handle SVE predicates.
3384 (aarch64_vectorize_preferred_vector_alignment): New function.
3385 (aarch64_simd_vector_alignment_reachable): Use it instead of
3387 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3388 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3390 (MAX_VECT_LEN): Delete.
3391 (expand_vec_perm_d): Add a vec_flags field.
3392 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3393 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3394 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3396 (aarch64_evpc_rev): Rename to...
3397 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3398 (aarch64_evpc_rev_global): New function.
3399 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3400 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3402 (aarch64_evpc_sve_tbl): New function.
3403 (aarch64_expand_vec_perm_const_1): Update after rename of
3404 aarch64_evpc_rev. Handle SVE permutes too, trying
3405 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3406 than aarch64_evpc_tbl.
3407 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3408 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3409 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3410 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3411 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3412 (aarch64_expand_sve_vcond): New functions.
3413 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3414 of aarch64_vector_mode_p.
3415 (aarch64_dwarf_poly_indeterminate_value): New function.
3416 (aarch64_compute_pressure_classes): Likewise.
3417 (aarch64_can_change_mode_class): Likewise.
3418 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3419 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3420 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3421 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3422 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3423 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3424 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3425 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3427 (Dn, Dl, Dr): Accept const as well as const_vector.
3428 (Dz): Likewise. Compare against CONST0_RTX.
3429 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3430 of "vector" where appropriate.
3431 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3432 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3433 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3434 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3435 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3436 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3437 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3438 (v_int_equiv): Extend to SVE modes.
3439 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3441 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3442 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3443 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3444 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3445 (SVE_COND_FP_CMP): New int iterators.
3446 (perm_hilo): Handle the new unpack unspecs.
3447 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3449 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3450 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3451 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3452 (aarch64_equality_operator, aarch64_constant_vector_operand)
3453 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3454 (aarch64_sve_nonimmediate_operand): Likewise.
3455 (aarch64_sve_general_operand): Likewise.
3456 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3457 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3458 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3459 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3460 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3461 (aarch64_sve_float_arith_immediate): Likewise.
3462 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3463 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3464 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3465 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3466 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3467 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3468 (aarch64_sve_float_arith_operand): Likewise.
3469 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3470 (aarch64_sve_float_mul_operand): Likewise.
3471 (aarch64_sve_vec_perm_operand): Likewise.
3472 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3473 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3474 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3475 as well as const_vector.
3476 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3477 in file. Use CONST0_RTX and CONSTM1_RTX.
3478 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3479 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3480 Use aarch64_simd_imm_zero.
3481 * config/aarch64/aarch64-sve.md: New file.
3482 * config/aarch64/aarch64.md: Include it.
3483 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3484 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3485 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3486 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3487 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3488 (sve): New attribute.
3489 (enabled): Disable instructions with the sve attribute unless
3491 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3492 aarch64_expand_mov_immediate.
3493 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3494 CNT[BHSD] immediates.
3495 (movti): Split CONST_POLY_INT moves into two halves.
3496 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3497 Split additions that need a temporary here if the destination
3498 is the stack pointer.
3499 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3500 (*add<mode>3_poly_1): New instruction.
3501 (set_clobber_cc): New expander.
3503 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3505 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3506 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3507 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3508 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3509 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3510 Change innermode from fixed_mode_size to machine_mode.
3511 (simplify_subreg): Update call accordingly. Handle a constant-sized
3512 subreg of a variable-length CONST_VECTOR.
3514 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3515 Alan Hayward <alan.hayward@arm.com>
3516 David Sherwood <david.sherwood@arm.com>
3518 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3519 (add_offset_to_base): New function, split out from...
3520 (create_mem_ref): ...here. When handling a scale other than 1,
3521 check first whether the address is valid without the offset.
3522 Add it into the base if so, leaving the index and scale as-is.
3524 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3527 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3528 fold_for_warn before checking if arg2 is INTEGER_CST.
3530 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3532 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3533 (store_multiple_operation): Delete.
3534 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3535 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3536 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3537 guarded by TARGET_STRING.
3538 (rs6000_output_load_multiple): Delete.
3539 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3540 OPTION_MASK_STRING / TARGET_STRING handling.
3541 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3542 (const rs6000_opt_masks) <"string">: Change mask to 0.
3543 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3544 (MASK_STRING): Delete.
3545 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3547 (load_multiple): Delete.
3554 (store_multiple): Delete.
3561 (movmemsi_8reg): Delete.
3562 (corresponding unnamed define_insn): Delete.
3563 (movmemsi_6reg): Delete.
3564 (corresponding unnamed define_insn): Delete.
3565 (movmemsi_4reg): Delete.
3566 (corresponding unnamed define_insn): Delete.
3567 (movmemsi_2reg): Delete.
3568 (corresponding unnamed define_insn): Delete.
3569 (movmemsi_1reg): Delete.
3570 (corresponding unnamed define_insn): Delete.
3571 * config/rs6000/rs6000.opt (mno-string): New.
3572 (mstring): Replace by deprecation warning stub.
3573 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3575 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3577 * regrename.c (regrename_do_replace): If replacing the same
3578 reg multiple times, try to reuse last created gen_raw_REG.
3581 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3582 main to workaround a bug in GDB.
3584 2018-01-12 Tom de Vries <tom@codesourcery.com>
3587 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3589 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3591 PR rtl-optimization/80481
3592 * ira-color.c (get_cap_member): New function.
3593 (allocnos_conflict_by_live_ranges_p): Use it.
3594 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3595 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3597 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3600 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3601 (*saddl_se_1): Ditto.
3603 (*ssubl_se_1): Ditto.
3605 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3607 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3608 rather than wi::to_widest for DR_INITs.
3609 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3610 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3611 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3613 (vect_analyze_group_access_1): Note that here.
3615 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3617 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3618 polynomial type sizes.
3620 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3622 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3623 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3624 (gimple_add_tmp_var): Likewise.
3626 2018-01-12 Martin Liska <mliska@suse.cz>
3628 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3629 (gimple_alloc_sizes): Likewise.
3630 (dump_gimple_statistics): Use PRIu64 in printf format.
3631 * gimple.h: Change uint64_t to int.
3633 2018-01-12 Martin Liska <mliska@suse.cz>
3635 * tree-core.h: Use uint64_t instead of int.
3636 * tree.c (tree_node_counts): Likewise.
3637 (tree_node_sizes): Likewise.
3638 (dump_tree_statistics): Use PRIu64 in printf format.
3640 2018-01-12 Martin Liska <mliska@suse.cz>
3642 * Makefile.in: As qsort_chk is implemented in vec.c, add
3643 vec.o to linkage of gencfn-macros.
3644 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3645 passing the info to record_node_allocation_statistics.
3646 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3648 * ggc-common.c (struct ggc_usage): Add operator== and use
3649 it in operator< and compare function.
3650 * mem-stats.h (struct mem_usage): Likewise.
3651 * vec.c (struct vec_usage): Remove operator< and compare
3652 function. Can be simply inherited.
3654 2018-01-12 Martin Jambor <mjambor@suse.cz>
3657 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3658 * tree-ssa-math-opts.c: Include domwalk.h.
3659 (convert_mult_to_fma_1): New function.
3660 (fma_transformation_info): New type.
3661 (fma_deferring_state): Likewise.
3662 (cancel_fma_deferring): New function.
3663 (result_of_phi): Likewise.
3664 (last_fma_candidate_feeds_initial_phi): Likewise.
3665 (convert_mult_to_fma): Added deferring logic, split actual
3666 transformation to convert_mult_to_fma_1.
3667 (math_opts_dom_walker): New type.
3668 (math_opts_dom_walker::after_dom_children): New method, body moved
3669 here from pass_optimize_widening_mul::execute, added deferring logic
3671 (pass_optimize_widening_mul::execute): Moved most of code to
3672 math_opts_dom_walker::after_dom_children.
3673 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3674 * config/i386/i386.c (ix86_option_override_internal): Added
3675 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3677 2018-01-12 Richard Biener <rguenther@suse.de>
3680 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3681 inline instance vars.
3683 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3686 * config/rx/rx.c (rx_is_restricted_memory_address):
3689 2018-01-12 Richard Biener <rguenther@suse.de>
3691 PR tree-optimization/80846
3692 * target.def (split_reduction): New target hook.
3693 * targhooks.c (default_split_reduction): New function.
3694 * targhooks.h (default_split_reduction): Declare.
3695 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3696 target requests first reduce vectors by combining low and high
3698 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3699 (get_vectype_for_scalar_type_and_size): Export.
3700 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3701 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3702 * doc/tm.texi: Regenerate.
3703 * config/i386/i386.c (ix86_split_reduction): Implement
3704 TARGET_VECTORIZE_SPLIT_REDUCTION.
3706 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3709 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3710 in PIC mode except for TARGET_VXWORKS_RTP.
3711 * config/sparc/sparc.c: Include cfgrtl.h.
3712 (TARGET_INIT_PIC_REG): Define.
3713 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3714 (sparc_pic_register_p): New predicate.
3715 (sparc_legitimate_address_p): Use it.
3716 (sparc_legitimize_pic_address): Likewise.
3717 (sparc_delegitimize_address): Likewise.
3718 (sparc_mode_dependent_address_p): Likewise.
3719 (gen_load_pcrel_sym): Remove 4th parameter.
3720 (load_got_register): Adjust call to above. Remove obsolete stuff.
3721 (sparc_expand_prologue): Do not call load_got_register here.
3722 (sparc_flat_expand_prologue): Likewise.
3723 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3724 (sparc_use_pseudo_pic_reg): New function.
3725 (sparc_init_pic_reg): Likewise.
3726 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3727 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3729 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3731 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3732 Add item for branch_cost.
3734 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3736 PR rtl-optimization/83565
3737 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3738 not extend the result to a larger mode for rotate operations.
3739 (num_sign_bit_copies1): Likewise.
3741 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3744 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3746 Use values-Xc.o for -pedantic.
3747 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3749 2018-01-12 Martin Liska <mliska@suse.cz>
3752 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3754 (possible_polymorphic_call_targets): Use it.
3755 (ipa_devirt): Likewise.
3757 2018-01-12 Martin Liska <mliska@suse.cz>
3759 * profile-count.h (enum profile_quality): Use 0 as invalid
3760 enum value of profile_quality.
3762 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3764 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3765 -mext-string options.
3767 2018-01-12 Richard Biener <rguenther@suse.de>
3769 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3770 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3771 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3773 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3775 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3777 * configure.ac (--with-long-double-format): Add support for the
3778 configuration option to change the default long double format on
3780 * config.gcc (powerpc*-linux*-*): Likewise.
3781 * configure: Regenerate.
3782 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3783 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3784 used without modification.
3786 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3788 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3789 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3790 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3791 MISC_BUILTIN_SPEC_BARRIER.
3792 (rs6000_init_builtins): Likewise.
3793 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3795 (speculation_barrier): New define_insn.
3796 * doc/extend.texi: Document __builtin_speculation_barrier.
3798 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3801 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3802 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3803 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3805 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3806 integral modes instead of "ss" and "sd".
3807 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3808 vectors with 32-bit and 64-bit elements.
3809 (vecdupssescalarmodesuffix): New mode attribute.
3810 (vec_dup<mode>): Use it.
3812 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3815 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3816 frame if argument is passed on stack.
3818 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3821 * ree.c (combine_reaching_defs): Optimize also
3822 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3823 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3825 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3828 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3830 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3833 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3834 after they are computed.
3836 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3838 PR tree-optimization/83695
3839 * gimple-loop-linterchange.cc
3840 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3841 reset cached scev information after interchange.
3842 (pass_linterchange::execute): Remove call to scev_reset_htab.
