1 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
4 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
5 (ASAN_MARK): Fix fnspec to account for return value, change pointer
6 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
8 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
11 * params.def (inline-min-speedup): Increase from 8 to 15.
12 (max-inline-insns-auto): Decrease from 40 to 30.
13 * ipa-split.c (consider_split): Add some buffer for function to
14 be considered inlining candidate.
15 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
18 2018-02-12 Richard Biener <rguenther@suse.de>
20 PR tree-optimization/84037
21 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
22 matched stmts if we cannot swap the non-matched ones.
24 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
26 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
27 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
28 _mm_maskz_scalef_round_ss): New intrinsics.
29 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
30 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
31 __builtin_ia32_scalefss_round): Remove.
32 (__builtin_ia32_scalefsd_mask_round,
33 __builtin_ia32_scalefss_mask_round): New intrinsics.
34 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
35 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
36 ((match_operand:VF_128 2 "<round_nimm_predicate>"
37 "<round_constraint>")): Changed to ...
38 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
39 "<round_scalar_constraint>")): ... this.
40 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
41 %0, %1, %2<round_op3>}"): Changed to ...
42 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
43 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
44 %2<round_scalar_mask_op3>}"): ... this.
45 * config/i386/subst.md (round_scalar_nimm_predicate): New.
47 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
49 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
50 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
51 (_mm_maskz_sqrt_round_ss): New intrinsics.
52 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
53 (__builtin_ia32_sqrtsd_mask_round)
54 (__builtin_ia32_sqrtss_mask_round): New builtins.
55 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
56 (__builtin_ia32_sqrtss_round): Remove.
57 (__builtin_ia32_sqrtsd_mask_round)
58 (__builtin_ia32_sqrtss_mask_round): New builtins.
59 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
60 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
61 ((match_operand:VF_128 1 "vector_operand"
62 "xBm,<round_constraint>")): Changed to ...
63 ((match_operand:VF_128 1 "vector_operand"
64 "xBm,<round_scalar_constraint>")): ... this.
65 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
66 %0, %2, %<iptr>1<round_op3>}): Changed to ...
67 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
68 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
69 %<iptr>1<round_scalar_mask_op3>}): ... this.
70 ((set_attr "prefix" "<round_prefix>")): Changed to ...
71 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
73 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
76 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
77 Cast vec_cmpeq result to correct type.
78 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
79 Cast vec_cmpgt result to correct type.
81 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
83 * final.c (final_scan_insn_1): Renamed from...
84 (final_scan_insn): ... this. New wrapper, to recover
85 seen from the outermost call in recursive ones.
86 * config/sparc/sparc.c (output_return): Drop seen from call.
87 (output_sibcall): Likewise.
88 * config/visium/visium.c (output_branch): Likewise.
90 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
92 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
95 2018-02-10 Alan Modra <amodra@gmail.com>
98 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
99 Specify LR as an input.
101 2018-02-10 Jakub Jelinek <jakub@redhat.com>
104 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
105 remove_member_access_dummy_vars): New functions.
106 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
107 lower_omp_1, execute_lower_omp): Use them.
109 PR rtl-optimization/84308
110 * shrink-wrap.c (spread_components): Release todo vector.
112 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
114 PR rtl-optimization/57193
115 * ira-color.c (struct allocno_color_data): Add member
116 conflict_allocno_hard_prefs.
117 (update_conflict_allocno_hard_prefs): New.
118 (bucket_allocno_compare_func): Add a preference based on
119 conflict_allocno_hard_prefs.
120 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
121 (color_allocnos): Remove a dead code. Initiate
122 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
124 2018-02-09 Jakub Jelinek <jakub@redhat.com>
127 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
128 constraint from =wa to wa. Avoid a subreg on the output operand,
129 instead use a pseudo and subreg it in a move.
130 (p9_xxbrd_<mode>): Changed to ...
131 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
132 (p9_xxbrd_v2df): New expander.
133 (p9_xxbrw_<mode>): Changed to ...
134 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
135 (p9_xxbrw_v4sf): New expander.
137 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
139 * config/rx.md: updated "movsicc" expand to be matched by GCC
140 * testsuite/gcc.target/rx/movsicc.c: new test case
142 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
145 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
146 multiply in 32-bit mode.
147 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
148 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
151 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
153 * config/rx/constraints.md: added new constraint CALL_OP_SYMBOL_REF
154 to allow or block "symbol_ref" depending on value of TARGET_JSR
155 * config/rx/rx.md: use CALL_OP_SYMBOL_REF in call_internal and
156 call_value_internal insns
158 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
161 * dwarf2out.c (is_trivial_indirect_ref): New function.
162 (dwarf2out_late_global_decl): Do not generate a location
163 attribute for variables that have a non-trivial DECL_VALUE_EXPR
164 and that are not defined in the current unit.
166 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
168 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
169 instead of a libcall for UNORDERED.
171 2018-02-09 Tamar Christina <tamar.christina@arm.com>
174 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
175 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
177 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
180 * config/s390/s390.c (s390_set_current_function): Invoke
181 s390_indirect_branch_settings also if fndecl didn't change.
183 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
185 * config/rs6000/rs6000.md (blockage): Set length to zero.
187 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
189 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
191 2018-02-09 Jakub Jelinek <jakub@redhat.com>
194 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
195 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
199 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
200 PARALLEL incoming that failed vt_get_decl_and_offset check.
203 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
204 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
206 (get_variable_section): For decls in named .bss* sections pass true as
207 second argument to bss_initializer_p.
209 2018-02-09 Marek Polacek <polacek@redhat.com>
210 Jakub Jelinek <jakub@redhat.com>
213 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
214 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
215 Sync some changes from cxx_fold_indirect_ref.
217 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
219 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
221 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
222 (BLOCK_INLINE_ENTRY_LABEL): New.
223 (dwarf2out_var_location): Disregard inline entry markers.
224 (inline_entry_data): New struct.
225 (inline_entry_data_hasher): New hashtable type.
226 (inline_entry_data_hasher::hash): New.
227 (inline_entry_data_hasher::equal): New.
228 (inline_entry_data_table): New variable.
229 (add_high_low_attributes): Add DW_AT_entry_pc and
230 DW_AT_GNU_entry_view attributes if a pending entry is found
231 in inline_entry_data_table. Add old entry_pc attribute only
232 if debug nonbinding markers are disabled.
233 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
235 (block_within_block_p, dwarf2out_inline_entry): New.
236 (dwarf2out_finish): Check that no entries remained in
237 inline_entry_data_table.
238 * final.c (reemit_insn_block_notes): Handle inline entry notes.
239 (final_scan_insn, notice_source_line): Likewise.
240 (rest_of_clean_state): Skip inline entry markers.
241 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
243 * gimple.c (gimple_build_debug_inline_entry): New.
244 * gimple.h (enum gimple_debug_subcode): Add
245 GIMPLE_DEBUG_INLINE_ENTRY.
246 (gimple_build_debug_inline_entry): Declare.
247 (gimple_debug_inline_entry_p): New.
248 (gimple_debug_nonbind_marker_p): Adjust.
249 * insn-notes.def (INLINE_ENTRY): New.
250 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
251 inline entry marker notes.
252 (print_insn): Likewise.
253 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
254 (INSN_DEBUG_MARKER_KIND): Likewise.
255 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
256 * tree-inline.c (expand_call_inline): Build and insert
257 debug_inline_entry stmt.
258 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
259 inline entry blocks early, if nonbind markers are enabled.
260 (dump_scope_block): Dump fragment info.
261 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
262 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
263 (gimple_build_debug_inline_entry): New.
264 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
265 Enable/disable inline entry points too.
266 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
267 (DEBUG_INSN): Describe inline entry markers.
269 * common.opt (gvariable-location-views): New.
270 (gvariable-location-views=incompat5): New.
271 * config.in: Rebuilt.
272 * configure: Rebuilt.
273 * configure.ac: Test assembler for view support.
274 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
275 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
276 * dwarf2out.c (var_loc_view): New typedef.
277 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
278 (dwarf2out_locviews_in_attribute): New.
279 (dwarf2out_locviews_in_loclist): New.
280 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
281 (enum dw_line_info_opcode): Add LI_adv_address.
282 (struct dw_line_info_table): Add view.
283 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
284 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
285 (zero_view_p): New variable.
286 (ZERO_VIEW_P): New macro.
287 (output_asm_line_debug_info): New.
288 (struct var_loc_node): Add view.
289 (add_AT_view_list, AT_loc_list): New.
290 (add_var_loc_to_decl): Add view param. Test it against last.
291 (new_loc_list): Add view params. Record them.
292 (AT_loc_list_ptr): Handle loc and view lists.
293 (view_list_to_loc_list_val_node): New.
294 (print_dw_val): Handle dw_val_class_view_list.
295 (size_of_die): Likewise.
296 (value_format): Likewise.
297 (loc_list_has_views): New.
298 (gen_llsym): Set vl_symbol too.
299 (maybe_gen_llsym, skip_loc_list_entry): New.
300 (dwarf2out_maybe_output_loclist_view_pair): New.
301 (output_loc_list): Output view list or entries too.
302 (output_view_list_offset): New.
303 (output_die): Handle dw_val_class_view_list.
304 (output_dwarf_version): New.
305 (output_compilation_unit_header): Use it.
306 (output_skeleton_debug_sections): Likewise.
307 (output_rnglists, output_line_info): Likewise.
308 (output_pubnames, output_aranges): Update version comments.
309 (output_one_line_info_table): Output view numbers in asm comments.
310 (dw_loc_list): Determine current endview, pass it to new_loc_list.
311 Call maybe_gen_llsym.
312 (loc_list_from_tree_1): Adjust.
313 (add_AT_location_description): Create view list attribute if
314 needed, check it's absent otherwise.
315 (convert_cfa_to_fb_loc_list): Adjust.
316 (maybe_emit_file): Call output_asm_line_debug_info for test.
317 (dwarf2out_var_location): Reset views as needed. Precompute
318 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
320 (new_line_info_table): Reset next view.
321 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
322 (dwarf2out_source_line): Likewise. Output view resets and labels to
323 the assembler, or select appropriate line info opcodes.
324 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
325 (optimize_string_length): Catch it. Adjust.
326 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
327 dw_val_class_view_list, and remove it if no longer needed.
328 (hash_loc_list): Hash view numbers.
329 (loc_list_hasher::equal): Compare them.
330 (optimize_location_lists): Check whether a view list symbol is
331 needed, and whether the locview attribute is present, and
332 whether they match. Remove the locview attribute if no longer
334 (index_location_lists): Call skip_loc_list_entry for test.
335 (dwarf2out_finish): Call output_asm_line_debug_info for test.
336 Use output_dwarf_version.
337 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
338 (struct dw_val_node): Add val_view_list.
339 * final.c (SEEN_NEXT_VIEW): New.
340 (set_next_view_needed): New.
341 (clear_next_view_needed): New.
342 (maybe_output_next_view): New.
343 (final_start_function): Rename to...
344 (final_start_function_1): ... this. Take pointer to FIRST,
345 add SEEN parameter. Emit param bindings in the initial view.
346 (final_start_function): Reintroduce SEEN-less interface.
347 (final): Rename to...
348 (final_1): ... this. Take SEEN parameter. Output final pending
349 next view at the end.
350 (final): Reintroduce seen-less interface.
351 (final_scan_insn): Output pending next view before switching
352 sections or ending a block. Mark the next view as needed when
353 outputting variable locations. Notify debug backend of section
354 changes, and of location view changes.
355 (rest_of_handle_final): Adjust.
356 * toplev.c (process_options): Autodetect value for debug variable
357 location views option. Warn on incompat5 without -gdwarf-5.
358 * doc/invoke.texi (gvariable-location-views): New.
359 (gvariable-location-views=incompat5): New.
360 (gno-variable-location-views): New.
362 2018-02-08 David Malcolm <dmalcolm@redhat.com>
364 PR tree-optimization/84136
365 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
366 that the result of find_edge is non-NULL.
368 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
371 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
372 storing integer register in SImode. Fix cost of 256 and 512
373 byte aligned SSE register store.
375 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
377 * config/i386/i386.c (ix86_multiplication_cost): Fix
378 multiplication cost for TARGET_AVX512DQ.
380 2018-02-08 Marek Polacek <polacek@redhat.com>
382 PR tree-optimization/84238
383 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
386 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
388 PR tree-optimization/84265
389 * tree-vect-stmts.c (vectorizable_store): Don't treat
390 VMAT_CONTIGUOUS accesses as grouped.
391 (vectorizable_load): Likewise.
393 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
395 PR tree-optimization/81635
396 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
397 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
398 (test_round_for_mask): New functions.
399 (wide_int_cc_tests): Call test_round_for_mask.
400 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
401 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
402 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
403 range returned by get_range_info.
405 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
408 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
409 * symtab.c: Include builtins.h
410 (symtab_node::output_to_lto_symbol_table_p): Move here
411 from lto-streamer-out.c:output_symbol_p.
412 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
413 (output_symbol_p): Move all logic to symtab.c
414 (produce_symtab): Update.
416 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
418 * config/s390/s390-opts.h (enum indirect_branch): Define.
419 * config/s390/s390-protos.h (s390_return_addr_from_memory)
420 (s390_indirect_branch_via_thunk)
421 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
422 (enum s390_indirect_branch_type): Define.
423 * config/s390/s390.c (struct s390_frame_layout, struct
424 machine_function): Remove.
425 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
426 (indirect_branch_table_label_no, indirect_branch_table_name):
428 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
429 (enum s390_indirect_branch_option): Define.
430 (s390_return_addr_from_memory): New function.
431 (s390_handle_string_attribute): New function.
432 (s390_attribute_table): Add new attribute handler.
433 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
434 (s390_indirect_branch_via_thunk): New function.
435 (s390_indirect_branch_via_inline_thunk): New function.
436 (s390_function_ok_for_sibcall): When jumping via thunk disallow
437 sibling call optimization for non z10 compiles.
438 (s390_emit_call): Force indirect branch target to be a single
439 register. Add r1 clobber for non-z10 compiles.
440 (s390_emit_epilogue): Emit return jump via return_use expander.
441 (s390_reorg): Handle JUMP_INSNs as execute targets.
442 (s390_option_override_internal): Perform validity checks for the
443 new command line options.
444 (s390_indirect_branch_attrvalue): New function.
445 (s390_indirect_branch_settings): New function.
446 (s390_set_current_function): Invoke s390_indirect_branch_settings.
447 (s390_output_indirect_thunk_function): New function.
448 (s390_code_end): Implement target hook.
449 (s390_case_values_threshold): Implement target hook.
450 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
452 * config/s390/s390.h (struct s390_frame_layout)
453 (struct machine_function): Move here from s390.c.
454 (TARGET_INDIRECT_BRANCH_NOBP_RET)
455 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
456 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
457 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
458 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
459 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
460 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
461 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
462 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
463 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
464 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
465 (mnemonic attribute): Add values which aren't recognized
467 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
468 pattern for branch conversion. Fix mnemonic attribute.
469 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
470 indirect branch via thunk if requested.
471 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
472 ("*indirect_jump"): Disable for branch conversion using out of
474 ("indirect_jump_via_thunk<mode>_z10")
475 ("indirect_jump_via_thunk<mode>")
476 ("indirect_jump_via_inlinethunk<mode>_z10")
477 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
478 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
479 ("casesi_jump_via_inlinethunk<mode>_z10")
480 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
481 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
482 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
483 ("*indirect2_jump"): Disable for branch conversion.
484 ("casesi_jump"): Turn into expander and expand patterns for branch
486 ("return_use"): New expander.
487 ("*return"): Emit return via thunk and rename it to ...
488 ("*return<mode>"): ... this one.
489 * config/s390/s390.opt: Add new options and and enum for the
492 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
494 * lra-constraints.c (match_reload): Unconditionally use
495 gen_lowpart_SUBREG, rather than selecting between that
496 and equivalent gen_rtx_SUBREG code.
498 2018-02-08 Richard Biener <rguenther@suse.de>
500 PR tree-optimization/84233
501 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
502 changed flag instead of boguously re-using phi_inserted.
504 2018-02-08 Martin Jambor <mjambor@suse.cz>
506 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
507 static local variables.
509 2018-02-08 Richard Biener <rguenther@suse.de>
511 PR tree-optimization/84278
512 * tree-vect-stmts.c (vectorizable_store): When looking for
513 smaller vector types to perform grouped strided loads/stores
514 make sure the mode is supported by the target.
515 (vectorizable_load): Likewise.
517 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
519 * config/aarch64/aarch64.c (aarch64_components_for_bb):
520 Increase LDP/STP opportunities by adding adjacent callee-saves.
522 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
524 PR rtl-optimization/84068
525 PR rtl-optimization/83459
526 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
528 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
530 PR tree-optimization/84224
531 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
532 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
535 2018-02-07 Iain Sandoe <iain@codesourcery.com>
538 * config/rs6000/altivec.md (*restore_world): Remove LR use.
539 * config/rs6000/predicates.md (restore_world_operation): Adjust op
540 count, remove one USE.
542 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
544 * doc/install.texi (Configuration): Document the
545 --with-long-double-format={ibm,ieee} PowerPC configuration
549 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
550 Convert from define_expand to be define_insn_and_split. Rework
551 float/double/_Float128 conversions to QI/HI/SImode to work with
552 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
553 conversions to QI/HImode types did a store and then a load to
554 truncate the value. For conversions to VSX registers, don't split
555 the insn, instead emit the code directly. Use the code iterator
556 any_fix to combine signed and unsigned conversions.
