1 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
3 *config/rx.md: updated "movsicc" expand to be matched by GCC
4 *testsuite/gcc.target/rx/movsicc.c: new test case
6 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
9 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
10 multiply in 32-bit mode.
11 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
12 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
15 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
17 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
18 to allow or block "symbol_ref" depending on value of TARGET_JSR
19 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
20 call_value_internal insns
22 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
25 * dwarf2out.c (is_trivial_indirect_ref): New function.
26 (dwarf2out_late_global_decl): Do not generate a location
27 attribute for variables that have a non-trivial DECL_VALUE_EXPR
28 and that are not defined in the current unit.
30 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
32 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
33 instead of a libcall for UNORDERED.
35 2018-02-09 Tamar Christina <tamar.christina@arm.com>
38 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
39 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
41 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
44 * config/s390/s390.c (s390_set_current_function): Invoke
45 s390_indirect_branch_settings also if fndecl didn't change.
47 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
49 * config/rs6000/rs6000.md (blockage): Set length to zero.
51 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
53 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
55 2018-02-09 Jakub Jelinek <jakub@redhat.com>
58 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
59 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
63 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
64 PARALLEL incoming that failed vt_get_decl_and_offset check.
67 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
68 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
70 (get_variable_section): For decls in named .bss* sections pass true as
71 second argument to bss_initializer_p.
73 2018-02-09 Marek Polacek <polacek@redhat.com>
74 Jakub Jelinek <jakub@redhat.com>
77 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
78 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
79 Sync some changes from cxx_fold_indirect_ref.
81 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
83 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
85 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
86 (BLOCK_INLINE_ENTRY_LABEL): New.
87 (dwarf2out_var_location): Disregard inline entry markers.
88 (inline_entry_data): New struct.
89 (inline_entry_data_hasher): New hashtable type.
90 (inline_entry_data_hasher::hash): New.
91 (inline_entry_data_hasher::equal): New.
92 (inline_entry_data_table): New variable.
93 (add_high_low_attributes): Add DW_AT_entry_pc and
94 DW_AT_GNU_entry_view attributes if a pending entry is found
95 in inline_entry_data_table. Add old entry_pc attribute only
96 if debug nonbinding markers are disabled.
97 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
99 (block_within_block_p, dwarf2out_inline_entry): New.
100 (dwarf2out_finish): Check that no entries remained in
101 inline_entry_data_table.
102 * final.c (reemit_insn_block_notes): Handle inline entry notes.
103 (final_scan_insn, notice_source_line): Likewise.
104 (rest_of_clean_state): Skip inline entry markers.
105 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
107 * gimple.c (gimple_build_debug_inline_entry): New.
108 * gimple.h (enum gimple_debug_subcode): Add
109 GIMPLE_DEBUG_INLINE_ENTRY.
110 (gimple_build_debug_inline_entry): Declare.
111 (gimple_debug_inline_entry_p): New.
112 (gimple_debug_nonbind_marker_p): Adjust.
113 * insn-notes.def (INLINE_ENTRY): New.
114 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
115 inline entry marker notes.
116 (print_insn): Likewise.
117 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
118 (INSN_DEBUG_MARKER_KIND): Likewise.
119 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
120 * tree-inline.c (expand_call_inline): Build and insert
121 debug_inline_entry stmt.
122 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
123 inline entry blocks early, if nonbind markers are enabled.
124 (dump_scope_block): Dump fragment info.
125 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
126 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
127 (gimple_build_debug_inline_entry): New.
128 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
129 Enable/disable inline entry points too.
130 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
131 (DEBUG_INSN): Describe inline entry markers.
133 * common.opt (gvariable-location-views): New.
134 (gvariable-location-views=incompat5): New.
135 * config.in: Rebuilt.
136 * configure: Rebuilt.
137 * configure.ac: Test assembler for view support.
138 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
139 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
140 * dwarf2out.c (var_loc_view): New typedef.
141 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
142 (dwarf2out_locviews_in_attribute): New.
143 (dwarf2out_locviews_in_loclist): New.
144 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
145 (enum dw_line_info_opcode): Add LI_adv_address.
146 (struct dw_line_info_table): Add view.
147 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
148 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
149 (zero_view_p): New variable.
150 (ZERO_VIEW_P): New macro.
151 (output_asm_line_debug_info): New.
152 (struct var_loc_node): Add view.
153 (add_AT_view_list, AT_loc_list): New.
154 (add_var_loc_to_decl): Add view param. Test it against last.
155 (new_loc_list): Add view params. Record them.
156 (AT_loc_list_ptr): Handle loc and view lists.
157 (view_list_to_loc_list_val_node): New.
158 (print_dw_val): Handle dw_val_class_view_list.
159 (size_of_die): Likewise.
160 (value_format): Likewise.
161 (loc_list_has_views): New.
162 (gen_llsym): Set vl_symbol too.
163 (maybe_gen_llsym, skip_loc_list_entry): New.
164 (dwarf2out_maybe_output_loclist_view_pair): New.
165 (output_loc_list): Output view list or entries too.
166 (output_view_list_offset): New.
167 (output_die): Handle dw_val_class_view_list.
168 (output_dwarf_version): New.
169 (output_compilation_unit_header): Use it.
170 (output_skeleton_debug_sections): Likewise.
171 (output_rnglists, output_line_info): Likewise.
172 (output_pubnames, output_aranges): Update version comments.
173 (output_one_line_info_table): Output view numbers in asm comments.
174 (dw_loc_list): Determine current endview, pass it to new_loc_list.
175 Call maybe_gen_llsym.
176 (loc_list_from_tree_1): Adjust.
177 (add_AT_location_description): Create view list attribute if
178 needed, check it's absent otherwise.
179 (convert_cfa_to_fb_loc_list): Adjust.
180 (maybe_emit_file): Call output_asm_line_debug_info for test.
181 (dwarf2out_var_location): Reset views as needed. Precompute
182 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
184 (new_line_info_table): Reset next view.
185 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
186 (dwarf2out_source_line): Likewise. Output view resets and labels to
187 the assembler, or select appropriate line info opcodes.
188 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
189 (optimize_string_length): Catch it. Adjust.
190 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
191 dw_val_class_view_list, and remove it if no longer needed.
192 (hash_loc_list): Hash view numbers.
193 (loc_list_hasher::equal): Compare them.
194 (optimize_location_lists): Check whether a view list symbol is
195 needed, and whether the locview attribute is present, and
196 whether they match. Remove the locview attribute if no longer
198 (index_location_lists): Call skip_loc_list_entry for test.
199 (dwarf2out_finish): Call output_asm_line_debug_info for test.
200 Use output_dwarf_version.
201 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
202 (struct dw_val_node): Add val_view_list.
203 * final.c (SEEN_NEXT_VIEW): New.
204 (set_next_view_needed): New.
205 (clear_next_view_needed): New.
206 (maybe_output_next_view): New.
207 (final_start_function): Rename to...
208 (final_start_function_1): ... this. Take pointer to FIRST,
209 add SEEN parameter. Emit param bindings in the initial view.
210 (final_start_function): Reintroduce SEEN-less interface.
211 (final): Rename to...
212 (final_1): ... this. Take SEEN parameter. Output final pending
213 next view at the end.
214 (final): Reintroduce seen-less interface.
215 (final_scan_insn): Output pending next view before switching
216 sections or ending a block. Mark the next view as needed when
217 outputting variable locations. Notify debug backend of section
218 changes, and of location view changes.
219 (rest_of_handle_final): Adjust.
220 * toplev.c (process_options): Autodetect value for debug variable
221 location views option. Warn on incompat5 without -gdwarf-5.
222 * doc/invoke.texi (gvariable-location-views): New.
223 (gvariable-location-views=incompat5): New.
224 (gno-variable-location-views): New.
226 2018-02-08 David Malcolm <dmalcolm@redhat.com>
228 PR tree-optimization/84136
229 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
230 that the result of find_edge is non-NULL.
232 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
235 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
236 storing integer register in SImode. Fix cost of 256 and 512
237 byte aligned SSE register store.
239 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
241 * config/i386/i386.c (ix86_multiplication_cost): Fix
242 multiplication cost for TARGET_AVX512DQ.
244 2018-02-08 Marek Polacek <polacek@redhat.com>
246 PR tree-optimization/84238
247 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
250 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
252 PR tree-optimization/84265
253 * tree-vect-stmts.c (vectorizable_store): Don't treat
254 VMAT_CONTIGUOUS accesses as grouped.
255 (vectorizable_load): Likewise.
257 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
259 PR tree-optimization/81635
260 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
261 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
262 (test_round_for_mask): New functions.
263 (wide_int_cc_tests): Call test_round_for_mask.
264 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
265 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
266 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
267 range returned by get_range_info.
269 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
272 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
273 * symtab.c: Include builtins.h
274 (symtab_node::output_to_lto_symbol_table_p): Move here
275 from lto-streamer-out.c:output_symbol_p.
276 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
277 (output_symbol_p): Move all logic to symtab.c
278 (produce_symtab): Update.
280 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
282 * config/s390/s390-opts.h (enum indirect_branch): Define.
283 * config/s390/s390-protos.h (s390_return_addr_from_memory)
284 (s390_indirect_branch_via_thunk)
285 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
286 (enum s390_indirect_branch_type): Define.
287 * config/s390/s390.c (struct s390_frame_layout, struct
288 machine_function): Remove.
289 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
290 (indirect_branch_table_label_no, indirect_branch_table_name):
292 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
293 (enum s390_indirect_branch_option): Define.
294 (s390_return_addr_from_memory): New function.
295 (s390_handle_string_attribute): New function.
296 (s390_attribute_table): Add new attribute handler.
297 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
298 (s390_indirect_branch_via_thunk): New function.
299 (s390_indirect_branch_via_inline_thunk): New function.
300 (s390_function_ok_for_sibcall): When jumping via thunk disallow
301 sibling call optimization for non z10 compiles.
302 (s390_emit_call): Force indirect branch target to be a single
303 register. Add r1 clobber for non-z10 compiles.
304 (s390_emit_epilogue): Emit return jump via return_use expander.
305 (s390_reorg): Handle JUMP_INSNs as execute targets.
306 (s390_option_override_internal): Perform validity checks for the
307 new command line options.
308 (s390_indirect_branch_attrvalue): New function.
309 (s390_indirect_branch_settings): New function.
310 (s390_set_current_function): Invoke s390_indirect_branch_settings.
311 (s390_output_indirect_thunk_function): New function.
312 (s390_code_end): Implement target hook.
313 (s390_case_values_threshold): Implement target hook.
314 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
316 * config/s390/s390.h (struct s390_frame_layout)
317 (struct machine_function): Move here from s390.c.
318 (TARGET_INDIRECT_BRANCH_NOBP_RET)
319 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
320 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
321 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
322 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
323 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
324 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
325 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
326 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
327 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
328 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
329 (mnemonic attribute): Add values which aren't recognized
331 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
332 pattern for branch conversion. Fix mnemonic attribute.
333 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
334 indirect branch via thunk if requested.
335 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
336 ("*indirect_jump"): Disable for branch conversion using out of
338 ("indirect_jump_via_thunk<mode>_z10")
339 ("indirect_jump_via_thunk<mode>")
340 ("indirect_jump_via_inlinethunk<mode>_z10")
341 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
342 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
343 ("casesi_jump_via_inlinethunk<mode>_z10")
344 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
345 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
346 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
347 ("*indirect2_jump"): Disable for branch conversion.
348 ("casesi_jump"): Turn into expander and expand patterns for branch
350 ("return_use"): New expander.
351 ("*return"): Emit return via thunk and rename it to ...
352 ("*return<mode>"): ... this one.
353 * config/s390/s390.opt: Add new options and and enum for the
356 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
358 * lra-constraints.c (match_reload): Unconditionally use
359 gen_lowpart_SUBREG, rather than selecting between that
360 and equivalent gen_rtx_SUBREG code.
362 2018-02-08 Richard Biener <rguenther@suse.de>
364 PR tree-optimization/84233
365 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
366 changed flag instead of boguously re-using phi_inserted.
368 2018-02-08 Martin Jambor <mjambor@suse.cz>
370 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
371 static local variables.
373 2018-02-08 Richard Biener <rguenther@suse.de>
375 PR tree-optimization/84278
376 * tree-vect-stmts.c (vectorizable_store): When looking for
377 smaller vector types to perform grouped strided loads/stores
378 make sure the mode is supported by the target.
379 (vectorizable_load): Likewise.
381 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
383 * config/aarch64/aarch64.c (aarch64_components_for_bb):
384 Increase LDP/STP opportunities by adding adjacent callee-saves.
386 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
388 PR rtl-optimization/84068
389 PR rtl-optimization/83459
390 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
392 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
394 PR tree-optimization/84224
395 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
396 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
399 2018-02-07 Iain Sandoe <iain@codesourcery.com>
402 * config/rs6000/altivec.md (*restore_world): Remove LR use.
403 * config/rs6000/predicates.md (restore_world_operation): Adjust op
404 count, remove one USE.
406 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
408 * doc/install.texi (Configuration): Document the
409 --with-long-double-format={ibm,ieee} PowerPC configuration
413 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
414 Convert from define_expand to be define_insn_and_split. Rework
415 float/double/_Float128 conversions to QI/HI/SImode to work with
416 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
417 conversions to QI/HImode types did a store and then a load to
418 truncate the value. For conversions to VSX registers, don't split
419 the insn, instead emit the code directly. Use the code iterator
420 any_fix to combine signed and unsigned conversions.
421 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
422 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
423 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
424 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
425 (fix_<mode>di2_hw): Likewise.
426 (fixuns_<mode>di2_hw): Likewise.
427 (fix_<mode>si2_hw): Likewise.
428 (fixuns_<mode>si2_hw): Likewise.
429 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
430 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
431 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
432 fix<uns>_trunc<SFDF:mode>si2_p8.
433 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
435 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
436 (fix<uns>_<mode>_mem): Likewise.
437 (fctiw<u>z_<mode>_mem): Likewise.
438 (fix<uns>_<mode>_mem): Likewise.
439 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
440 the register allocator from doing a direct move to the GPRs to do
441 a store, and instead use the ISA 3.0 store byte/half-word from
442 vector register instruction. For IEEE 128-bit floating point,
443 also optimize stores of 32-bit ints.
444 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
446 2018-02-07 Alan Hayward <alan.hayward@arm.com>
448 * genextract.c (push_pathstr_operand): New function to support
450 (walk_rtx): Call push_pathstr_operand.
451 (print_path): Support [a-zA-Z].
453 2018-02-07 Richard Biener <rguenther@suse.de>
455 PR tree-optimization/84037
456 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
457 (cse_and_gimplify_to_preheader): Declare.
458 (vect_get_place_in_interleaving_chain): Likewise.
459 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
461 (_loop_vec_info::~_loop_vec_info): Delete it.
462 (cse_and_gimplify_to_preheader): New function.
463 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
464 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
465 (vectorizable_load): Likewise. For grouped stores always base
466 the IV on the first element.
467 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
468 condition before gimplifying.
470 2018-02-07 Jakub Jelinek <jakub@redhat.com>
472 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
473 *DIV_EXPR and *MOD_EXPR.
475 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
478 * config/i386/i386.c (ix86_option_override_internal): Mask out
479 the CF_SET bit when checking -fcf-protection.
481 2018-02-07 Tom de Vries <tom@codesourcery.com>
484 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
487 2018-02-07 Richard Biener <rguenther@suse.de>
489 PR tree-optimization/84204
490 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
493 PR tree-optimization/84205
494 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
495 special-case isl_ast_op_zdiv_r.
497 PR tree-optimization/84223
498 * graphite-scop-detection.c (gather_bbs::before_dom_children):
499 Only add conditions from within the region.
500 (gather_bbs::after_dom_children): Adjust.
502 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
505 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
506 * config/avr/avr.md: Only post-reload split REG-REG moves if
507 either register is REGERAL_REG_P.
509 2018-02-07 Jakub Jelinek <jakub@redhat.com>
511 PR tree-optimization/84235
512 * tree-ssa-scopedtables.c
513 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
514 if the subtraction is performed in floating point type where NaNs are
515 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
516 build 1. Formatting fix.
518 2018-02-06 Jakub Jelinek <jakub@redhat.com>
521 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
522 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
523 and skip it regardless of bb boundaries. Use CALL_P macro,
524 don't test INSN_P (insn) together with CALL_P or JUMP_P check
525 unnecessarily, formatting fix.
527 2018-02-06 Michael Collison <michael.collison@arm.com>
529 * config/arm/thumb2.md:
530 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
531 (*thumb_mov_notscc): Ditto.
533 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
536 * config/rs6000/rs6000.md (su code attribute): Use "u" for
537 unsigned_fix, not "s".
539 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
541 * configure.ac (gcc_fn_eh_frame_ro): New function.
542 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
543 correct .eh_frame permissions.
544 * configure: Regenerate.
546 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
548 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
551 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
553 * config/rs6000/rs6000.c (rs6000_option_override_internal):
554 Display warning message for -mno-speculate-indirect-jumps.
556 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
558 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
560 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
562 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
564 PR tree-optimization/84225
565 * tree-eh.c (find_trapping_overflow): Only call
566 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
568 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
571 * config/i386/i386.c: Reimplement the check of possible options
572 -mibt/-mshstk conbination. Change error messages.
573 * doc/invoke.texi: Fix a typo: remove extra '='.
575 2018-02-06 Marek Polacek <polacek@redhat.com>
577 PR tree-optimization/84228
578 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
580 2018-02-06 Tamar Christina <tamar.christina@arm.com>
583 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
584 emitted arch directives.
585 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
586 __ARM_FEATURE_COPROC before changing architectures.
588 2018-02-06 Richard Biener <rguenther@suse.de>
590 * config/i386/i386.c (print_reg): Fix typo.
591 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
593 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
595 * configure: Regenerate.
597 2018-02-05 Martin Sebor <msebor@redhat.com>
599 PR tree-optimization/83369
600 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
603 2018-02-05 Martin Liska <mliska@suse.cz>
605 * doc/invoke.texi: Cherry-pick upstream r323995.
607 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
609 * ira.c (ira_init_register_move_cost): Adjust comment.
611 2018-02-05 Martin Liska <mliska@suse.cz>
613 PR gcov-profile/84137
614 * doc/gcov.texi: Fix typo in documentation.
616 2018-02-05 Martin Liska <mliska@suse.cz>
618 PR gcov-profile/83879
619 * doc/gcov.texi: Document necessity of --dynamic-list-data when
620 using dlopen functionality.
622 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
624 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
625 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
626 _mm_maskz_range_ss, _mm_mask_range_round_ss,
627 _mm_maskz_range_round_ss): New intrinsics.
628 (__builtin_ia32_rangesd128_round)
629 (__builtin_ia32_rangess128_round): Remove.
630 (__builtin_ia32_rangesd128_mask_round,
631 __builtin_ia32_rangess128_mask_round): New builtins.
632 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
633 __builtin_ia32_rangess128_round): Remove.
634 (__builtin_ia32_rangesd128_mask_round,
635 __builtin_ia32_rangess128_mask_round): New builtins.
636 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
637 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
638 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
639 "<round_saeonly_constraint>")): Changed to ...
640 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
641 "<round_saeonly_scalar_constraint>")): ... this.
642 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
643 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
644 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
645 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
646 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
648 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
650 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
652 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
653 Remove all values except native, 8540 and 8548.
655 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
657 * config/i386/i386.c (ix86_output_function_return): Pass
658 INVALID_REGNUM, instead of -1, as invalid register number to
659 indirect_thunk_name and output_indirect_thunk.
661 2018-02-02 Julia Koval <julia.koval@intel.com>
663 * config.gcc: Add -march=icelake.
664 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
665 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
666 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
667 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
668 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
669 (processor_target_table): Add icelake.
670 (ix86_option_override_internal): Handle new PTAs.
671 (get_builtin_code_for_version): Handle icelake.
672 (M_INTEL_COREI7_ICELAKE): New.
673 (fold_builtin_cpu): Handle icelake.
674 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
675 * doc/invoke.texi: Add -march=icelake.
677 2018-02-02 Julia Koval <julia.koval@intel.com>
679 * config/i386/i386.c (ix86_option_override_internal): Change flags type
681 * wide-int-bitmask.h: New.
683 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
686 * config/i386/i386.md: Replace Pmode with word_mode in
687 builtin_setjmp_setup and builtin_longjmp to support x32.
689 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
693 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
695 (rs6000_supported_cpu_names): New static variable.
696 (linux_cpu_translation_table): Likewise.
697 (elf_platform) <cpu>: Define new static variable and use it.
698 Translate kernel AT_PLATFORM name to canonical name if needed.
699 Error if platform name is unknown.
701 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
704 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
706 2018-02-01 Jeff Law <law@redhat.com>
709 * config/i386/i386.c (release_scratch_register_on_entry): Add new
710 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
711 the scratch if RELEASE_VIA_POP is false.
712 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
713 If we have to save a temporary register, decrement SIZE appropriately.
714 Pass new arguments to release_scratch_register_on_entry.
715 (ix86_adjust_stack_and_probe): Likewise.
716 (ix86_emit_probe_stack_range): Pass new arguments to
717 release_scratch_register_on_entry.
719 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
721 PR rtl-optimization/84157
722 * combine.c (change_zero_ext): Use REG_P predicate in
723 front of HARD_REGISTER_P predicate.
725 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
727 * config/avr/avr.c (avr_option_override): Move disabling of
728 -fdelete-null-pointer-checks to...
729 * common/config/avr/avr-common.c (avr_option_optimization_table):
732 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
734 PR tree-optimization/81635
735 * tree-data-ref.c (split_constant_offset_1): For types that
736 wrap on overflow, try to use range info to prove that wrapping
739 2018-02-01 Renlin Li <renlin.li@arm.com>
742 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
744 (aarch64_register_move_cost): Likewise.
745 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
747 (REG_CLASS_NAMES): Likewise.
748 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
749 TAILCALL_ADDR_REGS. Remove IP registers.
750 * config/aarch64/aarch64.md (Ucs): Update register constraint.
752 2018-02-01 Richard Biener <rguenther@suse.de>
754 * domwalk.h (dom_walker::dom_walker): Add additional constructor
755 for specifying RPO order and allow NULL for that.
756 * domwalk.c (dom_walker::dom_walker): Likewise.
757 (dom_walker::walk): Handle NULL RPO order.
758 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
760 (rewrite_update_dom_walker): Likewise.
761 (mark_def_dom_walker): Likewise.
763 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
765 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
766 (aarch64_maybe_expand_sve_subreg_move): Declare.
767 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
768 * config/aarch64/predicates.md (aarch64_any_register_operand): New
770 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
771 that are semantically a reverse operation.
772 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
773 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
774 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
776 (aarch64_can_change_mode_class): For big-endian, forbid changes
777 between two SVE modes if they have different element sizes.
779 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
781 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
782 the TImode handling for big-endian targets.
784 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
786 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
787 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
789 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
790 Remove BSWAP handing for big-endian targets and use the form of
791 LD1RQ appropariate for the mode.
793 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
795 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
796 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
799 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
802 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
803 check for operands that need to go through aarch64_sve_reload_be.
805 2018-02-01 Jakub Jelinek <jakub@redhat.com>
807 PR tree-optimization/81661
808 PR tree-optimization/84117
809 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
810 * tree-eh.c: Include gimplify.h.
811 (find_trapping_overflow, replace_trapping_overflow,
812 rewrite_to_non_trapping_overflow): New functions.
813 * tree-vect-loop.c: Include tree-eh.h.
814 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
815 * tree-data-ref.c: Include tree-eh.h.
816 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
818 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
820 PR rtl-optimization/84123
821 * combine.c (change_zero_ext): Check if hard register satisfies
822 can_change_dest_mode before calling gen_lowpart_SUBREG.