3844 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3846 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3847 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3848 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3849 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3850 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3851 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3852 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3853 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3854 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3855 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3856 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3857 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3858 (V_lane_reg): Likewise.
3859 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3861 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3862 (vfmal_lane_low<mode>_intrinsic,
3863 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3864 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3865 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3866 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3867 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3868 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3870 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3872 * config/arm/arm-cpus.in (fp16fml): New feature.
3873 (ALL_SIMD): Add fp16fml.
3874 (armv8.2-a): Add fp16fml as an option.
3875 (armv8.3-a): Likewise.
3876 (armv8.4-a): Add fp16fml as part of fp16.
3877 * config/arm/arm.h (TARGET_FP16FML): Define.
3878 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3880 * config/arm/arm-modes.def (V2HF): Define.
3881 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3882 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3883 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3884 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3885 vfmsl_low, vfmsl_high): New set of builtins.
3886 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3887 (vfml_op): New code attribute.
3888 (VFMLHALVES): New int iterator.
3889 (VFML, VFMLSEL): New mode attributes.
3890 (V_reg): Define mapping for V2HF.
3891 (V_hi, V_lo): New mode attributes.
3892 (VF_constraint): Likewise.
3893 (vfml_half, vfml_half_selector): New int attributes.
3894 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3896 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3897 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3899 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3900 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3901 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3902 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3904 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3905 Document new effective target and option set.
3907 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3909 * config/arm/arm-cpus.in (armv8_4): New feature.
3910 (ARMv8_4a): New fgroup.
3911 (armv8.4-a): New arch.
3912 * config/arm/arm-tables.opt: Regenerate.
3913 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3914 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3915 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3916 Add matching rules for -march=armv8.4-a and extensions.
3917 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3919 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3922 * config/rx/rx.md (BW): New mode attribute.
3923 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3925 2018-01-11 Richard Biener <rguenther@suse.de>
3927 PR tree-optimization/83435
3928 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3929 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3930 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3932 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3933 Alan Hayward <alan.hayward@arm.com>
3934 David Sherwood <david.sherwood@arm.com>
3936 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3938 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3939 (aarch64_print_address_internal): Use it to check for a zero offset.
3941 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3942 Alan Hayward <alan.hayward@arm.com>
3943 David Sherwood <david.sherwood@arm.com>
3945 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3946 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3947 Return a poly_int64 rather than a HOST_WIDE_INT.
3948 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3949 rather than a HOST_WIDE_INT.
3950 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3951 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3952 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3953 final_offset from HOST_WIDE_INT to poly_int64.
3954 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3955 to_constant when getting the number of units in an Advanced SIMD
3957 (aarch64_builtin_vectorized_function): Check for a constant number
3959 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3961 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3962 attribute instead of GET_MODE_NUNITS.
3963 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3964 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3965 GET_MODE_SIZE for fixed-size registers.
3966 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3967 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3968 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3969 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3970 (aarch64_print_operand, aarch64_print_address_internal)
3971 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3972 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3973 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3974 Handle polynomial GET_MODE_SIZE.
3975 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3976 wider than SImode without modification.
3977 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3978 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3979 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3980 passing and returning SVE modes.
3981 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3982 rather than GEN_INT.
3983 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3984 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3985 (aarch64_allocate_and_probe_stack_space): Likewise.
3986 (aarch64_layout_frame): Cope with polynomial offsets.
3987 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3988 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3990 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3991 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3992 poly_int64 rather than a HOST_WIDE_INT.
3993 (aarch64_get_separate_components, aarch64_process_components)
3994 (aarch64_expand_prologue, aarch64_expand_epilogue)
3995 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3996 (aarch64_anchor_offset): New function, split out from...
3997 (aarch64_legitimize_address): ...here.
3998 (aarch64_builtin_vectorization_cost): Handle polynomial
3999 TYPE_VECTOR_SUBPARTS.
4000 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4002 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4003 number of elements from the PARALLEL rather than the mode.
4004 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4005 rather than GET_MODE_BITSIZE.
4006 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4007 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4008 (aarch64_expand_vec_perm_const_1): Handle polynomial
4009 d->perm.length () and d->perm elements.
4010 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4011 Apply to_constant to d->perm elements.
4012 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4013 polynomial CONST_VECTOR_NUNITS.
4014 (aarch64_move_pointer): Take amount as a poly_int64 rather
4016 (aarch64_progress_pointer): Avoid temporary variable.
4017 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4018 the mode attribute instead of GET_MODE.
4020 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4021 Alan Hayward <alan.hayward@arm.com>
4022 David Sherwood <david.sherwood@arm.com>
4024 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4025 x exists before using it.
4026 (aarch64_add_constant_internal): Rename to...
4027 (aarch64_add_offset_1): ...this. Replace regnum with separate
4028 src and dest rtxes. Handle the case in which they're different,
4029 including when the offset is zero. Replace scratchreg with an rtx.
4030 Use 2 additions if there is no spare register into which we can
4031 move a 16-bit constant.
4032 (aarch64_add_constant): Delete.
4033 (aarch64_add_offset): Replace reg with separate src and dest
4034 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4035 Use aarch64_add_offset_1.
4036 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4037 an rtx rather than an int. Take the delta as a poly_int64
4038 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4039 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4040 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4041 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4042 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4044 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4045 aarch64_add_constant.
4047 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4049 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4050 Use scalar_float_mode.
4052 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4054 * config/aarch64/aarch64-simd.md
4055 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4056 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4057 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4058 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4059 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4060 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4061 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4062 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4063 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4064 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4066 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4069 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4070 targ_options->x_arm_arch_string is non NULL.
4072 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4074 * config/aarch64/aarch64.h
4075 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4077 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4080 * expmed.c (emit_store_flag_force): Swap if const op0
4081 and change VOIDmode to mode of op0.
4083 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4085 PR rtl-optimization/83761
4086 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4087 than bytes to mode_for_size.
4089 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4092 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4093 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4096 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4099 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4100 when in layout mode.
4101 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4102 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4105 2018-01-10 Michael Collison <michael.collison@arm.com>
4107 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4108 * config/aarch64/aarch64-option-extension.def: Add
4109 AARCH64_OPT_EXTENSION of 'fp16fml'.
4110 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4111 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4112 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4113 * config/aarch64/constraints.md (Ui7): New constraint.
4114 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4115 (VFMLA_SEL_W): Ditto.
4118 (VFMLA16_LOW): New int iterator.
4119 (VFMLA16_HIGH): Ditto.
4120 (UNSPEC_FMLAL): New unspec.
4121 (UNSPEC_FMLSL): Ditto.
4122 (UNSPEC_FMLAL2): Ditto.
4123 (UNSPEC_FMLSL2): Ditto.
4124 (f16mac): New code attribute.
4125 * config/aarch64/aarch64-simd-builtins.def
4126 (aarch64_fmlal_lowv2sf): Ditto.
4127 (aarch64_fmlsl_lowv2sf): Ditto.
4128 (aarch64_fmlalq_lowv4sf): Ditto.
4129 (aarch64_fmlslq_lowv4sf): Ditto.
4130 (aarch64_fmlal_highv2sf): Ditto.
4131 (aarch64_fmlsl_highv2sf): Ditto.
4132 (aarch64_fmlalq_highv4sf): Ditto.
4133 (aarch64_fmlslq_highv4sf): Ditto.
4134 (aarch64_fmlal_lane_lowv2sf): Ditto.
4135 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4136 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4137 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4138 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4139 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4140 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4141 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4142 (aarch64_fmlal_lane_highv2sf): Ditto.
4143 (aarch64_fmlsl_lane_highv2sf): Ditto.
4144 (aarch64_fmlal_laneq_highv2sf): Ditto.
4145 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4146 (aarch64_fmlalq_lane_highv4sf): Ditto.
4147 (aarch64_fmlsl_lane_highv4sf): Ditto.
4148 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4149 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4150 * config/aarch64/aarch64-simd.md:
4151 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4152 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4153 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4154 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4155 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4156 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4157 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4158 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4159 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4160 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4161 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4162 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4163 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4164 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4165 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4166 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4167 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4168 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4169 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4170 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4171 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4172 (vfmlsl_low_u32): Ditto.
4173 (vfmlalq_low_u32): Ditto.
4174 (vfmlslq_low_u32): Ditto.
4175 (vfmlal_high_u32): Ditto.
4176 (vfmlsl_high_u32): Ditto.
4177 (vfmlalq_high_u32): Ditto.
4178 (vfmlslq_high_u32): Ditto.
4179 (vfmlal_lane_low_u32): Ditto.
4180 (vfmlsl_lane_low_u32): Ditto.
4181 (vfmlal_laneq_low_u32): Ditto.
4182 (vfmlsl_laneq_low_u32): Ditto.
4183 (vfmlalq_lane_low_u32): Ditto.
4184 (vfmlslq_lane_low_u32): Ditto.
4185 (vfmlalq_laneq_low_u32): Ditto.
4186 (vfmlslq_laneq_low_u32): Ditto.
4187 (vfmlal_lane_high_u32): Ditto.
4188 (vfmlsl_lane_high_u32): Ditto.
4189 (vfmlal_laneq_high_u32): Ditto.
4190 (vfmlsl_laneq_high_u32): Ditto.
4191 (vfmlalq_lane_high_u32): Ditto.
4192 (vfmlslq_lane_high_u32): Ditto.
4193 (vfmlalq_laneq_high_u32): Ditto.
4194 (vfmlslq_laneq_high_u32): Ditto.
4195 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4196 (AARCH64_FL_FOR_ARCH8_4): New.
4197 (AARCH64_ISA_F16FML): New ISA flag.
4198 (TARGET_F16FML): New feature flag for fp16fml.
4199 (doc/invoke.texi): Document new fp16fml option.
4201 2018-01-10 Michael Collison <michael.collison@arm.com>
4203 * config/aarch64/aarch64-builtins.c:
4204 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4205 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4206 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4207 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4208 (AARCH64_ISA_SHA3): New ISA flag.
4209 (TARGET_SHA3): New feature flag for sha3.
4210 * config/aarch64/iterators.md (sha512_op): New int attribute.
4211 (CRYPTO_SHA512): New int iterator.
4212 (UNSPEC_SHA512H): New unspec.
4213 (UNSPEC_SHA512H2): Ditto.
4214 (UNSPEC_SHA512SU0): Ditto.
4215 (UNSPEC_SHA512SU1): Ditto.
4216 * config/aarch64/aarch64-simd-builtins.def
4217 (aarch64_crypto_sha512hqv2di): New builtin.
4218 (aarch64_crypto_sha512h2qv2di): Ditto.
4219 (aarch64_crypto_sha512su0qv2di): Ditto.
4220 (aarch64_crypto_sha512su1qv2di): Ditto.
4221 (aarch64_eor3qv8hi): Ditto.
4222 (aarch64_rax1qv2di): Ditto.
4223 (aarch64_xarqv2di): Ditto.
4224 (aarch64_bcaxqv8hi): Ditto.
4225 * config/aarch64/aarch64-simd.md:
4226 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4227 (aarch64_crypto_sha512su0qv2di): Ditto.
4228 (aarch64_crypto_sha512su1qv2di): Ditto.
4229 (aarch64_eor3qv8hi): Ditto.
4230 (aarch64_rax1qv2di): Ditto.
4231 (aarch64_xarqv2di): Ditto.
4232 (aarch64_bcaxqv8hi): Ditto.
4233 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4234 (vsha512h2q_u64): Ditto.
4235 (vsha512su0q_u64): Ditto.
4236 (vsha512su1q_u64): Ditto.
4237 (veor3q_u16): Ditto.
4238 (vrax1q_u64): Ditto.
4240 (vbcaxq_u16): Ditto.
4241 * config/arm/types.md (crypto_sha512): New type attribute.