557 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
558 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
559 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
560 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
561 (fix_<mode>di2_hw): Likewise.
562 (fixuns_<mode>di2_hw): Likewise.
563 (fix_<mode>si2_hw): Likewise.
564 (fixuns_<mode>si2_hw): Likewise.
565 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
566 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
567 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
568 fix<uns>_trunc<SFDF:mode>si2_p8.
569 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
571 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
572 (fix<uns>_<mode>_mem): Likewise.
573 (fctiw<u>z_<mode>_mem): Likewise.
574 (fix<uns>_<mode>_mem): Likewise.
575 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
576 the register allocator from doing a direct move to the GPRs to do
577 a store, and instead use the ISA 3.0 store byte/half-word from
578 vector register instruction. For IEEE 128-bit floating point,
579 also optimize stores of 32-bit ints.
580 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
582 2018-02-07 Alan Hayward <alan.hayward@arm.com>
584 * genextract.c (push_pathstr_operand): New function to support
586 (walk_rtx): Call push_pathstr_operand.
587 (print_path): Support [a-zA-Z].
589 2018-02-07 Richard Biener <rguenther@suse.de>
591 PR tree-optimization/84037
592 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
593 (cse_and_gimplify_to_preheader): Declare.
594 (vect_get_place_in_interleaving_chain): Likewise.
595 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
597 (_loop_vec_info::~_loop_vec_info): Delete it.
598 (cse_and_gimplify_to_preheader): New function.
599 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
600 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
601 (vectorizable_load): Likewise. For grouped stores always base
602 the IV on the first element.
603 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
604 condition before gimplifying.
606 2018-02-07 Jakub Jelinek <jakub@redhat.com>
608 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
609 *DIV_EXPR and *MOD_EXPR.
611 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
614 * config/i386/i386.c (ix86_option_override_internal): Mask out
615 the CF_SET bit when checking -fcf-protection.
617 2018-02-07 Tom de Vries <tom@codesourcery.com>
620 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
623 2018-02-07 Richard Biener <rguenther@suse.de>
625 PR tree-optimization/84204
626 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
629 PR tree-optimization/84205
630 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
631 special-case isl_ast_op_zdiv_r.
633 PR tree-optimization/84223
634 * graphite-scop-detection.c (gather_bbs::before_dom_children):
635 Only add conditions from within the region.
636 (gather_bbs::after_dom_children): Adjust.
638 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
641 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
642 * config/avr/avr.md: Only post-reload split REG-REG moves if
643 either register is REGERAL_REG_P.
645 2018-02-07 Jakub Jelinek <jakub@redhat.com>
647 PR tree-optimization/84235
648 * tree-ssa-scopedtables.c
649 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
650 if the subtraction is performed in floating point type where NaNs are
651 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
652 build 1. Formatting fix.
654 2018-02-06 Jakub Jelinek <jakub@redhat.com>
657 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
658 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
659 and skip it regardless of bb boundaries. Use CALL_P macro,
660 don't test INSN_P (insn) together with CALL_P or JUMP_P check
661 unnecessarily, formatting fix.
663 2018-02-06 Michael Collison <michael.collison@arm.com>
665 * config/arm/thumb2.md:
666 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
667 (*thumb_mov_notscc): Ditto.
669 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
672 * config/rs6000/rs6000.md (su code attribute): Use "u" for
673 unsigned_fix, not "s".
675 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
677 * configure.ac (gcc_fn_eh_frame_ro): New function.
678 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
679 correct .eh_frame permissions.
680 * configure: Regenerate.
682 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
684 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
687 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
689 * config/rs6000/rs6000.c (rs6000_option_override_internal):
690 Display warning message for -mno-speculate-indirect-jumps.
692 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
694 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
696 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
698 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
700 PR tree-optimization/84225
701 * tree-eh.c (find_trapping_overflow): Only call
702 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
704 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
707 * config/i386/i386.c: Reimplement the check of possible options
708 -mibt/-mshstk conbination. Change error messages.
709 * doc/invoke.texi: Fix a typo: remove extra '='.
711 2018-02-06 Marek Polacek <polacek@redhat.com>
713 PR tree-optimization/84228
714 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
716 2018-02-06 Tamar Christina <tamar.christina@arm.com>
719 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
720 emitted arch directives.
721 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
722 __ARM_FEATURE_COPROC before changing architectures.
724 2018-02-06 Richard Biener <rguenther@suse.de>
726 * config/i386/i386.c (print_reg): Fix typo.
727 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
729 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
731 * configure: Regenerate.
733 2018-02-05 Martin Sebor <msebor@redhat.com>
735 PR tree-optimization/83369
736 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
739 2018-02-05 Martin Liska <mliska@suse.cz>
741 * doc/invoke.texi: Cherry-pick upstream r323995.
743 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
745 * ira.c (ira_init_register_move_cost): Adjust comment.
747 2018-02-05 Martin Liska <mliska@suse.cz>
749 PR gcov-profile/84137
750 * doc/gcov.texi: Fix typo in documentation.
752 2018-02-05 Martin Liska <mliska@suse.cz>
754 PR gcov-profile/83879
755 * doc/gcov.texi: Document necessity of --dynamic-list-data when
756 using dlopen functionality.
758 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
760 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
761 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
762 _mm_maskz_range_ss, _mm_mask_range_round_ss,
763 _mm_maskz_range_round_ss): New intrinsics.
764 (__builtin_ia32_rangesd128_round)
765 (__builtin_ia32_rangess128_round): Remove.
766 (__builtin_ia32_rangesd128_mask_round,
767 __builtin_ia32_rangess128_mask_round): New builtins.
768 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
769 __builtin_ia32_rangess128_round): Remove.
770 (__builtin_ia32_rangesd128_mask_round,
771 __builtin_ia32_rangess128_mask_round): New builtins.
772 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
773 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
774 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
775 "<round_saeonly_constraint>")): Changed to ...
776 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
777 "<round_saeonly_scalar_constraint>")): ... this.
778 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
779 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
780 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
781 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
782 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
784 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
786 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
788 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
789 Remove all values except native, 8540 and 8548.
791 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
793 * config/i386/i386.c (ix86_output_function_return): Pass
794 INVALID_REGNUM, instead of -1, as invalid register number to
795 indirect_thunk_name and output_indirect_thunk.
797 2018-02-02 Julia Koval <julia.koval@intel.com>
799 * config.gcc: Add -march=icelake.
800 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
801 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
802 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
803 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
804 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
805 (processor_target_table): Add icelake.
806 (ix86_option_override_internal): Handle new PTAs.
807 (get_builtin_code_for_version): Handle icelake.
808 (M_INTEL_COREI7_ICELAKE): New.
809 (fold_builtin_cpu): Handle icelake.
810 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
811 * doc/invoke.texi: Add -march=icelake.
813 2018-02-02 Julia Koval <julia.koval@intel.com>
815 * config/i386/i386.c (ix86_option_override_internal): Change flags type
817 * wide-int-bitmask.h: New.
819 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
822 * config/i386/i386.md: Replace Pmode with word_mode in
823 builtin_setjmp_setup and builtin_longjmp to support x32.
825 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
829 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
831 (rs6000_supported_cpu_names): New static variable.
832 (linux_cpu_translation_table): Likewise.
833 (elf_platform) <cpu>: Define new static variable and use it.
834 Translate kernel AT_PLATFORM name to canonical name if needed.
835 Error if platform name is unknown.
837 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
840 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
842 2018-02-01 Jeff Law <law@redhat.com>
845 * config/i386/i386.c (release_scratch_register_on_entry): Add new
846 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
847 the scratch if RELEASE_VIA_POP is false.
848 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
849 If we have to save a temporary register, decrement SIZE appropriately.
850 Pass new arguments to release_scratch_register_on_entry.
851 (ix86_adjust_stack_and_probe): Likewise.
852 (ix86_emit_probe_stack_range): Pass new arguments to
853 release_scratch_register_on_entry.
855 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
857 PR rtl-optimization/84157
858 * combine.c (change_zero_ext): Use REG_P predicate in
859 front of HARD_REGISTER_P predicate.
861 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
863 * config/avr/avr.c (avr_option_override): Move disabling of
864 -fdelete-null-pointer-checks to...
865 * common/config/avr/avr-common.c (avr_option_optimization_table):
868 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
870 PR tree-optimization/81635
871 * tree-data-ref.c (split_constant_offset_1): For types that
872 wrap on overflow, try to use range info to prove that wrapping
875 2018-02-01 Renlin Li <renlin.li@arm.com>
878 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
880 (aarch64_register_move_cost): Likewise.
881 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
883 (REG_CLASS_NAMES): Likewise.
884 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
885 TAILCALL_ADDR_REGS. Remove IP registers.
886 * config/aarch64/aarch64.md (Ucs): Update register constraint.
888 2018-02-01 Richard Biener <rguenther@suse.de>
890 * domwalk.h (dom_walker::dom_walker): Add additional constructor
891 for specifying RPO order and allow NULL for that.
892 * domwalk.c (dom_walker::dom_walker): Likewise.
893 (dom_walker::walk): Handle NULL RPO order.
894 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
896 (rewrite_update_dom_walker): Likewise.
897 (mark_def_dom_walker): Likewise.
899 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
901 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
902 (aarch64_maybe_expand_sve_subreg_move): Declare.
903 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
904 * config/aarch64/predicates.md (aarch64_any_register_operand): New
906 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
907 that are semantically a reverse operation.
908 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
909 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
910 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
912 (aarch64_can_change_mode_class): For big-endian, forbid changes
913 between two SVE modes if they have different element sizes.
915 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
917 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
918 the TImode handling for big-endian targets.
920 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
922 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
923 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
925 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
926 Remove BSWAP handing for big-endian targets and use the form of
927 LD1RQ appropariate for the mode.
929 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
931 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
932 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
935 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
938 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
939 check for operands that need to go through aarch64_sve_reload_be.
941 2018-02-01 Jakub Jelinek <jakub@redhat.com>
943 PR tree-optimization/81661
944 PR tree-optimization/84117
945 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
946 * tree-eh.c: Include gimplify.h.
947 (find_trapping_overflow, replace_trapping_overflow,
948 rewrite_to_non_trapping_overflow): New functions.
949 * tree-vect-loop.c: Include tree-eh.h.
950 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
951 * tree-data-ref.c: Include tree-eh.h.
952 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
954 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
956 PR rtl-optimization/84123
957 * combine.c (change_zero_ext): Check if hard register satisfies
958 can_change_dest_mode before calling gen_lowpart_SUBREG.
960 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
963 * ira.c (ira_init_register_move_cost): Remove assert.
965 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
967 PR rtl-optimization/84071
968 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
969 * doc/tm.texi: Regenerate.
971 2018-01-31 Richard Biener <rguenther@suse.de>
973 PR tree-optimization/84132
974 * tree-data-ref.c (analyze_miv_subscript): Properly
975 check whether evolution_function_is_affine_multivariate_p
976 before calling gcd_of_steps_may_divide_p.
978 2018-01-31 Julia Koval <julia.koval@intel.com>
981 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
982 * config/i386/i386.md (rdpid_rex64) New.
983 (rdpid): Make 32bit only.
985 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
988 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
989 an IDENTIFIER_NODE for FUNCTION_TYPE's.
991 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
994 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
996 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
998 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1000 PR rtl-optimization/84071
1001 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1002 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1004 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1006 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1007 (arc_attribute_table): Add 'aux' attribute.
1008 (arc_in_small_data_p): Consider aux like variables.
1009 (arc_is_aux_reg_p): New function.
1010 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1011 (arc_get_aux_arg): New function.
1012 (prepare_move_operands): Handle aux-register access.
1013 (arc_handle_aux_attribute): New function.
1014 * doc/extend.texi (ARC Variable attributes): Add subsection.
1016 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1018 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1019 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1020 (arc_attribute_table): Add 'uncached' attribute.
1021 (arc_print_operand): Print '.di' flag for uncached memory
1023 (arc_in_small_data_p): Do not consider for small data the uncached
1025 (arc_is_uncached_mem_p): New function.
1026 * config/arc/predicates.md (compact_store_memory_operand): Check
1027 for uncached memory accesses.
1028 (nonvol_nonimm_operand): Likewise.
1029 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1031 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1034 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1035 falign-loops=): Add Optimization flag.
1037 2018-01-30 Jeff Law <law@redhat.com>
1040 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1041 INT_REGISTERS_SAVED. Check it prior to calling
1042 get_scratch_register_on_entry.
1043 (ix86_adjust_stack_and_probe): Similarly.
1044 (ix86_emit_probe_stack_range): Similarly.
1045 (ix86_expand_prologue): Corresponding changes.
1047 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1050 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1051 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1053 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1056 * lra-constraints.c (curr_insn_transform): Process AND in the
1059 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1061 PR rtl-optimization/83986
1062 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1063 dependence against last_pending_memory_flush in addition to
1066 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1068 PR tree-optimization/81611
1069 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1072 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1074 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1077 2018-01-30 Richard Biener <rguenther@suse.de>
1078 Jakub Jelinek <jakub@redhat.com>
1080 PR tree-optimization/84111
1081 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1082 inner loops added during recursion, as they don't have up-to-date
1085 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1088 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1089 (can_inline_edge_by_limits_p): ... here.
1090 (can_early_inline_edge_p, check_callers,
1091 update_caller_keys, update_callee_keys, recursive_inlining,
1092 add_new_edges_to_heap, speculation_useful_p,
1093 inline_small_functions,
1094 inline_small_functions, flatten_function,
1095 inline_to_all_callers_1): Update.
1097 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1099 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1102 2018-01-30 Richard Biener <rguenther@suse.de>
1104 PR tree-optimization/83008
1105 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1106 invariant and constant vector uses in stmts when they need
1109 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1112 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1113 * configure: Regenerate.
1115 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1117 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1119 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1120 Use gen_rtx_REG rather than gen_lowpart.
1122 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1124 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1125 rather than 0 when creating partial subregs.
1127 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1129 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1132 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1135 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1136 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1137 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1138 flags. This restores the settings used before the 2017-07-24.
1139 Turning off pre increment/decrement/modify allows IVOPTS to
1140 optimize DF/SF loops where the index is an int.
1142 2018-01-29 Richard Biener <rguenther@suse.de>
1143 Kelvin Nilsen <kelvin@gcc.gnu.org>
1146 * tree-vect-stmts.c (vectorizable_call): Don't call
1147 targetm.vectorize_builtin_md_vectorized_function if callee is
1150 2018-01-22 Carl Love <cel@us.ibm.com>
1152 * doc/extend.tex: Fix typo in second arg in
1153 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1155 2018-01-29 Richard Biener <rguenther@suse.de>
1157 PR tree-optimization/84086
1158 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1159 (flush_ssaname_freelist): When SSA names were released reset
1160 the SCEV hash table.
1162 2018-01-29 Richard Biener <rguenther@suse.de>
1164 PR tree-optimization/84057
1165 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1166 removed paths when removing edges.
1168 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1170 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1171 -mfunction-return=@var{choice}.
1173 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1176 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1178 (layout::print_source_line): Likewise.
1179 (test_get_line_width_without_trailing_whitespace): Add test cases.
1181 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1184 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1187 2018-01-26 Jim Wilson <jimw@sifive.com>
1189 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1191 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1194 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1196 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1197 and CMP + SUB-immediate -> SUBS.
1199 2018-01-26 Martin Sebor <msebor@redhat.com>
1201 PR tree-optimization/83896
1202 * tree-ssa-strlen.c (get_string_len): Rename...
1203 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1204 Avoid assuming length is constant.
1205 (handle_char_store): Use HOST_WIDE_INT for string length.
1207 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1210 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1211 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1213 2018-01-26 Richard Biener <rguenther@suse.de>
1215 PR rtl-optimization/84003
1216 * dse.c (record_store): Only record redundant stores when
1217 the earlier store aliases at least all accesses the later one does.
1219 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1221 PR rtl-optimization/83985
1222 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1223 REG_CFA_RESTORE insns.
1224 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1225 REG_CFA_RESTORE insns here.
1228 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1229 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1231 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1233 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1234 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1235 (arc_init): Likewise.
1236 (arc_override_options): Likewise.
1237 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1239 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1241 * config/arc/arc.h (TARGET_DBNZ): Define.
1242 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1243 properly set the tune attribute.
1244 (dbnz): Use TARGET_DBNZ guard.
1245 * config/arc/arc.opt (mtune): Add core3 option.
1247 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1249 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1250 recognize new pic like addresses.
1251 (arc_delegitimize_address): Clean up.
1253 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1255 * config/arc/arc-arches.def: Option mrf16 valid for all
1257 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1258 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1259 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1260 * config/arc/arc-tables.opt: Regenerate.
1261 * config/arc/arc.c (arc_conditional_register_usage): Handle
1262 reduced register file case.
1263 (arc_file_start): Set must have build attributes.
1264 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1266 * config/arc/arc.opt (mrf16): Add new option.
1267 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1268 * config/arc/genmultilib.awk: Handle new mrf16 option.
1269 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1270 * config/arc/t-multilib: Regenerate.
1271 * doc/invoke.texi (ARC Options): Document mrf16 option.
1273 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1275 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1276 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1277 (arc_attribute_table): Add 'secure_call' attribute.
1278 (arc_print_operand): Print secure call operand.
1279 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1281 (arc_is_secure_call_p): New function. * config/arc/arc.md
1282 (call_i): Add support for sjli instruction.
1283 (call_value_i): Likewise.