824 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
827 * ira.c (ira_init_register_move_cost): Remove assert.
829 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
831 PR rtl-optimization/84071
832 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
833 * doc/tm.texi: Regenerate.
835 2018-01-31 Richard Biener <rguenther@suse.de>
837 PR tree-optimization/84132
838 * tree-data-ref.c (analyze_miv_subscript): Properly
839 check whether evolution_function_is_affine_multivariate_p
840 before calling gcd_of_steps_may_divide_p.
842 2018-01-31 Julia Koval <julia.koval@intel.com>
845 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
846 * config/i386/i386.md (rdpid_rex64) New.
847 (rdpid): Make 32bit only.
849 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
852 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
853 an IDENTIFIER_NODE for FUNCTION_TYPE's.
855 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
858 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
860 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
862 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
864 PR rtl-optimization/84071
865 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
866 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
868 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
870 * config/arc/arc.c (arc_handle_aux_attribute): New function.
871 (arc_attribute_table): Add 'aux' attribute.
872 (arc_in_small_data_p): Consider aux like variables.
873 (arc_is_aux_reg_p): New function.
874 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
875 (arc_get_aux_arg): New function.
876 (prepare_move_operands): Handle aux-register access.
877 (arc_handle_aux_attribute): New function.
878 * doc/extend.texi (ARC Variable attributes): Add subsection.
880 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
882 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
883 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
884 (arc_attribute_table): Add 'uncached' attribute.
885 (arc_print_operand): Print '.di' flag for uncached memory
887 (arc_in_small_data_p): Do not consider for small data the uncached
889 (arc_is_uncached_mem_p): New function.
890 * config/arc/predicates.md (compact_store_memory_operand): Check
891 for uncached memory accesses.
892 (nonvol_nonimm_operand): Likewise.
893 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
895 2018-01-31 Jakub Jelinek <jakub@redhat.com>
898 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
899 falign-loops=): Add Optimization flag.
901 2018-01-30 Jeff Law <law@redhat.com>
904 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
905 INT_REGISTERS_SAVED. Check it prior to calling
906 get_scratch_register_on_entry.
907 (ix86_adjust_stack_and_probe): Similarly.
908 (ix86_emit_probe_stack_range): Similarly.
909 (ix86_expand_prologue): Corresponding changes.
911 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
914 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
915 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
917 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
920 * lra-constraints.c (curr_insn_transform): Process AND in the
923 2018-01-30 Jakub Jelinek <jakub@redhat.com>
925 PR rtl-optimization/83986
926 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
927 dependence against last_pending_memory_flush in addition to
930 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
932 PR tree-optimization/81611
933 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
936 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
938 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
941 2018-01-30 Richard Biener <rguenther@suse.de>
942 Jakub Jelinek <jakub@redhat.com>
944 PR tree-optimization/84111
945 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
946 inner loops added during recursion, as they don't have up-to-date
949 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
952 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
953 (can_inline_edge_by_limits_p): ... here.
954 (can_early_inline_edge_p, check_callers,
955 update_caller_keys, update_callee_keys, recursive_inlining,
956 add_new_edges_to_heap, speculation_useful_p,
957 inline_small_functions,
958 inline_small_functions, flatten_function,
959 inline_to_all_callers_1): Update.
961 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
963 * profile-count.c (profile_count::combine_with_ipa_count): Handle
966 2018-01-30 Richard Biener <rguenther@suse.de>
968 PR tree-optimization/83008
969 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
970 invariant and constant vector uses in stmts when they need
973 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
976 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
977 * configure: Regenerate.
979 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
981 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
983 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
984 Use gen_rtx_REG rather than gen_lowpart.
986 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
988 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
989 rather than 0 when creating partial subregs.
991 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
993 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
996 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
999 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1000 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1001 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1002 flags. This restores the settings used before the 2017-07-24.
1003 Turning off pre increment/decrement/modify allows IVOPTS to
1004 optimize DF/SF loops where the index is an int.
1006 2018-01-29 Richard Biener <rguenther@suse.de>
1007 Kelvin Nilsen <kelvin@gcc.gnu.org>
1010 * tree-vect-stmts.c (vectorizable_call): Don't call
1011 targetm.vectorize_builtin_md_vectorized_function if callee is
1014 2018-01-22 Carl Love <cel@us.ibm.com>
1016 * doc/extend.tex: Fix typo in second arg in
1017 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1019 2018-01-29 Richard Biener <rguenther@suse.de>
1021 PR tree-optimization/84086
1022 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1023 (flush_ssaname_freelist): When SSA names were released reset
1024 the SCEV hash table.
1026 2018-01-29 Richard Biener <rguenther@suse.de>
1028 PR tree-optimization/84057
1029 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1030 removed paths when removing edges.
1032 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1034 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1035 -mfunction-return=@var{choice}.
1037 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1040 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1042 (layout::print_source_line): Likewise.
1043 (test_get_line_width_without_trailing_whitespace): Add test cases.
1045 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1048 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1051 2018-01-26 Jim Wilson <jimw@sifive.com>
1053 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1055 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1058 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1060 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1061 and CMP + SUB-immediate -> SUBS.
1063 2018-01-26 Martin Sebor <msebor@redhat.com>
1065 PR tree-optimization/83896
1066 * tree-ssa-strlen.c (get_string_len): Rename...
1067 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1068 Avoid assuming length is constant.
1069 (handle_char_store): Use HOST_WIDE_INT for string length.
1071 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1074 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1075 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1077 2018-01-26 Richard Biener <rguenther@suse.de>
1079 PR rtl-optimization/84003
1080 * dse.c (record_store): Only record redundant stores when
1081 the earlier store aliases at least all accesses the later one does.
1083 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1085 PR rtl-optimization/83985
1086 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1087 REG_CFA_RESTORE insns.
1088 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1089 REG_CFA_RESTORE insns here.
1092 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1093 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1095 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1097 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1098 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1099 (arc_init): Likewise.
1100 (arc_override_options): Likewise.
1101 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1103 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1105 * config/arc/arc.h (TARGET_DBNZ): Define.
1106 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1107 properly set the tune attribute.
1108 (dbnz): Use TARGET_DBNZ guard.
1109 * config/arc/arc.opt (mtune): Add core3 option.
1111 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1113 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1114 recognize new pic like addresses.
1115 (arc_delegitimize_address): Clean up.
1117 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1119 * config/arc/arc-arches.def: Option mrf16 valid for all
1121 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1122 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1123 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1124 * config/arc/arc-tables.opt: Regenerate.
1125 * config/arc/arc.c (arc_conditional_register_usage): Handle
1126 reduced register file case.
1127 (arc_file_start): Set must have build attributes.
1128 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1130 * config/arc/arc.opt (mrf16): Add new option.
1131 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1132 * config/arc/genmultilib.awk: Handle new mrf16 option.
1133 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1134 * config/arc/t-multilib: Regenerate.
1135 * doc/invoke.texi (ARC Options): Document mrf16 option.
1137 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1139 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1140 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1141 (arc_attribute_table): Add 'secure_call' attribute.
1142 (arc_print_operand): Print secure call operand.
1143 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1145 (arc_is_secure_call_p): New function. * config/arc/arc.md
1146 (call_i): Add support for sjli instruction.
1147 (call_value_i): Likewise.
1148 * config/arc/constraints.md (Csc): New constraint.
1150 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1151 John Eric Martin <John.Martin@emmicro-us.com>
1153 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1154 * config/arc/arc.c (_arc_jli_section): New struct.
1155 (arc_jli_section): New type.
1156 (rc_jli_sections): New static variable.
1157 (arc_handle_jli_attribute): New function.
1158 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1159 (arc_file_end): New function.
1160 (TARGET_ASM_FILE_END): Define.
1161 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1162 (arc_add_jli_section): New function.
1163 (jli_call_scan): Likewise.
1164 (arc_reorg): Call jli_call_scan.
1165 (arc_output_addsi): Remove 'S' from printing asm operand.
1166 (arc_is_jli_call_p): New function.
1167 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1169 (movhi_insn): Likewise.
1170 (movsi_insn): Likewise.
1171 (movsi_set_cc_insn): Likewise.
1172 (loadqi_update): Likewise.
1173 (load_zeroextendqisi_update): Likewise.
1174 (load_signextendqisi_update): Likewise.
1175 (loadhi_update): Likewise.
1176 (load_zeroextendhisi_update): Likewise.
1177 (load_signextendhisi_update): Likewise.
1178 (loadsi_update): Likewise.
1179 (loadsf_update): Likewise.
1180 (movsicc_insn): Likewise.
1181 (bset_insn): Likewise.
1182 (bxor_insn): Likewise.
1183 (bclr_insn): Likewise.
1184 (bmsk_insn): Likewise.
1185 (bicsi3_insn): Likewise.
1186 (cmpsi_cc_c_insn): Likewise.
1187 (movsi_ne): Likewise.
1188 (movsi_cond_exec): Likewise.
1189 (clrsbsi2): Likewise.
1200 (call_i): Remove 'S' asm letter, add jli instruction.
1201 (call_value_i): Likewise.
1202 * config/arc/arc.op (mjli-always): New option.
1203 * config/arc/constraints.md (Cji): New constraint.
1204 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1206 (subsf3_fpx): Likewise.
1207 (mulsf3_fpx): Likewise.
1208 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1210 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1212 * doc/invoke.texi (ARC): Document mjli-always option.
1214 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1216 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1217 and use incw and decw where possible
1218 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1220 2018-01-26 Richard Biener <rguenther@suse.de>
1222 PR tree-optimization/81082
1223 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1224 association if it requires casting to unsigned.
1225 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1226 from fold_plusminus_mult_expr to catch important cases late when
1227 range info is available.
1229 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1231 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1232 * configure.ac (hidden_linkonce): New test.
1233 * configure: Regenerate.
1234 * config.in: Regenerate.
1236 2018-01-26 Julia Koval <julia.koval@intel.com>
1238 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1239 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1240 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1241 _mm_mask_bitshuffle_epi64_mask): Fix type.
1242 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1243 USI_FTYPE_V4DI_V4DI_USI): Remove.
1244 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1245 __builtin_ia32_vpshufbitqmb256_mask,
1246 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1247 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1248 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1250 2018-01-26 Alan Modra <amodra@gmail.com>
1253 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1254 UNSPEC_VBPERMQ. Sort other unspecs.
1256 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1258 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1260 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1263 * predict.c (drop_profile): Do not push/pop cfun; update also
1265 (handle_missing_profiles): Fix logic looking for zero profiles.
1267 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1270 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1271 on functions with #pragma omp declare simd or functions with simd
1273 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1274 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1275 Remove trailing \n from warning_at calls.
1277 2018-01-25 Tom de Vries <tom@codesourcery.com>
1280 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1281 for neutered workers.
1283 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1286 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1287 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1289 2018-01-24 Jeff Law <law@redhat.com>
1292 * i386.c (get_probe_interval): Move to earlier point.
1293 (ix86_compute_frame_layout): If -fstack-clash-protection and
1294 the frame is larger than the probe interval, then use pushes
1295 to save registers rather than reg->mem moves.
1296 (ix86_expand_prologue): Remove conditional for int_registers_saved
1299 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1302 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1303 min/max for never referenced object.
1305 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1308 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1310 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1311 attributes from DECL_ATTRIBUTES (decl) without affecting
1312 DECL_ATTRIBUTES (current_function_decl).
1313 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1314 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1316 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1318 PR tree-optimization/83979
1319 * fold-const.c (fold_comparison): Use constant_boolean_node
1320 instead of boolean_{true,false}_node.
1322 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1324 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1327 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1329 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1330 Simplify the clause that sets the length attribute.
1331 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1332 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1333 clause that sets the length attribute.
1334 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1336 2018-01-24 Tom de Vries <tom@codesourcery.com>
1339 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1340 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1341 Add strict parameter.
1342 (prevent_branch_around_nothing): Insert dummy insn between branch to
1343 label and label with no ptx insn inbetween.
1344 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1346 2018-01-24 Tom de Vries <tom@codesourcery.com>
1349 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1350 for neutered threads in warp.
1351 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1353 2018-01-24 Richard Biener <rguenther@suse.de>
1355 PR tree-optimization/83176
1356 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1359 2018-01-24 Richard Biener <rguenther@suse.de>
1361 PR tree-optimization/82819
1362 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1363 code generating pluses that are no-ops in the target precision.
1365 2018-01-24 Richard Biener <rguenther@suse.de>
1368 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1370 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1372 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1373 to merge probabilities.
1374 * predict.c (probably_never_executed): Also mark as cold functions
1375 with global 0 profile and guessed local profile.
1376 * profile-count.c (profile_probability::combine_with_count): New
1378 * profile-count.h (profile_probability::operator*,
1379 profile_probability::operator*=, profile_probability::operator/,
1380 profile_probability::operator/=): Reduce precision to adjusted
1381 and set value to guessed on contradictory divisions.
1382 (profile_probability::combine_with_freq): Remove.
1383 (profile_probability::combine_wiht_count): Declare.
1384 (profile_count::force_nonzero):: Set to adjusted.
1385 (profile_count::probability_in):: Set quality to adjusted.
1386 * tree-ssa-tail-merge.c (replace_block_by): Use
1389 2018-01-23 Andrew Waterman <andrew@sifive.com>
1390 Jim Wilson <jimw@sifive.com>
1392 * config/riscv/riscv.c (riscv_stack_boundary): New.
1393 (riscv_option_override): Set riscv_stack_boundary. Handle
1394 riscv_preferred_stack_boundary_arg.
1395 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1396 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1397 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1398 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1399 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1400 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1402 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1405 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1406 of struct ix86_frame.
1407 (ix86_expand_epilogue): Likewise. Add a local variable for
1408 the reg_save_offset field in struct ix86_frame.
1410 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1412 PR tree-optimization/82604
1413 * tree-loop-distribution.c (enum partition_kind): New enum item
1414 PKIND_PARTIAL_MEMSET.
1415 (partition_builtin_p): Support above new enum item.
1416 (generate_code_for_partition): Ditto.
1417 (compute_access_range): Differentiate cases that equality can be
1418 proven at all loops, the innermost loops or no loops.
1419 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1420 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1421 (finalize_partitions, distribute_loop): Don't fuse partition of
1422 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1423 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1426 2018-01-23 Martin Liska <mliska@suse.cz>
1428 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1429 order to ignore the predictor.
1430 (PRED_POLYMORPHIC_CALL): Likewise.
1431 (PRED_RECURSIVE_CALL): Likewise.
1433 2018-01-23 Martin Liska <mliska@suse.cz>
1435 * tree-profile.c (tree_profiling): Print function header to
1436 aware reader which function we are working on.
1437 * value-prof.c (gimple_find_values_to_profile): Do not print
1438 not interesting value histograms.
1440 2018-01-23 Martin Liska <mliska@suse.cz>
1442 * profile-count.h (enum profile_quality): Add
1443 profile_uninitialized as the first value. Do not number values
1444 as they are zero based.
1445 (profile_count::verify): Update sanity check.
1446 (profile_probability::verify): Likewise.
1448 2018-01-23 Nathan Sidwell <nathan@acm.org>
1450 * doc/invoke.texi (ffor-scope): Deprecate.
1452 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1454 PR tree-optimization/83510
1455 * domwalk.c (set_all_edges_as_executable): New function.
1456 (dom_walker::dom_walker): Convert bool param
1457 "skip_unreachable_blocks" to enum reachability. Move setup of
1458 edge flags to set_all_edges_as_executable and only do it when
1459 reachability is REACHABLE_BLOCKS.
1460 * domwalk.h (enum dom_walker::reachability): New enum.
1461 (dom_walker::dom_walker): Convert bool param
1462 "skip_unreachable_blocks" to enum reachability.
1463 (set_all_edges_as_executable): New decl.
1464 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1465 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1467 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1468 but converting true to REACHABLE_BLOCKS.
1469 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1471 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1472 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1473 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1475 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1476 if check_all_array_refs will be called.
1478 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1480 * tree.c (selftest::test_location_wrappers): Add more test
1483 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1485 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1486 (selftest::test_bit_in_range): Likewise.
1488 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1491 * doc/sourcebuild.texi (vect_float): Say that the selector
1492 only describes the situation when -funsafe-math-optimizations is on.
1493 (vect_float_strict): Document.
1495 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1497 PR tree-optimization/83965
1498 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1499 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1500 instead of checking only for a reduction.
1501 (vect_recog_widen_sum_pattern): Likewise.
1503 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1505 * predict.c (probably_never_executed): Only use precise profile info.
1506 (compute_function_frequency): Skip after inlining hack since we now
1507 have quality checking.
1509 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1511 * profile-count.h (profile_probability::very_unlikely,
1512 profile_probability::unlikely, profile_probability::even): Set
1513 precision to guessed.
1515 2018-01-23 Richard Biener <rguenther@suse.de>
1517 PR tree-optimization/83963
1518 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1519 Properly terminate dominator walk when crossing the exit edge not
1520 when visiting its source block.
1522 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1525 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1526 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1528 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1530 PR tree-optimization/83957
1531 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1532 semicolon after for body surrounded by braces.
1534 PR tree-optimization/83081
1535 * profile-count.h (profile_probability::split): New method.
1536 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1537 Use profile_probability::split.
1538 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1539 when splitting a single conditional jump into 2.
1541 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1543 PR tree-optimization/69452
1544 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1547 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1549 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1550 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1551 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1553 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1555 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1556 * config/rl78/rl78.md: New define_expand "movdi"
1557 * config/rl78/rl78.c: New function definition rl78_split_movdi
1559 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1562 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1564 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1565 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1566 128-bit to produce an UNSPEC move to get the double word with the
1567 signbit and then a shift directly to do signbit.
1568 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1569 implementation with a new version that just does either a direct
1570 move or a regular move. Move memory interface to separate insns.
1571 Move insns so they are next to the expander.
1572 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1573 with signbit move. Split big and little endian case.
1574 (signbit<mode>2_dm_mem_le): Likewise.
1575 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1576 (signbit<mode>2_dm2): Likewise.
1578 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1580 * config/rl78/rl78.md: New define_expand "anddi3".
1582 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1584 * config/rl78/rl78.md: New define_expand "umindi3".
1586 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1588 * config/rl78/rl78.md: New define_expand "smindi3".
1590 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1592 * config/rl78/rl78.md: New define_expand "smaxdi3".
1594 2018-01-22 Carl Love <cel@us.ibm.com>
1596 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1597 LVX_V1TI): Add macro expansion.
1598 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1599 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1600 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1601 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1602 Change check to determine if the instruction is a byte reversing
1603 entry. Fix typo in comment.
1604 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1605 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1606 Add def_builtin calls for new builtins.
1607 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1608 Add define_insn expansion.
1610 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1612 * config/rl78/rl78.md: New define_expand "umaxdi3".
1614 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1616 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1617 for non-QImode registers
1619 2018-01-22 Richard Biener <rguenther@suse.de>
1621 PR tree-optimization/83963
1622 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1623 including the loop exit block.
1624 (scop_detection::merge_sese): Likewise.
1625 (scop_detection::add_scop): Do it here instead.
1627 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1629 * doc/sourcebuild.texi (arm_softfloat): Document.
1631 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1634 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1635 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1636 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1638 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1639 David Edelsohn <dje.gcc@gmail.com>
1642 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1643 Change "crset eq" to "crset 2".
1644 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1645 (*call_indirect_aix<mode>_nospec): Likewise.
1646 (*call_value_indirect_aix<mode>_nospec): Likewise.
1647 (*call_indirect_elfv2<mode>_nospec): Likewise.
1648 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1649 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1650 change assembly output from . to $.
1651 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1652 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1653 (*tablejump<mode>_internal1_nospec): Likewise.
1655 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1658 * config/sh/sh_optimize_sett_clrt.cc:
1659 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1661 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1663 PR tree-optimization/83940
1664 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1665 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1666 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1667 use it to pass back the definition type.
1668 (vect_check_store_rhs): Likewise rhs_dt_out.
1669 (vect_build_gather_load_calls): Add a mask_dt argument and use
1670 it instead of a call to vect_is_simple_use.
1671 (vectorizable_store): Update calls to vect_check_load_store_mask
1672 and vect_check_store_rhs. Use the dt returned by the latter instead
1673 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1674 instead of calls to vect_is_simple_use. Pass the scalar rather
1675 than the vector operand to vect_is_simple_use when handling
1676 second and subsequent copies of an rhs value.
1677 (vectorizable_load): Update calls to vect_check_load_store_mask
1678 and vect_build_gather_load_calls. Use the cached mask_dt and
1679 gs_info.offset_dt instead of calls to vect_is_simple_use.
1681 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1684 * tree-emutls.c: Include gimplify.h.
1685 (lower_emutls_2): New function.
1686 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1687 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1688 it before further processing.
1691 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1692 UINTVAL (trueop1) instead of INTVAL (op1).
1694 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1698 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1699 INCOMING_FRAME_SP_OFFSET if not defined.
1700 (scan_trace): Add ENTRY argument. If true and
1701 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1702 emit a note to adjust the CFA offset.
1703 (create_cfi_notes): Adjust scan_trace callers.
1704 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1705 INCOMING_FRAME_SP_OFFSET in the CIE.
1706 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1707 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1709 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1710 * doc/tm.texi: Regenerated.
1712 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1714 PR rtl-optimization/83147
1715 * lra-constraints.c (remove_inheritance_pseudos): Use
1716 lra_substitute_pseudo_within_insn.
1718 2018-01-19 Tom de Vries <tom@codesourcery.com>
1719 Cesar Philippidis <cesar@codesourcery.com>
1722 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1724 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1727 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1728 spaces for function labels.
1730 2018-01-19 Martin Liska <mliska@suse.cz>
1732 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1733 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1734 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1735 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1736 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1737 (PRED_CONST_RETURN): Change from 69 to 65.
1738 (PRED_NULL_RETURN): Change from 91 to 71.
1739 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1740 (PRED_LOOP_GUARD): Change from 66 to 73.
1742 2018-01-19 Martin Liska <mliska@suse.cz>
1744 * predict.c (predict_insn_def): Add new assert.
1745 (struct branch_predictor): Change type to signed integer.
1746 (test_prediction_value_range): Amend test to cover
1748 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1749 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1750 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1751 (PRED_LOOP_IV_COMPARE): Likewise.
1752 * predict.h (PROB_UNINITIALIZED): Define new constant.
1754 2018-01-19 Martin Liska <mliska@suse.cz>
1756 * predict.c (dump_prediction): Add new format for
1757 analyze_brprob.py script which is enabled with -details
1759 * profile-count.h (precise_p): New function.
1761 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1763 PR tree-optimization/83922
1764 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1765 there are no statements that need masking.
1766 (vect_active_double_reduction_p): New function.
1767 (vect_analyze_loop_operations): Use it when handling phis that
1768 are not in the loop header.
1770 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1772 PR tree-optimization/83914
1773 * tree-vect-loop.c (vectorizable_induction): Don't convert
1774 init_expr or apply the peeling adjustment for inductions
1775 that are nested within the vectorized loop.
1777 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1779 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1782 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1786 * function.h (gimplify_parameters): Add gimple_seq * argument.
1787 * function.c: Include gimple.h and options.h.
1788 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1789 for the added local temporaries if needed.
1790 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1791 if there are any parameter cleanups, wrap whole body into a
1792 try/finally with the cleanups.
1794 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1797 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1798 Use GET_MODE_CLASS for scalar floating point.
1800 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1804 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1805 Fix call of call_cgraph_insertion_hooks.
1807 2018-01-18 Martin Sebor <msebor@redhat.com>
1809 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1811 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1814 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1817 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1820 * common.opt: (-ffile-prefix-map): New option.