4242 (crypto_sha3): Ditto.
4243 (doc/invoke.texi): Document new sha3 option.
4245 2018-01-10 Michael Collison <michael.collison@arm.com>
4247 * config/aarch64/aarch64-builtins.c:
4248 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4249 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4250 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4251 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4252 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4253 (AARCH64_ISA_SM4): New ISA flag.
4254 (TARGET_SM4): New feature flag for sm4.
4255 * config/aarch64/aarch64-simd-builtins.def
4256 (aarch64_sm3ss1qv4si): Ditto.
4257 (aarch64_sm3tt1aq4si): Ditto.
4258 (aarch64_sm3tt1bq4si): Ditto.
4259 (aarch64_sm3tt2aq4si): Ditto.
4260 (aarch64_sm3tt2bq4si): Ditto.
4261 (aarch64_sm3partw1qv4si): Ditto.
4262 (aarch64_sm3partw2qv4si): Ditto.
4263 (aarch64_sm4eqv4si): Ditto.
4264 (aarch64_sm4ekeyqv4si): Ditto.
4265 * config/aarch64/aarch64-simd.md:
4266 (aarch64_sm3ss1qv4si): Ditto.
4267 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4268 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4269 (aarch64_sm4eqv4si): Ditto.
4270 (aarch64_sm4ekeyqv4si): Ditto.
4271 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4272 (sm3part_op): Ditto.
4273 (CRYPTO_SM3TT): Ditto.
4274 (CRYPTO_SM3PART): Ditto.
4275 (UNSPEC_SM3SS1): New unspec.
4276 (UNSPEC_SM3TT1A): Ditto.
4277 (UNSPEC_SM3TT1B): Ditto.
4278 (UNSPEC_SM3TT2A): Ditto.
4279 (UNSPEC_SM3TT2B): Ditto.
4280 (UNSPEC_SM3PARTW1): Ditto.
4281 (UNSPEC_SM3PARTW2): Ditto.
4282 (UNSPEC_SM4E): Ditto.
4283 (UNSPEC_SM4EKEY): Ditto.
4284 * config/aarch64/constraints.md (Ui2): New constraint.
4285 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4286 * config/arm/types.md (crypto_sm3): New type attribute.
4287 (crypto_sm4): Ditto.
4288 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4289 (vsm3tt1aq_u32): Ditto.
4290 (vsm3tt1bq_u32): Ditto.
4291 (vsm3tt2aq_u32): Ditto.
4292 (vsm3tt2bq_u32): Ditto.
4293 (vsm3partw1q_u32): Ditto.
4294 (vsm3partw2q_u32): Ditto.
4295 (vsm4eq_u32): Ditto.
4296 (vsm4ekeyq_u32): Ditto.
4297 (doc/invoke.texi): Document new sm4 option.
4299 2018-01-10 Michael Collison <michael.collison@arm.com>
4301 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4302 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4303 (AARCH64_FL_FOR_ARCH8_4): New.
4304 (AARCH64_FL_V8_4): New flag.
4305 (doc/invoke.texi): Document new armv8.4-a option.
4307 2018-01-10 Michael Collison <michael.collison@arm.com>
4309 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4310 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4311 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4312 * config/aarch64/aarch64-option-extension.def: Add
4313 AARCH64_OPT_EXTENSION of 'sha2'.
4314 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4315 (crypto): Disable sha2 and aes if crypto disabled.
4316 (crypto): Enable aes and sha2 if enabled.
4317 (simd): Disable sha2 and aes if simd disabled.
4318 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4320 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4321 (TARGET_SHA2): New feature flag for sha2.
4322 (TARGET_AES): New feature flag for aes.
4323 * config/aarch64/aarch64-simd.md:
4324 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4325 conditional on TARGET_AES.
4326 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4327 (aarch64_crypto_sha1hsi): Make pattern conditional
4329 (aarch64_crypto_sha1hv4si): Ditto.
4330 (aarch64_be_crypto_sha1hv4si): Ditto.
4331 (aarch64_crypto_sha1su1v4si): Ditto.
4332 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4333 (aarch64_crypto_sha1su0v4si): Ditto.
4334 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4335 (aarch64_crypto_sha256su0v4si): Ditto.
4336 (aarch64_crypto_sha256su1v4si): Ditto.
4337 (doc/invoke.texi): Document new aes and sha2 options.
4339 2018-01-10 Martin Sebor <msebor@redhat.com>
4341 PR tree-optimization/83781
4342 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4345 2018-01-11 Martin Sebor <msebor@gmail.com>
4346 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4348 PR tree-optimization/83501
4349 PR tree-optimization/81703
4351 * tree-ssa-strlen.c (get_string_cst): Rename...
4352 (get_string_len): ...to this. Handle global constants.
4353 (handle_char_store): Adjust.
4355 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4356 Jim Wilson <jimw@sifive.com>
4358 * config/riscv/riscv-protos.h (riscv_output_return): New.
4359 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4360 (riscv_attribute_table, riscv_output_return),
4361 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4362 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4363 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4365 (riscv_expand_prologue): Add early return for naked function.
4366 (riscv_expand_epilogue): Likewise.
4367 (riscv_function_ok_for_sibcall): Return false for naked function.
4368 (riscv_set_current_function): New.
4369 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4370 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4371 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4372 * doc/extend.texi (RISC-V Function Attributes): New.
4374 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4376 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4377 check for 128-bit long double before checking TCmode.
4378 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4379 128-bit long doubles before checking TFmode or TCmode.
4380 (FLOAT128_IBM_P): Likewise.
4382 2018-01-10 Martin Sebor <msebor@redhat.com>
4384 PR tree-optimization/83671
4385 * builtins.c (c_strlen): Unconditionally return zero for the empty
4387 Use -Warray-bounds for warnings.
4388 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4389 for non-constant array indices with COMPONENT_REF, arrays of
4390 arrays, and pointers to arrays.
4391 (gimple_fold_builtin_strlen): Determine and set length range for
4392 non-constant character arrays.
4394 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4397 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4400 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4402 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4404 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4407 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4408 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4409 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4410 indexed_or_indirect_operand predicate.
4411 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4412 (*vsx_le_perm_load_v8hi): Likewise.
4413 (*vsx_le_perm_load_v16qi): Likewise.
4414 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4415 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4416 (*vsx_le_perm_store_v8hi): Likewise.
4417 (*vsx_le_perm_store_v16qi): Likewise.
4418 (eight unnamed splitters): Likewise.
4420 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4422 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4423 * config/rs6000/emmintrin.h: Likewise.
4424 * config/rs6000/mmintrin.h: Likewise.
4425 * config/rs6000/xmmintrin.h: Likewise.
4427 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4430 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4432 * tree.c (tree_nop_conversion): Return true for location wrapper
4434 (maybe_wrap_with_location): New function.
4435 (selftest::check_strip_nops): New function.
4436 (selftest::test_location_wrappers): New function.
4437 (selftest::tree_c_tests): Call it.
4438 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4439 (maybe_wrap_with_location): New decl.
4440 (EXPR_LOCATION_WRAPPER_P): New macro.
4441 (location_wrapper_p): New inline function.
4442 (tree_strip_any_location_wrapper): New inline function.
4444 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4447 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4448 stack_realign_offset for the largest alignment of stack slot
4450 (ix86_find_max_used_stack_alignment): New function.
4451 (ix86_finalize_stack_frame_flags): Use it. Set
4452 max_used_stack_alignment if we don't realign stack.
4453 * config/i386/i386.h (machine_function): Add
4454 max_used_stack_alignment.
4456 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4458 * config/arm/arm.opt (-mbranch-cost): New option.
4459 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4462 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4465 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4466 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4468 2018-01-10 Richard Biener <rguenther@suse.de>
4471 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4472 early out so it also covers the case where we have a non-NULL
4475 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4477 PR tree-optimization/83753
4478 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4479 for non-strided grouped accesses if the number of elements is 1.
4481 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4484 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4485 * i386.h (TARGET_USE_GATHER): Define.
4486 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4488 2018-01-10 Martin Liska <mliska@suse.cz>
4491 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4492 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4494 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4495 CLEANUP_NO_PARTITIONING is not set.
4497 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4499 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4500 for vectors, as a partial revert of r254296.
4501 * rtl.h (const_vec_p): Delete.
4502 (const_vec_duplicate_p): Don't test for vector CONSTs.
4503 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4504 * expmed.c (make_tree): Likewise.
4507 * common.md (E, F): Use CONSTANT_P instead of checking for
4509 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4510 checking for CONST_VECTOR.
4512 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4515 * predict.c (force_edge_cold): Handle in more sane way edges
4518 2018-01-09 Carl Love <cel@us.ibm.com>
4520 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4522 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4523 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4524 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4525 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4526 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4527 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4528 * config/rs6000/rs6000-protos.h: Add extern defition for
4529 rs6000_generate_float2_double_code.
4530 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4532 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4533 (float2_v2df): Add define_expand.
4535 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4538 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4539 op_mode in the force_to_mode call.
4541 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4543 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4544 instead of checking each element individually.
4545 (aarch64_evpc_uzp): Likewise.
4546 (aarch64_evpc_zip): Likewise.
4547 (aarch64_evpc_ext): Likewise.
4548 (aarch64_evpc_rev): Likewise.
4549 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4550 instead of checking each element individually. Return true without
4552 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4553 whether all selected elements come from the same input, instead of
4554 checking each element individually. Remove calls to gen_rtx_REG,
4555 start_sequence and end_sequence and instead assert that no rtl is
4558 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4560 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4561 order of HIGH and CONST checks.
4563 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4565 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4566 if the destination isn't an SSA_NAME.
4568 2018-01-09 Richard Biener <rguenther@suse.de>
4570 PR tree-optimization/83668
4571 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4573 (canonicalize_loop_form): ... here, renamed from ...
4574 (canonicalize_loop_closed_ssa_form): ... this and amended to
4575 swap successor edges for loop exit blocks to make us use
4576 the RPO order we need for initial schedule generation.
4578 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4580 PR tree-optimization/64811
4581 * match.pd: When optimizing comparisons with Inf, avoid
4582 introducing or losing exceptions from comparisons with NaN.
4584 2018-01-09 Martin Liska <mliska@suse.cz>
4587 * asan.c (shadow_mem_size): Add gcc_assert.
4589 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4591 Don't save registers in main().
4594 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4595 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4596 * config/avr/avr.c (avr_set_current_function): Don't error if
4597 naked, OS_task or OS_main are specified at the same time.
4598 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4600 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4602 * common/config/avr/avr-common.c (avr_option_optimization_table):
4603 Switch on -mmain-is-OS_task for optimizing compilations.
4605 2018-01-09 Richard Biener <rguenther@suse.de>
4607 PR tree-optimization/83572
4608 * graphite.c: Include cfganal.h.
4609 (graphite_transform_loops): Connect infinite loops to exit
4610 and remove fake edges at the end.
4612 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4614 * ipa-inline.c (edge_badness): Revert accidental checkin.
4616 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4619 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4620 symbols; not inline clones.
4622 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4625 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4626 hard registers. Formatting fixes.
4628 PR preprocessor/83722
4629 * gcc.c (try_generate_repro): Pass
4630 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4631 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4634 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4635 Kito Cheng <kito.cheng@gmail.com>
4637 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4638 (riscv_leaf_function_p): Delete.
4639 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4641 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4643 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4645 (do_ifelse): New function.
4646 (do_isel): New function.
4647 (do_sub3): New function.
4648 (do_add3): New function.
4649 (do_load_mask_compare): New function.
4650 (do_overlap_load_compare): New function.
4651 (expand_compare_loop): New function.
4652 (expand_block_compare): Call expand_compare_loop() when appropriate.