1284 * config/arc/constraints.md (Csc): New constraint.
1286 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1287 John Eric Martin <John.Martin@emmicro-us.com>
1289 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1290 * config/arc/arc.c (_arc_jli_section): New struct.
1291 (arc_jli_section): New type.
1292 (rc_jli_sections): New static variable.
1293 (arc_handle_jli_attribute): New function.
1294 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1295 (arc_file_end): New function.
1296 (TARGET_ASM_FILE_END): Define.
1297 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1298 (arc_add_jli_section): New function.
1299 (jli_call_scan): Likewise.
1300 (arc_reorg): Call jli_call_scan.
1301 (arc_output_addsi): Remove 'S' from printing asm operand.
1302 (arc_is_jli_call_p): New function.
1303 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1305 (movhi_insn): Likewise.
1306 (movsi_insn): Likewise.
1307 (movsi_set_cc_insn): Likewise.
1308 (loadqi_update): Likewise.
1309 (load_zeroextendqisi_update): Likewise.
1310 (load_signextendqisi_update): Likewise.
1311 (loadhi_update): Likewise.
1312 (load_zeroextendhisi_update): Likewise.
1313 (load_signextendhisi_update): Likewise.
1314 (loadsi_update): Likewise.
1315 (loadsf_update): Likewise.
1316 (movsicc_insn): Likewise.
1317 (bset_insn): Likewise.
1318 (bxor_insn): Likewise.
1319 (bclr_insn): Likewise.
1320 (bmsk_insn): Likewise.
1321 (bicsi3_insn): Likewise.
1322 (cmpsi_cc_c_insn): Likewise.
1323 (movsi_ne): Likewise.
1324 (movsi_cond_exec): Likewise.
1325 (clrsbsi2): Likewise.
1336 (call_i): Remove 'S' asm letter, add jli instruction.
1337 (call_value_i): Likewise.
1338 * config/arc/arc.op (mjli-always): New option.
1339 * config/arc/constraints.md (Cji): New constraint.
1340 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1342 (subsf3_fpx): Likewise.
1343 (mulsf3_fpx): Likewise.
1344 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1346 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1348 * doc/invoke.texi (ARC): Document mjli-always option.
1350 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1352 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1353 and use incw and decw where possible
1354 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1356 2018-01-26 Richard Biener <rguenther@suse.de>
1358 PR tree-optimization/81082
1359 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1360 association if it requires casting to unsigned.
1361 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1362 from fold_plusminus_mult_expr to catch important cases late when
1363 range info is available.
1365 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1367 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1368 * configure.ac (hidden_linkonce): New test.
1369 * configure: Regenerate.
1370 * config.in: Regenerate.
1372 2018-01-26 Julia Koval <julia.koval@intel.com>
1374 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1375 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1376 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1377 _mm_mask_bitshuffle_epi64_mask): Fix type.
1378 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1379 USI_FTYPE_V4DI_V4DI_USI): Remove.
1380 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1381 __builtin_ia32_vpshufbitqmb256_mask,
1382 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1383 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1384 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1386 2018-01-26 Alan Modra <amodra@gmail.com>
1389 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1390 UNSPEC_VBPERMQ. Sort other unspecs.
1392 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1394 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1396 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1399 * predict.c (drop_profile): Do not push/pop cfun; update also
1401 (handle_missing_profiles): Fix logic looking for zero profiles.
1403 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1406 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1407 on functions with #pragma omp declare simd or functions with simd
1409 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1410 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1411 Remove trailing \n from warning_at calls.
1413 2018-01-25 Tom de Vries <tom@codesourcery.com>
1416 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1417 for neutered workers.
1419 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1422 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1423 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1425 2018-01-24 Jeff Law <law@redhat.com>
1428 * i386.c (get_probe_interval): Move to earlier point.
1429 (ix86_compute_frame_layout): If -fstack-clash-protection and
1430 the frame is larger than the probe interval, then use pushes
1431 to save registers rather than reg->mem moves.
1432 (ix86_expand_prologue): Remove conditional for int_registers_saved
1435 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1438 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1439 min/max for never referenced object.
1441 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1444 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1446 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1447 attributes from DECL_ATTRIBUTES (decl) without affecting
1448 DECL_ATTRIBUTES (current_function_decl).
1449 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1450 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1452 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1454 PR tree-optimization/83979
1455 * fold-const.c (fold_comparison): Use constant_boolean_node
1456 instead of boolean_{true,false}_node.
1458 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1460 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1463 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1465 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1466 Simplify the clause that sets the length attribute.
1467 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1468 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1469 clause that sets the length attribute.
1470 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1472 2018-01-24 Tom de Vries <tom@codesourcery.com>
1475 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1476 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1477 Add strict parameter.
1478 (prevent_branch_around_nothing): Insert dummy insn between branch to
1479 label and label with no ptx insn inbetween.
1480 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1482 2018-01-24 Tom de Vries <tom@codesourcery.com>
1485 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1486 for neutered threads in warp.
1487 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1489 2018-01-24 Richard Biener <rguenther@suse.de>
1491 PR tree-optimization/83176
1492 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1495 2018-01-24 Richard Biener <rguenther@suse.de>
1497 PR tree-optimization/82819
1498 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1499 code generating pluses that are no-ops in the target precision.
1501 2018-01-24 Richard Biener <rguenther@suse.de>
1504 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1506 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1508 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1509 to merge probabilities.
1510 * predict.c (probably_never_executed): Also mark as cold functions
1511 with global 0 profile and guessed local profile.
1512 * profile-count.c (profile_probability::combine_with_count): New
1514 * profile-count.h (profile_probability::operator*,
1515 profile_probability::operator*=, profile_probability::operator/,
1516 profile_probability::operator/=): Reduce precision to adjusted
1517 and set value to guessed on contradictory divisions.
1518 (profile_probability::combine_with_freq): Remove.
1519 (profile_probability::combine_wiht_count): Declare.
1520 (profile_count::force_nonzero):: Set to adjusted.
1521 (profile_count::probability_in):: Set quality to adjusted.
1522 * tree-ssa-tail-merge.c (replace_block_by): Use
1525 2018-01-23 Andrew Waterman <andrew@sifive.com>
1526 Jim Wilson <jimw@sifive.com>
1528 * config/riscv/riscv.c (riscv_stack_boundary): New.
1529 (riscv_option_override): Set riscv_stack_boundary. Handle
1530 riscv_preferred_stack_boundary_arg.
1531 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1532 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1533 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1534 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1535 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1536 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1538 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1541 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1542 of struct ix86_frame.
1543 (ix86_expand_epilogue): Likewise. Add a local variable for
1544 the reg_save_offset field in struct ix86_frame.
1546 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1548 PR tree-optimization/82604
1549 * tree-loop-distribution.c (enum partition_kind): New enum item
1550 PKIND_PARTIAL_MEMSET.
1551 (partition_builtin_p): Support above new enum item.
1552 (generate_code_for_partition): Ditto.
1553 (compute_access_range): Differentiate cases that equality can be
1554 proven at all loops, the innermost loops or no loops.
1555 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1556 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1557 (finalize_partitions, distribute_loop): Don't fuse partition of
1558 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1559 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1562 2018-01-23 Martin Liska <mliska@suse.cz>
1564 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1565 order to ignore the predictor.
1566 (PRED_POLYMORPHIC_CALL): Likewise.
1567 (PRED_RECURSIVE_CALL): Likewise.
1569 2018-01-23 Martin Liska <mliska@suse.cz>
1571 * tree-profile.c (tree_profiling): Print function header to
1572 aware reader which function we are working on.
1573 * value-prof.c (gimple_find_values_to_profile): Do not print
1574 not interesting value histograms.
1576 2018-01-23 Martin Liska <mliska@suse.cz>
1578 * profile-count.h (enum profile_quality): Add
1579 profile_uninitialized as the first value. Do not number values
1580 as they are zero based.
1581 (profile_count::verify): Update sanity check.
1582 (profile_probability::verify): Likewise.
1584 2018-01-23 Nathan Sidwell <nathan@acm.org>
1586 * doc/invoke.texi (ffor-scope): Deprecate.
1588 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1590 PR tree-optimization/83510
1591 * domwalk.c (set_all_edges_as_executable): New function.
1592 (dom_walker::dom_walker): Convert bool param
1593 "skip_unreachable_blocks" to enum reachability. Move setup of
1594 edge flags to set_all_edges_as_executable and only do it when
1595 reachability is REACHABLE_BLOCKS.
1596 * domwalk.h (enum dom_walker::reachability): New enum.
1597 (dom_walker::dom_walker): Convert bool param
1598 "skip_unreachable_blocks" to enum reachability.
1599 (set_all_edges_as_executable): New decl.
1600 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1601 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1603 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1604 but converting true to REACHABLE_BLOCKS.
1605 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1607 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1608 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1609 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1611 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1612 if check_all_array_refs will be called.
1614 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1616 * tree.c (selftest::test_location_wrappers): Add more test
1619 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1621 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1622 (selftest::test_bit_in_range): Likewise.
1624 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1627 * doc/sourcebuild.texi (vect_float): Say that the selector
1628 only describes the situation when -funsafe-math-optimizations is on.
1629 (vect_float_strict): Document.
1631 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1633 PR tree-optimization/83965
1634 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1635 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1636 instead of checking only for a reduction.
1637 (vect_recog_widen_sum_pattern): Likewise.
1639 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1641 * predict.c (probably_never_executed): Only use precise profile info.
1642 (compute_function_frequency): Skip after inlining hack since we now
1643 have quality checking.
1645 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1647 * profile-count.h (profile_probability::very_unlikely,
1648 profile_probability::unlikely, profile_probability::even): Set
1649 precision to guessed.
1651 2018-01-23 Richard Biener <rguenther@suse.de>
1653 PR tree-optimization/83963
1654 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1655 Properly terminate dominator walk when crossing the exit edge not
1656 when visiting its source block.
1658 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1661 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1662 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1664 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1666 PR tree-optimization/83957
1667 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1668 semicolon after for body surrounded by braces.
1670 PR tree-optimization/83081
1671 * profile-count.h (profile_probability::split): New method.
1672 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1673 Use profile_probability::split.
1674 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1675 when splitting a single conditional jump into 2.
1677 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1679 PR tree-optimization/69452
1680 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1683 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1685 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1686 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1687 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1689 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1691 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1692 * config/rl78/rl78.md: New define_expand "movdi"
1693 * config/rl78/rl78.c: New function definition rl78_split_movdi
1695 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1698 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1700 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1701 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1702 128-bit to produce an UNSPEC move to get the double word with the
1703 signbit and then a shift directly to do signbit.
1704 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1705 implementation with a new version that just does either a direct
1706 move or a regular move. Move memory interface to separate insns.
1707 Move insns so they are next to the expander.
1708 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1709 with signbit move. Split big and little endian case.
1710 (signbit<mode>2_dm_mem_le): Likewise.
1711 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1712 (signbit<mode>2_dm2): Likewise.
1714 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1716 * config/rl78/rl78.md: New define_expand "anddi3".
1718 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1720 * config/rl78/rl78.md: New define_expand "umindi3".
1722 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1724 * config/rl78/rl78.md: New define_expand "smindi3".
1726 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1728 * config/rl78/rl78.md: New define_expand "smaxdi3".
1730 2018-01-22 Carl Love <cel@us.ibm.com>
1732 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1733 LVX_V1TI): Add macro expansion.
1734 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1735 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1736 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1737 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1738 Change check to determine if the instruction is a byte reversing
1739 entry. Fix typo in comment.
1740 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1741 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1742 Add def_builtin calls for new builtins.
1743 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1744 Add define_insn expansion.
1746 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1748 * config/rl78/rl78.md: New define_expand "umaxdi3".
1750 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1752 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1753 for non-QImode registers
1755 2018-01-22 Richard Biener <rguenther@suse.de>
1757 PR tree-optimization/83963
1758 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1759 including the loop exit block.
1760 (scop_detection::merge_sese): Likewise.
1761 (scop_detection::add_scop): Do it here instead.
1763 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1765 * doc/sourcebuild.texi (arm_softfloat): Document.
1767 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1770 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1771 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1772 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1774 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1775 David Edelsohn <dje.gcc@gmail.com>
1778 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1779 Change "crset eq" to "crset 2".
1780 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1781 (*call_indirect_aix<mode>_nospec): Likewise.
1782 (*call_value_indirect_aix<mode>_nospec): Likewise.
1783 (*call_indirect_elfv2<mode>_nospec): Likewise.
1784 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1785 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1786 change assembly output from . to $.
1787 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1788 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1789 (*tablejump<mode>_internal1_nospec): Likewise.
1791 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1794 * config/sh/sh_optimize_sett_clrt.cc:
1795 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1797 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1799 PR tree-optimization/83940
1800 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1801 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1802 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1803 use it to pass back the definition type.
1804 (vect_check_store_rhs): Likewise rhs_dt_out.
1805 (vect_build_gather_load_calls): Add a mask_dt argument and use
1806 it instead of a call to vect_is_simple_use.
1807 (vectorizable_store): Update calls to vect_check_load_store_mask
1808 and vect_check_store_rhs. Use the dt returned by the latter instead
1809 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1810 instead of calls to vect_is_simple_use. Pass the scalar rather
1811 than the vector operand to vect_is_simple_use when handling
1812 second and subsequent copies of an rhs value.
1813 (vectorizable_load): Update calls to vect_check_load_store_mask
1814 and vect_build_gather_load_calls. Use the cached mask_dt and
1815 gs_info.offset_dt instead of calls to vect_is_simple_use.
1817 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1820 * tree-emutls.c: Include gimplify.h.
1821 (lower_emutls_2): New function.
1822 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1823 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1824 it before further processing.
1827 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1828 UINTVAL (trueop1) instead of INTVAL (op1).
1830 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1834 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1835 INCOMING_FRAME_SP_OFFSET if not defined.
1836 (scan_trace): Add ENTRY argument. If true and
1837 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1838 emit a note to adjust the CFA offset.
1839 (create_cfi_notes): Adjust scan_trace callers.
1840 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1841 INCOMING_FRAME_SP_OFFSET in the CIE.
1842 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1843 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1845 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1846 * doc/tm.texi: Regenerated.
1848 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1850 PR rtl-optimization/83147
1851 * lra-constraints.c (remove_inheritance_pseudos): Use
1852 lra_substitute_pseudo_within_insn.
1854 2018-01-19 Tom de Vries <tom@codesourcery.com>
1855 Cesar Philippidis <cesar@codesourcery.com>
1858 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1860 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1863 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1864 spaces for function labels.
1866 2018-01-19 Martin Liska <mliska@suse.cz>
1868 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1869 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1870 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1871 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1872 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1873 (PRED_CONST_RETURN): Change from 69 to 65.
1874 (PRED_NULL_RETURN): Change from 91 to 71.
1875 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1876 (PRED_LOOP_GUARD): Change from 66 to 73.
1878 2018-01-19 Martin Liska <mliska@suse.cz>
1880 * predict.c (predict_insn_def): Add new assert.
1881 (struct branch_predictor): Change type to signed integer.
1882 (test_prediction_value_range): Amend test to cover
1884 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1885 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1886 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1887 (PRED_LOOP_IV_COMPARE): Likewise.
1888 * predict.h (PROB_UNINITIALIZED): Define new constant.
1890 2018-01-19 Martin Liska <mliska@suse.cz>
1892 * predict.c (dump_prediction): Add new format for
1893 analyze_brprob.py script which is enabled with -details
1895 * profile-count.h (precise_p): New function.
1897 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1899 PR tree-optimization/83922
1900 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1901 there are no statements that need masking.
1902 (vect_active_double_reduction_p): New function.
1903 (vect_analyze_loop_operations): Use it when handling phis that
1904 are not in the loop header.
1906 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1908 PR tree-optimization/83914
1909 * tree-vect-loop.c (vectorizable_induction): Don't convert
1910 init_expr or apply the peeling adjustment for inductions
1911 that are nested within the vectorized loop.
1913 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1915 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1918 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1922 * function.h (gimplify_parameters): Add gimple_seq * argument.
1923 * function.c: Include gimple.h and options.h.
1924 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1925 for the added local temporaries if needed.
1926 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1927 if there are any parameter cleanups, wrap whole body into a
1928 try/finally with the cleanups.
1930 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1933 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1934 Use GET_MODE_CLASS for scalar floating point.
1936 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1940 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1941 Fix call of call_cgraph_insertion_hooks.
1943 2018-01-18 Martin Sebor <msebor@redhat.com>
1945 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1947 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1950 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1953 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1956 * common.opt: (-ffile-prefix-map): New option.
1957 * opts.c (common_handle_option): Defer it.
1958 * opts-global.c (handle_common_deferred_options): Handle it.
1959 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1960 * file-prefix-map.h: New file.
1961 (remap_debug_filename, add_debug_prefix_map): ...here.
1962 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1963 * final.c (debug_prefix_map, add_debug_prefix_map
1964 remap_debug_filename): Move to...
1965 * file-prefix-map.c: New file.
1966 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1967 generalize, get rid of alloca(), use strrchr() instead of strchr().
1968 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1969 Implement in terms of add_prefix_map().
1970 (remap_macro_filename, remap_debug_filename): Implement in term of
1972 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1973 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1974 * dbxout.c: Include file-prefix-map.h.
1975 * varasm.c: Likewise.
1976 * vmsdbgout.c: Likewise.
1977 * xcoffout.c: Likewise.
1978 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1979 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1980 * doc/invoke.texi (-ffile-prefix-map): Document.