1821 * opts.c (common_handle_option): Defer it.
1822 * opts-global.c (handle_common_deferred_options): Handle it.
1823 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1824 * file-prefix-map.h: New file.
1825 (remap_debug_filename, add_debug_prefix_map): ...here.
1826 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1827 * final.c (debug_prefix_map, add_debug_prefix_map
1828 remap_debug_filename): Move to...
1829 * file-prefix-map.c: New file.
1830 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1831 generalize, get rid of alloca(), use strrchr() instead of strchr().
1832 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1833 Implement in terms of add_prefix_map().
1834 (remap_macro_filename, remap_debug_filename): Implement in term of
1836 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1837 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1838 * dbxout.c: Include file-prefix-map.h.
1839 * varasm.c: Likewise.
1840 * vmsdbgout.c: Likewise.
1841 * xcoffout.c: Likewise.
1842 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1843 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1844 * doc/invoke.texi (-ffile-prefix-map): Document.
1845 (-fdebug-prefix-map): Update description.
1847 2018-01-18 Martin Liska <mliska@suse.cz>
1849 * config/i386/i386.c (indirect_thunk_name): Document that also
1851 (output_indirect_thunk): Document why both instructions
1852 (pause and lfence) are generated.
1854 2018-01-18 Richard Biener <rguenther@suse.de>
1856 PR tree-optimization/83887
1857 * graphite-scop-detection.c
1858 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1859 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1860 (scop_detection::merge_sese): Re-implement with a flood-fill
1861 algorithm that properly finds a SESE region if it exists.
1863 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1866 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1867 pointer_diff optimizations use view_convert instead of convert.
1869 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1871 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1872 Generate different code for -mno-speculate-indirect-jumps.
1873 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1874 (*call_indirect_aix<mode>): Disable for
1875 -mno-speculate-indirect-jumps.
1876 (*call_indirect_aix<mode>_nospec): New define_insn.
1877 (*call_value_indirect_aix<mode>): Disable for
1878 -mno-speculate-indirect-jumps.
1879 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1880 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1881 -mno-speculate-indirect-jumps.
1882 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1884 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1886 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1887 long double type, set the flags for noting the default long double
1888 type, even if we don't pass or return a long double type.
1890 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1893 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1896 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1898 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1899 support for merge[hl].
1900 (fold_mergehl_helper): New helper function.
1901 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1902 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1903 (altivec_vmrglw_direct): Add xxmrglw insn.
1905 2018-01-17 Andrew Waterman <andrew@sifive.com>
1907 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1908 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1910 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1913 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1914 call the lto_location_cache before reading the
1915 DECL_SOURCE_LOCATION of the types.
1917 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1918 Richard Sandiford <richard.sandiford@linaro.org>
1920 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1921 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1922 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1923 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1924 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1926 * config/aarch64/constraints.md (aarch64_movti_operand):
1928 * config/aarch64/predicates.md (Uti): Add new constraint.
1930 2018-01-17 Carl Love <cel@us.ibm.com>
1931 * config/rs6000/vsx.md (define_expand xl_len_r,
1932 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1933 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1935 (define_expand, define_insn): Move the shift left from the
1936 define_insn to the define_expand for lxvl and stxvl instructions.
1937 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1938 and XL_LEN_R definitions to PURE.
1940 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1942 * config/i386/i386.c (indirect_thunk_name): Declare regno
1943 as unsigned int. Compare regno with INVALID_REGNUM.
1944 (output_indirect_thunk): Ditto.
1945 (output_indirect_thunk_function): Ditto.
1946 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1947 in the call to output_indirect_thunk_function.
1949 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1952 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1953 rather than the size of inner_type to determine the stack slot size
1954 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1956 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1959 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1962 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1964 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1965 endian Linux systems to optionally enable multilibs for selecting
1966 the long double type if the user configured an explicit type.
1967 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1968 have no long double multilibs if not defined.
1969 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1970 warn if the user used -mabi={ieee,ibm}longdouble and we built
1971 multilibs for long double.
1972 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1973 appropriate multilib option.
1974 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1976 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1977 for building long double multilibs.
1978 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1980 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1982 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1985 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1987 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1990 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1993 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1996 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1998 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1999 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2001 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2003 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2004 different rtl trees depending on TARGET_64BIT.
2005 (rs6000_gen_lvx): Likewise.
2007 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2009 * config/visium/visium.md (nop): Tweak comment.
2010 (hazard_nop): Likewise.
2012 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2014 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2015 -mspeculate-indirect-jumps.
2016 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2017 for -mno-speculate-indirect-jumps.
2018 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2019 (*call_value_indirect_elfv2<mode>): Disable for
2020 -mno-speculate-indirect-jumps.
2021 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2022 (indirect_jump): Emit different RTL for
2023 -mno-speculate-indirect-jumps.
2024 (*indirect_jump<mode>): Disable for
2025 -mno-speculate-indirect-jumps.
2026 (*indirect_jump<mode>_nospec): New define_insn.
2027 (tablejump): Emit different RTL for
2028 -mno-speculate-indirect-jumps.
2029 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2030 (tablejumpsi_nospec): New define_expand.
2031 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2032 (tablejumpdi_nospec): New define_expand.
2033 (*tablejump<mode>_internal1): Disable for
2034 -mno-speculate-indirect-jumps.
2035 (*tablejump<mode>_internal1_nospec): New define_insn.
2036 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2039 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2041 * caller-save.c (insert_save): Drop unnecessary parameter. All
2044 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2045 Richard Biener <rguenth@suse.de>
2048 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2049 return early, inline manually is_gimple_sizepos. Make sure if we
2050 call gimplify_expr we don't end up with a gimple constant.
2051 * tree.c (variably_modified_type_p): Don't return true for
2052 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2053 * gimplify.h (is_gimple_sizepos): Remove.
2055 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2057 PR tree-optimization/83857
2058 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2059 vectorizable_live_operation for pure SLP statements.
2060 (vectorizable_live_operation): Handle PHIs.
2062 2018-01-16 Richard Biener <rguenther@suse.de>
2064 PR tree-optimization/83867
2065 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2066 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2068 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2071 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2072 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2073 If off is not INTEGER_CST, issue a may not be aligned warning
2074 rather than isn't aligned. Use isn%'t rather than isn't.
2075 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2077 <case MULT_EXPR>: Improve the case when bottom and one of the
2078 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2079 operand, in that case check if the other operand is multiple of
2080 bottom divided by the INTEGER_CST operand.
2082 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2085 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2086 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2087 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2088 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2089 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2090 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2091 (pa_function_arg_size): New function.
2093 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2095 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2096 in a separate statement.
2098 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2100 PR tree-optimization/83847
2101 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2102 group gathers and scatters.
2104 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2106 PR rtl-optimization/86620
2107 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2109 PR rtl-optimization/83213
2110 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2111 to last if both are JUMP_INSNs.
2113 PR tree-optimization/83843
2114 * gimple-ssa-store-merging.c
2115 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2116 store_immediate_info for bswap/nop orig_stores.
2118 2018-01-15 Andrew Waterman <andrew@sifive.com>
2120 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2122 <UDIV>: Increase cost if !TARGET_DIV.
2124 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2126 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2127 (define_attr "cr_logical_3op"): New.
2128 (cceq_ior_compare): Adjust.
2129 (cceq_ior_compare_complement): Adjust.
2130 (*cceq_rev_compare): Adjust.
2131 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2132 (is_cracked_insn): Adjust.
2133 (insn_must_be_first_in_group): Adjust.
2134 * config/rs6000/40x.md: Adjust.
2135 * config/rs6000/440.md: Adjust.
2136 * config/rs6000/476.md: Adjust.
2137 * config/rs6000/601.md: Adjust.
2138 * config/rs6000/603.md: Adjust.
2139 * config/rs6000/6xx.md: Adjust.
2140 * config/rs6000/7450.md: Adjust.
2141 * config/rs6000/7xx.md: Adjust.
2142 * config/rs6000/8540.md: Adjust.
2143 * config/rs6000/cell.md: Adjust.
2144 * config/rs6000/e300c2c3.md: Adjust.
2145 * config/rs6000/e500mc.md: Adjust.
2146 * config/rs6000/e500mc64.md: Adjust.
2147 * config/rs6000/e5500.md: Adjust.
2148 * config/rs6000/e6500.md: Adjust.
2149 * config/rs6000/mpc.md: Adjust.
2150 * config/rs6000/power4.md: Adjust.
2151 * config/rs6000/power5.md: Adjust.
2152 * config/rs6000/power6.md: Adjust.
2153 * config/rs6000/power7.md: Adjust.
2154 * config/rs6000/power8.md: Adjust.
2155 * config/rs6000/power9.md: Adjust.
2156 * config/rs6000/rs64.md: Adjust.
2157 * config/rs6000/titan.md: Adjust.
2159 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2161 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2162 ix86_indirect_branch_register logic.
2164 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2166 * config/i386/constraints.md (Bs): Update
2167 ix86_indirect_branch_register check. Don't check
2168 ix86_indirect_branch_register with GOT_memory_operand.
2170 * config/i386/predicates.md (GOT_memory_operand): Don't check
2171 ix86_indirect_branch_register here.
2172 (GOT32_symbol_operand): Likewise.
2174 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2176 * config/i386/predicates.md (constant_call_address_operand):
2177 Rewrite ix86_indirect_branch_register logic.
2178 (sibcall_insn_operand): Likewise.
2180 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2182 * config/i386/constraints.md (Bs): Replace
2183 ix86_indirect_branch_thunk_register with
2184 ix86_indirect_branch_register.
2186 * config/i386/i386.md (indirect_jump): Likewise.
2187 (tablejump): Likewise.
2188 (*sibcall_memory): Likewise.
2189 (*sibcall_value_memory): Likewise.
2190 Peepholes of indirect call and jump via memory: Likewise.
2191 * config/i386/i386.opt: Likewise.
2192 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2193 (GOT_memory_operand): Likewise.
2194 (call_insn_operand): Likewise.
2195 (sibcall_insn_operand): Likewise.
2196 (GOT32_symbol_operand): Likewise.
2198 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2201 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2202 type rather than type addr's type points to.
2203 (expand_omp_atomic_mutex): Likewise.
2204 (expand_omp_atomic): Likewise.
2206 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2209 * config/i386/i386.c (output_indirect_thunk_function): Use
2210 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2211 for __x86_return_thunk.
2213 2018-01-15 Richard Biener <rguenther@suse.de>
2216 * expmed.c (extract_bit_field_1): Fix typo.
2218 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2221 * config/arm/iterators.md (VF): New mode iterator.
2222 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2223 Remove integer-related logic from pattern.
2224 (neon_vabd<mode>_3): Likewise.
2226 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2229 * common.opt (fstrict-overflow): No longer an alias.
2230 (fwrapv-pointer): New option.
2231 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2232 also for pointer types based on flag_wrapv_pointer.
2233 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2234 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2235 opts->x_flag_wrapv got set.
2236 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2237 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2238 POINTER_TYPE_OVERFLOW_UNDEFINED.
2239 * match.pd: Likewise in address comparison pattern.
2240 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2242 2018-01-15 Richard Biener <rguenther@suse.de>
2245 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2246 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2247 Reset type names to their identifier if their TYPE_DECL doesn't
2248 have linkage (and thus is used for ODR and devirt).
2249 (save_debug_info_for_decl): Remove.
2250 (save_debug_info_for_type): Likewise.
2251 (add_tree_to_fld_list): Adjust.
2252 * tree-pretty-print.c (dump_generic_node): Make dumping of
2253 type names more robust.
2255 2018-01-15 Richard Biener <rguenther@suse.de>
2257 * BASE-VER: Bump to 8.0.1.
2259 2018-01-14 Martin Sebor <msebor@redhat.com>
2262 * builtins.c (check_access): Avoid warning when the no-warning bit
2265 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2267 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2268 * ira-color (allocno_hard_regs_compare): Likewise.
2270 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2273 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2274 Use .pushsection/.popsection.
2276 2018-01-14 Martin Sebor <msebor@redhat.com>
2279 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2281 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2283 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2284 entry from extra_headers.
2285 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2286 extra_headers, make the list bitwise identical to the i?86-*-* one.
2288 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2290 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2291 -mcmodel=large with -mindirect-branch=thunk,
2292 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2293 -mfunction-return=thunk-extern.
2294 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2295 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2296 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2298 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2300 * config/i386/i386.c (print_reg): Print the name of the full
2301 integer register without '%'.
2302 (ix86_print_operand): Handle 'V'.
2303 * doc/extend.texi: Document 'V' modifier.
2305 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2307 * config/i386/constraints.md (Bs): Disallow memory operand for
2308 -mindirect-branch-register.
2310 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2311 (GOT_memory_operand): Likewise.
2312 (call_insn_operand): Likewise.
2313 (sibcall_insn_operand): Likewise.
2314 (GOT32_symbol_operand): Likewise.
2315 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2316 for -mindirect-branch-register.
2317 (tablejump): Likewise.
2318 (*sibcall_memory): Likewise.
2319 (*sibcall_value_memory): Likewise.
2320 Disallow peepholes of indirect call and jump via memory for
2321 -mindirect-branch-register.
2322 (*call_pop): Replace m with Bw.
2323 (*call_value_pop): Likewise.
2324 (*sibcall_pop_memory): Replace m with Bs.
2325 * config/i386/i386.opt (mindirect-branch-register): New option.
2326 * doc/invoke.texi: Document -mindirect-branch-register option.
2328 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2330 * config/i386/i386-protos.h (ix86_output_function_return): New.
2331 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2332 set function_return_type.
2333 (indirect_thunk_name): Add ret_p to indicate thunk for function
2335 (output_indirect_thunk_function): Pass false to
2336 indirect_thunk_name.
2337 (ix86_output_indirect_branch_via_reg): Likewise.
2338 (ix86_output_indirect_branch_via_push): Likewise.
2339 (output_indirect_thunk_function): Create alias for function
2340 return thunk if regno < 0.
2341 (ix86_output_function_return): New function.
2342 (ix86_handle_fndecl_attribute): Handle function_return.
2343 (ix86_attribute_table): Add function_return.
2344 * config/i386/i386.h (machine_function): Add
2345 function_return_type.
2346 * config/i386/i386.md (simple_return_internal): Use
2347 ix86_output_function_return.
2348 (simple_return_internal_long): Likewise.
2349 * config/i386/i386.opt (mfunction-return=): New option.
2350 (indirect_branch): Mention -mfunction-return=.
2351 * doc/extend.texi: Document function_return function attribute.
2352 * doc/invoke.texi: Document -mfunction-return= option.
2354 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2356 * config/i386/i386-opts.h (indirect_branch): New.
2357 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2358 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2359 with local indirect jump when converting indirect call and jump.
2360 (ix86_set_indirect_branch_type): New.
2361 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2362 (indirectlabelno): New.
2363 (indirect_thunk_needed): Likewise.
2364 (indirect_thunk_bnd_needed): Likewise.
2365 (indirect_thunks_used): Likewise.
2366 (indirect_thunks_bnd_used): Likewise.
2367 (INDIRECT_LABEL): Likewise.
2368 (indirect_thunk_name): Likewise.
2369 (output_indirect_thunk): Likewise.
2370 (output_indirect_thunk_function): Likewise.
2371 (ix86_output_indirect_branch_via_reg): Likewise.
2372 (ix86_output_indirect_branch_via_push): Likewise.
2373 (ix86_output_indirect_branch): Likewise.
2374 (ix86_output_indirect_jmp): Likewise.
2375 (ix86_code_end): Call output_indirect_thunk_function if needed.
2376 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2378 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2379 (ix86_attribute_table): Add indirect_branch.
2380 * config/i386/i386.h (machine_function): Add indirect_branch_type
2381 and has_local_indirect_jump.
2382 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2384 (tablejump): Likewise.
2385 (*indirect_jump): Use ix86_output_indirect_jmp.
2386 (*tablejump_1): Likewise.
2387 (simple_return_indirect_internal): Likewise.
2388 * config/i386/i386.opt (mindirect-branch=): New option.
2389 (indirect_branch): New.
2392 (thunk-inline): Likewise.
2393 (thunk-extern): Likewise.
2394 * doc/extend.texi: Document indirect_branch function attribute.
2395 * doc/invoke.texi: Document -mindirect-branch= option.
2397 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2400 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2402 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2404 * ipa-inline.c (want_inline_small_function_p): Return false if
2405 inlining has already failed with CIF_FINAL_ERROR.
2406 (update_caller_keys): Call want_inline_small_function_p before
2408 (update_callee_keys): Likewise.
2410 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2412 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2414 (rs6000_quadword_masked_address_p): Likewise.
2415 (quad_aligned_load_p): Likewise.
2416 (quad_aligned_store_p): Likewise.
2417 (const_load_sequence_p): Add comment to describe the outer-most loop.
2418 (mimic_memory_attributes_and_flags): New function.
2419 (rs6000_gen_stvx): Likewise.
2420 (replace_swapped_aligned_store): Likewise.
2421 (rs6000_gen_lvx): Likewise.
2422 (replace_swapped_aligned_load): Likewise.
2423 (replace_swapped_load_constant): Capitalize argument name in
2424 comment describing this function.
2425 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2426 and stores that access quad-word aligned addresses and replace
2427 with stvx or lvx instructions when appropriate.
2428 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2429 New function prototype.
2430 (rs6000_quadword_masked_address_p): Likewise.
2431 (rs6000_gen_lvx): Likewise.
2432 (rs6000_gen_stvx): Likewise.
2433 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2434 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2435 when memory address is aligned.
2436 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2437 this split to select lvx instruction when memory address is aligned.
2438 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2439 instruction when memory address is aligned.
2440 (*vsx_le_perm_load_v16qi): Likewise.
2441 (four unnamed splitters): Modify to select the stvx instruction
2442 when memory is aligned.
2444 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2446 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2447 which appears in the queue multiple times.
2449 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2450 Alan Hayward <alan.hayward@arm.com>
2451 David Sherwood <david.sherwood@arm.com>
2453 * tree-vectorizer.h (vec_lower_bound): New structure.
2454 (_loop_vec_info): Add check_nonzero and lower_bounds.
2455 (LOOP_VINFO_CHECK_NONZERO): New macro.
2456 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2457 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2458 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2459 fields. Make seg_len the distance travelled, not including the
2461 (dr_direction_indicator): Declare.
2462 (dr_zero_step_indicator): Likewise.
2463 (dr_known_forward_stride_p): Likewise.
2464 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2466 (runtime_alias_check_p): Allow runtime alias checks with
2468 (operator ==): Compare access_size and align.
2469 (prune_runtime_alias_test_list): Rework for new distinction between
2470 the access_size and seg_len.
2471 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2473 (get_segment_min_max): New function.
2474 (create_intersect_range_checks): Use it.
2475 (dr_step_indicator): New function.
2476 (dr_direction_indicator): Likewise.
2477 (dr_zero_step_indicator): Likewise.
2478 (dr_known_forward_stride_p): Likewise.
2479 * tree-loop-distribution.c (data_ref_segment_size): Return
2480 DR_STEP * (niters - 1).
2481 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2483 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2484 (vect_preserves_scalar_order_p): New function, split out from...
2485 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2486 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2487 (vect_vfa_access_size): New function.
2488 (vect_vfa_align): Likewise.
2489 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2490 (dump_lower_bound): New function.
2491 (vect_check_lower_bound): Likewise.
2492 (vect_small_gap_p): Likewise.
2493 (vectorizable_with_step_bound_p): Likewise.
2494 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2495 depencies if the vectorization factor is 1. Convert the checks
2496 for nonzero steps into checks on the bounds of DR_STEP. Try using
2497 a bunds check for variable steps if the minimum required step is
2498 relatively small. Update calls to the dr_with_seg_len
2499 constructor and to vect_compile_time_alias.
2500 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2502 (vect_loop_versioning): Call it.
2503 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2505 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2507 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2508 Alan Hayward <alan.hayward@arm.com>
2509 David Sherwood <david.sherwood@arm.com>
2511 * doc/sourcebuild.texi (vect_scatter_store): Document.
2512 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2514 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2516 * genopinit.c (main): Add supports_vec_scatter_store and
2517 supports_vec_scatter_store_cached to target_optabs.
2518 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2519 IFN_MASK_SCATTER_STORE.
2520 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2522 * internal-fn.h (internal_store_fn_p): Declare.
2523 (internal_fn_stored_value_index): Likewise.
2524 * internal-fn.c (scatter_store_direct): New macro.
2525 (expand_scatter_store_optab_fn): New function.
2526 (direct_scatter_store_optab_supported_p): New macro.
2527 (internal_store_fn_p): New function.
2528 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2529 IFN_MASK_SCATTER_STORE.
2530 (internal_fn_mask_index): Likewise.
2531 (internal_fn_stored_value_index): New function.
2532 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2534 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2535 * optabs-query.c (supports_vec_scatter_store_p): New function.
2536 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2537 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2538 true for scatter stores.
2539 (vect_gather_scatter_fn_p): Handle scatter stores too.
2540 (vect_check_gather_scatter): Consider using scatter stores if
2541 supports_vec_scatter_store_p.
2542 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2544 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2545 internal_fn_stored_value_index.
2546 (check_load_store_masking): Handle scatter stores too.
2547 (vect_get_store_rhs): Make public.
2548 (vectorizable_call): Use internal_store_fn_p.
2549 (vectorizable_store): Handle scatter store internal functions.
2550 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2551 when deciding whether the end of the group has been reached.
2552 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2553 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2554 (mask_scatter_store<mode>): New insns.
2556 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2557 Alan Hayward <alan.hayward@arm.com>
2558 David Sherwood <david.sherwood@arm.com>
2560 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2561 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2562 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2564 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2565 Use vect_truncate_gather_scatter_offset if we can't treat the
2566 operation as a normal gather load or scatter store.
2567 (get_group_load_store_type): Take the gather_scatter_info
2568 as argument. Try using a gather load or scatter store for
2569 single-element groups.
2570 (get_load_store_type): Update calls to get_group_load_store_type
2571 and vect_use_strided_gather_scatters_p.
2573 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2574 Alan Hayward <alan.hayward@arm.com>
2575 David Sherwood <david.sherwood@arm.com>
2577 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2578 optional tree argument.
2579 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2581 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2582 but continue to use the current value as a fallback.
2583 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2584 to compare the updates.
2585 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2586 (get_load_store_type): Use it when handling a strided access.
2587 (vect_get_strided_load_store_ops): New function.
2588 (vect_get_data_ptr_increment): Likewise.
2589 (vectorizable_load): Handle strided gather loads. Always pass
2590 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2592 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2593 Alan Hayward <alan.hayward@arm.com>
2594 David Sherwood <david.sherwood@arm.com>
2596 * doc/md.texi (gather_load@var{m}): Document.
2597 (mask_gather_load@var{m}): Likewise.
2598 * genopinit.c (main): Add supports_vec_gather_load and
2599 supports_vec_gather_load_cached to target_optabs.
2600 * optabs-tree.c (init_tree_optimization_optabs): Use
2601 ggc_cleared_alloc to allocate target_optabs.
2602 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2603 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2605 * internal-fn.h (internal_load_fn_p): Declare.
2606 (internal_gather_scatter_fn_p): Likewise.
2607 (internal_fn_mask_index): Likewise.
2608 (internal_gather_scatter_fn_supported_p): Likewise.
2609 * internal-fn.c (gather_load_direct): New macro.
2610 (expand_gather_load_optab_fn): New function.
2611 (direct_gather_load_optab_supported_p): New macro.
2612 (direct_internal_fn_optab): New function.
2613 (internal_load_fn_p): Likewise.
2614 (internal_gather_scatter_fn_p): Likewise.
2615 (internal_fn_mask_index): Likewise.