4653 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4655 (-mblock-compare-inline-loop-limit): New option.
4657 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4660 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4661 Reverse order of second and third operands in first alternative.
4662 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4663 of first and second elements in UNSPEC_VPERMR vector.
4664 (altivec_expand_vec_perm_le): Likewise.
4666 2018-01-08 Jeff Law <law@redhat.com>
4668 PR rtl-optimizatin/81308
4669 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4670 (process_switch): If group_case_labels makes a change, then set
4672 (pass_convert_switch::execute): If a switch is converted, then
4673 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4675 PR rtl-optimization/81308
4676 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4679 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4681 PR target/83663 - Revert r255946
4682 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4683 generation for cases where splatting a value is not useful.
4684 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4685 across a vec_duplicate and a paradoxical subreg forming a vector
4686 mode to a vec_concat.
4688 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4690 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4691 -march=armv8.3-a variants.
4692 * config/arm/t-multilib: Likewise.
4693 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4695 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4697 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4699 (cceq_ior_compare_complement): Give it a name so I can use it, and
4700 change boolean_or_operator predicate to boolean_operator so it can
4701 be used to generate a crand.
4702 (eqne): New code iterator.
4703 (bd/bd_neg): New code_attrs.
4704 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4705 a single define_insn.
4706 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4707 decrement (bdnzt/bdnzf/bdzt/bdzf).
4708 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4709 with the new names of the branch decrement patterns, and added the
4710 names of the branch decrement conditional patterns.
4712 2018-01-08 Richard Biener <rguenther@suse.de>
4714 PR tree-optimization/83563
4715 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4718 2018-01-08 Richard Biener <rguenther@suse.de>
4721 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4723 2018-01-08 Richard Biener <rguenther@suse.de>
4725 PR tree-optimization/83685
4726 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4727 references to abnormals.
4729 2018-01-08 Richard Biener <rguenther@suse.de>
4732 * dwarf2out.c (output_indirect_strings): Handle empty
4733 skeleton_debug_str_hash.
4734 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4736 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4738 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4739 (emit_store_direct): Likewise.
4740 (arc_trampoline_adjust_address): Likewise.
4741 (arc_asm_trampoline_template): New function.
4742 (arc_initialize_trampoline): Use asm_trampoline_template.
4743 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4744 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4745 * config/arc/arc.md (flush_icache): Delete pattern.
4747 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4749 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4750 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4753 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4756 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4757 by not USED_FOR_TARGET.
4758 (make_pass_resolve_sw_modes): Likewise.
4760 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4762 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4765 2018-01-08 Richard Biener <rguenther@suse.de>
4768 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4770 2018-01-08 Richard Biener <rguenther@suse.de>
4773 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4775 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4778 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4779 basic blocks with a small number of successors.
4780 (convert_control_dep_chain_into_preds): Improve handling of
4782 (dump_predicates): Split apart into...
4783 (dump_pred_chain): ...here...
4784 (dump_pred_info): ...and here.
4785 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4786 (can_chain_union_be_invalidated_p): Improve check for invalidation
4788 (uninit_uses_cannot_happen): Avoid unnecessary if
4789 convert_control_dep_chain_into_preds yielded nothing.
4791 2018-01-06 Martin Sebor <msebor@redhat.com>
4793 PR tree-optimization/83640
4794 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4795 subtracting negative offset from size.
4796 (builtin_access::overlap): Adjust offset bounds of the access to fall
4797 within the size of the object if possible.
4799 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4801 PR rtl-optimization/83699
4802 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4803 extract_bit_field_as_subreg to cases in which the extracted
4804 value is also a vector.
4806 * lra-constraints.c (process_alt_operands): Test for the equivalence
4807 substitutions when detecting a possible reload cycle.
4809 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4812 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4813 by default if flag_selective_schedling{,2}. Formatting fixes.
4815 PR rtl-optimization/83682
4816 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4817 if it has non-VECTOR_MODE element mode.
4818 (vec_duplicate_p): Likewise.
4821 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4822 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4824 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4827 * config/i386/i386-builtin.def
4828 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4829 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4830 Require also OPTION_MASK_ISA_AVX512F in addition to
4831 OPTION_MASK_ISA_GFNI.
4832 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4833 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4834 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4835 to OPTION_MASK_ISA_GFNI.
4836 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4837 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4838 OPTION_MASK_ISA_AVX512BW.
4839 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4840 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4841 addition to OPTION_MASK_ISA_GFNI.
4842 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4843 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4844 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4845 to OPTION_MASK_ISA_GFNI.
4846 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4847 a requirement for all ISAs rather than any of them with a few
4849 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4851 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4852 bitmasks to be enabled with 3 exceptions, instead of requiring any
4853 enabled ISA with lots of exceptions.
4854 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4855 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4856 Change avx512bw in isa attribute to avx512f.
4857 * config/i386/sgxintrin.h: Add license boilerplate.
4858 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4859 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4860 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4861 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4863 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4864 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4865 temporarily sse2 rather than sse if not enabled already.
4868 * config/i386/sse.md (VI248_VLBW): Rename to ...
4869 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4870 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4871 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4872 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4873 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4874 mode iterator instead of VI248_VLBW.
4876 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4878 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4879 (record_modified): Skip clobbers; add debug output.
4880 (param_change_prob): Use sreal frequencies.
4882 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4884 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4885 punt for user-aligned variables.
4887 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4889 * tree-chrec.c (chrec_contains_symbols): Return true for
4892 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4895 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4896 of (x|y) == x for BICS pattern.
4898 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4900 PR tree-optimization/83605
4901 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4902 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4905 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4907 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4908 * config/epiphany/rtems.h: New file.
4910 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4911 Uros Bizjak <ubizjak@gmail.com>
4914 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4915 QIreg_operand instead of register_operand predicate.
4916 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4917 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4918 comments instead of -fmitigate[-_]rop.
4920 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4923 * cgraphunit.c (symbol_table::compile): Switch to text_section
4924 before calling assembly_start debug hook.
4925 * run-rtl-passes.c (run_rtl_passes): Likewise.
4928 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4930 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4931 range_int_cst_p rather than !symbolic_range_p before calling
4932 extract_range_from_multiplicative_op_1.
4934 2018-01-04 Jeff Law <law@redhat.com>
4936 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4937 redundant test in assertion.
4939 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4941 * doc/rtl.texi: Document machine_mode wrapper classes.
4943 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4945 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4948 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4950 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4951 the VEC_PERM_EXPR fold to fail.
4953 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4956 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4957 to switched_sections.
4959 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4962 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4965 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4968 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4969 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4971 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4974 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4975 is BLKmode and bitpos not zero or mode change is needed.
4977 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4980 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4983 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4986 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4987 instead of MULT rtx. Update all corresponding splitters.
4989 (*ssub<modesuffix>): Ditto.
4991 (*cmp_sadd_di): Update split patterns.
4992 (*cmp_sadd_si): Ditto.
4993 (*cmp_sadd_sidi): Ditto.
4994 (*cmp_ssub_di): Ditto.
4995 (*cmp_ssub_si): Ditto.
4996 (*cmp_ssub_sidi): Ditto.
4997 * config/alpha/predicates.md (const23_operand): New predicate.
4998 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4999 Look for ASHIFT, not MULT inner operand.
5000 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5002 2018-01-04 Martin Liska <mliska@suse.cz>
5004 PR gcov-profile/83669
5005 * gcov.c (output_intermediate_file): Add version to intermediate
5007 * doc/gcov.texi: Document new field 'version' in intermediate
5008 file format. Fix location of '-k' option of gcov command.
5010 2018-01-04 Martin Liska <mliska@suse.cz>
5013 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5015 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5017 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5019 2018-01-03 Martin Sebor <msebor@redhat.com>
5021 PR tree-optimization/83655
5022 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5023 checking calls with invalid arguments.
5025 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5027 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5028 (vectorizable_mask_load_store): Delete.
5029 (vectorizable_call): Return false for masked loads and stores.
5030 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5031 instead of gimple_assign_rhs1.
5032 (vectorizable_load): Handle IFN_MASK_LOAD.
5033 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5037 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5039 (vectorizable_mask_load_store): ...here.
5040 (vectorizable_load): ...and here.
5042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5044 * tree-vect-stmts.c (vect_build_all_ones_mask)
5045 (vect_build_zero_merge_argument): New functions, split out from...
5046 (vectorizable_load): ...here.
5048 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5050 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5052 (vectorizable_mask_load_store): ...here.
5053 (vectorizable_store): ...and here.
5055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5057 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5059 (vectorizable_mask_load_store): ...here.
5061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5063 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5064 (vect_model_store_cost): Take a vec_load_store_type instead of a
5066 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5067 (vect_model_store_cost): Take a vec_load_store_type instead of a
5069 (vectorizable_mask_load_store): Update accordingly.
5070 (vectorizable_store): Likewise.
5071 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5075 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5076 IFN_MASK_LOAD calls here rather than...
5077 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5080 Alan Hayward <alan.hayward@arm.com>
5081 David Sherwood <david.sherwood@arm.com>
5083 * expmed.c (extract_bit_field_1): For vector extracts,
5084 fall back to extract_bit_field_as_subreg if vec_extract
5087 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5088 Alan Hayward <alan.hayward@arm.com>
5089 David Sherwood <david.sherwood@arm.com>
5091 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5092 they are variable or constant sized.
5093 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5094 slots for constant-sized data.
5096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5097 Alan Hayward <alan.hayward@arm.com>
5098 David Sherwood <david.sherwood@arm.com>
5100 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5101 handling COND_EXPRs with boolean comparisons, try to find a better
5102 basis for the mask type than the boolean itself.
5104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5106 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5107 is calculated and how it can be overridden.
5108 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5109 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5111 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5115 Alan Hayward <alan.hayward@arm.com>
5116 David Sherwood <david.sherwood@arm.com>
5118 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5119 Remove the mode argument.
5120 (aarch64_simd_valid_immediate): Remove the mode and inverse
5122 * config/aarch64/iterators.md (bitsize): New iterator.
5123 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5124 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5125 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5126 aarch64_simd_valid_immediate.
5127 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5128 (aarch64_reg_or_bic_imm): Likewise.
5129 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5130 with an insn_type enum and msl with a modifier_type enum.
5131 Replace element_width with a scalar_mode. Change the shift
5132 to unsigned int. Add constructors for scalar_float_mode and
5133 scalar_int_mode elements.
5134 (aarch64_vect_float_const_representable_p): Delete.
5135 (aarch64_can_const_movi_rtx_p)
5136 (aarch64_simd_scalar_immediate_valid_for_move)
5137 (aarch64_simd_make_constant): Update call to
5138 aarch64_simd_valid_immediate.
5139 (aarch64_advsimd_valid_immediate_hs): New function.
5140 (aarch64_advsimd_valid_immediate): Likewise.
5141 (aarch64_simd_valid_immediate): Remove mode and inverse
5142 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5143 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5144 and aarch64_float_const_representable_p on the result.
5145 (aarch64_output_simd_mov_immediate): Remove mode argument.
5146 Update call to aarch64_simd_valid_immediate and use of
5147 simd_immediate_info.
5148 (aarch64_output_scalar_simd_mov_immediate): Update call
5151 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5152 Alan Hayward <alan.hayward@arm.com>
5153 David Sherwood <david.sherwood@arm.com>
5155 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5156 (mode_nunits): Likewise CONST_MODE_NUNITS.
5157 * machmode.def (ADJUST_NUNITS): Document.
5158 * genmodes.c (mode_data::need_nunits_adj): New field.
5159 (blank_mode): Update accordingly.
5160 (adj_nunits): New variable.
5161 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5163 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5164 listed in adj_nunits.
5165 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5166 listed in adj_nunits. Don't emit case statements for such modes.