1981 (-fdebug-prefix-map): Update description.
1983 2018-01-18 Martin Liska <mliska@suse.cz>
1985 * config/i386/i386.c (indirect_thunk_name): Document that also
1987 (output_indirect_thunk): Document why both instructions
1988 (pause and lfence) are generated.
1990 2018-01-18 Richard Biener <rguenther@suse.de>
1992 PR tree-optimization/83887
1993 * graphite-scop-detection.c
1994 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1995 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1996 (scop_detection::merge_sese): Re-implement with a flood-fill
1997 algorithm that properly finds a SESE region if it exists.
1999 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2002 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2003 pointer_diff optimizations use view_convert instead of convert.
2005 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2007 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2008 Generate different code for -mno-speculate-indirect-jumps.
2009 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2010 (*call_indirect_aix<mode>): Disable for
2011 -mno-speculate-indirect-jumps.
2012 (*call_indirect_aix<mode>_nospec): New define_insn.
2013 (*call_value_indirect_aix<mode>): Disable for
2014 -mno-speculate-indirect-jumps.
2015 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2016 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2017 -mno-speculate-indirect-jumps.
2018 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2020 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2022 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2023 long double type, set the flags for noting the default long double
2024 type, even if we don't pass or return a long double type.
2026 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2029 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2032 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2034 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2035 support for merge[hl].
2036 (fold_mergehl_helper): New helper function.
2037 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2038 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2039 (altivec_vmrglw_direct): Add xxmrglw insn.
2041 2018-01-17 Andrew Waterman <andrew@sifive.com>
2043 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2044 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2046 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2049 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2050 call the lto_location_cache before reading the
2051 DECL_SOURCE_LOCATION of the types.
2053 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2054 Richard Sandiford <richard.sandiford@linaro.org>
2056 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2057 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2058 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2059 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2060 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2062 * config/aarch64/constraints.md (aarch64_movti_operand):
2064 * config/aarch64/predicates.md (Uti): Add new constraint.
2066 2018-01-17 Carl Love <cel@us.ibm.com>
2067 * config/rs6000/vsx.md (define_expand xl_len_r,
2068 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2069 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2071 (define_expand, define_insn): Move the shift left from the
2072 define_insn to the define_expand for lxvl and stxvl instructions.
2073 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2074 and XL_LEN_R definitions to PURE.
2076 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2078 * config/i386/i386.c (indirect_thunk_name): Declare regno
2079 as unsigned int. Compare regno with INVALID_REGNUM.
2080 (output_indirect_thunk): Ditto.
2081 (output_indirect_thunk_function): Ditto.
2082 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2083 in the call to output_indirect_thunk_function.
2085 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2088 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2089 rather than the size of inner_type to determine the stack slot size
2090 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2092 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2095 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2098 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2100 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2101 endian Linux systems to optionally enable multilibs for selecting
2102 the long double type if the user configured an explicit type.
2103 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2104 have no long double multilibs if not defined.
2105 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2106 warn if the user used -mabi={ieee,ibm}longdouble and we built
2107 multilibs for long double.
2108 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2109 appropriate multilib option.
2110 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2112 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2113 for building long double multilibs.
2114 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2116 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2118 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2121 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2123 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2126 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2129 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2132 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2134 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2135 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2137 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2139 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2140 different rtl trees depending on TARGET_64BIT.
2141 (rs6000_gen_lvx): Likewise.
2143 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2145 * config/visium/visium.md (nop): Tweak comment.
2146 (hazard_nop): Likewise.
2148 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2150 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2151 -mspeculate-indirect-jumps.
2152 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2153 for -mno-speculate-indirect-jumps.
2154 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2155 (*call_value_indirect_elfv2<mode>): Disable for
2156 -mno-speculate-indirect-jumps.
2157 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2158 (indirect_jump): Emit different RTL for
2159 -mno-speculate-indirect-jumps.
2160 (*indirect_jump<mode>): Disable for
2161 -mno-speculate-indirect-jumps.
2162 (*indirect_jump<mode>_nospec): New define_insn.
2163 (tablejump): Emit different RTL for
2164 -mno-speculate-indirect-jumps.
2165 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2166 (tablejumpsi_nospec): New define_expand.
2167 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2168 (tablejumpdi_nospec): New define_expand.
2169 (*tablejump<mode>_internal1): Disable for
2170 -mno-speculate-indirect-jumps.
2171 (*tablejump<mode>_internal1_nospec): New define_insn.
2172 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2175 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2177 * caller-save.c (insert_save): Drop unnecessary parameter. All
2180 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2181 Richard Biener <rguenth@suse.de>
2184 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2185 return early, inline manually is_gimple_sizepos. Make sure if we
2186 call gimplify_expr we don't end up with a gimple constant.
2187 * tree.c (variably_modified_type_p): Don't return true for
2188 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2189 * gimplify.h (is_gimple_sizepos): Remove.
2191 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2193 PR tree-optimization/83857
2194 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2195 vectorizable_live_operation for pure SLP statements.
2196 (vectorizable_live_operation): Handle PHIs.
2198 2018-01-16 Richard Biener <rguenther@suse.de>
2200 PR tree-optimization/83867
2201 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2202 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2204 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2207 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2208 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2209 If off is not INTEGER_CST, issue a may not be aligned warning
2210 rather than isn't aligned. Use isn%'t rather than isn't.
2211 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2213 <case MULT_EXPR>: Improve the case when bottom and one of the
2214 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2215 operand, in that case check if the other operand is multiple of
2216 bottom divided by the INTEGER_CST operand.
2218 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2221 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2222 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2223 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2224 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2225 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2226 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2227 (pa_function_arg_size): New function.
2229 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2231 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2232 in a separate statement.
2234 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2236 PR tree-optimization/83847
2237 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2238 group gathers and scatters.
2240 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2242 PR rtl-optimization/86620
2243 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2245 PR rtl-optimization/83213
2246 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2247 to last if both are JUMP_INSNs.
2249 PR tree-optimization/83843
2250 * gimple-ssa-store-merging.c
2251 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2252 store_immediate_info for bswap/nop orig_stores.
2254 2018-01-15 Andrew Waterman <andrew@sifive.com>
2256 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2258 <UDIV>: Increase cost if !TARGET_DIV.
2260 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2262 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2263 (define_attr "cr_logical_3op"): New.
2264 (cceq_ior_compare): Adjust.
2265 (cceq_ior_compare_complement): Adjust.
2266 (*cceq_rev_compare): Adjust.
2267 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2268 (is_cracked_insn): Adjust.
2269 (insn_must_be_first_in_group): Adjust.
2270 * config/rs6000/40x.md: Adjust.
2271 * config/rs6000/440.md: Adjust.
2272 * config/rs6000/476.md: Adjust.
2273 * config/rs6000/601.md: Adjust.
2274 * config/rs6000/603.md: Adjust.
2275 * config/rs6000/6xx.md: Adjust.
2276 * config/rs6000/7450.md: Adjust.
2277 * config/rs6000/7xx.md: Adjust.
2278 * config/rs6000/8540.md: Adjust.
2279 * config/rs6000/cell.md: Adjust.
2280 * config/rs6000/e300c2c3.md: Adjust.
2281 * config/rs6000/e500mc.md: Adjust.
2282 * config/rs6000/e500mc64.md: Adjust.
2283 * config/rs6000/e5500.md: Adjust.
2284 * config/rs6000/e6500.md: Adjust.
2285 * config/rs6000/mpc.md: Adjust.
2286 * config/rs6000/power4.md: Adjust.
2287 * config/rs6000/power5.md: Adjust.
2288 * config/rs6000/power6.md: Adjust.
2289 * config/rs6000/power7.md: Adjust.
2290 * config/rs6000/power8.md: Adjust.
2291 * config/rs6000/power9.md: Adjust.
2292 * config/rs6000/rs64.md: Adjust.
2293 * config/rs6000/titan.md: Adjust.
2295 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2297 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2298 ix86_indirect_branch_register logic.
2300 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2302 * config/i386/constraints.md (Bs): Update
2303 ix86_indirect_branch_register check. Don't check
2304 ix86_indirect_branch_register with GOT_memory_operand.
2306 * config/i386/predicates.md (GOT_memory_operand): Don't check
2307 ix86_indirect_branch_register here.
2308 (GOT32_symbol_operand): Likewise.
2310 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2312 * config/i386/predicates.md (constant_call_address_operand):
2313 Rewrite ix86_indirect_branch_register logic.
2314 (sibcall_insn_operand): Likewise.
2316 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2318 * config/i386/constraints.md (Bs): Replace
2319 ix86_indirect_branch_thunk_register with
2320 ix86_indirect_branch_register.
2322 * config/i386/i386.md (indirect_jump): Likewise.
2323 (tablejump): Likewise.
2324 (*sibcall_memory): Likewise.
2325 (*sibcall_value_memory): Likewise.
2326 Peepholes of indirect call and jump via memory: Likewise.
2327 * config/i386/i386.opt: Likewise.
2328 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2329 (GOT_memory_operand): Likewise.
2330 (call_insn_operand): Likewise.
2331 (sibcall_insn_operand): Likewise.
2332 (GOT32_symbol_operand): Likewise.
2334 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2337 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2338 type rather than type addr's type points to.
2339 (expand_omp_atomic_mutex): Likewise.
2340 (expand_omp_atomic): Likewise.
2342 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2345 * config/i386/i386.c (output_indirect_thunk_function): Use
2346 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2347 for __x86_return_thunk.
2349 2018-01-15 Richard Biener <rguenther@suse.de>
2352 * expmed.c (extract_bit_field_1): Fix typo.
2354 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2357 * config/arm/iterators.md (VF): New mode iterator.
2358 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2359 Remove integer-related logic from pattern.
2360 (neon_vabd<mode>_3): Likewise.
2362 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2365 * common.opt (fstrict-overflow): No longer an alias.
2366 (fwrapv-pointer): New option.
2367 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2368 also for pointer types based on flag_wrapv_pointer.
2369 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2370 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2371 opts->x_flag_wrapv got set.
2372 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2373 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2374 POINTER_TYPE_OVERFLOW_UNDEFINED.
2375 * match.pd: Likewise in address comparison pattern.
2376 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2378 2018-01-15 Richard Biener <rguenther@suse.de>
2381 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2382 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2383 Reset type names to their identifier if their TYPE_DECL doesn't
2384 have linkage (and thus is used for ODR and devirt).
2385 (save_debug_info_for_decl): Remove.
2386 (save_debug_info_for_type): Likewise.
2387 (add_tree_to_fld_list): Adjust.
2388 * tree-pretty-print.c (dump_generic_node): Make dumping of
2389 type names more robust.
2391 2018-01-15 Richard Biener <rguenther@suse.de>
2393 * BASE-VER: Bump to 8.0.1.
2395 2018-01-14 Martin Sebor <msebor@redhat.com>
2398 * builtins.c (check_access): Avoid warning when the no-warning bit
2401 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2403 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2404 * ira-color (allocno_hard_regs_compare): Likewise.
2406 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2409 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2410 Use .pushsection/.popsection.
2412 2018-01-14 Martin Sebor <msebor@redhat.com>
2415 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2417 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2419 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2420 entry from extra_headers.
2421 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2422 extra_headers, make the list bitwise identical to the i?86-*-* one.
2424 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2426 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2427 -mcmodel=large with -mindirect-branch=thunk,
2428 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2429 -mfunction-return=thunk-extern.
2430 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2431 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2432 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2434 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2436 * config/i386/i386.c (print_reg): Print the name of the full
2437 integer register without '%'.
2438 (ix86_print_operand): Handle 'V'.
2439 * doc/extend.texi: Document 'V' modifier.
2441 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2443 * config/i386/constraints.md (Bs): Disallow memory operand for
2444 -mindirect-branch-register.
2446 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2447 (GOT_memory_operand): Likewise.
2448 (call_insn_operand): Likewise.
2449 (sibcall_insn_operand): Likewise.
2450 (GOT32_symbol_operand): Likewise.
2451 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2452 for -mindirect-branch-register.
2453 (tablejump): Likewise.
2454 (*sibcall_memory): Likewise.
2455 (*sibcall_value_memory): Likewise.
2456 Disallow peepholes of indirect call and jump via memory for
2457 -mindirect-branch-register.
2458 (*call_pop): Replace m with Bw.
2459 (*call_value_pop): Likewise.
2460 (*sibcall_pop_memory): Replace m with Bs.
2461 * config/i386/i386.opt (mindirect-branch-register): New option.
2462 * doc/invoke.texi: Document -mindirect-branch-register option.
2464 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2466 * config/i386/i386-protos.h (ix86_output_function_return): New.
2467 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2468 set function_return_type.
2469 (indirect_thunk_name): Add ret_p to indicate thunk for function
2471 (output_indirect_thunk_function): Pass false to
2472 indirect_thunk_name.
2473 (ix86_output_indirect_branch_via_reg): Likewise.
2474 (ix86_output_indirect_branch_via_push): Likewise.
2475 (output_indirect_thunk_function): Create alias for function
2476 return thunk if regno < 0.
2477 (ix86_output_function_return): New function.
2478 (ix86_handle_fndecl_attribute): Handle function_return.
2479 (ix86_attribute_table): Add function_return.
2480 * config/i386/i386.h (machine_function): Add
2481 function_return_type.
2482 * config/i386/i386.md (simple_return_internal): Use
2483 ix86_output_function_return.
2484 (simple_return_internal_long): Likewise.
2485 * config/i386/i386.opt (mfunction-return=): New option.
2486 (indirect_branch): Mention -mfunction-return=.
2487 * doc/extend.texi: Document function_return function attribute.
2488 * doc/invoke.texi: Document -mfunction-return= option.
2490 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2492 * config/i386/i386-opts.h (indirect_branch): New.
2493 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2494 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2495 with local indirect jump when converting indirect call and jump.
2496 (ix86_set_indirect_branch_type): New.
2497 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2498 (indirectlabelno): New.
2499 (indirect_thunk_needed): Likewise.
2500 (indirect_thunk_bnd_needed): Likewise.
2501 (indirect_thunks_used): Likewise.
2502 (indirect_thunks_bnd_used): Likewise.
2503 (INDIRECT_LABEL): Likewise.
2504 (indirect_thunk_name): Likewise.
2505 (output_indirect_thunk): Likewise.
2506 (output_indirect_thunk_function): Likewise.
2507 (ix86_output_indirect_branch_via_reg): Likewise.
2508 (ix86_output_indirect_branch_via_push): Likewise.
2509 (ix86_output_indirect_branch): Likewise.
2510 (ix86_output_indirect_jmp): Likewise.
2511 (ix86_code_end): Call output_indirect_thunk_function if needed.
2512 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2514 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2515 (ix86_attribute_table): Add indirect_branch.
2516 * config/i386/i386.h (machine_function): Add indirect_branch_type
2517 and has_local_indirect_jump.
2518 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2520 (tablejump): Likewise.
2521 (*indirect_jump): Use ix86_output_indirect_jmp.
2522 (*tablejump_1): Likewise.
2523 (simple_return_indirect_internal): Likewise.
2524 * config/i386/i386.opt (mindirect-branch=): New option.
2525 (indirect_branch): New.
2528 (thunk-inline): Likewise.
2529 (thunk-extern): Likewise.
2530 * doc/extend.texi: Document indirect_branch function attribute.
2531 * doc/invoke.texi: Document -mindirect-branch= option.
2533 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2536 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2538 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2540 * ipa-inline.c (want_inline_small_function_p): Return false if
2541 inlining has already failed with CIF_FINAL_ERROR.
2542 (update_caller_keys): Call want_inline_small_function_p before
2544 (update_callee_keys): Likewise.
2546 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2548 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2550 (rs6000_quadword_masked_address_p): Likewise.
2551 (quad_aligned_load_p): Likewise.
2552 (quad_aligned_store_p): Likewise.
2553 (const_load_sequence_p): Add comment to describe the outer-most loop.
2554 (mimic_memory_attributes_and_flags): New function.
2555 (rs6000_gen_stvx): Likewise.
2556 (replace_swapped_aligned_store): Likewise.
2557 (rs6000_gen_lvx): Likewise.
2558 (replace_swapped_aligned_load): Likewise.
2559 (replace_swapped_load_constant): Capitalize argument name in
2560 comment describing this function.
2561 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2562 and stores that access quad-word aligned addresses and replace
2563 with stvx or lvx instructions when appropriate.
2564 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2565 New function prototype.
2566 (rs6000_quadword_masked_address_p): Likewise.
2567 (rs6000_gen_lvx): Likewise.
2568 (rs6000_gen_stvx): Likewise.
2569 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2570 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2571 when memory address is aligned.
2572 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2573 this split to select lvx instruction when memory address is aligned.
2574 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2575 instruction when memory address is aligned.
2576 (*vsx_le_perm_load_v16qi): Likewise.
2577 (four unnamed splitters): Modify to select the stvx instruction
2578 when memory is aligned.
2580 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2582 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2583 which appears in the queue multiple times.
2585 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2586 Alan Hayward <alan.hayward@arm.com>
2587 David Sherwood <david.sherwood@arm.com>
2589 * tree-vectorizer.h (vec_lower_bound): New structure.
2590 (_loop_vec_info): Add check_nonzero and lower_bounds.
2591 (LOOP_VINFO_CHECK_NONZERO): New macro.
2592 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2593 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2594 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2595 fields. Make seg_len the distance travelled, not including the
2597 (dr_direction_indicator): Declare.