2616 (internal_gather_scatter_fn_supported_p): Likewise.
2617 * optabs-query.c (supports_at_least_one_mode_p): New function.
2618 (supports_vec_gather_load_p): Likewise.
2619 * optabs-query.h (supports_vec_gather_load_p): Declare.
2620 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2621 and memory_type field.
2622 (NUM_PATTERNS): Bump to 15.
2623 * tree-vect-data-refs.c: Include internal-fn.h.
2624 (vect_gather_scatter_fn_p): New function.
2625 (vect_describe_gather_scatter_call): Likewise.
2626 (vect_check_gather_scatter): Try using internal functions for
2627 gather loads. Recognize existing calls to a gather load function.
2628 (vect_analyze_data_refs): Consider using gather loads if
2629 supports_vec_gather_load_p.
2630 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2631 (vect_get_gather_scatter_offset_type): Likewise.
2632 (vect_convert_mask_for_vectype): Likewise.
2633 (vect_add_conversion_to_patterm): Likewise.
2634 (vect_try_gather_scatter_pattern): Likewise.
2635 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2636 (vect_vect_recog_func_ptrs): Add it.
2637 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2638 internal_fn_mask_index and internal_gather_scatter_fn_p.
2639 (check_load_store_masking): Take the gather_scatter_info as an
2640 argument and handle gather loads.
2641 (vect_get_gather_scatter_ops): New function.
2642 (vectorizable_call): Check internal_load_fn_p.
2643 (vectorizable_load): Likewise. Handle gather load internal
2645 (vectorizable_store): Update call to check_load_store_masking.
2646 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2647 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2648 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2649 (aarch64_gather_scale_operand_d): New predicates.
2650 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2651 (mask_gather_load<mode>): New insns.
2653 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2654 Alan Hayward <alan.hayward@arm.com>
2655 David Sherwood <david.sherwood@arm.com>
2657 * optabs.def (fold_left_plus_optab): New optab.
2658 * doc/md.texi (fold_left_plus_@var{m}): Document.
2659 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2660 * internal-fn.c (fold_left_direct): Define.
2661 (expand_fold_left_optab_fn): Likewise.
2662 (direct_fold_left_optab_supported_p): Likewise.
2663 * fold-const-call.c (fold_const_fold_left): New function.
2664 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2665 * tree-parloops.c (valid_reduction_p): New function.
2666 (gather_scalar_reductions): Use it.
2667 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2668 (vect_finish_replace_stmt): Declare.
2669 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2670 (needs_fold_left_reduction_p): New function, split out from...
2671 (vect_is_simple_reduction): ...here. Accept reductions that
2672 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2673 (vect_force_simple_reduction): Also store the reduction type in
2674 the assignment's STMT_VINFO_REDUC_TYPE.
2675 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2676 (merge_with_identity): New function.
2677 (vect_expand_fold_left): Likewise.
2678 (vectorize_fold_left_reduction): Likewise.
2679 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2680 scalar phi in place for it. Check for target support and reject
2681 cases that would reassociate the operation. Defer the transform
2682 phase to vectorize_fold_left_reduction.
2683 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2684 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2685 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2687 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2689 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2690 call to ifc_temp_var.
2692 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2693 Alan Hayward <alan.hayward@arm.com>
2694 David Sherwood <david.sherwood@arm.com>
2696 * target.def (legitimize_address_displacement): Take the original
2697 offset as a poly_int.
2698 * targhooks.h (default_legitimize_address_displacement): Update
2700 * targhooks.c (default_legitimize_address_displacement): Likewise.
2701 * doc/tm.texi: Regenerate.
2702 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2703 as an argument, moving assert of ad->disp == ad->disp_term to...
2704 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2705 Try calling targetm.legitimize_address_displacement before expanding
2706 the address rather than afterwards, and adjust for the new interface.
2707 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2708 Match the new hook interface. Handle SVE addresses.
2709 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2714 * Makefile.in (OBJS): Add early-remat.o.
2715 * target.def (select_early_remat_modes): New hook.
2716 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2717 * doc/tm.texi: Regenerate.
2718 * targhooks.h (default_select_early_remat_modes): Declare.
2719 * targhooks.c (default_select_early_remat_modes): New function.
2720 * timevar.def (TV_EARLY_REMAT): New timevar.
2721 * passes.def (pass_early_remat): New pass.
2722 * tree-pass.h (make_pass_early_remat): Declare.
2723 * early-remat.c: New file.
2724 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2726 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2728 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2729 Alan Hayward <alan.hayward@arm.com>
2730 David Sherwood <david.sherwood@arm.com>
2732 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2733 vfm1 with a bound_epilog parameter.
2734 (vect_do_peeling): Update calls accordingly, and move the prologue
2735 call earlier in the function. Treat the base bound_epilog as 0 for
2736 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2737 this base when peeling for gaps.
2738 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2739 with fully-masked loops.
2740 (vect_estimate_min_profitable_iters): Handle the single peeled
2741 iteration in that case.
2743 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2744 Alan Hayward <alan.hayward@arm.com>
2745 David Sherwood <david.sherwood@arm.com>
2747 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2748 single-element interleaving even if the size is not a power of 2.
2749 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2750 accesses for single-element interleaving if the group size is
2753 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2754 Alan Hayward <alan.hayward@arm.com>
2755 David Sherwood <david.sherwood@arm.com>
2757 * doc/md.texi (fold_extract_last_@var{m}): Document.
2758 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2759 * optabs.def (fold_extract_last_optab): New optab.
2760 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2761 * internal-fn.c (fold_extract_direct): New macro.
2762 (expand_fold_extract_optab_fn): Likewise.
2763 (direct_fold_extract_optab_supported_p): Likewise.
2764 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2765 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2766 EXTRACT_LAST_REDUCTION.
2767 (get_initial_def_for_reduction): Do not create an initial vector
2768 for EXTRACT_LAST_REDUCTION reductions.
2769 (vectorizable_reduction): Leave the scalar phi in place for
2770 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2771 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2772 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2773 transform phase to vectorizable_condition.
2774 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2776 (vect_finish_stmt_generation): ...here.
2777 (vect_finish_replace_stmt): New function.
2778 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2779 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2781 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2783 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2784 Alan Hayward <alan.hayward@arm.com>
2785 David Sherwood <david.sherwood@arm.com>
2787 * doc/md.texi (extract_last_@var{m}): Document.
2788 * optabs.def (extract_last_optab): New optab.
2789 * internal-fn.def (EXTRACT_LAST): New internal function.
2790 * internal-fn.c (cond_unary_direct): New macro.
2791 (expand_cond_unary_optab_fn): Likewise.
2792 (direct_cond_unary_optab_supported_p): Likewise.
2793 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2794 loops using EXTRACT_LAST.
2795 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2796 (extract_last_<mode>): ...this optab.
2797 (vec_extract<mode><Vel>): Update accordingly.
2799 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2800 Alan Hayward <alan.hayward@arm.com>
2801 David Sherwood <david.sherwood@arm.com>
2803 * target.def (empty_mask_is_expensive): New hook.
2804 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2805 * doc/tm.texi: Regenerate.
2806 * targhooks.h (default_empty_mask_is_expensive): Declare.
2807 * targhooks.c (default_empty_mask_is_expensive): New function.
2808 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2809 if the target says that empty masks are expensive.
2810 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2812 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2814 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2815 Alan Hayward <alan.hayward@arm.com>
2816 David Sherwood <david.sherwood@arm.com>
2818 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2819 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2820 (vect_use_loop_mask_for_alignment_p): New function.
2821 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2822 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2823 niters_skip argument. Make sure that the first niters_skip elements
2824 of the first iteration are inactive.
2825 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2826 Update call to vect_set_loop_masks_directly.
2827 (get_misalign_in_elems): New function, split out from...
2828 (vect_gen_prolog_loop_niters): ...here.
2829 (vect_update_init_of_dr): Take a code argument that specifies whether
2830 the adjustment should be added or subtracted.
2831 (vect_update_init_of_drs): Likewise.
2832 (vect_prepare_for_masked_peels): New function.
2833 (vect_do_peeling): Skip prologue peeling if we're using a mask
2834 instead. Update call to vect_update_inits_of_drs.
2835 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2837 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2838 alignment. Do not include the number of peeled iterations in
2839 the minimum threshold in that case.
2840 (vectorizable_induction): Adjust the start value down by
2841 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2842 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2843 Take the number of skipped iterations into account when calculating
2845 * tree-vect-stmts.c (vect_gen_while_not): New function.
2847 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2848 Alan Hayward <alan.hayward@arm.com>
2849 David Sherwood <david.sherwood@arm.com>
2851 * doc/sourcebuild.texi (vect_fully_masked): Document.
2852 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2854 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2856 (vect_analyze_loop_2): ...here. Don't check the vectorization
2857 factor against the number of loop iterations if the loop is
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 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2865 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2866 (dump_groups): Update accordingly.
2867 (iv_use::mem_type): New member variable.
2868 (address_p): New function.
2869 (record_use): Add a mem_type argument and initialize the new
2871 (record_group_use): Add a mem_type argument. Use address_p.
2872 Remove obsolete null checks of base_object. Update call to record_use.
2873 (find_interesting_uses_op): Update call to record_group_use.
2874 (find_interesting_uses_cond): Likewise.
2875 (find_interesting_uses_address): Likewise.
2876 (get_mem_type_for_internal_fn): New function.
2877 (find_address_like_use): Likewise.
2878 (find_interesting_uses_stmt): Try find_address_like_use before
2879 calling find_interesting_uses_op.
2880 (addr_offset_valid_p): Use the iv mem_type field as the type
2881 of the addressed memory.
2882 (add_autoinc_candidates): Likewise.
2883 (get_address_cost): Likewise.
2884 (split_small_address_groups_p): Use address_p.
2885 (split_address_groups): Likewise.
2886 (add_iv_candidate_for_use): Likewise.
2887 (autoinc_possible_for_pair): Likewise.
2888 (rewrite_groups): Likewise.
2889 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2890 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2891 (get_alias_ptr_type_for_ptr_address): New function.
2892 (rewrite_use_address): Rewrite address uses in calls that were
2893 identified by find_address_like_use.
2895 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2896 Alan Hayward <alan.hayward@arm.com>
2897 David Sherwood <david.sherwood@arm.com>
2899 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2901 * gimple-expr.h (is_gimple_addressable: Likewise.
2902 * gimple-expr.c (is_gimple_address): Likewise.
2903 * internal-fn.c (expand_call_mem_ref): New function.
2904 (expand_mask_load_optab_fn): Use it.
2905 (expand_mask_store_optab_fn): Likewise.
2907 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2908 Alan Hayward <alan.hayward@arm.com>
2909 David Sherwood <david.sherwood@arm.com>
2911 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2912 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2913 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2914 (cond_umax@var{mode}): Document.
2915 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2916 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2917 (cond_umin_optab, cond_umax_optab): New optabs.
2918 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2919 (COND_IOR, COND_XOR): New internal functions.
2920 * internal-fn.h (get_conditional_internal_fn): Declare.
2921 * internal-fn.c (cond_binary_direct): New macro.
2922 (expand_cond_binary_optab_fn): Likewise.
2923 (direct_cond_binary_optab_supported_p): Likewise.
2924 (get_conditional_internal_fn): New function.
2925 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2926 Cope with reduction statements that are vectorized as calls rather
2928 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2929 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2930 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2931 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2932 (UNSPEC_COND_EOR): New unspecs.
2933 (optab): Add mappings for them.
2934 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2935 (sve_int_op, sve_fp_op): New int attributes.
2937 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2938 Alan Hayward <alan.hayward@arm.com>
2939 David Sherwood <david.sherwood@arm.com>
2941 * optabs.def (while_ult_optab): New optab.
2942 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2943 * internal-fn.def (WHILE_ULT): New internal function.
2944 * internal-fn.h (direct_internal_fn_supported_p): New override
2945 that takes two types as argument.
2946 * internal-fn.c (while_direct): New macro.
2947 (expand_while_optab_fn): New function.
2948 (convert_optab_supported_p): Likewise.
2949 (direct_while_optab_supported_p): New macro.
2950 * wide-int.h (wi::udiv_ceil): New function.
2951 * tree-vectorizer.h (rgroup_masks): New structure.
2952 (vec_loop_masks): New typedef.
2953 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2955 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2956 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2957 (vect_max_vf): New function.
2958 (slpeel_make_loop_iterate_ntimes): Delete.
2959 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2960 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2961 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2962 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2963 internal-fn.h, stor-layout.h and optabs-query.h.
2964 (vect_set_loop_mask): New function.
2965 (add_preheader_seq): Likewise.
2966 (add_header_seq): Likewise.
2967 (interleave_supported_p): Likewise.
2968 (vect_maybe_permute_loop_masks): Likewise.
2969 (vect_set_loop_masks_directly): Likewise.
2970 (vect_set_loop_condition_masked): Likewise.
2971 (vect_set_loop_condition_unmasked): New function, split out from
2972 slpeel_make_loop_iterate_ntimes.
2973 (slpeel_make_loop_iterate_ntimes): Rename to..
2974 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2975 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2976 (vect_do_peeling): Update call accordingly.
2977 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2979 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2980 mask_compare_type, can_fully_mask_p and fully_masked_p.
2981 (release_vec_loop_masks): New function.
2982 (_loop_vec_info): Use it to free the loop masks.
2983 (can_produce_all_loop_masks_p): New function.
2984 (vect_get_max_nscalars_per_iter): Likewise.
2985 (vect_verify_full_masking): Likewise.
2986 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2987 retries, and free the mask rgroups before retrying. Check loop-wide
2988 reasons for disallowing fully-masked loops. Make the final decision
2989 about whether use a fully-masked loop or not.
2990 (vect_estimate_min_profitable_iters): Do not assume that peeling
2991 for the number of iterations will be needed for fully-masked loops.
2992 (vectorizable_reduction): Disable fully-masked loops.
2993 (vectorizable_live_operation): Likewise.
2994 (vect_halve_mask_nunits): New function.
2995 (vect_double_mask_nunits): Likewise.
2996 (vect_record_loop_mask): Likewise.
2997 (vect_get_loop_mask): Likewise.
2998 (vect_transform_loop): Handle the case in which the final loop
2999 iteration might handle a partial vector. Call vect_set_loop_condition
3000 instead of slpeel_make_loop_iterate_ntimes.
3001 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3002 (check_load_store_masking): New function.
3003 (prepare_load_store_mask): Likewise.
3004 (vectorizable_store): Handle fully-masked loops.
3005 (vectorizable_load): Likewise.
3006 (supportable_widening_operation): Use vect_halve_mask_nunits for
3008 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3009 (vect_gen_while): New function.
3010 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3011 (aarch64_uqdec<mode>): New insn.
3013 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3014 Alan Hayward <alan.hayward@arm.com>
3015 David Sherwood <david.sherwood@arm.com>
3017 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3018 (reduc_xor_scal_optab): New optabs.
3019 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3020 (reduc_xor_scal_@var{m}): Document.
3021 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3022 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3024 * fold-const-call.c (fold_const_call): Handle them.
3025 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3026 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3027 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3028 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3029 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3030 (UNSPEC_XORV): New unspecs.
3031 (optab): Add entries for them.
3032 (BITWISEV): New int iterator.
3033 (bit_reduc_op): New int attributes.
3035 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3036 Alan Hayward <alan.hayward@arm.com>
3037 David Sherwood <david.sherwood@arm.com>
3039 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3040 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3041 * optabs.def (vec_shl_insert_optab): New optab.
3042 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3043 (duplicate_and_interleave): Likewise.
3044 * tree-vect-loop.c: Include internal-fn.h.
3045 (neutral_op_for_slp_reduction): New function, split out from
3046 get_initial_defs_for_reduction.
3047 (get_initial_def_for_reduction): Handle option 2 for variable-length
3048 vectors by loading the neutral value into a vector and then shifting
3049 the initial value into element 0.
3050 (get_initial_defs_for_reduction): Replace the code argument with
3051 the neutral value calculated by neutral_op_for_slp_reduction.
3052 Use gimple_build_vector for constant-length vectors.
3053 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3054 but the first group_size elements have a neutral value.
3055 Use duplicate_and_interleave otherwise.
3056 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3057 Update call to get_initial_defs_for_reduction. Handle SLP
3058 reductions for variable-length vectors by creating one vector
3059 result for each scalar result, with the elements associated
3060 with other scalar results stubbed out with the neutral value.
3061 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3062 Require IFN_VEC_SHL_INSERT for double reductions on
3063 variable-length vectors, or SLP reductions that have
3064 a neutral value. Require can_duplicate_and_interleave_p
3065 support for variable-length unchained SLP reductions if there
3066 is no neutral value, such as for MIN/MAX reductions. Also require
3067 the number of vector elements to be a multiple of the number of
3068 SLP statements when doing variable-length unchained SLP reductions.
3069 Update call to vect_create_epilog_for_reduction.
3070 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3071 and remove initial values.
3072 (duplicate_and_interleave): Make public.
3073 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3074 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3076 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3077 Alan Hayward <alan.hayward@arm.com>
3078 David Sherwood <david.sherwood@arm.com>
3080 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3081 (can_duplicate_and_interleave_p): New function.
3082 (vect_get_and_check_slp_defs): Take the vector of statements
3083 rather than just the current one. Remove excess parentheses.
3084 Restriction rejectinon of vect_constant_def and vect_external_def
3085 for variable-length vectors to boolean types, or types for which
3086 can_duplicate_and_interleave_p is false.
3087 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3088 (duplicate_and_interleave): New function.
3089 (vect_get_constant_vectors): Use gimple_build_vector for
3090 constant-length vectors and suitable variable-length constant
3091 vectors. Use duplicate_and_interleave for other variable-length
3092 vectors. Don't defer the update when inserting new statements.
3094 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3095 Alan Hayward <alan.hayward@arm.com>
3096 David Sherwood <david.sherwood@arm.com>
3098 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3099 min_profitable_iters doesn't go negative.
3101 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3102 Alan Hayward <alan.hayward@arm.com>
3103 David Sherwood <david.sherwood@arm.com>
3105 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3106 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3107 * optabs.def (vec_mask_load_lanes_optab): New optab.
3108 (vec_mask_store_lanes_optab): Likewise.
3109 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3110 (MASK_STORE_LANES): Likewise.
3111 * internal-fn.c (mask_load_lanes_direct): New macro.
3112 (mask_store_lanes_direct): Likewise.
3113 (expand_mask_load_optab_fn): Handle masked operations.
3114 (expand_mask_load_lanes_optab_fn): New macro.
3115 (expand_mask_store_optab_fn): Handle masked operations.
3116 (expand_mask_store_lanes_optab_fn): New macro.
3117 (direct_mask_load_lanes_optab_supported_p): Likewise.
3118 (direct_mask_store_lanes_optab_supported_p): Likewise.
3119 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3121 (vect_load_lanes_supported): Likewise.
3122 * tree-vect-data-refs.c (strip_conversion): New function.
3123 (can_group_stmts_p): Likewise.
3124 (vect_analyze_data_ref_accesses): Use it instead of checking
3125 for a pair of assignments.
3126 (vect_store_lanes_supported): Take a masked_p parameter.
3127 (vect_load_lanes_supported): Likewise.
3128 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3129 vect_store_lanes_supported and vect_load_lanes_supported.
3130 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3131 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3132 parameter. Don't allow gaps for masked accesses.
3133 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3134 and vect_load_lanes_supported.
3135 (get_load_store_type): Take a masked_p parameter and update
3136 call to get_group_load_store_type.
3137 (vectorizable_store): Update call to get_load_store_type.
3138 Handle IFN_MASK_STORE_LANES.
3139 (vectorizable_load): Update call to get_load_store_type.
3140 Handle IFN_MASK_LOAD_LANES.
3142 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3143 Alan Hayward <alan.hayward@arm.com>
3144 David Sherwood <david.sherwood@arm.com>
3146 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3148 * config/aarch64/aarch64-protos.h
3149 (aarch64_sve_struct_memory_operand_p): Declare.
3150 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3151 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3152 (VPRED, vpred): Handle SVE structure modes.
3153 * config/aarch64/constraints.md (Utx): New constraint.
3154 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3155 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3156 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3157 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3158 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3159 structure modes. Split into pieces after RA.
3160 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3161 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3163 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3164 SVE structure modes.
3165 (aarch64_classify_address): Likewise.
3166 (sizetochar): Move earlier in file.
3167 (aarch64_print_operand): Handle SVE register lists.
3168 (aarch64_array_mode): New function.
3169 (aarch64_sve_struct_memory_operand_p): Likewise.
3170 (TARGET_ARRAY_MODE): Redefine.
3172 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3173 Alan Hayward <alan.hayward@arm.com>
3174 David Sherwood <david.sherwood@arm.com>
3176 * target.def (array_mode): New target hook.
3177 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3178 * doc/tm.texi: Regenerate.
3179 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3180 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3181 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3183 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3186 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3187 Alan Hayward <alan.hayward@arm.com>
3188 David Sherwood <david.sherwood@arm.com>
3190 * fold-const.c (fold_binary_loc): Check the argument types
3191 rather than the result type when testing for a vector operation.
3193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3195 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3196 * doc/tm.texi: Regenerate.
3198 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3199 Alan Hayward <alan.hayward@arm.com>
3200 David Sherwood <david.sherwood@arm.com>
3202 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3203 (sve): Document new AArch64 extension.
3204 * doc/md.texi (w): Extend the description of the AArch64
3205 constraint to include SVE vectors.
3206 (Upl, Upa): Document new AArch64 predicate constraints.
3207 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3209 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3210 (msve-vector-bits=): New option.
3211 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3212 SVE when these are disabled.
3213 (sve): New extension.
3214 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3215 modes. Adjust their number of units based on aarch64_sve_vg.
3216 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3217 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3218 aarch64_addr_query_type.
3219 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3220 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3221 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3222 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3223 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3224 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3225 (aarch64_simd_imm_zero_p): Delete.
3226 (aarch64_check_zero_based_sve_index_immediate): Declare.
3227 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3228 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3229 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3230 (aarch64_sve_float_mul_immediate_p): Likewise.
3231 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3233 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3234 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3235 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3236 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3237 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3238 (aarch64_regmode_natural_size): Likewise.
3239 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3240 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3242 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3243 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3244 for VG and the SVE predicate registers.
3245 (V_ALIASES): Add a "z"-prefixed alias.
3246 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3247 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3248 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3249 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3250 (REG_CLASS_NAMES): Add entries for them.
3251 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3252 and the predicate registers.
3253 (aarch64_sve_vg): Declare.
3254 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3255 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3256 (REGMODE_NATURAL_SIZE): Define.
3257 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3259 * config/aarch64/aarch64.c: Include cfgrtl.h.
3260 (simd_immediate_info): Add a constructor for series vectors,
3261 and an associated step field.
3262 (aarch64_sve_vg): New variable.
3263 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3264 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3265 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3266 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3267 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3268 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3269 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3270 (aarch64_get_mask_mode): New functions.
3271 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3272 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3273 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3274 predicate modes and predicate registers. Explicitly restrict
3275 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3276 to store a vector mode if it is recognized by
3277 aarch64_classify_vector_mode.
3278 (aarch64_regmode_natural_size): New function.
3279 (aarch64_hard_regno_caller_save_mode): Return the original mode
3281 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3282 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3283 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3284 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3286 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3287 does not overlap dest if the function is frame-related. Handle
3289 (aarch64_split_add_offset): New function.
3290 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3291 them aarch64_add_offset.
3292 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3293 and update call to aarch64_sub_sp.
3294 (aarch64_add_cfa_expression): New function.