5167 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5168 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5169 nothing if adj_nunits is nonnull.
5170 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5171 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5172 (emit_mode_fbit): Update use of print_maybe_const_decl.
5173 (emit_move_size): Likewise. Treat the array as non-const
5175 (emit_mode_adjustments): Handle adj_nunits.
5177 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5179 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5180 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5181 (VECTOR_MODES): Use it.
5182 (make_vector_modes): Take the prefix as an argument.
5184 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5185 Alan Hayward <alan.hayward@arm.com>
5186 David Sherwood <david.sherwood@arm.com>
5188 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5189 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5190 for MODE_VECTOR_BOOL.
5191 * machmode.def (VECTOR_BOOL_MODE): Document.
5192 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5193 (make_vector_bool_mode): New function.
5194 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5196 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5197 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5199 * stor-layout.c (int_mode_for_mode): Likewise.
5200 * tree.c (build_vector_type_for_mode): Likewise.
5201 * varasm.c (output_constant_pool_2): Likewise.
5202 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5203 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5204 for MODE_VECTOR_BOOL.
5205 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5206 of mode class checks.
5207 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5208 instead of a list of mode class checks.
5209 (expand_vector_scalar_condition): Likewise.
5210 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5213 Alan Hayward <alan.hayward@arm.com>
5214 David Sherwood <david.sherwood@arm.com>
5216 * machmode.h (mode_size): Change from unsigned short to
5218 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5219 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5220 or if measurement_type is not polynomial.
5221 (fixed_size_mode::includes_p): Check for constant-sized modes.
5222 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5223 return a poly_uint16 rather than an unsigned short.
5224 (emit_mode_size): Change the type of mode_size from unsigned short
5225 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5226 (emit_mode_adjustments): Cope with polynomial vector sizes.
5227 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5229 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5231 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5232 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5233 * caller-save.c (setup_save_areas): Likewise.
5234 (replace_reg_with_saved_mem): Likewise.
5235 * calls.c (emit_library_call_value_1): Likewise.
5236 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5237 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5238 (gen_lowpart_for_combine): Likewise.
5239 * convert.c (convert_to_integer_1): Likewise.
5240 * cse.c (equiv_constant, cse_insn): Likewise.
5241 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5242 (cselib_subst_to_values): Likewise.
5243 * dce.c (word_dce_process_block): Likewise.
5244 * df-problems.c (df_word_lr_mark_ref): Likewise.
5245 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5246 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5247 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5248 (rtl_for_decl_location): Likewise.
5249 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5250 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5251 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5252 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5253 (expand_expr_real_1): Likewise.
5254 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5255 (pad_below): Likewise.
5256 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5257 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5258 * ira.c (get_subreg_tracking_sizes): Likewise.
5259 * ira-build.c (ira_create_allocno_objects): Likewise.
5260 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5261 (ira_sort_regnos_for_alter_reg): Likewise.
5262 * ira-costs.c (record_operand_costs): Likewise.
5263 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5264 (resolve_simple_move): Likewise.
5265 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5266 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5267 (lra_constraints): Likewise.
5268 (CONST_POOL_OK_P): Reject variable-sized modes.
5269 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5270 (add_pseudo_to_slot, lra_spill): Likewise.
5271 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5272 * optabs-query.c (get_best_extraction_insn): Likewise.
5273 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5274 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5275 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5276 * recog.c (offsettable_address_addr_space_p): Likewise.
5277 * regcprop.c (maybe_mode_change): Likewise.
5278 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5279 * regrename.c (build_def_use): Likewise.
5280 * regstat.c (dump_reg_info): Likewise.
5281 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5282 (find_reloads, find_reloads_subreg_address): Likewise.
5283 * reload1.c (eliminate_regs_1): Likewise.
5284 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5285 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5286 (simplify_binary_operation_1, simplify_subreg): Likewise.
5287 * targhooks.c (default_function_arg_padding): Likewise.
5288 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5289 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5290 (verify_gimple_assign_ternary): Likewise.
5291 * tree-inline.c (estimate_move_cost): Likewise.
5292 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5293 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5294 (get_address_cost_ainc): Likewise.
5295 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5296 (vect_supportable_dr_alignment): Likewise.
5297 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5298 (vectorizable_reduction): Likewise.
5299 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5300 (vectorizable_operation, vectorizable_load): Likewise.
5301 * tree.c (build_same_sized_truth_vector_type): Likewise.
5302 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5303 * var-tracking.c (emit_note_insn_var_location): Likewise.
5304 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5305 (ADDR_VEC_ALIGN): Likewise.
5307 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5308 Alan Hayward <alan.hayward@arm.com>
5309 David Sherwood <david.sherwood@arm.com>
5311 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5313 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5314 or if measurement_type is polynomial.
5315 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5316 * combine.c (make_extraction): Likewise.
5317 * dse.c (find_shift_sequence): Likewise.
5318 * dwarf2out.c (mem_loc_descriptor): Likewise.
5319 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5320 (extract_bit_field, extract_low_bits): Likewise.
5321 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5322 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5323 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5324 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5325 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5326 * reload.c (find_reloads): Likewise.
5327 * reload1.c (alter_reg): Likewise.
5328 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5329 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5330 * tree-if-conv.c (predicate_mem_writes): Likewise.
5331 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5332 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5333 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5334 * valtrack.c (dead_debug_insert_temp): Likewise.
5335 * varasm.c (mergeable_constant_section): Likewise.
5336 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5338 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5339 Alan Hayward <alan.hayward@arm.com>
5340 David Sherwood <david.sherwood@arm.com>
5342 * expr.c (expand_assignment): Cope with polynomial mode sizes
5343 when assigning to a CONCAT.
5345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5346 Alan Hayward <alan.hayward@arm.com>
5347 David Sherwood <david.sherwood@arm.com>
5349 * machmode.h (mode_precision): Change from unsigned short to
5351 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5353 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5354 or if measurement_type is not polynomial.
5355 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5356 in which the mode is already known to be a scalar_int_mode.
5357 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5358 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5360 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5361 for GET_MODE_PRECISION.
5362 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5363 for GET_MODE_PRECISION.
5364 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5366 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5367 (expand_field_assignment, make_extraction): Likewise.
5368 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5369 (get_last_value): Likewise.
5370 * convert.c (convert_to_integer_1): Likewise.
5371 * cse.c (cse_insn): Likewise.
5372 * expr.c (expand_expr_real_1): Likewise.
5373 * lra-constraints.c (simplify_operand_subreg): Likewise.
5374 * optabs-query.c (can_atomic_load_p): Likewise.
5375 * optabs.c (expand_atomic_load): Likewise.
5376 (expand_atomic_store): Likewise.
5377 * ree.c (combine_reaching_defs): Likewise.
5378 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5379 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5380 * tree.h (type_has_mode_precision_p): Likewise.
5381 * ubsan.c (instrument_si_overflow): Likewise.
5383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5384 Alan Hayward <alan.hayward@arm.com>
5385 David Sherwood <david.sherwood@arm.com>
5387 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5388 polynomial numbers of units.
5389 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5390 (valid_vector_subparts_p): New function.
5391 (build_vector_type): Remove temporary shim and take the number
5392 of units as a poly_uint64 rather than an int.
5393 (build_opaque_vector_type): Take the number of units as a
5394 poly_uint64 rather than an int.
5395 * tree.c (build_vector_from_ctor): Handle polynomial
5396 TYPE_VECTOR_SUBPARTS.
5397 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5398 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5399 (build_vector_from_val): If the number of units is variable,
5400 use build_vec_duplicate_cst for constant operands and
5401 VEC_DUPLICATE_EXPR otherwise.
5402 (make_vector_type): Remove temporary is_constant ().
5403 (build_vector_type, build_opaque_vector_type): Take the number of
5404 units as a poly_uint64 rather than an int.
5405 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5407 * cfgexpand.c (expand_debug_expr): Likewise.
5408 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5409 (store_constructor, expand_expr_real_1): Likewise.
5410 (const_scalar_mask_from_tree): Likewise.
5411 * fold-const-call.c (fold_const_reduction): Likewise.
5412 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5413 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5414 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5415 (fold_relational_const): Likewise.
5416 (native_interpret_vector): Likewise. Change the size from an
5417 int to an unsigned int.
5418 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5419 TYPE_VECTOR_SUBPARTS.
5420 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5421 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5422 duplicating a non-constant operand into a variable-length vector.
5423 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5424 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5425 * ipa-icf.c (sem_variable::equals): Likewise.
5426 * match.pd: Likewise.
5427 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5428 * print-tree.c (print_node): Likewise.
5429 * stor-layout.c (layout_type): Likewise.
5430 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5431 * tree-cfg.c (verify_gimple_comparison): Likewise.
5432 (verify_gimple_assign_binary): Likewise.
5433 (verify_gimple_assign_ternary): Likewise.
5434 (verify_gimple_assign_single): Likewise.
5435 * tree-pretty-print.c (dump_generic_node): Likewise.
5436 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5437 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5438 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5439 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5440 (vect_shift_permute_load_chain): Likewise.
5441 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5442 (expand_vector_condition, optimize_vector_constructor): Likewise.
5443 (lower_vec_perm, get_compute_type): Likewise.
5444 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5445 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5446 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5447 (vect_recog_mask_conversion_pattern): Likewise.
5448 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5449 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5450 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5451 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5452 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5453 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5454 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5455 (supportable_widening_operation): Likewise.
5456 (supportable_narrowing_operation): Likewise.
5457 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5459 * varasm.c (output_constant): Likewise.
5461 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5462 Alan Hayward <alan.hayward@arm.com>
5463 David Sherwood <david.sherwood@arm.com>
5465 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5466 so that both the length == 3 and length != 3 cases set up their
5467 own permute vectors. Add comments explaining why we know the
5468 number of elements is constant.
5469 (vect_permute_load_chain): Likewise.
5471 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5472 Alan Hayward <alan.hayward@arm.com>
5473 David Sherwood <david.sherwood@arm.com>
5475 * machmode.h (mode_nunits): Change from unsigned char to
5477 (ONLY_FIXED_SIZE_MODES): New macro.
5478 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5479 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5480 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5482 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5483 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5484 or if measurement_type is not polynomial.
5485 * genmodes.c (ZERO_COEFFS): New macro.
5486 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5488 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5489 Use ZERO_COEFFS when emitting initializers.
5490 * data-streamer.h (bp_pack_poly_value): New function.
5491 (bp_unpack_poly_value): Likewise.
5492 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5493 for GET_MODE_NUNITS.
5494 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5495 for GET_MODE_NUNITS.
5496 * tree.c (make_vector_type): Remove temporary shim and make
5497 the real function take the number of units as a poly_uint64
5499 (build_vector_type_for_mode): Handle polynomial nunits.
5500 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5501 * emit-rtl.c (const_vec_series_p_1): Likewise.
5502 (gen_rtx_CONST_VECTOR): Likewise.
5503 * fold-const.c (test_vec_duplicate_folding): Likewise.
5504 * genrecog.c (validate_pattern): Likewise.
5505 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5506 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5507 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5508 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5509 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5510 * rtlanal.c (subreg_get_info): Likewise.
5511 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5512 (vect_grouped_load_supported): Likewise.
5513 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5514 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5515 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5516 (simplify_const_unary_operation, simplify_binary_operation_1)
5517 (simplify_const_binary_operation, simplify_ternary_operation)
5518 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5519 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5520 instead of CONST_VECTOR_NUNITS.
5521 * varasm.c (output_constant_pool_2): Likewise.
5522 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5523 explicit-encoded elements in the XVEC for variable-length vectors.
5525 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5527 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5529 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5530 Alan Hayward <alan.hayward@arm.com>
5531 David Sherwood <david.sherwood@arm.com>
5533 * coretypes.h (fixed_size_mode): Declare.