2598 (dr_zero_step_indicator): Likewise.
2599 (dr_known_forward_stride_p): Likewise.
2600 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2602 (runtime_alias_check_p): Allow runtime alias checks with
2604 (operator ==): Compare access_size and align.
2605 (prune_runtime_alias_test_list): Rework for new distinction between
2606 the access_size and seg_len.
2607 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2609 (get_segment_min_max): New function.
2610 (create_intersect_range_checks): Use it.
2611 (dr_step_indicator): New function.
2612 (dr_direction_indicator): Likewise.
2613 (dr_zero_step_indicator): Likewise.
2614 (dr_known_forward_stride_p): Likewise.
2615 * tree-loop-distribution.c (data_ref_segment_size): Return
2616 DR_STEP * (niters - 1).
2617 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2619 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2620 (vect_preserves_scalar_order_p): New function, split out from...
2621 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2622 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2623 (vect_vfa_access_size): New function.
2624 (vect_vfa_align): Likewise.
2625 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2626 (dump_lower_bound): New function.
2627 (vect_check_lower_bound): Likewise.
2628 (vect_small_gap_p): Likewise.
2629 (vectorizable_with_step_bound_p): Likewise.
2630 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2631 depencies if the vectorization factor is 1. Convert the checks
2632 for nonzero steps into checks on the bounds of DR_STEP. Try using
2633 a bunds check for variable steps if the minimum required step is
2634 relatively small. Update calls to the dr_with_seg_len
2635 constructor and to vect_compile_time_alias.
2636 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2638 (vect_loop_versioning): Call it.
2639 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2641 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2643 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2644 Alan Hayward <alan.hayward@arm.com>
2645 David Sherwood <david.sherwood@arm.com>
2647 * doc/sourcebuild.texi (vect_scatter_store): Document.
2648 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2650 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2652 * genopinit.c (main): Add supports_vec_scatter_store and
2653 supports_vec_scatter_store_cached to target_optabs.
2654 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2655 IFN_MASK_SCATTER_STORE.
2656 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2658 * internal-fn.h (internal_store_fn_p): Declare.
2659 (internal_fn_stored_value_index): Likewise.
2660 * internal-fn.c (scatter_store_direct): New macro.
2661 (expand_scatter_store_optab_fn): New function.
2662 (direct_scatter_store_optab_supported_p): New macro.
2663 (internal_store_fn_p): New function.
2664 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2665 IFN_MASK_SCATTER_STORE.
2666 (internal_fn_mask_index): Likewise.
2667 (internal_fn_stored_value_index): New function.
2668 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2670 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2671 * optabs-query.c (supports_vec_scatter_store_p): New function.
2672 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2673 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2674 true for scatter stores.
2675 (vect_gather_scatter_fn_p): Handle scatter stores too.
2676 (vect_check_gather_scatter): Consider using scatter stores if
2677 supports_vec_scatter_store_p.
2678 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2680 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2681 internal_fn_stored_value_index.
2682 (check_load_store_masking): Handle scatter stores too.
2683 (vect_get_store_rhs): Make public.
2684 (vectorizable_call): Use internal_store_fn_p.
2685 (vectorizable_store): Handle scatter store internal functions.
2686 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2687 when deciding whether the end of the group has been reached.
2688 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2689 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2690 (mask_scatter_store<mode>): New insns.
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 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2697 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2698 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2700 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2701 Use vect_truncate_gather_scatter_offset if we can't treat the
2702 operation as a normal gather load or scatter store.
2703 (get_group_load_store_type): Take the gather_scatter_info
2704 as argument. Try using a gather load or scatter store for
2705 single-element groups.
2706 (get_load_store_type): Update calls to get_group_load_store_type
2707 and vect_use_strided_gather_scatters_p.
2709 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2710 Alan Hayward <alan.hayward@arm.com>
2711 David Sherwood <david.sherwood@arm.com>
2713 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2714 optional tree argument.
2715 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2717 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2718 but continue to use the current value as a fallback.
2719 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2720 to compare the updates.
2721 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2722 (get_load_store_type): Use it when handling a strided access.
2723 (vect_get_strided_load_store_ops): New function.
2724 (vect_get_data_ptr_increment): Likewise.
2725 (vectorizable_load): Handle strided gather loads. Always pass
2726 a step to vect_create_data_ref_ptr and bump_vector_ptr.
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 * doc/md.texi (gather_load@var{m}): Document.
2733 (mask_gather_load@var{m}): Likewise.
2734 * genopinit.c (main): Add supports_vec_gather_load and
2735 supports_vec_gather_load_cached to target_optabs.
2736 * optabs-tree.c (init_tree_optimization_optabs): Use
2737 ggc_cleared_alloc to allocate target_optabs.
2738 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2739 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2741 * internal-fn.h (internal_load_fn_p): Declare.
2742 (internal_gather_scatter_fn_p): Likewise.
2743 (internal_fn_mask_index): Likewise.
2744 (internal_gather_scatter_fn_supported_p): Likewise.
2745 * internal-fn.c (gather_load_direct): New macro.
2746 (expand_gather_load_optab_fn): New function.
2747 (direct_gather_load_optab_supported_p): New macro.
2748 (direct_internal_fn_optab): New function.
2749 (internal_load_fn_p): Likewise.
2750 (internal_gather_scatter_fn_p): Likewise.
2751 (internal_fn_mask_index): Likewise.
2752 (internal_gather_scatter_fn_supported_p): Likewise.
2753 * optabs-query.c (supports_at_least_one_mode_p): New function.
2754 (supports_vec_gather_load_p): Likewise.
2755 * optabs-query.h (supports_vec_gather_load_p): Declare.
2756 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2757 and memory_type field.
2758 (NUM_PATTERNS): Bump to 15.
2759 * tree-vect-data-refs.c: Include internal-fn.h.
2760 (vect_gather_scatter_fn_p): New function.
2761 (vect_describe_gather_scatter_call): Likewise.
2762 (vect_check_gather_scatter): Try using internal functions for
2763 gather loads. Recognize existing calls to a gather load function.
2764 (vect_analyze_data_refs): Consider using gather loads if
2765 supports_vec_gather_load_p.
2766 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2767 (vect_get_gather_scatter_offset_type): Likewise.
2768 (vect_convert_mask_for_vectype): Likewise.
2769 (vect_add_conversion_to_patterm): Likewise.
2770 (vect_try_gather_scatter_pattern): Likewise.
2771 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2772 (vect_vect_recog_func_ptrs): Add it.
2773 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2774 internal_fn_mask_index and internal_gather_scatter_fn_p.
2775 (check_load_store_masking): Take the gather_scatter_info as an
2776 argument and handle gather loads.
2777 (vect_get_gather_scatter_ops): New function.
2778 (vectorizable_call): Check internal_load_fn_p.
2779 (vectorizable_load): Likewise. Handle gather load internal
2781 (vectorizable_store): Update call to check_load_store_masking.
2782 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2783 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2784 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2785 (aarch64_gather_scale_operand_d): New predicates.
2786 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2787 (mask_gather_load<mode>): New insns.
2789 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2790 Alan Hayward <alan.hayward@arm.com>
2791 David Sherwood <david.sherwood@arm.com>
2793 * optabs.def (fold_left_plus_optab): New optab.
2794 * doc/md.texi (fold_left_plus_@var{m}): Document.
2795 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2796 * internal-fn.c (fold_left_direct): Define.
2797 (expand_fold_left_optab_fn): Likewise.
2798 (direct_fold_left_optab_supported_p): Likewise.
2799 * fold-const-call.c (fold_const_fold_left): New function.
2800 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2801 * tree-parloops.c (valid_reduction_p): New function.
2802 (gather_scalar_reductions): Use it.
2803 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2804 (vect_finish_replace_stmt): Declare.
2805 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2806 (needs_fold_left_reduction_p): New function, split out from...
2807 (vect_is_simple_reduction): ...here. Accept reductions that
2808 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2809 (vect_force_simple_reduction): Also store the reduction type in
2810 the assignment's STMT_VINFO_REDUC_TYPE.
2811 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2812 (merge_with_identity): New function.
2813 (vect_expand_fold_left): Likewise.
2814 (vectorize_fold_left_reduction): Likewise.
2815 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2816 scalar phi in place for it. Check for target support and reject
2817 cases that would reassociate the operation. Defer the transform
2818 phase to vectorize_fold_left_reduction.
2819 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2820 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2821 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2823 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2825 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2826 call to ifc_temp_var.
2828 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2829 Alan Hayward <alan.hayward@arm.com>
2830 David Sherwood <david.sherwood@arm.com>
2832 * target.def (legitimize_address_displacement): Take the original
2833 offset as a poly_int.
2834 * targhooks.h (default_legitimize_address_displacement): Update
2836 * targhooks.c (default_legitimize_address_displacement): Likewise.
2837 * doc/tm.texi: Regenerate.
2838 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2839 as an argument, moving assert of ad->disp == ad->disp_term to...
2840 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2841 Try calling targetm.legitimize_address_displacement before expanding
2842 the address rather than afterwards, and adjust for the new interface.
2843 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2844 Match the new hook interface. Handle SVE addresses.
2845 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2848 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2850 * Makefile.in (OBJS): Add early-remat.o.
2851 * target.def (select_early_remat_modes): New hook.
2852 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2853 * doc/tm.texi: Regenerate.
2854 * targhooks.h (default_select_early_remat_modes): Declare.
2855 * targhooks.c (default_select_early_remat_modes): New function.
2856 * timevar.def (TV_EARLY_REMAT): New timevar.
2857 * passes.def (pass_early_remat): New pass.
2858 * tree-pass.h (make_pass_early_remat): Declare.
2859 * early-remat.c: New file.
2860 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2862 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2864 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2865 Alan Hayward <alan.hayward@arm.com>
2866 David Sherwood <david.sherwood@arm.com>
2868 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2869 vfm1 with a bound_epilog parameter.
2870 (vect_do_peeling): Update calls accordingly, and move the prologue
2871 call earlier in the function. Treat the base bound_epilog as 0 for
2872 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2873 this base when peeling for gaps.
2874 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2875 with fully-masked loops.
2876 (vect_estimate_min_profitable_iters): Handle the single peeled
2877 iteration in that case.
2879 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2880 Alan Hayward <alan.hayward@arm.com>
2881 David Sherwood <david.sherwood@arm.com>
2883 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2884 single-element interleaving even if the size is not a power of 2.
2885 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2886 accesses for single-element interleaving if the group size is
2889 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2890 Alan Hayward <alan.hayward@arm.com>
2891 David Sherwood <david.sherwood@arm.com>
2893 * doc/md.texi (fold_extract_last_@var{m}): Document.
2894 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2895 * optabs.def (fold_extract_last_optab): New optab.
2896 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2897 * internal-fn.c (fold_extract_direct): New macro.
2898 (expand_fold_extract_optab_fn): Likewise.
2899 (direct_fold_extract_optab_supported_p): Likewise.
2900 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2901 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2902 EXTRACT_LAST_REDUCTION.
2903 (get_initial_def_for_reduction): Do not create an initial vector
2904 for EXTRACT_LAST_REDUCTION reductions.
2905 (vectorizable_reduction): Leave the scalar phi in place for
2906 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2907 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2908 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2909 transform phase to vectorizable_condition.
2910 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2912 (vect_finish_stmt_generation): ...here.
2913 (vect_finish_replace_stmt): New function.
2914 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2915 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2917 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2919 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2920 Alan Hayward <alan.hayward@arm.com>
2921 David Sherwood <david.sherwood@arm.com>
2923 * doc/md.texi (extract_last_@var{m}): Document.
2924 * optabs.def (extract_last_optab): New optab.
2925 * internal-fn.def (EXTRACT_LAST): New internal function.
2926 * internal-fn.c (cond_unary_direct): New macro.
2927 (expand_cond_unary_optab_fn): Likewise.
2928 (direct_cond_unary_optab_supported_p): Likewise.
2929 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2930 loops using EXTRACT_LAST.
2931 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2932 (extract_last_<mode>): ...this optab.
2933 (vec_extract<mode><Vel>): Update accordingly.
2935 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2936 Alan Hayward <alan.hayward@arm.com>
2937 David Sherwood <david.sherwood@arm.com>
2939 * target.def (empty_mask_is_expensive): New hook.
2940 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2941 * doc/tm.texi: Regenerate.
2942 * targhooks.h (default_empty_mask_is_expensive): Declare.
2943 * targhooks.c (default_empty_mask_is_expensive): New function.
2944 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2945 if the target says that empty masks are expensive.
2946 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2948 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2950 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2951 Alan Hayward <alan.hayward@arm.com>
2952 David Sherwood <david.sherwood@arm.com>
2954 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2955 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2956 (vect_use_loop_mask_for_alignment_p): New function.
2957 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2958 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2959 niters_skip argument. Make sure that the first niters_skip elements
2960 of the first iteration are inactive.
2961 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2962 Update call to vect_set_loop_masks_directly.
2963 (get_misalign_in_elems): New function, split out from...
2964 (vect_gen_prolog_loop_niters): ...here.
2965 (vect_update_init_of_dr): Take a code argument that specifies whether
2966 the adjustment should be added or subtracted.
2967 (vect_update_init_of_drs): Likewise.
2968 (vect_prepare_for_masked_peels): New function.
2969 (vect_do_peeling): Skip prologue peeling if we're using a mask
2970 instead. Update call to vect_update_inits_of_drs.
2971 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2973 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2974 alignment. Do not include the number of peeled iterations in
2975 the minimum threshold in that case.
2976 (vectorizable_induction): Adjust the start value down by
2977 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2978 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2979 Take the number of skipped iterations into account when calculating
2981 * tree-vect-stmts.c (vect_gen_while_not): New function.
2983 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2984 Alan Hayward <alan.hayward@arm.com>
2985 David Sherwood <david.sherwood@arm.com>
2987 * doc/sourcebuild.texi (vect_fully_masked): Document.
2988 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2990 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2992 (vect_analyze_loop_2): ...here. Don't check the vectorization
2993 factor against the number of loop iterations if the loop is
2996 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2997 Alan Hayward <alan.hayward@arm.com>
2998 David Sherwood <david.sherwood@arm.com>
3000 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3001 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3002 (dump_groups): Update accordingly.
3003 (iv_use::mem_type): New member variable.
3004 (address_p): New function.
3005 (record_use): Add a mem_type argument and initialize the new
3007 (record_group_use): Add a mem_type argument. Use address_p.
3008 Remove obsolete null checks of base_object. Update call to record_use.
3009 (find_interesting_uses_op): Update call to record_group_use.
3010 (find_interesting_uses_cond): Likewise.
3011 (find_interesting_uses_address): Likewise.
3012 (get_mem_type_for_internal_fn): New function.
3013 (find_address_like_use): Likewise.
3014 (find_interesting_uses_stmt): Try find_address_like_use before
3015 calling find_interesting_uses_op.
3016 (addr_offset_valid_p): Use the iv mem_type field as the type
3017 of the addressed memory.
3018 (add_autoinc_candidates): Likewise.
3019 (get_address_cost): Likewise.
3020 (split_small_address_groups_p): Use address_p.
3021 (split_address_groups): Likewise.
3022 (add_iv_candidate_for_use): Likewise.
3023 (autoinc_possible_for_pair): Likewise.
3024 (rewrite_groups): Likewise.
3025 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3026 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3027 (get_alias_ptr_type_for_ptr_address): New function.
3028 (rewrite_use_address): Rewrite address uses in calls that were
3029 identified by find_address_like_use.
3031 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3032 Alan Hayward <alan.hayward@arm.com>
3033 David Sherwood <david.sherwood@arm.com>
3035 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3037 * gimple-expr.h (is_gimple_addressable: Likewise.
3038 * gimple-expr.c (is_gimple_address): Likewise.
3039 * internal-fn.c (expand_call_mem_ref): New function.
3040 (expand_mask_load_optab_fn): Use it.
3041 (expand_mask_store_optab_fn): Likewise.
3043 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3044 Alan Hayward <alan.hayward@arm.com>
3045 David Sherwood <david.sherwood@arm.com>
3047 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3048 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3049 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3050 (cond_umax@var{mode}): Document.
3051 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3052 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3053 (cond_umin_optab, cond_umax_optab): New optabs.
3054 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3055 (COND_IOR, COND_XOR): New internal functions.
3056 * internal-fn.h (get_conditional_internal_fn): Declare.
3057 * internal-fn.c (cond_binary_direct): New macro.
3058 (expand_cond_binary_optab_fn): Likewise.
3059 (direct_cond_binary_optab_supported_p): Likewise.
3060 (get_conditional_internal_fn): New function.
3061 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3062 Cope with reduction statements that are vectorized as calls rather
3064 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3065 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3066 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3067 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3068 (UNSPEC_COND_EOR): New unspecs.
3069 (optab): Add mappings for them.
3070 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3071 (sve_int_op, sve_fp_op): New int attributes.
3073 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3074 Alan Hayward <alan.hayward@arm.com>
3075 David Sherwood <david.sherwood@arm.com>
3077 * optabs.def (while_ult_optab): New optab.
3078 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3079 * internal-fn.def (WHILE_ULT): New internal function.
3080 * internal-fn.h (direct_internal_fn_supported_p): New override
3081 that takes two types as argument.
3082 * internal-fn.c (while_direct): New macro.
3083 (expand_while_optab_fn): New function.
3084 (convert_optab_supported_p): Likewise.
3085 (direct_while_optab_supported_p): New macro.
3086 * wide-int.h (wi::udiv_ceil): New function.