3295 (aarch64_expand_prologue): Pass extra temporary registers to the
3296 functions above. Handle the case in which we need to emit new
3297 DW_CFA_expressions for registers that were originally saved
3298 relative to the stack pointer, but now have to be expressed
3299 relative to the frame pointer.
3300 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3302 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3303 IP0 and IP1 values for SVE frames.
3304 (aarch64_expand_vec_series): New function.
3305 (aarch64_expand_sve_widened_duplicate): Likewise.
3306 (aarch64_expand_sve_const_vector): Likewise.
3307 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3308 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3309 into the register, rather than emitting a SET directly.
3310 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3311 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3312 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3313 (offset_9bit_signed_scaled_p): New functions.
3314 (aarch64_replicate_bitmask_imm): New function.
3315 (aarch64_bitmask_imm): Use it.
3316 (aarch64_cannot_force_const_mem): Reject expressions involving
3317 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3318 (aarch64_classify_index): Handle SVE indices, by requiring
3319 a plain register index with a scale that matches the element size.
3320 (aarch64_classify_address): Handle SVE addresses. Assert that
3321 the mode of the address is VOIDmode or an integer mode.
3322 Update call to aarch64_classify_symbol.
3323 (aarch64_classify_symbolic_expression): Update call to
3324 aarch64_classify_symbol.
3325 (aarch64_const_vec_all_in_range_p): New function.
3326 (aarch64_print_vector_float_operand): Likewise.
3327 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3328 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3329 and the FP immediates 1.0 and 0.5.
3330 (aarch64_print_address_internal): Handle SVE addresses.
3331 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3332 (aarch64_regno_regclass): Handle predicate registers.
3333 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3335 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3336 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3337 (aarch64_convert_sve_vector_bits): New function.
3338 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3339 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3341 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3342 Handle SVE vector and predicate modes. Accept VL-based constants
3343 that need only one temporary register, and VL offsets that require
3344 no temporary registers.
3345 (aarch64_conditional_register_usage): Mark the predicate registers
3346 as fixed if SVE isn't available.
3347 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3348 Return true for SVE vector and predicate modes.
3349 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3350 rather than an unsigned int. Handle SVE modes.
3351 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3353 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3355 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3356 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3357 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3358 (aarch64_sve_float_mul_immediate_p): New functions.
3359 (aarch64_sve_valid_immediate): New function.
3360 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3361 Explicitly reject structure modes. Check for INDEX constants.
3362 Handle PTRUE and PFALSE constants.
3363 (aarch64_check_zero_based_sve_index_immediate): New function.
3364 (aarch64_simd_imm_zero_p): Delete.
3365 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3366 vector modes. Accept constants in the range of CNT[BHWD].
3367 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3368 ask for an Advanced SIMD mode.
3369 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3370 (aarch64_simd_vector_alignment): Handle SVE predicates.
3371 (aarch64_vectorize_preferred_vector_alignment): New function.
3372 (aarch64_simd_vector_alignment_reachable): Use it instead of
3374 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3375 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3377 (MAX_VECT_LEN): Delete.
3378 (expand_vec_perm_d): Add a vec_flags field.
3379 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3380 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3381 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3383 (aarch64_evpc_rev): Rename to...
3384 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3385 (aarch64_evpc_rev_global): New function.
3386 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3387 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3389 (aarch64_evpc_sve_tbl): New function.
3390 (aarch64_expand_vec_perm_const_1): Update after rename of
3391 aarch64_evpc_rev. Handle SVE permutes too, trying
3392 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3393 than aarch64_evpc_tbl.
3394 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3395 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3396 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3397 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3398 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3399 (aarch64_expand_sve_vcond): New functions.
3400 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3401 of aarch64_vector_mode_p.
3402 (aarch64_dwarf_poly_indeterminate_value): New function.
3403 (aarch64_compute_pressure_classes): Likewise.
3404 (aarch64_can_change_mode_class): Likewise.
3405 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3406 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3407 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3408 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3409 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3410 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3411 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3412 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3414 (Dn, Dl, Dr): Accept const as well as const_vector.
3415 (Dz): Likewise. Compare against CONST0_RTX.
3416 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3417 of "vector" where appropriate.
3418 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3419 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3420 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3421 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3422 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3423 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3424 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3425 (v_int_equiv): Extend to SVE modes.
3426 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3428 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3429 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3430 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3431 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3432 (SVE_COND_FP_CMP): New int iterators.
3433 (perm_hilo): Handle the new unpack unspecs.
3434 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3436 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3437 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3438 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3439 (aarch64_equality_operator, aarch64_constant_vector_operand)
3440 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3441 (aarch64_sve_nonimmediate_operand): Likewise.
3442 (aarch64_sve_general_operand): Likewise.
3443 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3444 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3445 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3446 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3447 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3448 (aarch64_sve_float_arith_immediate): Likewise.
3449 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3450 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3451 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3452 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3453 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3454 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3455 (aarch64_sve_float_arith_operand): Likewise.
3456 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3457 (aarch64_sve_float_mul_operand): Likewise.
3458 (aarch64_sve_vec_perm_operand): Likewise.
3459 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3460 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3461 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3462 as well as const_vector.
3463 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3464 in file. Use CONST0_RTX and CONSTM1_RTX.
3465 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3466 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3467 Use aarch64_simd_imm_zero.
3468 * config/aarch64/aarch64-sve.md: New file.
3469 * config/aarch64/aarch64.md: Include it.
3470 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3471 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3472 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3473 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3474 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3475 (sve): New attribute.
3476 (enabled): Disable instructions with the sve attribute unless
3478 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3479 aarch64_expand_mov_immediate.
3480 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3481 CNT[BHSD] immediates.
3482 (movti): Split CONST_POLY_INT moves into two halves.
3483 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3484 Split additions that need a temporary here if the destination
3485 is the stack pointer.
3486 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3487 (*add<mode>3_poly_1): New instruction.
3488 (set_clobber_cc): New expander.
3490 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3492 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3493 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3494 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3495 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3496 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3497 Change innermode from fixed_mode_size to machine_mode.
3498 (simplify_subreg): Update call accordingly. Handle a constant-sized
3499 subreg of a variable-length CONST_VECTOR.
3501 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3502 Alan Hayward <alan.hayward@arm.com>
3503 David Sherwood <david.sherwood@arm.com>
3505 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3506 (add_offset_to_base): New function, split out from...
3507 (create_mem_ref): ...here. When handling a scale other than 1,
3508 check first whether the address is valid without the offset.
3509 Add it into the base if so, leaving the index and scale as-is.
3511 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3514 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3515 fold_for_warn before checking if arg2 is INTEGER_CST.
3517 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3519 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3520 (store_multiple_operation): Delete.
3521 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3522 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3523 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3524 guarded by TARGET_STRING.
3525 (rs6000_output_load_multiple): Delete.
3526 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3527 OPTION_MASK_STRING / TARGET_STRING handling.
3528 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3529 (const rs6000_opt_masks) <"string">: Change mask to 0.
3530 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3531 (MASK_STRING): Delete.
3532 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3534 (load_multiple): Delete.
3541 (store_multiple): Delete.
3548 (movmemsi_8reg): Delete.
3549 (corresponding unnamed define_insn): Delete.
3550 (movmemsi_6reg): Delete.
3551 (corresponding unnamed define_insn): Delete.
3552 (movmemsi_4reg): Delete.
3553 (corresponding unnamed define_insn): Delete.
3554 (movmemsi_2reg): Delete.
3555 (corresponding unnamed define_insn): Delete.
3556 (movmemsi_1reg): Delete.
3557 (corresponding unnamed define_insn): Delete.
3558 * config/rs6000/rs6000.opt (mno-string): New.
3559 (mstring): Replace by deprecation warning stub.
3560 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3562 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3564 * regrename.c (regrename_do_replace): If replacing the same
3565 reg multiple times, try to reuse last created gen_raw_REG.
3568 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3569 main to workaround a bug in GDB.
3571 2018-01-12 Tom de Vries <tom@codesourcery.com>
3574 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3576 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3578 PR rtl-optimization/80481
3579 * ira-color.c (get_cap_member): New function.
3580 (allocnos_conflict_by_live_ranges_p): Use it.
3581 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3582 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3584 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3587 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3588 (*saddl_se_1): Ditto.
3590 (*ssubl_se_1): Ditto.
3592 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3594 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3595 rather than wi::to_widest for DR_INITs.
3596 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3597 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3598 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3600 (vect_analyze_group_access_1): Note that here.
3602 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3604 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3605 polynomial type sizes.
3607 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3609 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3610 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3611 (gimple_add_tmp_var): Likewise.
3613 2018-01-12 Martin Liska <mliska@suse.cz>
3615 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3616 (gimple_alloc_sizes): Likewise.
3617 (dump_gimple_statistics): Use PRIu64 in printf format.
3618 * gimple.h: Change uint64_t to int.
3620 2018-01-12 Martin Liska <mliska@suse.cz>
3622 * tree-core.h: Use uint64_t instead of int.
3623 * tree.c (tree_node_counts): Likewise.
3624 (tree_node_sizes): Likewise.
3625 (dump_tree_statistics): Use PRIu64 in printf format.
3627 2018-01-12 Martin Liska <mliska@suse.cz>
3629 * Makefile.in: As qsort_chk is implemented in vec.c, add
3630 vec.o to linkage of gencfn-macros.
3631 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3632 passing the info to record_node_allocation_statistics.
3633 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3635 * ggc-common.c (struct ggc_usage): Add operator== and use
3636 it in operator< and compare function.
3637 * mem-stats.h (struct mem_usage): Likewise.
3638 * vec.c (struct vec_usage): Remove operator< and compare
3639 function. Can be simply inherited.
3641 2018-01-12 Martin Jambor <mjambor@suse.cz>
3644 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3645 * tree-ssa-math-opts.c: Include domwalk.h.
3646 (convert_mult_to_fma_1): New function.
3647 (fma_transformation_info): New type.
3648 (fma_deferring_state): Likewise.
3649 (cancel_fma_deferring): New function.
3650 (result_of_phi): Likewise.
3651 (last_fma_candidate_feeds_initial_phi): Likewise.
3652 (convert_mult_to_fma): Added deferring logic, split actual
3653 transformation to convert_mult_to_fma_1.
3654 (math_opts_dom_walker): New type.
3655 (math_opts_dom_walker::after_dom_children): New method, body moved
3656 here from pass_optimize_widening_mul::execute, added deferring logic
3658 (pass_optimize_widening_mul::execute): Moved most of code to
3659 math_opts_dom_walker::after_dom_children.
3660 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3661 * config/i386/i386.c (ix86_option_override_internal): Added
3662 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3664 2018-01-12 Richard Biener <rguenther@suse.de>
3667 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3668 inline instance vars.
3670 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3673 * config/rx/rx.c (rx_is_restricted_memory_address):
3676 2018-01-12 Richard Biener <rguenther@suse.de>
3678 PR tree-optimization/80846
3679 * target.def (split_reduction): New target hook.
3680 * targhooks.c (default_split_reduction): New function.
3681 * targhooks.h (default_split_reduction): Declare.
3682 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3683 target requests first reduce vectors by combining low and high
3685 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3686 (get_vectype_for_scalar_type_and_size): Export.
3687 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3688 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3689 * doc/tm.texi: Regenerate.
3690 * config/i386/i386.c (ix86_split_reduction): Implement
3691 TARGET_VECTORIZE_SPLIT_REDUCTION.
3693 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3696 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3697 in PIC mode except for TARGET_VXWORKS_RTP.
3698 * config/sparc/sparc.c: Include cfgrtl.h.
3699 (TARGET_INIT_PIC_REG): Define.
3700 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3701 (sparc_pic_register_p): New predicate.
3702 (sparc_legitimate_address_p): Use it.
3703 (sparc_legitimize_pic_address): Likewise.
3704 (sparc_delegitimize_address): Likewise.
3705 (sparc_mode_dependent_address_p): Likewise.
3706 (gen_load_pcrel_sym): Remove 4th parameter.
3707 (load_got_register): Adjust call to above. Remove obsolete stuff.
3708 (sparc_expand_prologue): Do not call load_got_register here.
3709 (sparc_flat_expand_prologue): Likewise.
3710 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3711 (sparc_use_pseudo_pic_reg): New function.
3712 (sparc_init_pic_reg): Likewise.
3713 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3714 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3716 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3718 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3719 Add item for branch_cost.
3721 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3723 PR rtl-optimization/83565
3724 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3725 not extend the result to a larger mode for rotate operations.
3726 (num_sign_bit_copies1): Likewise.
3728 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3731 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3733 Use values-Xc.o for -pedantic.
3734 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3736 2018-01-12 Martin Liska <mliska@suse.cz>
3739 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3741 (possible_polymorphic_call_targets): Use it.
3742 (ipa_devirt): Likewise.
3744 2018-01-12 Martin Liska <mliska@suse.cz>
3746 * profile-count.h (enum profile_quality): Use 0 as invalid
3747 enum value of profile_quality.
3749 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3751 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3752 -mext-string options.
3754 2018-01-12 Richard Biener <rguenther@suse.de>
3756 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3757 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3758 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3760 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3762 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3764 * configure.ac (--with-long-double-format): Add support for the
3765 configuration option to change the default long double format on
3767 * config.gcc (powerpc*-linux*-*): Likewise.
3768 * configure: Regenerate.
3769 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3770 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3771 used without modification.
3773 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3775 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3776 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3777 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3778 MISC_BUILTIN_SPEC_BARRIER.
3779 (rs6000_init_builtins): Likewise.
3780 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3782 (speculation_barrier): New define_insn.
3783 * doc/extend.texi: Document __builtin_speculation_barrier.
3785 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3788 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3789 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3790 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3792 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3793 integral modes instead of "ss" and "sd".
3794 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3795 vectors with 32-bit and 64-bit elements.
3796 (vecdupssescalarmodesuffix): New mode attribute.
3797 (vec_dup<mode>): Use it.
3799 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3802 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3803 frame if argument is passed on stack.
3805 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3808 * ree.c (combine_reaching_defs): Optimize also
3809 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3810 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3812 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3815 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3817 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3820 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3821 after they are computed.
3823 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3825 PR tree-optimization/83695
3826 * gimple-loop-linterchange.cc
3827 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3828 reset cached scev information after interchange.
3829 (pass_linterchange::execute): Remove call to scev_reset_htab.
3831 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3833 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3834 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3835 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3836 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3837 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3838 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3839 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3840 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3841 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3842 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3843 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3844 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3845 (V_lane_reg): Likewise.
3846 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3848 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3849 (vfmal_lane_low<mode>_intrinsic,
3850 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3851 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3852 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3853 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3854 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3855 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3857 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3859 * config/arm/arm-cpus.in (fp16fml): New feature.
3860 (ALL_SIMD): Add fp16fml.
3861 (armv8.2-a): Add fp16fml as an option.
3862 (armv8.3-a): Likewise.
3863 (armv8.4-a): Add fp16fml as part of fp16.
3864 * config/arm/arm.h (TARGET_FP16FML): Define.
3865 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3867 * config/arm/arm-modes.def (V2HF): Define.
3868 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3869 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3870 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3871 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3872 vfmsl_low, vfmsl_high): New set of builtins.
3873 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3874 (vfml_op): New code attribute.
3875 (VFMLHALVES): New int iterator.
3876 (VFML, VFMLSEL): New mode attributes.
3877 (V_reg): Define mapping for V2HF.
3878 (V_hi, V_lo): New mode attributes.
3879 (VF_constraint): Likewise.
3880 (vfml_half, vfml_half_selector): New int attributes.
3881 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3883 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3884 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3886 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3887 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3888 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3889 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3891 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3892 Document new effective target and option set.
3894 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3896 * config/arm/arm-cpus.in (armv8_4): New feature.
3897 (ARMv8_4a): New fgroup.
3898 (armv8.4-a): New arch.
3899 * config/arm/arm-tables.opt: Regenerate.
3900 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3901 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3902 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3903 Add matching rules for -march=armv8.4-a and extensions.
3904 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3906 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3909 * config/rx/rx.md (BW): New mode attribute.
3910 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3912 2018-01-11 Richard Biener <rguenther@suse.de>
3914 PR tree-optimization/83435
3915 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3916 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3917 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3919 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3920 Alan Hayward <alan.hayward@arm.com>
3921 David Sherwood <david.sherwood@arm.com>
3923 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3925 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3926 (aarch64_print_address_internal): Use it to check for a zero offset.
3928 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3929 Alan Hayward <alan.hayward@arm.com>
3930 David Sherwood <david.sherwood@arm.com>
3932 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3933 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3934 Return a poly_int64 rather than a HOST_WIDE_INT.
3935 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3936 rather than a HOST_WIDE_INT.
3937 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3938 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3939 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3940 final_offset from HOST_WIDE_INT to poly_int64.
3941 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3942 to_constant when getting the number of units in an Advanced SIMD
3944 (aarch64_builtin_vectorized_function): Check for a constant number
3946 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3948 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3949 attribute instead of GET_MODE_NUNITS.
3950 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3951 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3952 GET_MODE_SIZE for fixed-size registers.
3953 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3954 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3955 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3956 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3957 (aarch64_print_operand, aarch64_print_address_internal)
3958 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3959 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3960 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3961 Handle polynomial GET_MODE_SIZE.
3962 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3963 wider than SImode without modification.
3964 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3965 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3966 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3967 passing and returning SVE modes.
3968 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3969 rather than GEN_INT.
3970 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3971 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3972 (aarch64_allocate_and_probe_stack_space): Likewise.
3973 (aarch64_layout_frame): Cope with polynomial offsets.
3974 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3975 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3977 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3978 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3979 poly_int64 rather than a HOST_WIDE_INT.
3980 (aarch64_get_separate_components, aarch64_process_components)
3981 (aarch64_expand_prologue, aarch64_expand_epilogue)
3982 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3983 (aarch64_anchor_offset): New function, split out from...
3984 (aarch64_legitimize_address): ...here.
3985 (aarch64_builtin_vectorization_cost): Handle polynomial
3986 TYPE_VECTOR_SUBPARTS.
3987 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3989 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3990 number of elements from the PARALLEL rather than the mode.
3991 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3992 rather than GET_MODE_BITSIZE.
3993 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3994 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3995 (aarch64_expand_vec_perm_const_1): Handle polynomial
3996 d->perm.length () and d->perm elements.
3997 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3998 Apply to_constant to d->perm elements.
3999 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4000 polynomial CONST_VECTOR_NUNITS.
4001 (aarch64_move_pointer): Take amount as a poly_int64 rather
4003 (aarch64_progress_pointer): Avoid temporary variable.
4004 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4005 the mode attribute instead of GET_MODE.
4007 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4008 Alan Hayward <alan.hayward@arm.com>
4009 David Sherwood <david.sherwood@arm.com>
4011 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4012 x exists before using it.
4013 (aarch64_add_constant_internal): Rename to...
4014 (aarch64_add_offset_1): ...this. Replace regnum with separate
4015 src and dest rtxes. Handle the case in which they're different,
4016 including when the offset is zero. Replace scratchreg with an rtx.
4017 Use 2 additions if there is no spare register into which we can
4018 move a 16-bit constant.
4019 (aarch64_add_constant): Delete.
4020 (aarch64_add_offset): Replace reg with separate src and dest
4021 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4022 Use aarch64_add_offset_1.
4023 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4024 an rtx rather than an int. Take the delta as a poly_int64
4025 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4026 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4027 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4028 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4029 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4031 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4032 aarch64_add_constant.
4034 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4036 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4037 Use scalar_float_mode.
4039 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4041 * config/aarch64/aarch64-simd.md
4042 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4043 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4044 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4045 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4046 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4047 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4048 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4049 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4050 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4051 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4053 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4056 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4057 targ_options->x_arm_arch_string is non NULL.
4059 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4061 * config/aarch64/aarch64.h
4062 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4064 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4067 * expmed.c (emit_store_flag_force): Swap if const op0
4068 and change VOIDmode to mode of op0.
4070 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4072 PR rtl-optimization/83761
4073 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4074 than bytes to mode_for_size.
4076 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4079 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4080 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4083 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4086 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4087 when in layout mode.
4088 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4089 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4092 2018-01-10 Michael Collison <michael.collison@arm.com>
4094 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4095 * config/aarch64/aarch64-option-extension.def: Add
4096 AARCH64_OPT_EXTENSION of 'fp16fml'.
4097 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4098 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4099 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4100 * config/aarch64/constraints.md (Ui7): New constraint.
4101 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4102 (VFMLA_SEL_W): Ditto.
4105 (VFMLA16_LOW): New int iterator.
4106 (VFMLA16_HIGH): Ditto.
4107 (UNSPEC_FMLAL): New unspec.
4108 (UNSPEC_FMLSL): Ditto.
4109 (UNSPEC_FMLAL2): Ditto.
4110 (UNSPEC_FMLSL2): Ditto.
4111 (f16mac): New code attribute.
4112 * config/aarch64/aarch64-simd-builtins.def
4113 (aarch64_fmlal_lowv2sf): Ditto.
4114 (aarch64_fmlsl_lowv2sf): Ditto.
4115 (aarch64_fmlalq_lowv4sf): Ditto.
4116 (aarch64_fmlslq_lowv4sf): Ditto.
4117 (aarch64_fmlal_highv2sf): Ditto.
4118 (aarch64_fmlsl_highv2sf): Ditto.
4119 (aarch64_fmlalq_highv4sf): Ditto.
4120 (aarch64_fmlslq_highv4sf): Ditto.
4121 (aarch64_fmlal_lane_lowv2sf): Ditto.
4122 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4123 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4124 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4125 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4126 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4127 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4128 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4129 (aarch64_fmlal_lane_highv2sf): Ditto.
4130 (aarch64_fmlsl_lane_highv2sf): Ditto.
4131 (aarch64_fmlal_laneq_highv2sf): Ditto.
4132 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4133 (aarch64_fmlalq_lane_highv4sf): Ditto.
4134 (aarch64_fmlsl_lane_highv4sf): Ditto.
4135 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4136 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4137 * config/aarch64/aarch64-simd.md:
4138 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4139 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4140 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4141 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4142 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4143 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4144 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4145 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4146 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4147 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4148 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4149 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4150 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4151 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4152 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4153 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4154 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4155 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4156 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4157 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4158 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4159 (vfmlsl_low_u32): Ditto.
4160 (vfmlalq_low_u32): Ditto.
4161 (vfmlslq_low_u32): Ditto.
4162 (vfmlal_high_u32): Ditto.
4163 (vfmlsl_high_u32): Ditto.
4164 (vfmlalq_high_u32): Ditto.
4165 (vfmlslq_high_u32): Ditto.
4166 (vfmlal_lane_low_u32): Ditto.
4167 (vfmlsl_lane_low_u32): Ditto.
4168 (vfmlal_laneq_low_u32): Ditto.
4169 (vfmlsl_laneq_low_u32): Ditto.
4170 (vfmlalq_lane_low_u32): Ditto.
4171 (vfmlslq_lane_low_u32): Ditto.
4172 (vfmlalq_laneq_low_u32): Ditto.
4173 (vfmlslq_laneq_low_u32): Ditto.
4174 (vfmlal_lane_high_u32): Ditto.
4175 (vfmlsl_lane_high_u32): Ditto.
4176 (vfmlal_laneq_high_u32): Ditto.
4177 (vfmlsl_laneq_high_u32): Ditto.
4178 (vfmlalq_lane_high_u32): Ditto.
4179 (vfmlslq_lane_high_u32): Ditto.
4180 (vfmlalq_laneq_high_u32): Ditto.
4181 (vfmlslq_laneq_high_u32): Ditto.
4182 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4183 (AARCH64_FL_FOR_ARCH8_4): New.
4184 (AARCH64_ISA_F16FML): New ISA flag.
4185 (TARGET_F16FML): New feature flag for fp16fml.