5534 (fixed_size_mode_pod): New typedef.
5535 * builtins.h (target_builtins::x_apply_args_mode)
5536 (target_builtins::x_apply_result_mode): Change type to
5537 fixed_size_mode_pod.
5538 * builtins.c (apply_args_size, apply_result_size, result_vector)
5539 (expand_builtin_apply_args_1, expand_builtin_apply)
5540 (expand_builtin_return): Update accordingly.
5542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5544 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5545 * cselib.c (cselib_hash_rtx): Likewise.
5546 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5547 CONST_VECTOR encoding.
5549 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5550 Jeff Law <law@redhat.com>
5553 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5554 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5555 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5556 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5559 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5560 explicitly probe *sp in a noreturn function if there were any callee
5561 register saves or frame pointer is needed.
5563 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5566 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5567 BLKmode for ternary, binary or unary expressions.
5570 * var-tracking.c (delete_vta_debug_insn): New inline function.
5571 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5572 insns from get_insns () to NULL instead of each bb separately.
5573 Use delete_vta_debug_insn. No longer static.
5574 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5575 delete_vta_debug_insns callers.
5576 * rtl.h (delete_vta_debug_insns): Declare.
5577 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5578 instead of variable_tracking_main.
5580 2018-01-03 Martin Sebor <msebor@redhat.com>
5582 PR tree-optimization/83603
5583 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5584 arguments past the endof the argument list in functions declared
5585 without a prototype.
5586 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5587 Avoid checking when arguments are null.
5589 2018-01-03 Martin Sebor <msebor@redhat.com>
5592 * doc/extend.texi (attribute const): Fix a typo.
5593 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5594 issuing -Wsuggest-attribute for void functions.
5596 2018-01-03 Martin Sebor <msebor@redhat.com>
5598 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5599 offset_int::from instead of wide_int::to_shwi.
5600 (maybe_diag_overlap): Remove assertion.
5601 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5602 * gimple-ssa-sprintf.c (format_directive): Same.
5603 (parse_directive): Same.
5604 (sprintf_dom_walker::compute_format_length): Same.
5605 (try_substitute_return_value): Same.
5607 2018-01-03 Jeff Law <law@redhat.com>
5610 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5611 non-constant residual for zero at runtime and avoid probing in
5612 that case. Reorganize code for trailing problem to mirror handling
5615 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5617 PR tree-optimization/83501
5618 * tree-ssa-strlen.c (get_string_cst): New.
5619 (handle_char_store): Call get_string_cst.
5621 2018-01-03 Martin Liska <mliska@suse.cz>
5623 PR tree-optimization/83593
5624 * tree-ssa-strlen.c: Include tree-cfg.h.
5625 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5626 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5627 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5629 (strlen_dom_walker::before_dom_children): Call
5630 gimple_purge_dead_eh_edges. Dump tranformation with details
5632 (strlen_dom_walker::before_dom_children): Update call by adding
5633 new argument cleanup_eh.
5634 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5636 2018-01-03 Martin Liska <mliska@suse.cz>
5639 * cif-code.def (VARIADIC_THUNK): New enum value.
5640 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5643 2018-01-03 Jan Beulich <jbeulich@suse.com>
5645 * sse.md (mov<mode>_internal): Tighten condition for when to use
5646 vmovdqu<ssescalarsize> for TI and OI modes.
5648 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5650 Update copyright years.
5652 2018-01-03 Martin Liska <mliska@suse.cz>
5655 * ipa-visibility.c (function_and_variable_visibility): Skip
5656 functions with noipa attribure.
5658 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5660 * gcc.c (process_command): Update copyright notice dates.
5661 * gcov-dump.c (print_version): Ditto.
5662 * gcov.c (print_version): Ditto.
5663 * gcov-tool.c (print_version): Ditto.
5664 * gengtype.c (create_file): Ditto.
5665 * doc/cpp.texi: Bump @copying's copyright year.
5666 * doc/cppinternals.texi: Ditto.
5667 * doc/gcc.texi: Ditto.
5668 * doc/gccint.texi: Ditto.
5669 * doc/gcov.texi: Ditto.
5670 * doc/install.texi: Ditto.
5671 * doc/invoke.texi: Ditto.
5673 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5675 * vector-builder.h (vector_builder::m_full_nelts): Change from
5676 unsigned int to poly_uint64.
5677 (vector_builder::full_nelts): Update prototype accordingly.
5678 (vector_builder::new_vector): Likewise.
5679 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5680 (vector_builder::operator ==): Likewise.
5681 (vector_builder::finalize): Likewise.
5682 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5683 Take the number of elements as a poly_uint64 rather than an
5685 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5686 from unsigned int to poly_uint64.
5687 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5688 (vec_perm_indices::new_vector): Likewise.
5689 (vec_perm_indices::length): Likewise.
5690 (vec_perm_indices::nelts_per_input): Likewise.
5691 (vec_perm_indices::input_nelts): Likewise.
5692 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5693 number of elements per input as a poly_uint64 rather than an
5694 unsigned int. Use the original encoding for variable-length
5695 vectors, rather than clamping each individual element.
5696 For the second and subsequent elements in each pattern,
5697 clamp the step and base before clamping their sum.
5698 (vec_perm_indices::series_p): Handle polynomial element counts.
5699 (vec_perm_indices::all_in_range_p): Likewise.
5700 (vec_perm_indices_to_tree): Likewise.
5701 (vec_perm_indices_to_rtx): Likewise.
5702 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5703 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5704 (tree_vector_builder::new_binary_operation): Handle polynomial
5705 element counts. Return false if we need to know the number
5706 of elements at compile time.
5707 * fold-const.c (fold_vec_perm): Punt if the number of elements
5708 isn't known at compile time.
5710 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5712 * vec-perm-indices.h (vec_perm_builder): Change element type
5713 from HOST_WIDE_INT to poly_int64.
5714 (vec_perm_indices::element_type): Update accordingly.
5715 (vec_perm_indices::clamp): Handle polynomial element_types.
5716 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5717 (vec_perm_indices::all_in_range_p): Likewise.
5718 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5720 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5721 polynomial vec_perm_indices element types.
5722 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5723 * fold-const.c (fold_vec_perm): Likewise.
5724 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5725 * tree-vect-generic.c (lower_vec_perm): Likewise.
5726 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5727 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5728 element type to HOST_WIDE_INT.
5730 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5731 Alan Hayward <alan.hayward@arm.com>
5732 David Sherwood <david.sherwood@arm.com>
5734 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5735 rather than an int. Use plus_constant.
5736 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5737 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5740 Alan Hayward <alan.hayward@arm.com>
5741 David Sherwood <david.sherwood@arm.com>
5743 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5744 a HOST_WIDE_INT to a poly_int64.
5746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5747 Alan Hayward <alan.hayward@arm.com>
5748 David Sherwood <david.sherwood@arm.com>
5750 * calls.c (load_register_parameters): Cope with polynomial
5751 mode sizes. Require a constant size for BLKmode parameters
5752 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5753 forces a parameter to be padded at the lsb end in order to
5754 fill a complete number of words, require the parameter size
5755 to be ordered wrt UNITS_PER_WORD.
5757 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5758 Alan Hayward <alan.hayward@arm.com>
5759 David Sherwood <david.sherwood@arm.com>
5761 * reload1.c (spill_stack_slot_width): Change element type
5762 from unsigned int to poly_uint64_pod.
5763 (alter_reg): Treat mode sizes as polynomial.
5765 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5766 Alan Hayward <alan.hayward@arm.com>
5767 David Sherwood <david.sherwood@arm.com>
5769 * reload.c (complex_word_subreg_p): New function.
5770 (reload_inner_reg_of_subreg, push_reload): Use it.
5772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5773 Alan Hayward <alan.hayward@arm.com>
5774 David Sherwood <david.sherwood@arm.com>
5776 * lra-constraints.c (process_alt_operands): Reject matched
5777 operands whose sizes aren't ordered.
5778 (match_reload): Refer to this check here.
5780 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5781 Alan Hayward <alan.hayward@arm.com>
5782 David Sherwood <david.sherwood@arm.com>
5784 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5785 that the mode size is in the set {1, 2, 4, 8, 16}.
5787 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5788 Alan Hayward <alan.hayward@arm.com>
5789 David Sherwood <david.sherwood@arm.com>
5791 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5792 Use plus_constant instead of gen_rtx_PLUS.
5794 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5795 Alan Hayward <alan.hayward@arm.com>
5796 David Sherwood <david.sherwood@arm.com>
5798 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5799 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5800 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5801 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5802 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5803 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5804 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5805 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5806 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5807 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5809 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5810 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5811 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5812 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5813 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5814 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5815 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5816 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5817 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5818 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5820 * expr.c (emit_move_resolve_push): Treat the input and result
5821 of PUSH_ROUNDING as a poly_int64.
5822 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5823 (emit_push_insn): Likewise.
5824 * lra-eliminations.c (mark_not_eliminable): Likewise.
5825 * recog.c (push_operand): Likewise.
5826 * reload1.c (elimination_effects): Likewise.
5827 * rtlanal.c (nonzero_bits1): Likewise.
5828 * calls.c (store_one_arg): Likewise. Require the padding to be
5829 known at compile time.
5831 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5832 Alan Hayward <alan.hayward@arm.com>
5833 David Sherwood <david.sherwood@arm.com>
5835 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5836 Use plus_constant instead of gen_rtx_PLUS.
5838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5839 Alan Hayward <alan.hayward@arm.com>
5840 David Sherwood <david.sherwood@arm.com>
5842 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5845 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5846 Alan Hayward <alan.hayward@arm.com>
5847 David Sherwood <david.sherwood@arm.com>
5849 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5850 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5851 via stack temporaries. Treat the mode size as polynomial too.
5853 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5854 Alan Hayward <alan.hayward@arm.com>
5855 David Sherwood <david.sherwood@arm.com>
5857 * expr.c (expand_expr_real_2): When handling conversions involving
5858 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5859 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5860 as a poly_uint64 too.
5862 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5863 Alan Hayward <alan.hayward@arm.com>
5864 David Sherwood <david.sherwood@arm.com>
5866 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5868 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5869 Alan Hayward <alan.hayward@arm.com>
5870 David Sherwood <david.sherwood@arm.com>
5872 * combine.c (can_change_dest_mode): Handle polynomial
5873 REGMODE_NATURAL_SIZE.
5874 * expmed.c (store_bit_field_1): Likewise.
5875 * expr.c (store_constructor): Likewise.
5876 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5877 and polynomial REGMODE_NATURAL_SIZE.
5878 (gen_lowpart_common): Likewise.
5879 * reginfo.c (record_subregs_of_mode): Likewise.
5880 * rtlanal.c (read_modify_subreg_p): Likewise.
5882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5883 Alan Hayward <alan.hayward@arm.com>
5884 David Sherwood <david.sherwood@arm.com>
5886 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5887 numbers of elements.
5889 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5890 Alan Hayward <alan.hayward@arm.com>
5891 David Sherwood <david.sherwood@arm.com>
5893 * match.pd: Cope with polynomial numbers of vector elements.
5895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5896 Alan Hayward <alan.hayward@arm.com>
5897 David Sherwood <david.sherwood@arm.com>
5899 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5900 in a POINTER_PLUS_EXPR.
5902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5903 Alan Hayward <alan.hayward@arm.com>
5904 David Sherwood <david.sherwood@arm.com>
5906 * omp-simd-clone.c (simd_clone_subparts): New function.
5907 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5908 (ipa_simd_modify_function_body): Likewise.
5910 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5911 Alan Hayward <alan.hayward@arm.com>
5912 David Sherwood <david.sherwood@arm.com>
5914 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5915 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5916 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5917 (expand_vector_condition, vector_element): Likewise.