3087 * tree-vectorizer.h (rgroup_masks): New structure.
3088 (vec_loop_masks): New typedef.
3089 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3091 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3092 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3093 (vect_max_vf): New function.
3094 (slpeel_make_loop_iterate_ntimes): Delete.
3095 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3096 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3097 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3098 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3099 internal-fn.h, stor-layout.h and optabs-query.h.
3100 (vect_set_loop_mask): New function.
3101 (add_preheader_seq): Likewise.
3102 (add_header_seq): Likewise.
3103 (interleave_supported_p): Likewise.
3104 (vect_maybe_permute_loop_masks): Likewise.
3105 (vect_set_loop_masks_directly): Likewise.
3106 (vect_set_loop_condition_masked): Likewise.
3107 (vect_set_loop_condition_unmasked): New function, split out from
3108 slpeel_make_loop_iterate_ntimes.
3109 (slpeel_make_loop_iterate_ntimes): Rename to..
3110 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3111 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3112 (vect_do_peeling): Update call accordingly.
3113 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3115 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3116 mask_compare_type, can_fully_mask_p and fully_masked_p.
3117 (release_vec_loop_masks): New function.
3118 (_loop_vec_info): Use it to free the loop masks.
3119 (can_produce_all_loop_masks_p): New function.
3120 (vect_get_max_nscalars_per_iter): Likewise.
3121 (vect_verify_full_masking): Likewise.
3122 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3123 retries, and free the mask rgroups before retrying. Check loop-wide
3124 reasons for disallowing fully-masked loops. Make the final decision
3125 about whether use a fully-masked loop or not.
3126 (vect_estimate_min_profitable_iters): Do not assume that peeling
3127 for the number of iterations will be needed for fully-masked loops.
3128 (vectorizable_reduction): Disable fully-masked loops.
3129 (vectorizable_live_operation): Likewise.
3130 (vect_halve_mask_nunits): New function.
3131 (vect_double_mask_nunits): Likewise.
3132 (vect_record_loop_mask): Likewise.
3133 (vect_get_loop_mask): Likewise.
3134 (vect_transform_loop): Handle the case in which the final loop
3135 iteration might handle a partial vector. Call vect_set_loop_condition
3136 instead of slpeel_make_loop_iterate_ntimes.
3137 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3138 (check_load_store_masking): New function.
3139 (prepare_load_store_mask): Likewise.
3140 (vectorizable_store): Handle fully-masked loops.
3141 (vectorizable_load): Likewise.
3142 (supportable_widening_operation): Use vect_halve_mask_nunits for
3144 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3145 (vect_gen_while): New function.
3146 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3147 (aarch64_uqdec<mode>): New insn.
3149 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3150 Alan Hayward <alan.hayward@arm.com>
3151 David Sherwood <david.sherwood@arm.com>
3153 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3154 (reduc_xor_scal_optab): New optabs.
3155 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3156 (reduc_xor_scal_@var{m}): Document.
3157 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3158 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3160 * fold-const-call.c (fold_const_call): Handle them.
3161 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3162 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3163 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3164 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3165 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3166 (UNSPEC_XORV): New unspecs.
3167 (optab): Add entries for them.
3168 (BITWISEV): New int iterator.
3169 (bit_reduc_op): New int attributes.
3171 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3172 Alan Hayward <alan.hayward@arm.com>
3173 David Sherwood <david.sherwood@arm.com>
3175 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3176 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3177 * optabs.def (vec_shl_insert_optab): New optab.
3178 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3179 (duplicate_and_interleave): Likewise.
3180 * tree-vect-loop.c: Include internal-fn.h.
3181 (neutral_op_for_slp_reduction): New function, split out from
3182 get_initial_defs_for_reduction.
3183 (get_initial_def_for_reduction): Handle option 2 for variable-length
3184 vectors by loading the neutral value into a vector and then shifting
3185 the initial value into element 0.
3186 (get_initial_defs_for_reduction): Replace the code argument with
3187 the neutral value calculated by neutral_op_for_slp_reduction.
3188 Use gimple_build_vector for constant-length vectors.
3189 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3190 but the first group_size elements have a neutral value.
3191 Use duplicate_and_interleave otherwise.
3192 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3193 Update call to get_initial_defs_for_reduction. Handle SLP
3194 reductions for variable-length vectors by creating one vector
3195 result for each scalar result, with the elements associated
3196 with other scalar results stubbed out with the neutral value.
3197 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3198 Require IFN_VEC_SHL_INSERT for double reductions on
3199 variable-length vectors, or SLP reductions that have
3200 a neutral value. Require can_duplicate_and_interleave_p
3201 support for variable-length unchained SLP reductions if there
3202 is no neutral value, such as for MIN/MAX reductions. Also require
3203 the number of vector elements to be a multiple of the number of
3204 SLP statements when doing variable-length unchained SLP reductions.
3205 Update call to vect_create_epilog_for_reduction.
3206 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3207 and remove initial values.
3208 (duplicate_and_interleave): Make public.
3209 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3210 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3212 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3213 Alan Hayward <alan.hayward@arm.com>
3214 David Sherwood <david.sherwood@arm.com>
3216 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3217 (can_duplicate_and_interleave_p): New function.
3218 (vect_get_and_check_slp_defs): Take the vector of statements
3219 rather than just the current one. Remove excess parentheses.
3220 Restriction rejectinon of vect_constant_def and vect_external_def
3221 for variable-length vectors to boolean types, or types for which
3222 can_duplicate_and_interleave_p is false.
3223 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3224 (duplicate_and_interleave): New function.
3225 (vect_get_constant_vectors): Use gimple_build_vector for
3226 constant-length vectors and suitable variable-length constant
3227 vectors. Use duplicate_and_interleave for other variable-length
3228 vectors. Don't defer the update when inserting new statements.
3230 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3231 Alan Hayward <alan.hayward@arm.com>
3232 David Sherwood <david.sherwood@arm.com>
3234 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3235 min_profitable_iters doesn't go negative.
3237 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3238 Alan Hayward <alan.hayward@arm.com>
3239 David Sherwood <david.sherwood@arm.com>
3241 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3242 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3243 * optabs.def (vec_mask_load_lanes_optab): New optab.
3244 (vec_mask_store_lanes_optab): Likewise.
3245 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3246 (MASK_STORE_LANES): Likewise.
3247 * internal-fn.c (mask_load_lanes_direct): New macro.
3248 (mask_store_lanes_direct): Likewise.
3249 (expand_mask_load_optab_fn): Handle masked operations.
3250 (expand_mask_load_lanes_optab_fn): New macro.
3251 (expand_mask_store_optab_fn): Handle masked operations.
3252 (expand_mask_store_lanes_optab_fn): New macro.
3253 (direct_mask_load_lanes_optab_supported_p): Likewise.
3254 (direct_mask_store_lanes_optab_supported_p): Likewise.
3255 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3257 (vect_load_lanes_supported): Likewise.
3258 * tree-vect-data-refs.c (strip_conversion): New function.
3259 (can_group_stmts_p): Likewise.
3260 (vect_analyze_data_ref_accesses): Use it instead of checking
3261 for a pair of assignments.
3262 (vect_store_lanes_supported): Take a masked_p parameter.
3263 (vect_load_lanes_supported): Likewise.
3264 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3265 vect_store_lanes_supported and vect_load_lanes_supported.
3266 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3267 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3268 parameter. Don't allow gaps for masked accesses.
3269 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3270 and vect_load_lanes_supported.
3271 (get_load_store_type): Take a masked_p parameter and update
3272 call to get_group_load_store_type.
3273 (vectorizable_store): Update call to get_load_store_type.
3274 Handle IFN_MASK_STORE_LANES.
3275 (vectorizable_load): Update call to get_load_store_type.
3276 Handle IFN_MASK_LOAD_LANES.
3278 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3279 Alan Hayward <alan.hayward@arm.com>
3280 David Sherwood <david.sherwood@arm.com>
3282 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3284 * config/aarch64/aarch64-protos.h
3285 (aarch64_sve_struct_memory_operand_p): Declare.
3286 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3287 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3288 (VPRED, vpred): Handle SVE structure modes.
3289 * config/aarch64/constraints.md (Utx): New constraint.
3290 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3291 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3292 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3293 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3294 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3295 structure modes. Split into pieces after RA.
3296 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3297 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3299 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3300 SVE structure modes.
3301 (aarch64_classify_address): Likewise.
3302 (sizetochar): Move earlier in file.
3303 (aarch64_print_operand): Handle SVE register lists.
3304 (aarch64_array_mode): New function.
3305 (aarch64_sve_struct_memory_operand_p): Likewise.
3306 (TARGET_ARRAY_MODE): Redefine.
3308 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3309 Alan Hayward <alan.hayward@arm.com>
3310 David Sherwood <david.sherwood@arm.com>
3312 * target.def (array_mode): New target hook.
3313 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3314 * doc/tm.texi: Regenerate.
3315 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3316 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3317 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3319 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3322 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3323 Alan Hayward <alan.hayward@arm.com>
3324 David Sherwood <david.sherwood@arm.com>
3326 * fold-const.c (fold_binary_loc): Check the argument types
3327 rather than the result type when testing for a vector operation.
3329 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3331 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3332 * doc/tm.texi: Regenerate.
3334 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3335 Alan Hayward <alan.hayward@arm.com>
3336 David Sherwood <david.sherwood@arm.com>
3338 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3339 (sve): Document new AArch64 extension.
3340 * doc/md.texi (w): Extend the description of the AArch64
3341 constraint to include SVE vectors.
3342 (Upl, Upa): Document new AArch64 predicate constraints.
3343 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3345 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3346 (msve-vector-bits=): New option.
3347 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3348 SVE when these are disabled.
3349 (sve): New extension.
3350 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3351 modes. Adjust their number of units based on aarch64_sve_vg.
3352 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3353 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3354 aarch64_addr_query_type.
3355 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3356 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3357 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3358 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3359 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3360 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3361 (aarch64_simd_imm_zero_p): Delete.
3362 (aarch64_check_zero_based_sve_index_immediate): Declare.
3363 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3364 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3365 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3366 (aarch64_sve_float_mul_immediate_p): Likewise.
3367 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3369 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3370 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3371 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3372 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3373 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3374 (aarch64_regmode_natural_size): Likewise.
3375 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3376 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3378 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3379 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3380 for VG and the SVE predicate registers.
3381 (V_ALIASES): Add a "z"-prefixed alias.
3382 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3383 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3384 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3385 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3386 (REG_CLASS_NAMES): Add entries for them.
3387 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3388 and the predicate registers.
3389 (aarch64_sve_vg): Declare.
3390 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3391 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3392 (REGMODE_NATURAL_SIZE): Define.
3393 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3395 * config/aarch64/aarch64.c: Include cfgrtl.h.
3396 (simd_immediate_info): Add a constructor for series vectors,
3397 and an associated step field.
3398 (aarch64_sve_vg): New variable.
3399 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3400 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3401 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3402 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3403 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3404 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3405 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3406 (aarch64_get_mask_mode): New functions.
3407 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3408 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3409 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3410 predicate modes and predicate registers. Explicitly restrict
3411 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3412 to store a vector mode if it is recognized by
3413 aarch64_classify_vector_mode.
3414 (aarch64_regmode_natural_size): New function.
3415 (aarch64_hard_regno_caller_save_mode): Return the original mode
3417 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3418 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3419 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3420 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3422 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3423 does not overlap dest if the function is frame-related. Handle
3425 (aarch64_split_add_offset): New function.
3426 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3427 them aarch64_add_offset.
3428 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3429 and update call to aarch64_sub_sp.
3430 (aarch64_add_cfa_expression): New function.
3431 (aarch64_expand_prologue): Pass extra temporary registers to the
3432 functions above. Handle the case in which we need to emit new
3433 DW_CFA_expressions for registers that were originally saved
3434 relative to the stack pointer, but now have to be expressed
3435 relative to the frame pointer.
3436 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3438 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3439 IP0 and IP1 values for SVE frames.
3440 (aarch64_expand_vec_series): New function.
3441 (aarch64_expand_sve_widened_duplicate): Likewise.
3442 (aarch64_expand_sve_const_vector): Likewise.
3443 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3444 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3445 into the register, rather than emitting a SET directly.
3446 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3447 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3448 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3449 (offset_9bit_signed_scaled_p): New functions.
3450 (aarch64_replicate_bitmask_imm): New function.
3451 (aarch64_bitmask_imm): Use it.
3452 (aarch64_cannot_force_const_mem): Reject expressions involving
3453 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3454 (aarch64_classify_index): Handle SVE indices, by requiring
3455 a plain register index with a scale that matches the element size.
3456 (aarch64_classify_address): Handle SVE addresses. Assert that
3457 the mode of the address is VOIDmode or an integer mode.
3458 Update call to aarch64_classify_symbol.
3459 (aarch64_classify_symbolic_expression): Update call to
3460 aarch64_classify_symbol.
3461 (aarch64_const_vec_all_in_range_p): New function.
3462 (aarch64_print_vector_float_operand): Likewise.
3463 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3464 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3465 and the FP immediates 1.0 and 0.5.
3466 (aarch64_print_address_internal): Handle SVE addresses.
3467 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3468 (aarch64_regno_regclass): Handle predicate registers.
3469 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3471 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3472 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3473 (aarch64_convert_sve_vector_bits): New function.
3474 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3475 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3477 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3478 Handle SVE vector and predicate modes. Accept VL-based constants
3479 that need only one temporary register, and VL offsets that require
3480 no temporary registers.
3481 (aarch64_conditional_register_usage): Mark the predicate registers
3482 as fixed if SVE isn't available.
3483 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3484 Return true for SVE vector and predicate modes.
3485 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3486 rather than an unsigned int. Handle SVE modes.
3487 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3489 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3491 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3492 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3493 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3494 (aarch64_sve_float_mul_immediate_p): New functions.
3495 (aarch64_sve_valid_immediate): New function.
3496 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3497 Explicitly reject structure modes. Check for INDEX constants.
3498 Handle PTRUE and PFALSE constants.
3499 (aarch64_check_zero_based_sve_index_immediate): New function.
3500 (aarch64_simd_imm_zero_p): Delete.
3501 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3502 vector modes. Accept constants in the range of CNT[BHWD].
3503 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3504 ask for an Advanced SIMD mode.
3505 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3506 (aarch64_simd_vector_alignment): Handle SVE predicates.
3507 (aarch64_vectorize_preferred_vector_alignment): New function.
3508 (aarch64_simd_vector_alignment_reachable): Use it instead of
3510 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3511 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3513 (MAX_VECT_LEN): Delete.
3514 (expand_vec_perm_d): Add a vec_flags field.
3515 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3516 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3517 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3519 (aarch64_evpc_rev): Rename to...
3520 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3521 (aarch64_evpc_rev_global): New function.
3522 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3523 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3525 (aarch64_evpc_sve_tbl): New function.
3526 (aarch64_expand_vec_perm_const_1): Update after rename of
3527 aarch64_evpc_rev. Handle SVE permutes too, trying
3528 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3529 than aarch64_evpc_tbl.
3530 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3531 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3532 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3533 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3534 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3535 (aarch64_expand_sve_vcond): New functions.
3536 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3537 of aarch64_vector_mode_p.
3538 (aarch64_dwarf_poly_indeterminate_value): New function.
3539 (aarch64_compute_pressure_classes): Likewise.
3540 (aarch64_can_change_mode_class): Likewise.
3541 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3542 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3543 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3544 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3545 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3546 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3547 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3548 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3550 (Dn, Dl, Dr): Accept const as well as const_vector.
3551 (Dz): Likewise. Compare against CONST0_RTX.
3552 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3553 of "vector" where appropriate.
3554 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3555 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3556 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3557 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3558 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3559 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3560 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3561 (v_int_equiv): Extend to SVE modes.
3562 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3564 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3565 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3566 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3567 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3568 (SVE_COND_FP_CMP): New int iterators.
3569 (perm_hilo): Handle the new unpack unspecs.
3570 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3572 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3573 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3574 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3575 (aarch64_equality_operator, aarch64_constant_vector_operand)
3576 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3577 (aarch64_sve_nonimmediate_operand): Likewise.
3578 (aarch64_sve_general_operand): Likewise.
3579 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3580 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3581 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3582 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3583 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3584 (aarch64_sve_float_arith_immediate): Likewise.
3585 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3586 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3587 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3588 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3589 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3590 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3591 (aarch64_sve_float_arith_operand): Likewise.
3592 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3593 (aarch64_sve_float_mul_operand): Likewise.
3594 (aarch64_sve_vec_perm_operand): Likewise.
3595 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3596 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3597 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3598 as well as const_vector.
3599 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3600 in file. Use CONST0_RTX and CONSTM1_RTX.
3601 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3602 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3603 Use aarch64_simd_imm_zero.
3604 * config/aarch64/aarch64-sve.md: New file.
3605 * config/aarch64/aarch64.md: Include it.
3606 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3607 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3608 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3609 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3610 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3611 (sve): New attribute.
3612 (enabled): Disable instructions with the sve attribute unless
3614 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3615 aarch64_expand_mov_immediate.
3616 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3617 CNT[BHSD] immediates.
3618 (movti): Split CONST_POLY_INT moves into two halves.
3619 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3620 Split additions that need a temporary here if the destination
3621 is the stack pointer.
3622 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3623 (*add<mode>3_poly_1): New instruction.
3624 (set_clobber_cc): New expander.
3626 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3628 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3629 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3630 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3631 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3632 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3633 Change innermode from fixed_mode_size to machine_mode.