4186 (doc/invoke.texi): Document new fp16fml option.
4188 2018-01-10 Michael Collison <michael.collison@arm.com>
4190 * config/aarch64/aarch64-builtins.c:
4191 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4192 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4193 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4194 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4195 (AARCH64_ISA_SHA3): New ISA flag.
4196 (TARGET_SHA3): New feature flag for sha3.
4197 * config/aarch64/iterators.md (sha512_op): New int attribute.
4198 (CRYPTO_SHA512): New int iterator.
4199 (UNSPEC_SHA512H): New unspec.
4200 (UNSPEC_SHA512H2): Ditto.
4201 (UNSPEC_SHA512SU0): Ditto.
4202 (UNSPEC_SHA512SU1): Ditto.
4203 * config/aarch64/aarch64-simd-builtins.def
4204 (aarch64_crypto_sha512hqv2di): New builtin.
4205 (aarch64_crypto_sha512h2qv2di): Ditto.
4206 (aarch64_crypto_sha512su0qv2di): Ditto.
4207 (aarch64_crypto_sha512su1qv2di): Ditto.
4208 (aarch64_eor3qv8hi): Ditto.
4209 (aarch64_rax1qv2di): Ditto.
4210 (aarch64_xarqv2di): Ditto.
4211 (aarch64_bcaxqv8hi): Ditto.
4212 * config/aarch64/aarch64-simd.md:
4213 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4214 (aarch64_crypto_sha512su0qv2di): Ditto.
4215 (aarch64_crypto_sha512su1qv2di): Ditto.
4216 (aarch64_eor3qv8hi): Ditto.
4217 (aarch64_rax1qv2di): Ditto.
4218 (aarch64_xarqv2di): Ditto.
4219 (aarch64_bcaxqv8hi): Ditto.
4220 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4221 (vsha512h2q_u64): Ditto.
4222 (vsha512su0q_u64): Ditto.
4223 (vsha512su1q_u64): Ditto.
4224 (veor3q_u16): Ditto.
4225 (vrax1q_u64): Ditto.
4227 (vbcaxq_u16): Ditto.
4228 * config/arm/types.md (crypto_sha512): New type attribute.
4229 (crypto_sha3): Ditto.
4230 (doc/invoke.texi): Document new sha3 option.
4232 2018-01-10 Michael Collison <michael.collison@arm.com>
4234 * config/aarch64/aarch64-builtins.c:
4235 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4236 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4237 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4238 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4239 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4240 (AARCH64_ISA_SM4): New ISA flag.
4241 (TARGET_SM4): New feature flag for sm4.
4242 * config/aarch64/aarch64-simd-builtins.def
4243 (aarch64_sm3ss1qv4si): Ditto.
4244 (aarch64_sm3tt1aq4si): Ditto.
4245 (aarch64_sm3tt1bq4si): Ditto.
4246 (aarch64_sm3tt2aq4si): Ditto.
4247 (aarch64_sm3tt2bq4si): Ditto.
4248 (aarch64_sm3partw1qv4si): Ditto.
4249 (aarch64_sm3partw2qv4si): Ditto.
4250 (aarch64_sm4eqv4si): Ditto.
4251 (aarch64_sm4ekeyqv4si): Ditto.
4252 * config/aarch64/aarch64-simd.md:
4253 (aarch64_sm3ss1qv4si): Ditto.
4254 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4255 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4256 (aarch64_sm4eqv4si): Ditto.
4257 (aarch64_sm4ekeyqv4si): Ditto.
4258 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4259 (sm3part_op): Ditto.
4260 (CRYPTO_SM3TT): Ditto.
4261 (CRYPTO_SM3PART): Ditto.
4262 (UNSPEC_SM3SS1): New unspec.
4263 (UNSPEC_SM3TT1A): Ditto.
4264 (UNSPEC_SM3TT1B): Ditto.
4265 (UNSPEC_SM3TT2A): Ditto.
4266 (UNSPEC_SM3TT2B): Ditto.
4267 (UNSPEC_SM3PARTW1): Ditto.
4268 (UNSPEC_SM3PARTW2): Ditto.
4269 (UNSPEC_SM4E): Ditto.
4270 (UNSPEC_SM4EKEY): Ditto.
4271 * config/aarch64/constraints.md (Ui2): New constraint.
4272 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4273 * config/arm/types.md (crypto_sm3): New type attribute.
4274 (crypto_sm4): Ditto.
4275 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4276 (vsm3tt1aq_u32): Ditto.
4277 (vsm3tt1bq_u32): Ditto.
4278 (vsm3tt2aq_u32): Ditto.
4279 (vsm3tt2bq_u32): Ditto.
4280 (vsm3partw1q_u32): Ditto.
4281 (vsm3partw2q_u32): Ditto.
4282 (vsm4eq_u32): Ditto.
4283 (vsm4ekeyq_u32): Ditto.
4284 (doc/invoke.texi): Document new sm4 option.
4286 2018-01-10 Michael Collison <michael.collison@arm.com>
4288 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4289 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4290 (AARCH64_FL_FOR_ARCH8_4): New.
4291 (AARCH64_FL_V8_4): New flag.
4292 (doc/invoke.texi): Document new armv8.4-a option.
4294 2018-01-10 Michael Collison <michael.collison@arm.com>
4296 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4297 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4298 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4299 * config/aarch64/aarch64-option-extension.def: Add
4300 AARCH64_OPT_EXTENSION of 'sha2'.
4301 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4302 (crypto): Disable sha2 and aes if crypto disabled.
4303 (crypto): Enable aes and sha2 if enabled.
4304 (simd): Disable sha2 and aes if simd disabled.
4305 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4307 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4308 (TARGET_SHA2): New feature flag for sha2.
4309 (TARGET_AES): New feature flag for aes.
4310 * config/aarch64/aarch64-simd.md:
4311 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4312 conditional on TARGET_AES.
4313 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4314 (aarch64_crypto_sha1hsi): Make pattern conditional
4316 (aarch64_crypto_sha1hv4si): Ditto.
4317 (aarch64_be_crypto_sha1hv4si): Ditto.
4318 (aarch64_crypto_sha1su1v4si): Ditto.
4319 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4320 (aarch64_crypto_sha1su0v4si): Ditto.
4321 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4322 (aarch64_crypto_sha256su0v4si): Ditto.
4323 (aarch64_crypto_sha256su1v4si): Ditto.
4324 (doc/invoke.texi): Document new aes and sha2 options.
4326 2018-01-10 Martin Sebor <msebor@redhat.com>
4328 PR tree-optimization/83781
4329 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4332 2018-01-11 Martin Sebor <msebor@gmail.com>
4333 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4335 PR tree-optimization/83501
4336 PR tree-optimization/81703
4338 * tree-ssa-strlen.c (get_string_cst): Rename...
4339 (get_string_len): ...to this. Handle global constants.
4340 (handle_char_store): Adjust.
4342 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4343 Jim Wilson <jimw@sifive.com>
4345 * config/riscv/riscv-protos.h (riscv_output_return): New.
4346 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4347 (riscv_attribute_table, riscv_output_return),
4348 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4349 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4350 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4352 (riscv_expand_prologue): Add early return for naked function.
4353 (riscv_expand_epilogue): Likewise.
4354 (riscv_function_ok_for_sibcall): Return false for naked function.
4355 (riscv_set_current_function): New.
4356 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4357 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4358 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4359 * doc/extend.texi (RISC-V Function Attributes): New.
4361 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4363 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4364 check for 128-bit long double before checking TCmode.
4365 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4366 128-bit long doubles before checking TFmode or TCmode.
4367 (FLOAT128_IBM_P): Likewise.
4369 2018-01-10 Martin Sebor <msebor@redhat.com>
4371 PR tree-optimization/83671
4372 * builtins.c (c_strlen): Unconditionally return zero for the empty
4374 Use -Warray-bounds for warnings.
4375 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4376 for non-constant array indices with COMPONENT_REF, arrays of
4377 arrays, and pointers to arrays.
4378 (gimple_fold_builtin_strlen): Determine and set length range for
4379 non-constant character arrays.
4381 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4384 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4387 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4389 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4391 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4394 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4395 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4396 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4397 indexed_or_indirect_operand predicate.
4398 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4399 (*vsx_le_perm_load_v8hi): Likewise.
4400 (*vsx_le_perm_load_v16qi): Likewise.
4401 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4402 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4403 (*vsx_le_perm_store_v8hi): Likewise.
4404 (*vsx_le_perm_store_v16qi): Likewise.
4405 (eight unnamed splitters): Likewise.
4407 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4409 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4410 * config/rs6000/emmintrin.h: Likewise.
4411 * config/rs6000/mmintrin.h: Likewise.
4412 * config/rs6000/xmmintrin.h: Likewise.
4414 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4417 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4419 * tree.c (tree_nop_conversion): Return true for location wrapper
4421 (maybe_wrap_with_location): New function.
4422 (selftest::check_strip_nops): New function.
4423 (selftest::test_location_wrappers): New function.
4424 (selftest::tree_c_tests): Call it.
4425 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4426 (maybe_wrap_with_location): New decl.
4427 (EXPR_LOCATION_WRAPPER_P): New macro.
4428 (location_wrapper_p): New inline function.
4429 (tree_strip_any_location_wrapper): New inline function.
4431 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4434 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4435 stack_realign_offset for the largest alignment of stack slot
4437 (ix86_find_max_used_stack_alignment): New function.
4438 (ix86_finalize_stack_frame_flags): Use it. Set
4439 max_used_stack_alignment if we don't realign stack.
4440 * config/i386/i386.h (machine_function): Add
4441 max_used_stack_alignment.
4443 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4445 * config/arm/arm.opt (-mbranch-cost): New option.
4446 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4449 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4452 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4453 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4455 2018-01-10 Richard Biener <rguenther@suse.de>
4458 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4459 early out so it also covers the case where we have a non-NULL
4462 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4464 PR tree-optimization/83753
4465 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4466 for non-strided grouped accesses if the number of elements is 1.
4468 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4471 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4472 * i386.h (TARGET_USE_GATHER): Define.
4473 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4475 2018-01-10 Martin Liska <mliska@suse.cz>
4478 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4479 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4481 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4482 CLEANUP_NO_PARTITIONING is not set.
4484 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4486 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4487 for vectors, as a partial revert of r254296.
4488 * rtl.h (const_vec_p): Delete.
4489 (const_vec_duplicate_p): Don't test for vector CONSTs.
4490 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4491 * expmed.c (make_tree): Likewise.
4494 * common.md (E, F): Use CONSTANT_P instead of checking for
4496 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4497 checking for CONST_VECTOR.
4499 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4502 * predict.c (force_edge_cold): Handle in more sane way edges
4505 2018-01-09 Carl Love <cel@us.ibm.com>
4507 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4509 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4510 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4511 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4512 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4513 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4514 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4515 * config/rs6000/rs6000-protos.h: Add extern defition for
4516 rs6000_generate_float2_double_code.
4517 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4519 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4520 (float2_v2df): Add define_expand.
4522 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4525 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4526 op_mode in the force_to_mode call.
4528 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4530 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4531 instead of checking each element individually.
4532 (aarch64_evpc_uzp): Likewise.
4533 (aarch64_evpc_zip): Likewise.
4534 (aarch64_evpc_ext): Likewise.
4535 (aarch64_evpc_rev): Likewise.
4536 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4537 instead of checking each element individually. Return true without
4539 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4540 whether all selected elements come from the same input, instead of
4541 checking each element individually. Remove calls to gen_rtx_REG,
4542 start_sequence and end_sequence and instead assert that no rtl is
4545 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4547 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4548 order of HIGH and CONST checks.
4550 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4552 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4553 if the destination isn't an SSA_NAME.
4555 2018-01-09 Richard Biener <rguenther@suse.de>
4557 PR tree-optimization/83668
4558 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4560 (canonicalize_loop_form): ... here, renamed from ...
4561 (canonicalize_loop_closed_ssa_form): ... this and amended to
4562 swap successor edges for loop exit blocks to make us use
4563 the RPO order we need for initial schedule generation.
4565 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4567 PR tree-optimization/64811
4568 * match.pd: When optimizing comparisons with Inf, avoid
4569 introducing or losing exceptions from comparisons with NaN.
4571 2018-01-09 Martin Liska <mliska@suse.cz>
4574 * asan.c (shadow_mem_size): Add gcc_assert.
4576 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4578 Don't save registers in main().
4581 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4582 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4583 * config/avr/avr.c (avr_set_current_function): Don't error if
4584 naked, OS_task or OS_main are specified at the same time.
4585 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4587 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4589 * common/config/avr/avr-common.c (avr_option_optimization_table):
4590 Switch on -mmain-is-OS_task for optimizing compilations.
4592 2018-01-09 Richard Biener <rguenther@suse.de>
4594 PR tree-optimization/83572
4595 * graphite.c: Include cfganal.h.
4596 (graphite_transform_loops): Connect infinite loops to exit
4597 and remove fake edges at the end.
4599 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4601 * ipa-inline.c (edge_badness): Revert accidental checkin.
4603 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4606 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4607 symbols; not inline clones.
4609 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4612 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4613 hard registers. Formatting fixes.
4615 PR preprocessor/83722
4616 * gcc.c (try_generate_repro): Pass
4617 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4618 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4621 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4622 Kito Cheng <kito.cheng@gmail.com>
4624 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4625 (riscv_leaf_function_p): Delete.
4626 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4628 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4630 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4632 (do_ifelse): New function.
4633 (do_isel): New function.
4634 (do_sub3): New function.
4635 (do_add3): New function.
4636 (do_load_mask_compare): New function.
4637 (do_overlap_load_compare): New function.
4638 (expand_compare_loop): New function.
4639 (expand_block_compare): Call expand_compare_loop() when appropriate.
4640 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4642 (-mblock-compare-inline-loop-limit): New option.
4644 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4647 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4648 Reverse order of second and third operands in first alternative.
4649 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4650 of first and second elements in UNSPEC_VPERMR vector.
4651 (altivec_expand_vec_perm_le): Likewise.
4653 2018-01-08 Jeff Law <law@redhat.com>
4655 PR rtl-optimizatin/81308
4656 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4657 (process_switch): If group_case_labels makes a change, then set
4659 (pass_convert_switch::execute): If a switch is converted, then
4660 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4662 PR rtl-optimization/81308
4663 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4666 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4668 PR target/83663 - Revert r255946
4669 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4670 generation for cases where splatting a value is not useful.
4671 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4672 across a vec_duplicate and a paradoxical subreg forming a vector
4673 mode to a vec_concat.
4675 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4677 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4678 -march=armv8.3-a variants.
4679 * config/arm/t-multilib: Likewise.
4680 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4682 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4684 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4686 (cceq_ior_compare_complement): Give it a name so I can use it, and
4687 change boolean_or_operator predicate to boolean_operator so it can
4688 be used to generate a crand.
4689 (eqne): New code iterator.
4690 (bd/bd_neg): New code_attrs.
4691 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4692 a single define_insn.
4693 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4694 decrement (bdnzt/bdnzf/bdzt/bdzf).
4695 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4696 with the new names of the branch decrement patterns, and added the
4697 names of the branch decrement conditional patterns.
4699 2018-01-08 Richard Biener <rguenther@suse.de>
4701 PR tree-optimization/83563
4702 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4705 2018-01-08 Richard Biener <rguenther@suse.de>
4708 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4710 2018-01-08 Richard Biener <rguenther@suse.de>
4712 PR tree-optimization/83685
4713 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4714 references to abnormals.
4716 2018-01-08 Richard Biener <rguenther@suse.de>
4719 * dwarf2out.c (output_indirect_strings): Handle empty
4720 skeleton_debug_str_hash.
4721 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4723 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4725 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4726 (emit_store_direct): Likewise.
4727 (arc_trampoline_adjust_address): Likewise.
4728 (arc_asm_trampoline_template): New function.
4729 (arc_initialize_trampoline): Use asm_trampoline_template.
4730 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4731 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4732 * config/arc/arc.md (flush_icache): Delete pattern.
4734 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4736 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4737 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4740 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4743 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4744 by not USED_FOR_TARGET.
4745 (make_pass_resolve_sw_modes): Likewise.
4747 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4749 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4752 2018-01-08 Richard Biener <rguenther@suse.de>
4755 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4757 2018-01-08 Richard Biener <rguenther@suse.de>
4760 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4762 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4765 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4766 basic blocks with a small number of successors.
4767 (convert_control_dep_chain_into_preds): Improve handling of
4769 (dump_predicates): Split apart into...
4770 (dump_pred_chain): ...here...
4771 (dump_pred_info): ...and here.
4772 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4773 (can_chain_union_be_invalidated_p): Improve check for invalidation
4775 (uninit_uses_cannot_happen): Avoid unnecessary if
4776 convert_control_dep_chain_into_preds yielded nothing.
4778 2018-01-06 Martin Sebor <msebor@redhat.com>
4780 PR tree-optimization/83640
4781 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4782 subtracting negative offset from size.
4783 (builtin_access::overlap): Adjust offset bounds of the access to fall
4784 within the size of the object if possible.
4786 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4788 PR rtl-optimization/83699
4789 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4790 extract_bit_field_as_subreg to cases in which the extracted
4791 value is also a vector.
4793 * lra-constraints.c (process_alt_operands): Test for the equivalence
4794 substitutions when detecting a possible reload cycle.
4796 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4799 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4800 by default if flag_selective_schedling{,2}. Formatting fixes.
4802 PR rtl-optimization/83682
4803 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4804 if it has non-VECTOR_MODE element mode.
4805 (vec_duplicate_p): Likewise.
4808 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4809 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4811 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4814 * config/i386/i386-builtin.def
4815 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4816 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4817 Require also OPTION_MASK_ISA_AVX512F in addition to
4818 OPTION_MASK_ISA_GFNI.
4819 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4820 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4821 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4822 to OPTION_MASK_ISA_GFNI.
4823 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4824 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4825 OPTION_MASK_ISA_AVX512BW.
4826 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4827 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4828 addition to OPTION_MASK_ISA_GFNI.
4829 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4830 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4831 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4832 to OPTION_MASK_ISA_GFNI.
4833 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4834 a requirement for all ISAs rather than any of them with a few
4836 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4838 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4839 bitmasks to be enabled with 3 exceptions, instead of requiring any
4840 enabled ISA with lots of exceptions.
4841 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4842 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4843 Change avx512bw in isa attribute to avx512f.
4844 * config/i386/sgxintrin.h: Add license boilerplate.
4845 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4846 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4847 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4848 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4850 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4851 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4852 temporarily sse2 rather than sse if not enabled already.
4855 * config/i386/sse.md (VI248_VLBW): Rename to ...
4856 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4857 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4858 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4859 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4860 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4861 mode iterator instead of VI248_VLBW.
4863 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4865 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4866 (record_modified): Skip clobbers; add debug output.
4867 (param_change_prob): Use sreal frequencies.
4869 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4871 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4872 punt for user-aligned variables.
4874 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4876 * tree-chrec.c (chrec_contains_symbols): Return true for
4879 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4882 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4883 of (x|y) == x for BICS pattern.
4885 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4887 PR tree-optimization/83605
4888 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4889 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4892 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4894 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4895 * config/epiphany/rtems.h: New file.
4897 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4898 Uros Bizjak <ubizjak@gmail.com>
4901 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4902 QIreg_operand instead of register_operand predicate.
4903 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4904 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4905 comments instead of -fmitigate[-_]rop.
4907 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4910 * cgraphunit.c (symbol_table::compile): Switch to text_section
4911 before calling assembly_start debug hook.
4912 * run-rtl-passes.c (run_rtl_passes): Likewise.
4915 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4917 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4918 range_int_cst_p rather than !symbolic_range_p before calling
4919 extract_range_from_multiplicative_op_1.
4921 2018-01-04 Jeff Law <law@redhat.com>
4923 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4924 redundant test in assertion.
4926 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4928 * doc/rtl.texi: Document machine_mode wrapper classes.
4930 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4932 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4935 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4937 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4938 the VEC_PERM_EXPR fold to fail.
4940 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4943 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4944 to switched_sections.
4946 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4949 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4952 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4955 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4956 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4958 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4961 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4962 is BLKmode and bitpos not zero or mode change is needed.
4964 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4967 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4970 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4973 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4974 instead of MULT rtx. Update all corresponding splitters.
4976 (*ssub<modesuffix>): Ditto.
4978 (*cmp_sadd_di): Update split patterns.
4979 (*cmp_sadd_si): Ditto.
4980 (*cmp_sadd_sidi): Ditto.
4981 (*cmp_ssub_di): Ditto.
4982 (*cmp_ssub_si): Ditto.
4983 (*cmp_ssub_sidi): Ditto.
4984 * config/alpha/predicates.md (const23_operand): New predicate.
4985 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4986 Look for ASHIFT, not MULT inner operand.
4987 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4989 2018-01-04 Martin Liska <mliska@suse.cz>
4991 PR gcov-profile/83669
4992 * gcov.c (output_intermediate_file): Add version to intermediate
4994 * doc/gcov.texi: Document new field 'version' in intermediate
4995 file format. Fix location of '-k' option of gcov command.
4997 2018-01-04 Martin Liska <mliska@suse.cz>
5000 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5002 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5004 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5006 2018-01-03 Martin Sebor <msebor@redhat.com>
5008 PR tree-optimization/83655
5009 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5010 checking calls with invalid arguments.
5012 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5014 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5015 (vectorizable_mask_load_store): Delete.
5016 (vectorizable_call): Return false for masked loads and stores.
5017 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5018 instead of gimple_assign_rhs1.
5019 (vectorizable_load): Handle IFN_MASK_LOAD.
5020 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5024 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5026 (vectorizable_mask_load_store): ...here.
5027 (vectorizable_load): ...and here.
5029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5031 * tree-vect-stmts.c (vect_build_all_ones_mask)
5032 (vect_build_zero_merge_argument): New functions, split out from...
5033 (vectorizable_load): ...here.
5035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5037 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5039 (vectorizable_mask_load_store): ...here.
5040 (vectorizable_store): ...and here.
5042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5044 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5046 (vectorizable_mask_load_store): ...here.
5048 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5050 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5051 (vect_model_store_cost): Take a vec_load_store_type instead of a
5053 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5054 (vect_model_store_cost): Take a vec_load_store_type instead of a
5056 (vectorizable_mask_load_store): Update accordingly.
5057 (vectorizable_store): Likewise.
5058 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5062 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5063 IFN_MASK_LOAD calls here rather than...
5064 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5067 Alan Hayward <alan.hayward@arm.com>
5068 David Sherwood <david.sherwood@arm.com>
5070 * expmed.c (extract_bit_field_1): For vector extracts,
5071 fall back to extract_bit_field_as_subreg if vec_extract
5074 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5075 Alan Hayward <alan.hayward@arm.com>
5076 David Sherwood <david.sherwood@arm.com>
5078 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5079 they are variable or constant sized.
5080 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5081 slots for constant-sized data.
5083 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5084 Alan Hayward <alan.hayward@arm.com>
5085 David Sherwood <david.sherwood@arm.com>
5087 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5088 handling COND_EXPRs with boolean comparisons, try to find a better
5089 basis for the mask type than the boolean itself.
5091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5093 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5094 is calculated and how it can be overridden.
5095 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5096 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5098 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5101 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5102 Alan Hayward <alan.hayward@arm.com>
5103 David Sherwood <david.sherwood@arm.com>
5105 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5106 Remove the mode argument.
5107 (aarch64_simd_valid_immediate): Remove the mode and inverse
5109 * config/aarch64/iterators.md (bitsize): New iterator.
5110 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5111 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5112 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5113 aarch64_simd_valid_immediate.
5114 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5115 (aarch64_reg_or_bic_imm): Likewise.