5918 (subparts_gt): New function.
5919 (get_compute_type): Use subparts_gt.
5920 (count_type_subparts): Delete.
5921 (expand_vector_operations_1): Use subparts_gt instead of
5922 count_type_subparts.
5924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5925 Alan Hayward <alan.hayward@arm.com>
5926 David Sherwood <david.sherwood@arm.com>
5928 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5929 (vect_compile_time_alias): ...this new function. Do the calculation
5930 on poly_ints rather than trees.
5931 (vect_prune_runtime_alias_test_list): Update call accordingly.
5933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5934 Alan Hayward <alan.hayward@arm.com>
5935 David Sherwood <david.sherwood@arm.com>
5937 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5939 (vect_schedule_slp_instance): Likewise.
5941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5942 Alan Hayward <alan.hayward@arm.com>
5943 David Sherwood <david.sherwood@arm.com>
5945 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5946 constant and extern definitions for variable-length vectors.
5947 (vect_get_constant_vectors): Note that the number of units
5948 is known to be constant.
5950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5951 Alan Hayward <alan.hayward@arm.com>
5952 David Sherwood <david.sherwood@arm.com>
5954 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5955 of units as polynomial. Choose between WIDE and NARROW based
5958 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5959 Alan Hayward <alan.hayward@arm.com>
5960 David Sherwood <david.sherwood@arm.com>
5962 * tree-vect-stmts.c (simd_clone_subparts): New function.
5963 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5965 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5966 Alan Hayward <alan.hayward@arm.com>
5967 David Sherwood <david.sherwood@arm.com>
5969 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5970 vectors as polynomial. Use build_index_vector for
5973 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5974 Alan Hayward <alan.hayward@arm.com>
5975 David Sherwood <david.sherwood@arm.com>
5977 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5978 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5979 for variable-length vectors.
5980 (vectorizable_mask_load_store): Treat the number of units as
5981 polynomial, asserting that it is constant if the condition has
5982 already been enforced.
5983 (vectorizable_store, vectorizable_load): Likewise.
5985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5986 Alan Hayward <alan.hayward@arm.com>
5987 David Sherwood <david.sherwood@arm.com>
5989 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5990 of units as polynomial. Punt if we can't tell at compile time
5991 which vector contains the final result.
5993 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5994 Alan Hayward <alan.hayward@arm.com>
5995 David Sherwood <david.sherwood@arm.com>
5997 * tree-vect-loop.c (vectorizable_induction): Treat the number
5998 of units as polynomial. Punt on SLP inductions. Use an integer
5999 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6000 cast of such a series for variable-length floating-point
6003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6004 Alan Hayward <alan.hayward@arm.com>
6005 David Sherwood <david.sherwood@arm.com>
6007 * tree.h (build_index_vector): Declare.
6008 * tree.c (build_index_vector): New function.
6009 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6010 of units as polynomial, forcibly converting it to a constant if
6011 vectorizable_reduction has already enforced the condition.
6012 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6013 to create a {1,2,3,...} vector.
6014 (vectorizable_reduction): Treat the number of units as polynomial.
6015 Choose vectype_in based on the largest scalar element size rather
6016 than the smallest number of units. Enforce the restrictions
6019 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6020 Alan Hayward <alan.hayward@arm.com>
6021 David Sherwood <david.sherwood@arm.com>
6023 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6024 number of units as polynomial.
6026 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6027 Alan Hayward <alan.hayward@arm.com>
6028 David Sherwood <david.sherwood@arm.com>
6030 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6031 * target.def (autovectorize_vector_sizes): Return the vector sizes
6032 by pointer, using vector_sizes rather than a bitmask.
6033 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6034 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6035 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6037 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6038 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6039 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6040 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6041 * omp-general.c (omp_max_vf): Likewise.
6042 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6043 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6044 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6045 * tree-vect-slp.c (vect_slp_bb): Likewise.
6046 * doc/tm.texi: Regenerate.
6047 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6049 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6050 the vector size as a poly_uint64 rather than an unsigned int.
6051 (current_vector_size): Change from an unsigned int to a poly_uint64.
6052 (get_vectype_for_scalar_type): Update accordingly.
6053 * tree.h (build_truth_vector_type): Take the size and number of
6054 units as a poly_uint64 rather than an unsigned int.
6055 (build_vector_type): Add a temporary overload that takes
6056 the number of units as a poly_uint64 rather than an unsigned int.
6057 * tree.c (make_vector_type): Likewise.
6058 (build_truth_vector_type): Take the number of units as a poly_uint64
6059 rather than an unsigned int.
6061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6062 Alan Hayward <alan.hayward@arm.com>
6063 David Sherwood <david.sherwood@arm.com>
6065 * target.def (get_mask_mode): Take the number of units and length
6066 as poly_uint64s rather than unsigned ints.
6067 * targhooks.h (default_get_mask_mode): Update accordingly.
6068 * targhooks.c (default_get_mask_mode): Likewise.
6069 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6070 * doc/tm.texi: Regenerate.
6072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6073 Alan Hayward <alan.hayward@arm.com>
6074 David Sherwood <david.sherwood@arm.com>
6076 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6077 * omp-general.c (omp_max_vf): Likewise.
6078 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6079 (expand_omp_simd): Handle polynomial safelen.
6080 * omp-low.c (omplow_simd_context): Add a default constructor.
6081 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6082 (lower_rec_simd_input_clauses): Update accordingly.
6083 (lower_rec_input_clauses): Likewise.
6085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6086 Alan Hayward <alan.hayward@arm.com>
6087 David Sherwood <david.sherwood@arm.com>
6089 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6090 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6091 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6092 (vect_analyze_slp_cost): Likewise.
6093 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6094 (vect_model_load_cost): Likewise.
6096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6097 Alan Hayward <alan.hayward@arm.com>
6098 David Sherwood <david.sherwood@arm.com>
6100 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6101 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6102 from an unsigned int * to a poly_uint64_pod *.
6103 (calculate_unrolling_factor): New function.
6104 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6107 Alan Hayward <alan.hayward@arm.com>
6108 David Sherwood <david.sherwood@arm.com>
6110 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6111 from an unsigned int to a poly_uint64.
6112 (_loop_vec_info::slp_unrolling_factor): Likewise.
6113 (_loop_vec_info::vectorization_factor): Change from an int
6115 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6116 (vect_get_num_vectors): New function.
6117 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6118 (vect_get_num_copies): Use vect_get_num_vectors.
6119 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6120 to an unsigned int *.
6121 (vect_analyze_data_refs): Change min_vf from an int * to a
6123 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6124 than an unsigned HOST_WIDE_INT.
6125 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6126 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6127 to an unsigned int *.
6128 (vect_analyze_data_ref_dependences): Likewise.
6129 (vect_compute_data_ref_alignment): Handle polynomial vf.
6130 (vect_enhance_data_refs_alignment): Likewise.
6131 (vect_prune_runtime_alias_test_list): Likewise.
6132 (vect_shift_permute_load_chain): Likewise.
6133 (vect_supportable_dr_alignment): Likewise.
6134 (dependence_distance_ge_vf): Take the vectorization factor as a
6135 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6136 (vect_analyze_data_refs): Change min_vf from an int * to a
6138 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6139 vfm1 as a poly_uint64 rather than an int. Make the same change
6140 for the returned bound_scalar.
6141 (vect_gen_vector_loop_niters): Handle polynomial vf.
6142 (vect_do_peeling): Likewise. Update call to
6143 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6144 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6146 * tree-vect-loop.c (vect_determine_vectorization_factor)
6147 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6148 (vect_get_known_peeling_cost): Likewise.
6149 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6150 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6151 (vect_transform_loop): Likewise. Use the lowest possible VF when
6152 updating the upper bounds of the loop.
6153 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6155 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6156 polynomial unroll factors.
6157 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6158 (vect_make_slp_decision): Likewise.
6159 (vect_supported_load_permutation_p): Likewise, and polynomial
6161 (vect_analyze_slp_cost): Handle polynomial vf.
6162 (vect_slp_analyze_node_operations): Likewise.
6163 (vect_slp_analyze_bb_1): Likewise.
6164 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6165 than an unsigned HOST_WIDE_INT.
6166 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6167 (vectorizable_load): Handle polynomial vf.
6168 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6170 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6172 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6173 Alan Hayward <alan.hayward@arm.com>
6174 David Sherwood <david.sherwood@arm.com>
6176 * match.pd: Handle bit operations involving three constants
6177 and try to fold one pair.
6179 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6181 * tree-vect-loop-manip.c: Include gimple-fold.h.
6182 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6183 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6184 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6185 Add a path that uses a step of VF instead of 1, but disable it
6187 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6188 and niters_no_overflow parameters. Update calls to
6189 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6190 Create a new SSA name if the latter choses to use a ste other
6191 than zero, and return it via niters_vector_mult_vf_var.
6192 * tree-vect-loop.c (vect_transform_loop): Update calls to
6193 vect_do_peeling, vect_gen_vector_loop_niters and
6194 slpeel_make_loop_iterate_ntimes.
6195 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6196 (vect_gen_vector_loop_niters): Update declarations after above changes.
6198 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6200 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6201 128-bit round to integer instructions.
6202 (ceil<mode>2): Likewise.
6203 (btrunc<mode>2): Likewise.
6204 (round<mode>2): Likewise.
6206 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6208 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6209 unaligned VSX load/store on P8/P9.
6210 (expand_block_clear): Allow the use of unaligned VSX
6211 load/store on P8/P9.
6213 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6215 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6217 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6218 swap associated with both a load and a store.
6220 2018-01-02 Andrew Waterman <andrew@sifive.com>
6222 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6223 * config/riscv/riscv.md (clear_cache): Use it.
6225 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6227 * web.c: Remove out-of-date comment.
6229 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6231 * expr.c (fixup_args_size_notes): Check that any existing
6232 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6233 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6234 (emit_single_push_insn): ...here.
6236 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6238 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6239 (const_vector_encoded_nelts): New function.
6240 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6241 (const_vector_int_elt, const_vector_elt): Declare.
6242 * emit-rtl.c (const_vector_int_elt_1): New function.
6243 (const_vector_elt): Likewise.
6244 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6245 of CONST_VECTOR_ELT.
6247 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6249 * expr.c: Include rtx-vector-builder.h.
6250 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6251 directly on the tree encoding.
6252 (const_vector_from_tree): Likewise.
6253 * optabs.c: Include rtx-vector-builder.h.
6254 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6255 sequence of "u" values.
6256 * vec-perm-indices.c: Include rtx-vector-builder.h.
6257 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6258 directly on the vec_perm_indices encoding.
6260 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6262 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6263 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6264 * rtx-vector-builder.h: New file.
6265 * rtx-vector-builder.c: Likewise.
6266 * rtl.h (rtx_def::u2): Add a const_vector field.
6267 (CONST_VECTOR_NPATTERNS): New macro.
6268 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6269 (CONST_VECTOR_DUPLICATE_P): Likewise.
6270 (CONST_VECTOR_STEPPED_P): Likewise.
6271 (CONST_VECTOR_ENCODED_ELT): Likewise.
6272 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6273 (unwrap_const_vec_duplicate): Likewise.
6274 (const_vec_series_p): Check for a non-duplicated vector encoding.
6275 Say that the function only returns true for integer vectors.
6276 * emit-rtl.c: Include rtx-vector-builder.h.
6277 (gen_const_vec_duplicate_1): Delete.
6278 (gen_const_vector): Call gen_const_vec_duplicate instead of
6279 gen_const_vec_duplicate_1.
6280 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6281 (gen_const_vec_duplicate): Use rtx_vector_builder.