3634 (simplify_subreg): Update call accordingly. Handle a constant-sized
3635 subreg of a variable-length CONST_VECTOR.
3637 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3638 Alan Hayward <alan.hayward@arm.com>
3639 David Sherwood <david.sherwood@arm.com>
3641 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3642 (add_offset_to_base): New function, split out from...
3643 (create_mem_ref): ...here. When handling a scale other than 1,
3644 check first whether the address is valid without the offset.
3645 Add it into the base if so, leaving the index and scale as-is.
3647 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3650 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3651 fold_for_warn before checking if arg2 is INTEGER_CST.
3653 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3655 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3656 (store_multiple_operation): Delete.
3657 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3658 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3659 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3660 guarded by TARGET_STRING.
3661 (rs6000_output_load_multiple): Delete.
3662 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3663 OPTION_MASK_STRING / TARGET_STRING handling.
3664 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3665 (const rs6000_opt_masks) <"string">: Change mask to 0.
3666 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3667 (MASK_STRING): Delete.
3668 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3670 (load_multiple): Delete.
3677 (store_multiple): Delete.
3684 (movmemsi_8reg): Delete.
3685 (corresponding unnamed define_insn): Delete.
3686 (movmemsi_6reg): Delete.
3687 (corresponding unnamed define_insn): Delete.
3688 (movmemsi_4reg): Delete.
3689 (corresponding unnamed define_insn): Delete.
3690 (movmemsi_2reg): Delete.
3691 (corresponding unnamed define_insn): Delete.
3692 (movmemsi_1reg): Delete.
3693 (corresponding unnamed define_insn): Delete.
3694 * config/rs6000/rs6000.opt (mno-string): New.
3695 (mstring): Replace by deprecation warning stub.
3696 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3698 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3700 * regrename.c (regrename_do_replace): If replacing the same
3701 reg multiple times, try to reuse last created gen_raw_REG.
3704 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3705 main to workaround a bug in GDB.
3707 2018-01-12 Tom de Vries <tom@codesourcery.com>
3710 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3712 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3714 PR rtl-optimization/80481
3715 * ira-color.c (get_cap_member): New function.
3716 (allocnos_conflict_by_live_ranges_p): Use it.
3717 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3718 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3720 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3723 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3724 (*saddl_se_1): Ditto.
3726 (*ssubl_se_1): Ditto.
3728 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3730 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3731 rather than wi::to_widest for DR_INITs.
3732 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3733 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3734 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3736 (vect_analyze_group_access_1): Note that here.
3738 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3740 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3741 polynomial type sizes.
3743 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3745 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3746 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3747 (gimple_add_tmp_var): Likewise.
3749 2018-01-12 Martin Liska <mliska@suse.cz>
3751 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3752 (gimple_alloc_sizes): Likewise.
3753 (dump_gimple_statistics): Use PRIu64 in printf format.
3754 * gimple.h: Change uint64_t to int.
3756 2018-01-12 Martin Liska <mliska@suse.cz>
3758 * tree-core.h: Use uint64_t instead of int.
3759 * tree.c (tree_node_counts): Likewise.
3760 (tree_node_sizes): Likewise.
3761 (dump_tree_statistics): Use PRIu64 in printf format.
3763 2018-01-12 Martin Liska <mliska@suse.cz>
3765 * Makefile.in: As qsort_chk is implemented in vec.c, add
3766 vec.o to linkage of gencfn-macros.
3767 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3768 passing the info to record_node_allocation_statistics.
3769 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3771 * ggc-common.c (struct ggc_usage): Add operator== and use
3772 it in operator< and compare function.
3773 * mem-stats.h (struct mem_usage): Likewise.
3774 * vec.c (struct vec_usage): Remove operator< and compare
3775 function. Can be simply inherited.
3777 2018-01-12 Martin Jambor <mjambor@suse.cz>
3780 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3781 * tree-ssa-math-opts.c: Include domwalk.h.
3782 (convert_mult_to_fma_1): New function.
3783 (fma_transformation_info): New type.
3784 (fma_deferring_state): Likewise.
3785 (cancel_fma_deferring): New function.
3786 (result_of_phi): Likewise.
3787 (last_fma_candidate_feeds_initial_phi): Likewise.
3788 (convert_mult_to_fma): Added deferring logic, split actual
3789 transformation to convert_mult_to_fma_1.
3790 (math_opts_dom_walker): New type.
3791 (math_opts_dom_walker::after_dom_children): New method, body moved
3792 here from pass_optimize_widening_mul::execute, added deferring logic
3794 (pass_optimize_widening_mul::execute): Moved most of code to
3795 math_opts_dom_walker::after_dom_children.
3796 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3797 * config/i386/i386.c (ix86_option_override_internal): Added
3798 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3800 2018-01-12 Richard Biener <rguenther@suse.de>
3803 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3804 inline instance vars.
3806 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3809 * config/rx/rx.c (rx_is_restricted_memory_address):
3812 2018-01-12 Richard Biener <rguenther@suse.de>
3814 PR tree-optimization/80846
3815 * target.def (split_reduction): New target hook.
3816 * targhooks.c (default_split_reduction): New function.
3817 * targhooks.h (default_split_reduction): Declare.
3818 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3819 target requests first reduce vectors by combining low and high
3821 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3822 (get_vectype_for_scalar_type_and_size): Export.
3823 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3824 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3825 * doc/tm.texi: Regenerate.
3826 * config/i386/i386.c (ix86_split_reduction): Implement
3827 TARGET_VECTORIZE_SPLIT_REDUCTION.
3829 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3832 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3833 in PIC mode except for TARGET_VXWORKS_RTP.
3834 * config/sparc/sparc.c: Include cfgrtl.h.
3835 (TARGET_INIT_PIC_REG): Define.
3836 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3837 (sparc_pic_register_p): New predicate.
3838 (sparc_legitimate_address_p): Use it.
3839 (sparc_legitimize_pic_address): Likewise.
3840 (sparc_delegitimize_address): Likewise.
3841 (sparc_mode_dependent_address_p): Likewise.
3842 (gen_load_pcrel_sym): Remove 4th parameter.
3843 (load_got_register): Adjust call to above. Remove obsolete stuff.
3844 (sparc_expand_prologue): Do not call load_got_register here.
3845 (sparc_flat_expand_prologue): Likewise.
3846 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3847 (sparc_use_pseudo_pic_reg): New function.
3848 (sparc_init_pic_reg): Likewise.
3849 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3850 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3852 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3854 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3855 Add item for branch_cost.
3857 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3859 PR rtl-optimization/83565
3860 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3861 not extend the result to a larger mode for rotate operations.
3862 (num_sign_bit_copies1): Likewise.
3864 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3867 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3869 Use values-Xc.o for -pedantic.
3870 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3872 2018-01-12 Martin Liska <mliska@suse.cz>
3875 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3877 (possible_polymorphic_call_targets): Use it.
3878 (ipa_devirt): Likewise.
3880 2018-01-12 Martin Liska <mliska@suse.cz>
3882 * profile-count.h (enum profile_quality): Use 0 as invalid
3883 enum value of profile_quality.
3885 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3887 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3888 -mext-string options.
3890 2018-01-12 Richard Biener <rguenther@suse.de>
3892 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3893 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3894 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3896 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3898 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3900 * configure.ac (--with-long-double-format): Add support for the
3901 configuration option to change the default long double format on
3903 * config.gcc (powerpc*-linux*-*): Likewise.
3904 * configure: Regenerate.
3905 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3906 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3907 used without modification.
3909 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3911 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3912 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3913 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3914 MISC_BUILTIN_SPEC_BARRIER.
3915 (rs6000_init_builtins): Likewise.
3916 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3918 (speculation_barrier): New define_insn.
3919 * doc/extend.texi: Document __builtin_speculation_barrier.
3921 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3924 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3925 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3926 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3928 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3929 integral modes instead of "ss" and "sd".
3930 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3931 vectors with 32-bit and 64-bit elements.
3932 (vecdupssescalarmodesuffix): New mode attribute.
3933 (vec_dup<mode>): Use it.
3935 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3938 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3939 frame if argument is passed on stack.
3941 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3944 * ree.c (combine_reaching_defs): Optimize also
3945 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3946 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3948 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3951 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3953 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3956 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3957 after they are computed.
3959 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3961 PR tree-optimization/83695
3962 * gimple-loop-linterchange.cc
3963 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3964 reset cached scev information after interchange.
3965 (pass_linterchange::execute): Remove call to scev_reset_htab.
3967 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3969 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3970 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3971 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3972 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3973 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3974 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3975 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3976 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3977 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3978 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3979 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3980 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3981 (V_lane_reg): Likewise.
3982 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3984 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3985 (vfmal_lane_low<mode>_intrinsic,
3986 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3987 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3988 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3989 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3990 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3991 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3993 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3995 * config/arm/arm-cpus.in (fp16fml): New feature.
3996 (ALL_SIMD): Add fp16fml.
3997 (armv8.2-a): Add fp16fml as an option.
3998 (armv8.3-a): Likewise.
3999 (armv8.4-a): Add fp16fml as part of fp16.
4000 * config/arm/arm.h (TARGET_FP16FML): Define.
4001 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4003 * config/arm/arm-modes.def (V2HF): Define.
4004 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4005 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4006 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4007 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4008 vfmsl_low, vfmsl_high): New set of builtins.
4009 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4010 (vfml_op): New code attribute.
4011 (VFMLHALVES): New int iterator.
4012 (VFML, VFMLSEL): New mode attributes.
4013 (V_reg): Define mapping for V2HF.
4014 (V_hi, V_lo): New mode attributes.
4015 (VF_constraint): Likewise.
4016 (vfml_half, vfml_half_selector): New int attributes.
4017 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4019 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4020 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4022 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4023 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4024 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4025 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4027 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4028 Document new effective target and option set.
4030 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4032 * config/arm/arm-cpus.in (armv8_4): New feature.
4033 (ARMv8_4a): New fgroup.
4034 (armv8.4-a): New arch.
4035 * config/arm/arm-tables.opt: Regenerate.
4036 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4037 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4038 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4039 Add matching rules for -march=armv8.4-a and extensions.
4040 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4042 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4045 * config/rx/rx.md (BW): New mode attribute.
4046 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4048 2018-01-11 Richard Biener <rguenther@suse.de>
4050 PR tree-optimization/83435
4051 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4052 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4053 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4055 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4056 Alan Hayward <alan.hayward@arm.com>
4057 David Sherwood <david.sherwood@arm.com>
4059 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4061 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4062 (aarch64_print_address_internal): Use it to check for a zero offset.
4064 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4065 Alan Hayward <alan.hayward@arm.com>
4066 David Sherwood <david.sherwood@arm.com>
4068 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4069 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4070 Return a poly_int64 rather than a HOST_WIDE_INT.
4071 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4072 rather than a HOST_WIDE_INT.
4073 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4074 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4075 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4076 final_offset from HOST_WIDE_INT to poly_int64.
4077 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4078 to_constant when getting the number of units in an Advanced SIMD
4080 (aarch64_builtin_vectorized_function): Check for a constant number
4082 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4084 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4085 attribute instead of GET_MODE_NUNITS.
4086 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4087 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4088 GET_MODE_SIZE for fixed-size registers.
4089 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4090 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4091 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4092 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4093 (aarch64_print_operand, aarch64_print_address_internal)
4094 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4095 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4096 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4097 Handle polynomial GET_MODE_SIZE.
4098 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4099 wider than SImode without modification.
4100 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4101 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4102 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4103 passing and returning SVE modes.
4104 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4105 rather than GEN_INT.
4106 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4107 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4108 (aarch64_allocate_and_probe_stack_space): Likewise.
4109 (aarch64_layout_frame): Cope with polynomial offsets.
4110 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4111 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4113 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4114 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4115 poly_int64 rather than a HOST_WIDE_INT.
4116 (aarch64_get_separate_components, aarch64_process_components)
4117 (aarch64_expand_prologue, aarch64_expand_epilogue)
4118 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4119 (aarch64_anchor_offset): New function, split out from...
4120 (aarch64_legitimize_address): ...here.
4121 (aarch64_builtin_vectorization_cost): Handle polynomial
4122 TYPE_VECTOR_SUBPARTS.
4123 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4125 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4126 number of elements from the PARALLEL rather than the mode.
4127 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4128 rather than GET_MODE_BITSIZE.
4129 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4130 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4131 (aarch64_expand_vec_perm_const_1): Handle polynomial
4132 d->perm.length () and d->perm elements.
4133 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4134 Apply to_constant to d->perm elements.
4135 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4136 polynomial CONST_VECTOR_NUNITS.
4137 (aarch64_move_pointer): Take amount as a poly_int64 rather
4139 (aarch64_progress_pointer): Avoid temporary variable.
4140 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4141 the mode attribute instead of GET_MODE.
4143 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4144 Alan Hayward <alan.hayward@arm.com>
4145 David Sherwood <david.sherwood@arm.com>
4147 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4148 x exists before using it.
4149 (aarch64_add_constant_internal): Rename to...
4150 (aarch64_add_offset_1): ...this. Replace regnum with separate
4151 src and dest rtxes. Handle the case in which they're different,
4152 including when the offset is zero. Replace scratchreg with an rtx.
4153 Use 2 additions if there is no spare register into which we can
4154 move a 16-bit constant.
4155 (aarch64_add_constant): Delete.
4156 (aarch64_add_offset): Replace reg with separate src and dest
4157 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4158 Use aarch64_add_offset_1.
4159 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4160 an rtx rather than an int. Take the delta as a poly_int64
4161 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4162 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4163 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4164 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4165 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4167 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4168 aarch64_add_constant.
4170 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4172 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4173 Use scalar_float_mode.
4175 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4177 * config/aarch64/aarch64-simd.md
4178 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4179 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4180 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4181 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4182 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4183 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4184 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4185 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4186 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4187 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4189 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4192 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4193 targ_options->x_arm_arch_string is non NULL.
4195 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4197 * config/aarch64/aarch64.h
4198 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4200 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4203 * expmed.c (emit_store_flag_force): Swap if const op0
4204 and change VOIDmode to mode of op0.
4206 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4208 PR rtl-optimization/83761
4209 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4210 than bytes to mode_for_size.
4212 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4215 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4216 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4219 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4222 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4223 when in layout mode.
4224 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4225 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4228 2018-01-10 Michael Collison <michael.collison@arm.com>
4230 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4231 * config/aarch64/aarch64-option-extension.def: Add
4232 AARCH64_OPT_EXTENSION of 'fp16fml'.
4233 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4234 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4235 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4236 * config/aarch64/constraints.md (Ui7): New constraint.
4237 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4238 (VFMLA_SEL_W): Ditto.
4241 (VFMLA16_LOW): New int iterator.
4242 (VFMLA16_HIGH): Ditto.
4243 (UNSPEC_FMLAL): New unspec.
4244 (UNSPEC_FMLSL): Ditto.
4245 (UNSPEC_FMLAL2): Ditto.
4246 (UNSPEC_FMLSL2): Ditto.
4247 (f16mac): New code attribute.
4248 * config/aarch64/aarch64-simd-builtins.def
4249 (aarch64_fmlal_lowv2sf): Ditto.
4250 (aarch64_fmlsl_lowv2sf): Ditto.
4251 (aarch64_fmlalq_lowv4sf): Ditto.
4252 (aarch64_fmlslq_lowv4sf): Ditto.
4253 (aarch64_fmlal_highv2sf): Ditto.
4254 (aarch64_fmlsl_highv2sf): Ditto.
4255 (aarch64_fmlalq_highv4sf): Ditto.
4256 (aarch64_fmlslq_highv4sf): Ditto.
4257 (aarch64_fmlal_lane_lowv2sf): Ditto.
4258 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4259 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4260 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4261 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4262 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4263 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4264 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4265 (aarch64_fmlal_lane_highv2sf): Ditto.
4266 (aarch64_fmlsl_lane_highv2sf): Ditto.
4267 (aarch64_fmlal_laneq_highv2sf): Ditto.
4268 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4269 (aarch64_fmlalq_lane_highv4sf): Ditto.
4270 (aarch64_fmlsl_lane_highv4sf): Ditto.
4271 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4272 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4273 * config/aarch64/aarch64-simd.md:
4274 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4275 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4276 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4277 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4278 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4279 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4280 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4281 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4282 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4283 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4284 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4285 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4286 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4287 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4288 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4289 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4290 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4291 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4292 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4293 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4294 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4295 (vfmlsl_low_u32): Ditto.
4296 (vfmlalq_low_u32): Ditto.
4297 (vfmlslq_low_u32): Ditto.
4298 (vfmlal_high_u32): Ditto.
4299 (vfmlsl_high_u32): Ditto.
4300 (vfmlalq_high_u32): Ditto.
4301 (vfmlslq_high_u32): Ditto.
4302 (vfmlal_lane_low_u32): Ditto.
4303 (vfmlsl_lane_low_u32): Ditto.
4304 (vfmlal_laneq_low_u32): Ditto.
4305 (vfmlsl_laneq_low_u32): Ditto.
4306 (vfmlalq_lane_low_u32): Ditto.
4307 (vfmlslq_lane_low_u32): Ditto.
4308 (vfmlalq_laneq_low_u32): Ditto.
4309 (vfmlslq_laneq_low_u32): Ditto.
4310 (vfmlal_lane_high_u32): Ditto.
4311 (vfmlsl_lane_high_u32): Ditto.
4312 (vfmlal_laneq_high_u32): Ditto.