5116 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5117 with an insn_type enum and msl with a modifier_type enum.
5118 Replace element_width with a scalar_mode. Change the shift
5119 to unsigned int. Add constructors for scalar_float_mode and
5120 scalar_int_mode elements.
5121 (aarch64_vect_float_const_representable_p): Delete.
5122 (aarch64_can_const_movi_rtx_p)
5123 (aarch64_simd_scalar_immediate_valid_for_move)
5124 (aarch64_simd_make_constant): Update call to
5125 aarch64_simd_valid_immediate.
5126 (aarch64_advsimd_valid_immediate_hs): New function.
5127 (aarch64_advsimd_valid_immediate): Likewise.
5128 (aarch64_simd_valid_immediate): Remove mode and inverse
5129 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5130 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5131 and aarch64_float_const_representable_p on the result.
5132 (aarch64_output_simd_mov_immediate): Remove mode argument.
5133 Update call to aarch64_simd_valid_immediate and use of
5134 simd_immediate_info.
5135 (aarch64_output_scalar_simd_mov_immediate): Update call
5138 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5139 Alan Hayward <alan.hayward@arm.com>
5140 David Sherwood <david.sherwood@arm.com>
5142 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5143 (mode_nunits): Likewise CONST_MODE_NUNITS.
5144 * machmode.def (ADJUST_NUNITS): Document.
5145 * genmodes.c (mode_data::need_nunits_adj): New field.
5146 (blank_mode): Update accordingly.
5147 (adj_nunits): New variable.
5148 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5150 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5151 listed in adj_nunits.
5152 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5153 listed in adj_nunits. Don't emit case statements for such modes.
5154 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5155 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5156 nothing if adj_nunits is nonnull.
5157 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5158 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5159 (emit_mode_fbit): Update use of print_maybe_const_decl.
5160 (emit_move_size): Likewise. Treat the array as non-const
5162 (emit_mode_adjustments): Handle adj_nunits.
5164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5166 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5167 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5168 (VECTOR_MODES): Use it.
5169 (make_vector_modes): Take the prefix as an argument.
5171 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5172 Alan Hayward <alan.hayward@arm.com>
5173 David Sherwood <david.sherwood@arm.com>
5175 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5176 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5177 for MODE_VECTOR_BOOL.
5178 * machmode.def (VECTOR_BOOL_MODE): Document.
5179 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5180 (make_vector_bool_mode): New function.
5181 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5183 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5184 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5186 * stor-layout.c (int_mode_for_mode): Likewise.
5187 * tree.c (build_vector_type_for_mode): Likewise.
5188 * varasm.c (output_constant_pool_2): Likewise.
5189 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5190 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5191 for MODE_VECTOR_BOOL.
5192 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5193 of mode class checks.
5194 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5195 instead of a list of mode class checks.
5196 (expand_vector_scalar_condition): Likewise.
5197 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5199 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5200 Alan Hayward <alan.hayward@arm.com>
5201 David Sherwood <david.sherwood@arm.com>
5203 * machmode.h (mode_size): Change from unsigned short to
5205 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5206 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5207 or if measurement_type is not polynomial.
5208 (fixed_size_mode::includes_p): Check for constant-sized modes.
5209 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5210 return a poly_uint16 rather than an unsigned short.
5211 (emit_mode_size): Change the type of mode_size from unsigned short
5212 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5213 (emit_mode_adjustments): Cope with polynomial vector sizes.
5214 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5216 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5218 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5219 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5220 * caller-save.c (setup_save_areas): Likewise.
5221 (replace_reg_with_saved_mem): Likewise.
5222 * calls.c (emit_library_call_value_1): Likewise.
5223 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5224 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5225 (gen_lowpart_for_combine): Likewise.
5226 * convert.c (convert_to_integer_1): Likewise.
5227 * cse.c (equiv_constant, cse_insn): Likewise.
5228 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5229 (cselib_subst_to_values): Likewise.
5230 * dce.c (word_dce_process_block): Likewise.
5231 * df-problems.c (df_word_lr_mark_ref): Likewise.
5232 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5233 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5234 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5235 (rtl_for_decl_location): Likewise.
5236 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5237 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5238 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5239 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5240 (expand_expr_real_1): Likewise.
5241 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5242 (pad_below): Likewise.
5243 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5244 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5245 * ira.c (get_subreg_tracking_sizes): Likewise.
5246 * ira-build.c (ira_create_allocno_objects): Likewise.
5247 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5248 (ira_sort_regnos_for_alter_reg): Likewise.
5249 * ira-costs.c (record_operand_costs): Likewise.
5250 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5251 (resolve_simple_move): Likewise.
5252 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5253 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5254 (lra_constraints): Likewise.
5255 (CONST_POOL_OK_P): Reject variable-sized modes.
5256 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5257 (add_pseudo_to_slot, lra_spill): Likewise.
5258 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5259 * optabs-query.c (get_best_extraction_insn): Likewise.
5260 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5261 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5262 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5263 * recog.c (offsettable_address_addr_space_p): Likewise.
5264 * regcprop.c (maybe_mode_change): Likewise.
5265 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5266 * regrename.c (build_def_use): Likewise.
5267 * regstat.c (dump_reg_info): Likewise.
5268 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5269 (find_reloads, find_reloads_subreg_address): Likewise.
5270 * reload1.c (eliminate_regs_1): Likewise.
5271 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5272 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5273 (simplify_binary_operation_1, simplify_subreg): Likewise.
5274 * targhooks.c (default_function_arg_padding): Likewise.
5275 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5276 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5277 (verify_gimple_assign_ternary): Likewise.
5278 * tree-inline.c (estimate_move_cost): Likewise.
5279 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5280 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5281 (get_address_cost_ainc): Likewise.
5282 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5283 (vect_supportable_dr_alignment): Likewise.
5284 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5285 (vectorizable_reduction): Likewise.
5286 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5287 (vectorizable_operation, vectorizable_load): Likewise.
5288 * tree.c (build_same_sized_truth_vector_type): Likewise.
5289 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5290 * var-tracking.c (emit_note_insn_var_location): Likewise.
5291 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5292 (ADDR_VEC_ALIGN): Likewise.
5294 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5295 Alan Hayward <alan.hayward@arm.com>
5296 David Sherwood <david.sherwood@arm.com>
5298 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5300 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5301 or if measurement_type is polynomial.
5302 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5303 * combine.c (make_extraction): Likewise.
5304 * dse.c (find_shift_sequence): Likewise.
5305 * dwarf2out.c (mem_loc_descriptor): Likewise.
5306 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5307 (extract_bit_field, extract_low_bits): Likewise.
5308 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5309 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5310 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5311 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5312 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5313 * reload.c (find_reloads): Likewise.
5314 * reload1.c (alter_reg): Likewise.
5315 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5316 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5317 * tree-if-conv.c (predicate_mem_writes): Likewise.
5318 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5319 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5320 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5321 * valtrack.c (dead_debug_insert_temp): Likewise.
5322 * varasm.c (mergeable_constant_section): Likewise.
5323 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5325 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5326 Alan Hayward <alan.hayward@arm.com>
5327 David Sherwood <david.sherwood@arm.com>
5329 * expr.c (expand_assignment): Cope with polynomial mode sizes
5330 when assigning to a CONCAT.
5332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5333 Alan Hayward <alan.hayward@arm.com>
5334 David Sherwood <david.sherwood@arm.com>
5336 * machmode.h (mode_precision): Change from unsigned short to
5338 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5340 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5341 or if measurement_type is not polynomial.
5342 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5343 in which the mode is already known to be a scalar_int_mode.
5344 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5345 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5347 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5348 for GET_MODE_PRECISION.
5349 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5350 for GET_MODE_PRECISION.
5351 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5353 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5354 (expand_field_assignment, make_extraction): Likewise.
5355 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5356 (get_last_value): Likewise.
5357 * convert.c (convert_to_integer_1): Likewise.
5358 * cse.c (cse_insn): Likewise.
5359 * expr.c (expand_expr_real_1): Likewise.
5360 * lra-constraints.c (simplify_operand_subreg): Likewise.
5361 * optabs-query.c (can_atomic_load_p): Likewise.
5362 * optabs.c (expand_atomic_load): Likewise.
5363 (expand_atomic_store): Likewise.
5364 * ree.c (combine_reaching_defs): Likewise.
5365 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5366 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5367 * tree.h (type_has_mode_precision_p): Likewise.
5368 * ubsan.c (instrument_si_overflow): Likewise.
5370 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5371 Alan Hayward <alan.hayward@arm.com>
5372 David Sherwood <david.sherwood@arm.com>
5374 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5375 polynomial numbers of units.
5376 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5377 (valid_vector_subparts_p): New function.
5378 (build_vector_type): Remove temporary shim and take the number
5379 of units as a poly_uint64 rather than an int.
5380 (build_opaque_vector_type): Take the number of units as a
5381 poly_uint64 rather than an int.
5382 * tree.c (build_vector_from_ctor): Handle polynomial
5383 TYPE_VECTOR_SUBPARTS.
5384 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5385 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5386 (build_vector_from_val): If the number of units is variable,
5387 use build_vec_duplicate_cst for constant operands and
5388 VEC_DUPLICATE_EXPR otherwise.
5389 (make_vector_type): Remove temporary is_constant ().
5390 (build_vector_type, build_opaque_vector_type): Take the number of
5391 units as a poly_uint64 rather than an int.
5392 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5394 * cfgexpand.c (expand_debug_expr): Likewise.
5395 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5396 (store_constructor, expand_expr_real_1): Likewise.
5397 (const_scalar_mask_from_tree): Likewise.
5398 * fold-const-call.c (fold_const_reduction): Likewise.
5399 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5400 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5401 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5402 (fold_relational_const): Likewise.
5403 (native_interpret_vector): Likewise. Change the size from an
5404 int to an unsigned int.
5405 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5406 TYPE_VECTOR_SUBPARTS.
5407 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5408 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5409 duplicating a non-constant operand into a variable-length vector.
5410 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5411 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5412 * ipa-icf.c (sem_variable::equals): Likewise.
5413 * match.pd: Likewise.
5414 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5415 * print-tree.c (print_node): Likewise.
5416 * stor-layout.c (layout_type): Likewise.
5417 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5418 * tree-cfg.c (verify_gimple_comparison): Likewise.
5419 (verify_gimple_assign_binary): Likewise.
5420 (verify_gimple_assign_ternary): Likewise.
5421 (verify_gimple_assign_single): Likewise.
5422 * tree-pretty-print.c (dump_generic_node): Likewise.
5423 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5424 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5425 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5426 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5427 (vect_shift_permute_load_chain): Likewise.
5428 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5429 (expand_vector_condition, optimize_vector_constructor): Likewise.
5430 (lower_vec_perm, get_compute_type): Likewise.
5431 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5432 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5433 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5434 (vect_recog_mask_conversion_pattern): Likewise.
5435 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5436 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5437 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5438 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5439 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5440 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5441 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5442 (supportable_widening_operation): Likewise.
5443 (supportable_narrowing_operation): Likewise.
5444 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5446 * varasm.c (output_constant): Likewise.
5448 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5449 Alan Hayward <alan.hayward@arm.com>
5450 David Sherwood <david.sherwood@arm.com>
5452 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5453 so that both the length == 3 and length != 3 cases set up their
5454 own permute vectors. Add comments explaining why we know the
5455 number of elements is constant.
5456 (vect_permute_load_chain): Likewise.
5458 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5459 Alan Hayward <alan.hayward@arm.com>
5460 David Sherwood <david.sherwood@arm.com>
5462 * machmode.h (mode_nunits): Change from unsigned char to
5464 (ONLY_FIXED_SIZE_MODES): New macro.
5465 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5466 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5467 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5469 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5470 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5471 or if measurement_type is not polynomial.
5472 * genmodes.c (ZERO_COEFFS): New macro.
5473 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5475 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5476 Use ZERO_COEFFS when emitting initializers.
5477 * data-streamer.h (bp_pack_poly_value): New function.
5478 (bp_unpack_poly_value): Likewise.
5479 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5480 for GET_MODE_NUNITS.
5481 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5482 for GET_MODE_NUNITS.
5483 * tree.c (make_vector_type): Remove temporary shim and make
5484 the real function take the number of units as a poly_uint64
5486 (build_vector_type_for_mode): Handle polynomial nunits.
5487 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5488 * emit-rtl.c (const_vec_series_p_1): Likewise.
5489 (gen_rtx_CONST_VECTOR): Likewise.
5490 * fold-const.c (test_vec_duplicate_folding): Likewise.
5491 * genrecog.c (validate_pattern): Likewise.
5492 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5493 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5494 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5495 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5496 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5497 * rtlanal.c (subreg_get_info): Likewise.
5498 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5499 (vect_grouped_load_supported): Likewise.
5500 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5501 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5502 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5503 (simplify_const_unary_operation, simplify_binary_operation_1)
5504 (simplify_const_binary_operation, simplify_ternary_operation)
5505 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5506 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5507 instead of CONST_VECTOR_NUNITS.
5508 * varasm.c (output_constant_pool_2): Likewise.
5509 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5510 explicit-encoded elements in the XVEC for variable-length vectors.
5512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5514 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5517 Alan Hayward <alan.hayward@arm.com>
5518 David Sherwood <david.sherwood@arm.com>
5520 * coretypes.h (fixed_size_mode): Declare.
5521 (fixed_size_mode_pod): New typedef.
5522 * builtins.h (target_builtins::x_apply_args_mode)
5523 (target_builtins::x_apply_result_mode): Change type to
5524 fixed_size_mode_pod.
5525 * builtins.c (apply_args_size, apply_result_size, result_vector)
5526 (expand_builtin_apply_args_1, expand_builtin_apply)
5527 (expand_builtin_return): Update accordingly.
5529 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5531 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5532 * cselib.c (cselib_hash_rtx): Likewise.
5533 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5534 CONST_VECTOR encoding.
5536 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5537 Jeff Law <law@redhat.com>
5540 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5541 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5542 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5543 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5546 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5547 explicitly probe *sp in a noreturn function if there were any callee
5548 register saves or frame pointer is needed.
5550 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5553 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5554 BLKmode for ternary, binary or unary expressions.
5557 * var-tracking.c (delete_vta_debug_insn): New inline function.
5558 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5559 insns from get_insns () to NULL instead of each bb separately.
5560 Use delete_vta_debug_insn. No longer static.
5561 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5562 delete_vta_debug_insns callers.
5563 * rtl.h (delete_vta_debug_insns): Declare.
5564 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5565 instead of variable_tracking_main.
5567 2018-01-03 Martin Sebor <msebor@redhat.com>
5569 PR tree-optimization/83603
5570 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5571 arguments past the endof the argument list in functions declared
5572 without a prototype.
5573 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5574 Avoid checking when arguments are null.
5576 2018-01-03 Martin Sebor <msebor@redhat.com>
5579 * doc/extend.texi (attribute const): Fix a typo.
5580 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5581 issuing -Wsuggest-attribute for void functions.
5583 2018-01-03 Martin Sebor <msebor@redhat.com>
5585 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5586 offset_int::from instead of wide_int::to_shwi.
5587 (maybe_diag_overlap): Remove assertion.
5588 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5589 * gimple-ssa-sprintf.c (format_directive): Same.
5590 (parse_directive): Same.
5591 (sprintf_dom_walker::compute_format_length): Same.
5592 (try_substitute_return_value): Same.
5594 2018-01-03 Jeff Law <law@redhat.com>
5597 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5598 non-constant residual for zero at runtime and avoid probing in
5599 that case. Reorganize code for trailing problem to mirror handling
5602 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5604 PR tree-optimization/83501
5605 * tree-ssa-strlen.c (get_string_cst): New.
5606 (handle_char_store): Call get_string_cst.
5608 2018-01-03 Martin Liska <mliska@suse.cz>
5610 PR tree-optimization/83593
5611 * tree-ssa-strlen.c: Include tree-cfg.h.
5612 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5613 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5614 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5616 (strlen_dom_walker::before_dom_children): Call
5617 gimple_purge_dead_eh_edges. Dump tranformation with details
5619 (strlen_dom_walker::before_dom_children): Update call by adding
5620 new argument cleanup_eh.
5621 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5623 2018-01-03 Martin Liska <mliska@suse.cz>
5626 * cif-code.def (VARIADIC_THUNK): New enum value.
5627 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5630 2018-01-03 Jan Beulich <jbeulich@suse.com>
5632 * sse.md (mov<mode>_internal): Tighten condition for when to use
5633 vmovdqu<ssescalarsize> for TI and OI modes.
5635 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5637 Update copyright years.
5639 2018-01-03 Martin Liska <mliska@suse.cz>
5642 * ipa-visibility.c (function_and_variable_visibility): Skip
5643 functions with noipa attribure.
5645 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5647 * gcc.c (process_command): Update copyright notice dates.
5648 * gcov-dump.c (print_version): Ditto.
5649 * gcov.c (print_version): Ditto.
5650 * gcov-tool.c (print_version): Ditto.
5651 * gengtype.c (create_file): Ditto.
5652 * doc/cpp.texi: Bump @copying's copyright year.
5653 * doc/cppinternals.texi: Ditto.
5654 * doc/gcc.texi: Ditto.
5655 * doc/gccint.texi: Ditto.
5656 * doc/gcov.texi: Ditto.
5657 * doc/install.texi: Ditto.
5658 * doc/invoke.texi: Ditto.
5660 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5662 * vector-builder.h (vector_builder::m_full_nelts): Change from
5663 unsigned int to poly_uint64.
5664 (vector_builder::full_nelts): Update prototype accordingly.
5665 (vector_builder::new_vector): Likewise.
5666 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5667 (vector_builder::operator ==): Likewise.
5668 (vector_builder::finalize): Likewise.
5669 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5670 Take the number of elements as a poly_uint64 rather than an
5672 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5673 from unsigned int to poly_uint64.
5674 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5675 (vec_perm_indices::new_vector): Likewise.
5676 (vec_perm_indices::length): Likewise.
5677 (vec_perm_indices::nelts_per_input): Likewise.
5678 (vec_perm_indices::input_nelts): Likewise.
5679 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5680 number of elements per input as a poly_uint64 rather than an
5681 unsigned int. Use the original encoding for variable-length
5682 vectors, rather than clamping each individual element.
5683 For the second and subsequent elements in each pattern,
5684 clamp the step and base before clamping their sum.
5685 (vec_perm_indices::series_p): Handle polynomial element counts.
5686 (vec_perm_indices::all_in_range_p): Likewise.
5687 (vec_perm_indices_to_tree): Likewise.
5688 (vec_perm_indices_to_rtx): Likewise.
5689 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5690 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5691 (tree_vector_builder::new_binary_operation): Handle polynomial
5692 element counts. Return false if we need to know the number
5693 of elements at compile time.
5694 * fold-const.c (fold_vec_perm): Punt if the number of elements
5695 isn't known at compile time.
5697 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5699 * vec-perm-indices.h (vec_perm_builder): Change element type
5700 from HOST_WIDE_INT to poly_int64.
5701 (vec_perm_indices::element_type): Update accordingly.
5702 (vec_perm_indices::clamp): Handle polynomial element_types.
5703 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5704 (vec_perm_indices::all_in_range_p): Likewise.
5705 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5707 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5708 polynomial vec_perm_indices element types.
5709 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5710 * fold-const.c (fold_vec_perm): Likewise.
5711 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5712 * tree-vect-generic.c (lower_vec_perm): Likewise.
5713 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5714 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5715 element type to HOST_WIDE_INT.
5717 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5718 Alan Hayward <alan.hayward@arm.com>
5719 David Sherwood <david.sherwood@arm.com>
5721 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5722 rather than an int. Use plus_constant.
5723 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5724 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5726 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5727 Alan Hayward <alan.hayward@arm.com>
5728 David Sherwood <david.sherwood@arm.com>
5730 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5731 a HOST_WIDE_INT to a poly_int64.
5733 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5734 Alan Hayward <alan.hayward@arm.com>
5735 David Sherwood <david.sherwood@arm.com>
5737 * calls.c (load_register_parameters): Cope with polynomial
5738 mode sizes. Require a constant size for BLKmode parameters
5739 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5740 forces a parameter to be padded at the lsb end in order to
5741 fill a complete number of words, require the parameter size
5742 to be ordered wrt UNITS_PER_WORD.
5744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5745 Alan Hayward <alan.hayward@arm.com>
5746 David Sherwood <david.sherwood@arm.com>
5748 * reload1.c (spill_stack_slot_width): Change element type
5749 from unsigned int to poly_uint64_pod.
5750 (alter_reg): Treat mode sizes as polynomial.
5752 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5753 Alan Hayward <alan.hayward@arm.com>
5754 David Sherwood <david.sherwood@arm.com>
5756 * reload.c (complex_word_subreg_p): New function.
5757 (reload_inner_reg_of_subreg, push_reload): Use it.
5759 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5760 Alan Hayward <alan.hayward@arm.com>
5761 David Sherwood <david.sherwood@arm.com>
5763 * lra-constraints.c (process_alt_operands): Reject matched
5764 operands whose sizes aren't ordered.
5765 (match_reload): Refer to this check here.
5767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5768 Alan Hayward <alan.hayward@arm.com>
5769 David Sherwood <david.sherwood@arm.com>
5771 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5772 that the mode size is in the set {1, 2, 4, 8, 16}.
5774 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5775 Alan Hayward <alan.hayward@arm.com>
5776 David Sherwood <david.sherwood@arm.com>
5778 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5779 Use plus_constant instead of gen_rtx_PLUS.
5781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5782 Alan Hayward <alan.hayward@arm.com>
5783 David Sherwood <david.sherwood@arm.com>
5785 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5786 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5787 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5788 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5789 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5790 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5791 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5792 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5793 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5794 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5796 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5797 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5798 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5799 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5800 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5801 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5802 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5803 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5804 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5805 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5807 * expr.c (emit_move_resolve_push): Treat the input and result
5808 of PUSH_ROUNDING as a poly_int64.
5809 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5810 (emit_push_insn): Likewise.
5811 * lra-eliminations.c (mark_not_eliminable): Likewise.
5812 * recog.c (push_operand): Likewise.
5813 * reload1.c (elimination_effects): Likewise.
5814 * rtlanal.c (nonzero_bits1): Likewise.
5815 * calls.c (store_one_arg): Likewise. Require the padding to be
5816 known at compile time.
5818 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5819 Alan Hayward <alan.hayward@arm.com>
5820 David Sherwood <david.sherwood@arm.com>
5822 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5823 Use plus_constant instead of gen_rtx_PLUS.
5825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5826 Alan Hayward <alan.hayward@arm.com>
5827 David Sherwood <david.sherwood@arm.com>
5829 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5833 Alan Hayward <alan.hayward@arm.com>
5834 David Sherwood <david.sherwood@arm.com>
5836 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5837 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5838 via stack temporaries. Treat the mode size as polynomial too.
5840 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5841 Alan Hayward <alan.hayward@arm.com>
5842 David Sherwood <david.sherwood@arm.com>
5844 * expr.c (expand_expr_real_2): When handling conversions involving
5845 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5846 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5847 as a poly_uint64 too.
5849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5850 Alan Hayward <alan.hayward@arm.com>
5851 David Sherwood <david.sherwood@arm.com>
5853 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5856 Alan Hayward <alan.hayward@arm.com>
5857 David Sherwood <david.sherwood@arm.com>
5859 * combine.c (can_change_dest_mode): Handle polynomial
5860 REGMODE_NATURAL_SIZE.
5861 * expmed.c (store_bit_field_1): Likewise.
5862 * expr.c (store_constructor): Likewise.
5863 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5864 and polynomial REGMODE_NATURAL_SIZE.
5865 (gen_lowpart_common): Likewise.
5866 * reginfo.c (record_subregs_of_mode): Likewise.