6282 (gen_const_vec_series): Likewise.
6283 (gen_rtx_CONST_VECTOR): Likewise.
6284 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6285 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6286 Build a new vector rather than modifying a CONST_VECTOR in-place.
6287 (handle_special_swappables): Update call accordingly.
6288 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6289 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6290 Build a new vector rather than modifying a CONST_VECTOR in-place.
6291 (handle_special_swappables): Update call accordingly.
6293 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6295 * simplify-rtx.c (simplify_const_binary_operation): Use
6296 CONST_VECTOR_ELT instead of XVECEXP.
6298 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6300 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6301 the selector elements to be different from the data elements
6302 if the selector is a VECTOR_CST.
6303 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6304 ssizetype for the selector.
6306 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6308 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6309 before testing each element individually.
6310 * tree-vect-generic.c (lower_vec_perm): Likewise.
6312 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6314 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6315 * selftest-run-tests.c (selftest::run_tests): Call it.
6316 * vector-builder.h (vector_builder::operator ==): New function.
6317 (vector_builder::operator !=): Likewise.
6318 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6319 (vec_perm_indices::all_from_input_p): New function.
6320 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6321 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6322 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6323 instead of reading the VECTOR_CST directly. Detect whether both
6324 vector inputs are the same before constructing the vec_perm_indices,
6325 and update the number of inputs argument accordingly. Use the
6326 utility functions added above. Only construct sel2 if we need to.
6328 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6330 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6331 the broadcast of the low byte.
6332 (expand_mult_highpart): Use an explicit encoding for the permutes.
6333 * optabs-query.c (can_mult_highpart_p): Likewise.
6334 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6335 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6336 (vectorizable_bswap): Likewise.
6337 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6338 explicit encoding for the power-of-2 permutes.
6339 (vect_permute_store_chain): Likewise.
6340 (vect_grouped_load_supported): Likewise.
6341 (vect_permute_load_chain): Likewise.
6343 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6345 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6346 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6347 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6348 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6349 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6350 (vect_gen_perm_mask_any): Likewise.
6352 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6354 * int-vector-builder.h: New file.
6355 * vec-perm-indices.h: Include int-vector-builder.h.
6356 (vec_perm_indices): Redefine as an int_vector_builder.
6357 (auto_vec_perm_indices): Delete.
6358 (vec_perm_builder): Redefine as a stand-alone class.
6359 (vec_perm_indices::vec_perm_indices): New function.
6360 (vec_perm_indices::clamp): Likewise.
6361 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6362 (vec_perm_indices::new_vector): New function.
6363 (vec_perm_indices::new_expanded_vector): Update for new
6364 vec_perm_indices class.
6365 (vec_perm_indices::rotate_inputs): New function.
6366 (vec_perm_indices::all_in_range_p): Operate directly on the
6367 encoded form, without computing elided elements.
6368 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6369 encoding. Update for new vec_perm_indices class.
6370 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6371 the given vec_perm_builder.
6372 (expand_vec_perm_var): Update vec_perm_builder constructor.
6373 (expand_mult_highpart): Use vec_perm_builder instead of
6374 auto_vec_perm_indices.
6375 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6376 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6377 or double series encoding as appropriate.
6378 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6379 vec_perm_indices instead of auto_vec_perm_indices.
6380 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6381 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6382 (vect_permute_store_chain): Likewise.
6383 (vect_grouped_load_supported): Likewise.
6384 (vect_permute_load_chain): Likewise.
6385 (vect_shift_permute_load_chain): Likewise.
6386 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6387 (vect_transform_slp_perm_load): Likewise.
6388 (vect_schedule_slp_instance): Likewise.
6389 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6390 (vectorizable_mask_load_store): Likewise.
6391 (vectorizable_bswap): Likewise.
6392 (vectorizable_store): Likewise.
6393 (vectorizable_load): Likewise.
6394 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6395 vec_perm_indices instead of auto_vec_perm_indices. Use
6396 tree_to_vec_perm_builder to read the vector from a tree.
6397 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6398 vec_perm_builder instead of a vec_perm_indices.
6399 (have_whole_vector_shift): Use vec_perm_builder and
6400 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6401 truncation to calc_vec_perm_mask_for_shift.
6402 (vect_create_epilog_for_reduction): Likewise.
6403 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6404 from auto_vec_perm_indices to vec_perm_indices.
6405 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6406 instead of changing individual elements.
6407 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6408 the vector in d.perm.
6409 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6410 from auto_vec_perm_indices to vec_perm_indices.
6411 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6412 instead of changing individual elements.
6413 (arm_vectorize_vec_perm_const): Use new_vector to install
6414 the vector in d.perm.
6415 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6416 Update vec_perm_builder constructor.
6417 (rs6000_expand_interleave): Likewise.
6418 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6419 (rs6000_expand_interleave): Likewise.
6421 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6423 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6424 to qimode could truncate the indices.
6425 * optabs.c (expand_vec_perm_var): Likewise.
6427 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6429 * Makefile.in (OBJS): Add vec-perm-indices.o.
6430 * vec-perm-indices.h: New file.
6431 * vec-perm-indices.c: Likewise.
6432 * target.h (vec_perm_indices): Replace with a forward class
6434 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6435 * optabs.h: Include vec-perm-indices.h.
6436 (expand_vec_perm): Delete.
6437 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6438 (expand_vec_perm_const): Declare.
6439 * target.def (vec_perm_const_ok): Replace with...
6440 (vec_perm_const): ...this new hook.
6441 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6442 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6443 * doc/tm.texi: Regenerate.
6444 * optabs.def (vec_perm_const): Delete.
6445 * doc/md.texi (vec_perm_const): Likewise.
6446 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6447 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6448 expand_vec_perm for constant permutation vectors. Assert that
6449 the mode of variable permutation vectors is the integer equivalent
6450 of the mode that is being permuted.
6451 * optabs-query.h (selector_fits_mode_p): Declare.
6452 * optabs-query.c: Include vec-perm-indices.h.
6453 (selector_fits_mode_p): New function.
6454 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6455 is defined, instead of checking whether the vec_perm_const_optab
6456 exists. Use targetm.vectorize.vec_perm_const instead of
6457 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6458 fit in the vector mode before using a variable permute.
6459 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6460 vec_perm_indices instead of an rtx.
6461 (expand_vec_perm): Replace with...
6462 (expand_vec_perm_const): ...this new function. Take the selector
6463 as a vec_perm_indices rather than an rtx. Also take the mode of
6464 the selector. Update call to shift_amt_for_vec_perm_mask.
6465 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6466 Use vec_perm_indices::new_expanded_vector to expand the original
6467 selector into bytes. Check whether the indices fit in the vector
6468 mode before using a variable permute.
6469 (expand_vec_perm_var): Make global.
6470 (expand_mult_highpart): Use expand_vec_perm_const.
6471 * fold-const.c: Includes vec-perm-indices.h.
6472 * tree-ssa-forwprop.c: Likewise.
6473 * tree-vect-data-refs.c: Likewise.
6474 * tree-vect-generic.c: Likewise.
6475 * tree-vect-loop.c: Likewise.
6476 * tree-vect-slp.c: Likewise.
6477 * tree-vect-stmts.c: Likewise.
6478 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6480 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6481 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6482 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6483 (aarch64_vectorize_vec_perm_const): ...this new function.
6484 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6485 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6486 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6487 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6488 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6489 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6490 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6492 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6493 check for NEON modes.
6494 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6495 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6496 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6497 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6499 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6500 the old VEC_PERM_CONST conditions.
6501 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6502 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6503 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6504 (ia64_vectorize_vec_perm_const_ok): Merge into...
6505 (ia64_vectorize_vec_perm_const): ...this new function.
6506 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6507 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6508 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6509 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6510 * config/mips/mips.c (mips_expand_vec_perm_const)
6511 (mips_vectorize_vec_perm_const_ok): Merge into...
6512 (mips_vectorize_vec_perm_const): ...this new function.
6513 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6514 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6515 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6516 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6517 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6518 (rs6000_expand_vec_perm_const): Delete.
6519 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6521 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6522 (altivec_expand_vec_perm_const_le): Take each operand individually.
6523 Operate on constant selectors rather than rtxes.
6524 (altivec_expand_vec_perm_const): Likewise. Update call to
6525 altivec_expand_vec_perm_const_le.
6526 (rs6000_expand_vec_perm_const): Delete.
6527 (rs6000_vectorize_vec_perm_const_ok): Delete.
6528 (rs6000_vectorize_vec_perm_const): New function.
6529 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6530 an element count and rtx array.
6531 (rs6000_expand_extract_even): Update call accordingly.
6532 (rs6000_expand_interleave): Likewise.
6533 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6534 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6535 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6536 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6537 (rs6000_expand_vec_perm_const): Delete.
6538 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6539 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6540 (altivec_expand_vec_perm_const_le): Take each operand individually.
6541 Operate on constant selectors rather than rtxes.
6542 (altivec_expand_vec_perm_const): Likewise. Update call to
6543 altivec_expand_vec_perm_const_le.
6544 (rs6000_expand_vec_perm_const): Delete.
6545 (rs6000_vectorize_vec_perm_const_ok): Delete.
6546 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6547 reference to the SPE evmerge intructions.
6548 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6549 an element count and rtx array.
6550 (rs6000_expand_extract_even): Update call accordingly.
6551 (rs6000_expand_interleave): Likewise.
6552 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6553 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6555 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6557 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6559 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6560 vector mode and that that mode matches the mode of the data
6562 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6563 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6564 directly using expand_vec_perm_1 when forcing selectors into
6566 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6568 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6570 * optabs-query.h (can_vec_perm_p): Delete.
6571 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6572 * optabs-query.c (can_vec_perm_p): Split into...
6573 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6574 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6575 particular selector is valid.
6576 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6577 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6578 (vect_grouped_load_supported): Likewise.
6579 (vect_shift_permute_load_chain): Likewise.
6580 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6581 (vect_transform_slp_perm_load): Likewise.
6582 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6583 (vectorizable_bswap): Likewise.
6584 (vect_gen_perm_mask_checked): Likewise.
6585 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6586 implementations of variable permutation vectors into account
6587 when deciding which selector to use.
6588 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6589 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6590 with a false third argument.
6591 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6592 to test whether the constant selector is valid and can_vec_perm_var_p
6593 to test whether a variable selector is valid.
6595 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6597 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6598 * optabs-query.c (can_vec_perm_p): Likewise.
6599 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6600 instead of vec_perm_indices.
6601 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6602 (vect_gen_perm_mask_checked): Likewise,
6603 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6604 (vect_gen_perm_mask_checked): Likewise,
6606 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6608 * optabs-query.h (qimode_for_vec_perm): Declare.
6609 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6610 (qimode_for_vec_perm): ...this new function.
6611 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6613 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6615 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6616 does not have a conditional at the top.
6618 2018-01-02 Richard Biener <rguenther@suse.de>
6620 * ipa-inline.c (big_speedup_p): Fix expression.
6622 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6625 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6628 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6632 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6633 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6634 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6635 cond_taken_branch_cost 3->4.
6637 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6639 PR tree-optimization/83581
6640 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6641 TODO_cleanup_cfg if any changes have been made.
6644 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6645 convert_modes if target mode has the right side, but different mode
6649 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6650 last argument when extracting from CONCAT. If either from_real or
6651 from_imag is NULL, use expansion through memory. If result is not
6652 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6653 the parts directly to inner mode, if even that fails, use expansion
6657 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6658 check for bswap in mode rather than HImode and use that in expand_unop
6661 Copyright (C) 2018 Free Software Foundation, Inc.
6663 Copying and distribution of this file, with or without modification,
6664 are permitted in any medium without royalty provided the copyright
6665 notice and this notice are preserved.