5867 * rtlanal.c (read_modify_subreg_p): Likewise.
5869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5870 Alan Hayward <alan.hayward@arm.com>
5871 David Sherwood <david.sherwood@arm.com>
5873 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5874 numbers of elements.
5876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5877 Alan Hayward <alan.hayward@arm.com>
5878 David Sherwood <david.sherwood@arm.com>
5880 * match.pd: Cope with polynomial numbers of vector elements.
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 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5887 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
5894 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5895 (ipa_simd_modify_function_body): Likewise.
5897 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5898 Alan Hayward <alan.hayward@arm.com>
5899 David Sherwood <david.sherwood@arm.com>
5901 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5902 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5903 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5904 (expand_vector_condition, vector_element): Likewise.
5905 (subparts_gt): New function.
5906 (get_compute_type): Use subparts_gt.
5907 (count_type_subparts): Delete.
5908 (expand_vector_operations_1): Use subparts_gt instead of
5909 count_type_subparts.
5911 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5912 Alan Hayward <alan.hayward@arm.com>
5913 David Sherwood <david.sherwood@arm.com>
5915 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5916 (vect_compile_time_alias): ...this new function. Do the calculation
5917 on poly_ints rather than trees.
5918 (vect_prune_runtime_alias_test_list): Update call accordingly.
5920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5921 Alan Hayward <alan.hayward@arm.com>
5922 David Sherwood <david.sherwood@arm.com>
5924 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5926 (vect_schedule_slp_instance): Likewise.
5928 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5929 Alan Hayward <alan.hayward@arm.com>
5930 David Sherwood <david.sherwood@arm.com>
5932 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5933 constant and extern definitions for variable-length vectors.
5934 (vect_get_constant_vectors): Note that the number of units
5935 is known to be constant.
5937 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5938 Alan Hayward <alan.hayward@arm.com>
5939 David Sherwood <david.sherwood@arm.com>
5941 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5942 of units as polynomial. Choose between WIDE and NARROW based
5945 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5946 Alan Hayward <alan.hayward@arm.com>
5947 David Sherwood <david.sherwood@arm.com>
5949 * tree-vect-stmts.c (simd_clone_subparts): New function.
5950 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5952 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5953 Alan Hayward <alan.hayward@arm.com>
5954 David Sherwood <david.sherwood@arm.com>
5956 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5957 vectors as polynomial. Use build_index_vector for
5960 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5961 Alan Hayward <alan.hayward@arm.com>
5962 David Sherwood <david.sherwood@arm.com>
5964 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5965 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5966 for variable-length vectors.
5967 (vectorizable_mask_load_store): Treat the number of units as
5968 polynomial, asserting that it is constant if the condition has
5969 already been enforced.
5970 (vectorizable_store, vectorizable_load): Likewise.
5972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5973 Alan Hayward <alan.hayward@arm.com>
5974 David Sherwood <david.sherwood@arm.com>
5976 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5977 of units as polynomial. Punt if we can't tell at compile time
5978 which vector contains the final result.
5980 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5981 Alan Hayward <alan.hayward@arm.com>
5982 David Sherwood <david.sherwood@arm.com>
5984 * tree-vect-loop.c (vectorizable_induction): Treat the number
5985 of units as polynomial. Punt on SLP inductions. Use an integer
5986 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5987 cast of such a series for variable-length floating-point
5990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5991 Alan Hayward <alan.hayward@arm.com>
5992 David Sherwood <david.sherwood@arm.com>
5994 * tree.h (build_index_vector): Declare.
5995 * tree.c (build_index_vector): New function.
5996 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5997 of units as polynomial, forcibly converting it to a constant if
5998 vectorizable_reduction has already enforced the condition.
5999 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6000 to create a {1,2,3,...} vector.
6001 (vectorizable_reduction): Treat the number of units as polynomial.
6002 Choose vectype_in based on the largest scalar element size rather
6003 than the smallest number of units. Enforce the restrictions
6006 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6007 Alan Hayward <alan.hayward@arm.com>
6008 David Sherwood <david.sherwood@arm.com>
6010 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6011 number of units as polynomial.
6013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6014 Alan Hayward <alan.hayward@arm.com>
6015 David Sherwood <david.sherwood@arm.com>
6017 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6018 * target.def (autovectorize_vector_sizes): Return the vector sizes
6019 by pointer, using vector_sizes rather than a bitmask.
6020 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6021 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6022 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6024 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6025 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6026 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6027 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6028 * omp-general.c (omp_max_vf): Likewise.
6029 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6030 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6031 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6032 * tree-vect-slp.c (vect_slp_bb): Likewise.
6033 * doc/tm.texi: Regenerate.
6034 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6036 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6037 the vector size as a poly_uint64 rather than an unsigned int.
6038 (current_vector_size): Change from an unsigned int to a poly_uint64.
6039 (get_vectype_for_scalar_type): Update accordingly.
6040 * tree.h (build_truth_vector_type): Take the size and number of
6041 units as a poly_uint64 rather than an unsigned int.
6042 (build_vector_type): Add a temporary overload that takes
6043 the number of units as a poly_uint64 rather than an unsigned int.
6044 * tree.c (make_vector_type): Likewise.
6045 (build_truth_vector_type): Take the number of units as a poly_uint64
6046 rather than an unsigned int.
6048 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6049 Alan Hayward <alan.hayward@arm.com>
6050 David Sherwood <david.sherwood@arm.com>
6052 * target.def (get_mask_mode): Take the number of units and length
6053 as poly_uint64s rather than unsigned ints.
6054 * targhooks.h (default_get_mask_mode): Update accordingly.
6055 * targhooks.c (default_get_mask_mode): Likewise.
6056 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6057 * doc/tm.texi: Regenerate.
6059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6060 Alan Hayward <alan.hayward@arm.com>
6061 David Sherwood <david.sherwood@arm.com>
6063 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6064 * omp-general.c (omp_max_vf): Likewise.
6065 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6066 (expand_omp_simd): Handle polynomial safelen.
6067 * omp-low.c (omplow_simd_context): Add a default constructor.
6068 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6069 (lower_rec_simd_input_clauses): Update accordingly.
6070 (lower_rec_input_clauses): Likewise.
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 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6077 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6078 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6079 (vect_analyze_slp_cost): Likewise.
6080 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6081 (vect_model_load_cost): Likewise.
6083 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6084 Alan Hayward <alan.hayward@arm.com>
6085 David Sherwood <david.sherwood@arm.com>
6087 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6088 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6089 from an unsigned int * to a poly_uint64_pod *.
6090 (calculate_unrolling_factor): New function.
6091 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6094 Alan Hayward <alan.hayward@arm.com>
6095 David Sherwood <david.sherwood@arm.com>
6097 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6098 from an unsigned int to a poly_uint64.
6099 (_loop_vec_info::slp_unrolling_factor): Likewise.
6100 (_loop_vec_info::vectorization_factor): Change from an int
6102 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6103 (vect_get_num_vectors): New function.
6104 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6105 (vect_get_num_copies): Use vect_get_num_vectors.
6106 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6107 to an unsigned int *.
6108 (vect_analyze_data_refs): Change min_vf from an int * to a
6110 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6111 than an unsigned HOST_WIDE_INT.
6112 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6113 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6114 to an unsigned int *.
6115 (vect_analyze_data_ref_dependences): Likewise.
6116 (vect_compute_data_ref_alignment): Handle polynomial vf.
6117 (vect_enhance_data_refs_alignment): Likewise.
6118 (vect_prune_runtime_alias_test_list): Likewise.
6119 (vect_shift_permute_load_chain): Likewise.
6120 (vect_supportable_dr_alignment): Likewise.
6121 (dependence_distance_ge_vf): Take the vectorization factor as a
6122 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6123 (vect_analyze_data_refs): Change min_vf from an int * to a
6125 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6126 vfm1 as a poly_uint64 rather than an int. Make the same change
6127 for the returned bound_scalar.
6128 (vect_gen_vector_loop_niters): Handle polynomial vf.
6129 (vect_do_peeling): Likewise. Update call to
6130 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6131 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6133 * tree-vect-loop.c (vect_determine_vectorization_factor)
6134 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6135 (vect_get_known_peeling_cost): Likewise.
6136 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6137 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6138 (vect_transform_loop): Likewise. Use the lowest possible VF when
6139 updating the upper bounds of the loop.
6140 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6142 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6143 polynomial unroll factors.
6144 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6145 (vect_make_slp_decision): Likewise.
6146 (vect_supported_load_permutation_p): Likewise, and polynomial
6148 (vect_analyze_slp_cost): Handle polynomial vf.
6149 (vect_slp_analyze_node_operations): Likewise.
6150 (vect_slp_analyze_bb_1): Likewise.
6151 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6152 than an unsigned HOST_WIDE_INT.
6153 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6154 (vectorizable_load): Handle polynomial vf.
6155 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6157 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6159 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6160 Alan Hayward <alan.hayward@arm.com>
6161 David Sherwood <david.sherwood@arm.com>
6163 * match.pd: Handle bit operations involving three constants
6164 and try to fold one pair.
6166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6168 * tree-vect-loop-manip.c: Include gimple-fold.h.
6169 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6170 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6171 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6172 Add a path that uses a step of VF instead of 1, but disable it
6174 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6175 and niters_no_overflow parameters. Update calls to
6176 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6177 Create a new SSA name if the latter choses to use a ste other
6178 than zero, and return it via niters_vector_mult_vf_var.
6179 * tree-vect-loop.c (vect_transform_loop): Update calls to
6180 vect_do_peeling, vect_gen_vector_loop_niters and
6181 slpeel_make_loop_iterate_ntimes.
6182 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6183 (vect_gen_vector_loop_niters): Update declarations after above changes.
6185 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6187 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6188 128-bit round to integer instructions.
6189 (ceil<mode>2): Likewise.
6190 (btrunc<mode>2): Likewise.
6191 (round<mode>2): Likewise.
6193 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6195 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6196 unaligned VSX load/store on P8/P9.
6197 (expand_block_clear): Allow the use of unaligned VSX
6198 load/store on P8/P9.
6200 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6202 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6204 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6205 swap associated with both a load and a store.
6207 2018-01-02 Andrew Waterman <andrew@sifive.com>
6209 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6210 * config/riscv/riscv.md (clear_cache): Use it.
6212 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6214 * web.c: Remove out-of-date comment.
6216 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6218 * expr.c (fixup_args_size_notes): Check that any existing
6219 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6220 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6221 (emit_single_push_insn): ...here.
6223 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6225 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6226 (const_vector_encoded_nelts): New function.
6227 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6228 (const_vector_int_elt, const_vector_elt): Declare.
6229 * emit-rtl.c (const_vector_int_elt_1): New function.
6230 (const_vector_elt): Likewise.
6231 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6232 of CONST_VECTOR_ELT.
6234 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6236 * expr.c: Include rtx-vector-builder.h.
6237 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6238 directly on the tree encoding.
6239 (const_vector_from_tree): Likewise.
6240 * optabs.c: Include rtx-vector-builder.h.
6241 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6242 sequence of "u" values.
6243 * vec-perm-indices.c: Include rtx-vector-builder.h.
6244 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6245 directly on the vec_perm_indices encoding.
6247 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6249 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6250 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6251 * rtx-vector-builder.h: New file.
6252 * rtx-vector-builder.c: Likewise.
6253 * rtl.h (rtx_def::u2): Add a const_vector field.
6254 (CONST_VECTOR_NPATTERNS): New macro.
6255 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6256 (CONST_VECTOR_DUPLICATE_P): Likewise.
6257 (CONST_VECTOR_STEPPED_P): Likewise.
6258 (CONST_VECTOR_ENCODED_ELT): Likewise.
6259 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6260 (unwrap_const_vec_duplicate): Likewise.
6261 (const_vec_series_p): Check for a non-duplicated vector encoding.
6262 Say that the function only returns true for integer vectors.
6263 * emit-rtl.c: Include rtx-vector-builder.h.
6264 (gen_const_vec_duplicate_1): Delete.
6265 (gen_const_vector): Call gen_const_vec_duplicate instead of
6266 gen_const_vec_duplicate_1.
6267 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6268 (gen_const_vec_duplicate): Use rtx_vector_builder.
6269 (gen_const_vec_series): Likewise.
6270 (gen_rtx_CONST_VECTOR): Likewise.
6271 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6272 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6273 Build a new vector rather than modifying a CONST_VECTOR in-place.
6274 (handle_special_swappables): Update call accordingly.
6275 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6276 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6277 Build a new vector rather than modifying a CONST_VECTOR in-place.
6278 (handle_special_swappables): Update call accordingly.
6280 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6282 * simplify-rtx.c (simplify_const_binary_operation): Use
6283 CONST_VECTOR_ELT instead of XVECEXP.
6285 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6287 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6288 the selector elements to be different from the data elements
6289 if the selector is a VECTOR_CST.
6290 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6291 ssizetype for the selector.
6293 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6295 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6296 before testing each element individually.
6297 * tree-vect-generic.c (lower_vec_perm): Likewise.
6299 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6301 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6302 * selftest-run-tests.c (selftest::run_tests): Call it.
6303 * vector-builder.h (vector_builder::operator ==): New function.
6304 (vector_builder::operator !=): Likewise.
6305 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6306 (vec_perm_indices::all_from_input_p): New function.
6307 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6308 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6309 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6310 instead of reading the VECTOR_CST directly. Detect whether both
6311 vector inputs are the same before constructing the vec_perm_indices,
6312 and update the number of inputs argument accordingly. Use the
6313 utility functions added above. Only construct sel2 if we need to.
6315 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6317 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6318 the broadcast of the low byte.
6319 (expand_mult_highpart): Use an explicit encoding for the permutes.
6320 * optabs-query.c (can_mult_highpart_p): Likewise.
6321 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6322 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6323 (vectorizable_bswap): Likewise.
6324 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6325 explicit encoding for the power-of-2 permutes.
6326 (vect_permute_store_chain): Likewise.
6327 (vect_grouped_load_supported): Likewise.
6328 (vect_permute_load_chain): Likewise.
6330 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6332 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6333 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6334 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6335 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6336 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6337 (vect_gen_perm_mask_any): Likewise.
6339 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6341 * int-vector-builder.h: New file.
6342 * vec-perm-indices.h: Include int-vector-builder.h.
6343 (vec_perm_indices): Redefine as an int_vector_builder.
6344 (auto_vec_perm_indices): Delete.
6345 (vec_perm_builder): Redefine as a stand-alone class.
6346 (vec_perm_indices::vec_perm_indices): New function.
6347 (vec_perm_indices::clamp): Likewise.
6348 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6349 (vec_perm_indices::new_vector): New function.
6350 (vec_perm_indices::new_expanded_vector): Update for new
6351 vec_perm_indices class.
6352 (vec_perm_indices::rotate_inputs): New function.
6353 (vec_perm_indices::all_in_range_p): Operate directly on the
6354 encoded form, without computing elided elements.
6355 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6356 encoding. Update for new vec_perm_indices class.
6357 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6358 the given vec_perm_builder.
6359 (expand_vec_perm_var): Update vec_perm_builder constructor.
6360 (expand_mult_highpart): Use vec_perm_builder instead of
6361 auto_vec_perm_indices.
6362 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6363 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6364 or double series encoding as appropriate.
6365 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6366 vec_perm_indices instead of auto_vec_perm_indices.
6367 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6368 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6369 (vect_permute_store_chain): Likewise.
6370 (vect_grouped_load_supported): Likewise.
6371 (vect_permute_load_chain): Likewise.
6372 (vect_shift_permute_load_chain): Likewise.
6373 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6374 (vect_transform_slp_perm_load): Likewise.
6375 (vect_schedule_slp_instance): Likewise.
6376 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6377 (vectorizable_mask_load_store): Likewise.
6378 (vectorizable_bswap): Likewise.
6379 (vectorizable_store): Likewise.
6380 (vectorizable_load): Likewise.
6381 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6382 vec_perm_indices instead of auto_vec_perm_indices. Use
6383 tree_to_vec_perm_builder to read the vector from a tree.
6384 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6385 vec_perm_builder instead of a vec_perm_indices.
6386 (have_whole_vector_shift): Use vec_perm_builder and
6387 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6388 truncation to calc_vec_perm_mask_for_shift.
6389 (vect_create_epilog_for_reduction): Likewise.
6390 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6391 from auto_vec_perm_indices to vec_perm_indices.
6392 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6393 instead of changing individual elements.
6394 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6395 the vector in d.perm.
6396 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6397 from auto_vec_perm_indices to vec_perm_indices.
6398 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6399 instead of changing individual elements.
6400 (arm_vectorize_vec_perm_const): Use new_vector to install
6401 the vector in d.perm.
6402 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6403 Update vec_perm_builder constructor.
6404 (rs6000_expand_interleave): Likewise.
6405 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6406 (rs6000_expand_interleave): Likewise.
6408 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6410 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6411 to qimode could truncate the indices.
6412 * optabs.c (expand_vec_perm_var): Likewise.
6414 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6416 * Makefile.in (OBJS): Add vec-perm-indices.o.
6417 * vec-perm-indices.h: New file.
6418 * vec-perm-indices.c: Likewise.
6419 * target.h (vec_perm_indices): Replace with a forward class
6421 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6422 * optabs.h: Include vec-perm-indices.h.
6423 (expand_vec_perm): Delete.
6424 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6425 (expand_vec_perm_const): Declare.
6426 * target.def (vec_perm_const_ok): Replace with...
6427 (vec_perm_const): ...this new hook.
6428 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6429 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6430 * doc/tm.texi: Regenerate.
6431 * optabs.def (vec_perm_const): Delete.
6432 * doc/md.texi (vec_perm_const): Likewise.
6433 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6434 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6435 expand_vec_perm for constant permutation vectors. Assert that
6436 the mode of variable permutation vectors is the integer equivalent
6437 of the mode that is being permuted.
6438 * optabs-query.h (selector_fits_mode_p): Declare.
6439 * optabs-query.c: Include vec-perm-indices.h.
6440 (selector_fits_mode_p): New function.
6441 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6442 is defined, instead of checking whether the vec_perm_const_optab
6443 exists. Use targetm.vectorize.vec_perm_const instead of
6444 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6445 fit in the vector mode before using a variable permute.
6446 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6447 vec_perm_indices instead of an rtx.
6448 (expand_vec_perm): Replace with...
6449 (expand_vec_perm_const): ...this new function. Take the selector
6450 as a vec_perm_indices rather than an rtx. Also take the mode of
6451 the selector. Update call to shift_amt_for_vec_perm_mask.
6452 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6453 Use vec_perm_indices::new_expanded_vector to expand the original
6454 selector into bytes. Check whether the indices fit in the vector
6455 mode before using a variable permute.
6456 (expand_vec_perm_var): Make global.
6457 (expand_mult_highpart): Use expand_vec_perm_const.
6458 * fold-const.c: Includes vec-perm-indices.h.
6459 * tree-ssa-forwprop.c: Likewise.
6460 * tree-vect-data-refs.c: Likewise.
6461 * tree-vect-generic.c: Likewise.
6462 * tree-vect-loop.c: Likewise.
6463 * tree-vect-slp.c: Likewise.
6464 * tree-vect-stmts.c: Likewise.
6465 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6467 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6468 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6469 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6470 (aarch64_vectorize_vec_perm_const): ...this new function.
6471 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6472 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6473 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6474 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6475 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6476 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6477 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6479 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6480 check for NEON modes.
6481 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6482 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6483 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6484 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6486 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6487 the old VEC_PERM_CONST conditions.
6488 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6489 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6490 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6491 (ia64_vectorize_vec_perm_const_ok): Merge into...
6492 (ia64_vectorize_vec_perm_const): ...this new function.
6493 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6494 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6495 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6496 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6497 * config/mips/mips.c (mips_expand_vec_perm_const)
6498 (mips_vectorize_vec_perm_const_ok): Merge into...
6499 (mips_vectorize_vec_perm_const): ...this new function.
6500 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6501 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6502 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6503 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6504 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6505 (rs6000_expand_vec_perm_const): Delete.
6506 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6508 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6509 (altivec_expand_vec_perm_const_le): Take each operand individually.
6510 Operate on constant selectors rather than rtxes.
6511 (altivec_expand_vec_perm_const): Likewise. Update call to
6512 altivec_expand_vec_perm_const_le.
6513 (rs6000_expand_vec_perm_const): Delete.
6514 (rs6000_vectorize_vec_perm_const_ok): Delete.
6515 (rs6000_vectorize_vec_perm_const): New function.
6516 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6517 an element count and rtx array.
6518 (rs6000_expand_extract_even): Update call accordingly.
6519 (rs6000_expand_interleave): Likewise.
6520 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6521 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6522 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6523 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6524 (rs6000_expand_vec_perm_const): Delete.
6525 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6526 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6527 (altivec_expand_vec_perm_const_le): Take each operand individually.
6528 Operate on constant selectors rather than rtxes.
6529 (altivec_expand_vec_perm_const): Likewise. Update call to
6530 altivec_expand_vec_perm_const_le.
6531 (rs6000_expand_vec_perm_const): Delete.
6532 (rs6000_vectorize_vec_perm_const_ok): Delete.
6533 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6534 reference to the SPE evmerge intructions.
6535 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6536 an element count and rtx array.
6537 (rs6000_expand_extract_even): Update call accordingly.
6538 (rs6000_expand_interleave): Likewise.
6539 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6540 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6542 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6544 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6546 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6547 vector mode and that that mode matches the mode of the data
6549 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6550 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6551 directly using expand_vec_perm_1 when forcing selectors into
6553 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6555 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6557 * optabs-query.h (can_vec_perm_p): Delete.
6558 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6559 * optabs-query.c (can_vec_perm_p): Split into...
6560 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6561 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6562 particular selector is valid.
6563 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6564 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6565 (vect_grouped_load_supported): Likewise.
6566 (vect_shift_permute_load_chain): Likewise.
6567 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6568 (vect_transform_slp_perm_load): Likewise.
6569 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6570 (vectorizable_bswap): Likewise.
6571 (vect_gen_perm_mask_checked): Likewise.
6572 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6573 implementations of variable permutation vectors into account
6574 when deciding which selector to use.
6575 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6576 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6577 with a false third argument.
6578 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6579 to test whether the constant selector is valid and can_vec_perm_var_p
6580 to test whether a variable selector is valid.
6582 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6584 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6585 * optabs-query.c (can_vec_perm_p): Likewise.
6586 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6587 instead of vec_perm_indices.
6588 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6589 (vect_gen_perm_mask_checked): Likewise,
6590 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6591 (vect_gen_perm_mask_checked): Likewise,
6593 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6595 * optabs-query.h (qimode_for_vec_perm): Declare.
6596 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6597 (qimode_for_vec_perm): ...this new function.
6598 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6600 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6602 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6603 does not have a conditional at the top.
6605 2018-01-02 Richard Biener <rguenther@suse.de>
6607 * ipa-inline.c (big_speedup_p): Fix expression.
6609 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6612 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6615 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6619 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6620 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6621 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6622 cond_taken_branch_cost 3->4.
6624 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6626 PR tree-optimization/83581
6627 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6628 TODO_cleanup_cfg if any changes have been made.
6631 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6632 convert_modes if target mode has the right side, but different mode
6636 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6637 last argument when extracting from CONCAT. If either from_real or
6638 from_imag is NULL, use expansion through memory. If result is not
6639 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6640 the parts directly to inner mode, if even that fails, use expansion
6644 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6645 check for bswap in mode rather than HImode and use that in expand_unop
6648 Copyright (C) 2018 Free Software Foundation, Inc.
6650 Copying and distribution of this file, with or without modification,
6651 are permitted in any medium without royalty provided the copyright
6652 notice and this notice are preserved.