4313 (vfmlsl_laneq_high_u32): Ditto.
4314 (vfmlalq_lane_high_u32): Ditto.
4315 (vfmlslq_lane_high_u32): Ditto.
4316 (vfmlalq_laneq_high_u32): Ditto.
4317 (vfmlslq_laneq_high_u32): Ditto.
4318 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4319 (AARCH64_FL_FOR_ARCH8_4): New.
4320 (AARCH64_ISA_F16FML): New ISA flag.
4321 (TARGET_F16FML): New feature flag for fp16fml.
4322 (doc/invoke.texi): Document new fp16fml option.
4324 2018-01-10 Michael Collison <michael.collison@arm.com>
4326 * config/aarch64/aarch64-builtins.c:
4327 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4328 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4329 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4330 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4331 (AARCH64_ISA_SHA3): New ISA flag.
4332 (TARGET_SHA3): New feature flag for sha3.
4333 * config/aarch64/iterators.md (sha512_op): New int attribute.
4334 (CRYPTO_SHA512): New int iterator.
4335 (UNSPEC_SHA512H): New unspec.
4336 (UNSPEC_SHA512H2): Ditto.
4337 (UNSPEC_SHA512SU0): Ditto.
4338 (UNSPEC_SHA512SU1): Ditto.
4339 * config/aarch64/aarch64-simd-builtins.def
4340 (aarch64_crypto_sha512hqv2di): New builtin.
4341 (aarch64_crypto_sha512h2qv2di): Ditto.
4342 (aarch64_crypto_sha512su0qv2di): Ditto.
4343 (aarch64_crypto_sha512su1qv2di): Ditto.
4344 (aarch64_eor3qv8hi): Ditto.
4345 (aarch64_rax1qv2di): Ditto.
4346 (aarch64_xarqv2di): Ditto.
4347 (aarch64_bcaxqv8hi): Ditto.
4348 * config/aarch64/aarch64-simd.md:
4349 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4350 (aarch64_crypto_sha512su0qv2di): Ditto.
4351 (aarch64_crypto_sha512su1qv2di): Ditto.
4352 (aarch64_eor3qv8hi): Ditto.
4353 (aarch64_rax1qv2di): Ditto.
4354 (aarch64_xarqv2di): Ditto.
4355 (aarch64_bcaxqv8hi): Ditto.
4356 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4357 (vsha512h2q_u64): Ditto.
4358 (vsha512su0q_u64): Ditto.
4359 (vsha512su1q_u64): Ditto.
4360 (veor3q_u16): Ditto.
4361 (vrax1q_u64): Ditto.
4363 (vbcaxq_u16): Ditto.
4364 * config/arm/types.md (crypto_sha512): New type attribute.
4365 (crypto_sha3): Ditto.
4366 (doc/invoke.texi): Document new sha3 option.
4368 2018-01-10 Michael Collison <michael.collison@arm.com>
4370 * config/aarch64/aarch64-builtins.c:
4371 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4372 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4373 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4374 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4375 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4376 (AARCH64_ISA_SM4): New ISA flag.
4377 (TARGET_SM4): New feature flag for sm4.
4378 * config/aarch64/aarch64-simd-builtins.def
4379 (aarch64_sm3ss1qv4si): Ditto.
4380 (aarch64_sm3tt1aq4si): Ditto.
4381 (aarch64_sm3tt1bq4si): Ditto.
4382 (aarch64_sm3tt2aq4si): Ditto.
4383 (aarch64_sm3tt2bq4si): Ditto.
4384 (aarch64_sm3partw1qv4si): Ditto.
4385 (aarch64_sm3partw2qv4si): Ditto.
4386 (aarch64_sm4eqv4si): Ditto.
4387 (aarch64_sm4ekeyqv4si): Ditto.
4388 * config/aarch64/aarch64-simd.md:
4389 (aarch64_sm3ss1qv4si): Ditto.
4390 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4391 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4392 (aarch64_sm4eqv4si): Ditto.
4393 (aarch64_sm4ekeyqv4si): Ditto.
4394 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4395 (sm3part_op): Ditto.
4396 (CRYPTO_SM3TT): Ditto.
4397 (CRYPTO_SM3PART): Ditto.
4398 (UNSPEC_SM3SS1): New unspec.
4399 (UNSPEC_SM3TT1A): Ditto.
4400 (UNSPEC_SM3TT1B): Ditto.
4401 (UNSPEC_SM3TT2A): Ditto.
4402 (UNSPEC_SM3TT2B): Ditto.
4403 (UNSPEC_SM3PARTW1): Ditto.
4404 (UNSPEC_SM3PARTW2): Ditto.
4405 (UNSPEC_SM4E): Ditto.
4406 (UNSPEC_SM4EKEY): Ditto.
4407 * config/aarch64/constraints.md (Ui2): New constraint.
4408 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4409 * config/arm/types.md (crypto_sm3): New type attribute.
4410 (crypto_sm4): Ditto.
4411 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4412 (vsm3tt1aq_u32): Ditto.
4413 (vsm3tt1bq_u32): Ditto.
4414 (vsm3tt2aq_u32): Ditto.
4415 (vsm3tt2bq_u32): Ditto.
4416 (vsm3partw1q_u32): Ditto.
4417 (vsm3partw2q_u32): Ditto.
4418 (vsm4eq_u32): Ditto.
4419 (vsm4ekeyq_u32): Ditto.
4420 (doc/invoke.texi): Document new sm4 option.
4422 2018-01-10 Michael Collison <michael.collison@arm.com>
4424 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4425 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4426 (AARCH64_FL_FOR_ARCH8_4): New.
4427 (AARCH64_FL_V8_4): New flag.
4428 (doc/invoke.texi): Document new armv8.4-a option.
4430 2018-01-10 Michael Collison <michael.collison@arm.com>
4432 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4433 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4434 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4435 * config/aarch64/aarch64-option-extension.def: Add
4436 AARCH64_OPT_EXTENSION of 'sha2'.
4437 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4438 (crypto): Disable sha2 and aes if crypto disabled.
4439 (crypto): Enable aes and sha2 if enabled.
4440 (simd): Disable sha2 and aes if simd disabled.
4441 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4443 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4444 (TARGET_SHA2): New feature flag for sha2.
4445 (TARGET_AES): New feature flag for aes.
4446 * config/aarch64/aarch64-simd.md:
4447 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4448 conditional on TARGET_AES.
4449 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4450 (aarch64_crypto_sha1hsi): Make pattern conditional
4452 (aarch64_crypto_sha1hv4si): Ditto.
4453 (aarch64_be_crypto_sha1hv4si): Ditto.
4454 (aarch64_crypto_sha1su1v4si): Ditto.
4455 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4456 (aarch64_crypto_sha1su0v4si): Ditto.
4457 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4458 (aarch64_crypto_sha256su0v4si): Ditto.
4459 (aarch64_crypto_sha256su1v4si): Ditto.
4460 (doc/invoke.texi): Document new aes and sha2 options.
4462 2018-01-10 Martin Sebor <msebor@redhat.com>
4464 PR tree-optimization/83781
4465 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4468 2018-01-11 Martin Sebor <msebor@gmail.com>
4469 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4471 PR tree-optimization/83501
4472 PR tree-optimization/81703
4474 * tree-ssa-strlen.c (get_string_cst): Rename...
4475 (get_string_len): ...to this. Handle global constants.
4476 (handle_char_store): Adjust.
4478 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4479 Jim Wilson <jimw@sifive.com>
4481 * config/riscv/riscv-protos.h (riscv_output_return): New.
4482 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4483 (riscv_attribute_table, riscv_output_return),
4484 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4485 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4486 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4488 (riscv_expand_prologue): Add early return for naked function.
4489 (riscv_expand_epilogue): Likewise.
4490 (riscv_function_ok_for_sibcall): Return false for naked function.
4491 (riscv_set_current_function): New.
4492 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4493 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4494 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4495 * doc/extend.texi (RISC-V Function Attributes): New.
4497 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4499 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4500 check for 128-bit long double before checking TCmode.
4501 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4502 128-bit long doubles before checking TFmode or TCmode.
4503 (FLOAT128_IBM_P): Likewise.
4505 2018-01-10 Martin Sebor <msebor@redhat.com>
4507 PR tree-optimization/83671
4508 * builtins.c (c_strlen): Unconditionally return zero for the empty
4510 Use -Warray-bounds for warnings.
4511 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4512 for non-constant array indices with COMPONENT_REF, arrays of
4513 arrays, and pointers to arrays.
4514 (gimple_fold_builtin_strlen): Determine and set length range for
4515 non-constant character arrays.
4517 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4520 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4523 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4525 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4527 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4530 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4531 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4532 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4533 indexed_or_indirect_operand predicate.
4534 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4535 (*vsx_le_perm_load_v8hi): Likewise.
4536 (*vsx_le_perm_load_v16qi): Likewise.
4537 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4538 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4539 (*vsx_le_perm_store_v8hi): Likewise.
4540 (*vsx_le_perm_store_v16qi): Likewise.
4541 (eight unnamed splitters): Likewise.
4543 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4545 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4546 * config/rs6000/emmintrin.h: Likewise.
4547 * config/rs6000/mmintrin.h: Likewise.
4548 * config/rs6000/xmmintrin.h: Likewise.
4550 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4553 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4555 * tree.c (tree_nop_conversion): Return true for location wrapper
4557 (maybe_wrap_with_location): New function.
4558 (selftest::check_strip_nops): New function.
4559 (selftest::test_location_wrappers): New function.
4560 (selftest::tree_c_tests): Call it.
4561 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4562 (maybe_wrap_with_location): New decl.
4563 (EXPR_LOCATION_WRAPPER_P): New macro.
4564 (location_wrapper_p): New inline function.
4565 (tree_strip_any_location_wrapper): New inline function.
4567 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4570 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4571 stack_realign_offset for the largest alignment of stack slot
4573 (ix86_find_max_used_stack_alignment): New function.
4574 (ix86_finalize_stack_frame_flags): Use it. Set
4575 max_used_stack_alignment if we don't realign stack.
4576 * config/i386/i386.h (machine_function): Add
4577 max_used_stack_alignment.
4579 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4581 * config/arm/arm.opt (-mbranch-cost): New option.
4582 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4585 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4588 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4589 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4591 2018-01-10 Richard Biener <rguenther@suse.de>
4594 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4595 early out so it also covers the case where we have a non-NULL
4598 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4600 PR tree-optimization/83753
4601 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4602 for non-strided grouped accesses if the number of elements is 1.
4604 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4607 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4608 * i386.h (TARGET_USE_GATHER): Define.
4609 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4611 2018-01-10 Martin Liska <mliska@suse.cz>
4614 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4615 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4617 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4618 CLEANUP_NO_PARTITIONING is not set.
4620 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4622 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4623 for vectors, as a partial revert of r254296.
4624 * rtl.h (const_vec_p): Delete.
4625 (const_vec_duplicate_p): Don't test for vector CONSTs.
4626 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4627 * expmed.c (make_tree): Likewise.
4630 * common.md (E, F): Use CONSTANT_P instead of checking for
4632 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4633 checking for CONST_VECTOR.
4635 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4638 * predict.c (force_edge_cold): Handle in more sane way edges
4641 2018-01-09 Carl Love <cel@us.ibm.com>
4643 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4645 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4646 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4647 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4648 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4649 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4650 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4651 * config/rs6000/rs6000-protos.h: Add extern defition for
4652 rs6000_generate_float2_double_code.
4653 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4655 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4656 (float2_v2df): Add define_expand.
4658 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4661 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4662 op_mode in the force_to_mode call.
4664 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4666 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4667 instead of checking each element individually.
4668 (aarch64_evpc_uzp): Likewise.
4669 (aarch64_evpc_zip): Likewise.
4670 (aarch64_evpc_ext): Likewise.
4671 (aarch64_evpc_rev): Likewise.
4672 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4673 instead of checking each element individually. Return true without
4675 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4676 whether all selected elements come from the same input, instead of
4677 checking each element individually. Remove calls to gen_rtx_REG,
4678 start_sequence and end_sequence and instead assert that no rtl is
4681 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4683 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4684 order of HIGH and CONST checks.
4686 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4688 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4689 if the destination isn't an SSA_NAME.
4691 2018-01-09 Richard Biener <rguenther@suse.de>
4693 PR tree-optimization/83668
4694 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4696 (canonicalize_loop_form): ... here, renamed from ...
4697 (canonicalize_loop_closed_ssa_form): ... this and amended to
4698 swap successor edges for loop exit blocks to make us use
4699 the RPO order we need for initial schedule generation.
4701 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4703 PR tree-optimization/64811
4704 * match.pd: When optimizing comparisons with Inf, avoid
4705 introducing or losing exceptions from comparisons with NaN.
4707 2018-01-09 Martin Liska <mliska@suse.cz>
4710 * asan.c (shadow_mem_size): Add gcc_assert.
4712 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4714 Don't save registers in main().
4717 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4718 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4719 * config/avr/avr.c (avr_set_current_function): Don't error if
4720 naked, OS_task or OS_main are specified at the same time.
4721 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4723 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4725 * common/config/avr/avr-common.c (avr_option_optimization_table):
4726 Switch on -mmain-is-OS_task for optimizing compilations.
4728 2018-01-09 Richard Biener <rguenther@suse.de>
4730 PR tree-optimization/83572
4731 * graphite.c: Include cfganal.h.
4732 (graphite_transform_loops): Connect infinite loops to exit
4733 and remove fake edges at the end.
4735 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4737 * ipa-inline.c (edge_badness): Revert accidental checkin.
4739 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4742 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4743 symbols; not inline clones.
4745 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4748 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4749 hard registers. Formatting fixes.
4751 PR preprocessor/83722
4752 * gcc.c (try_generate_repro): Pass
4753 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4754 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4757 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4758 Kito Cheng <kito.cheng@gmail.com>
4760 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4761 (riscv_leaf_function_p): Delete.
4762 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4764 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4766 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4768 (do_ifelse): New function.
4769 (do_isel): New function.
4770 (do_sub3): New function.
4771 (do_add3): New function.
4772 (do_load_mask_compare): New function.
4773 (do_overlap_load_compare): New function.
4774 (expand_compare_loop): New function.
4775 (expand_block_compare): Call expand_compare_loop() when appropriate.
4776 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4778 (-mblock-compare-inline-loop-limit): New option.
4780 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4783 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4784 Reverse order of second and third operands in first alternative.
4785 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4786 of first and second elements in UNSPEC_VPERMR vector.
4787 (altivec_expand_vec_perm_le): Likewise.
4789 2018-01-08 Jeff Law <law@redhat.com>
4791 PR rtl-optimizatin/81308
4792 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4793 (process_switch): If group_case_labels makes a change, then set
4795 (pass_convert_switch::execute): If a switch is converted, then
4796 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4798 PR rtl-optimization/81308
4799 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4802 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4804 PR target/83663 - Revert r255946
4805 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4806 generation for cases where splatting a value is not useful.
4807 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4808 across a vec_duplicate and a paradoxical subreg forming a vector
4809 mode to a vec_concat.
4811 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4813 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4814 -march=armv8.3-a variants.
4815 * config/arm/t-multilib: Likewise.
4816 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4818 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4820 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4822 (cceq_ior_compare_complement): Give it a name so I can use it, and
4823 change boolean_or_operator predicate to boolean_operator so it can
4824 be used to generate a crand.
4825 (eqne): New code iterator.
4826 (bd/bd_neg): New code_attrs.
4827 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4828 a single define_insn.
4829 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4830 decrement (bdnzt/bdnzf/bdzt/bdzf).
4831 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4832 with the new names of the branch decrement patterns, and added the
4833 names of the branch decrement conditional patterns.
4835 2018-01-08 Richard Biener <rguenther@suse.de>
4837 PR tree-optimization/83563
4838 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4841 2018-01-08 Richard Biener <rguenther@suse.de>
4844 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4846 2018-01-08 Richard Biener <rguenther@suse.de>
4848 PR tree-optimization/83685
4849 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4850 references to abnormals.
4852 2018-01-08 Richard Biener <rguenther@suse.de>
4855 * dwarf2out.c (output_indirect_strings): Handle empty
4856 skeleton_debug_str_hash.
4857 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4859 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4861 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4862 (emit_store_direct): Likewise.
4863 (arc_trampoline_adjust_address): Likewise.
4864 (arc_asm_trampoline_template): New function.
4865 (arc_initialize_trampoline): Use asm_trampoline_template.
4866 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4867 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4868 * config/arc/arc.md (flush_icache): Delete pattern.
4870 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4872 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4873 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4876 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4879 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4880 by not USED_FOR_TARGET.
4881 (make_pass_resolve_sw_modes): Likewise.
4883 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4885 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4888 2018-01-08 Richard Biener <rguenther@suse.de>
4891 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4893 2018-01-08 Richard Biener <rguenther@suse.de>
4896 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4898 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4901 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4902 basic blocks with a small number of successors.
4903 (convert_control_dep_chain_into_preds): Improve handling of
4905 (dump_predicates): Split apart into...
4906 (dump_pred_chain): ...here...
4907 (dump_pred_info): ...and here.
4908 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4909 (can_chain_union_be_invalidated_p): Improve check for invalidation
4911 (uninit_uses_cannot_happen): Avoid unnecessary if
4912 convert_control_dep_chain_into_preds yielded nothing.
4914 2018-01-06 Martin Sebor <msebor@redhat.com>
4916 PR tree-optimization/83640
4917 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4918 subtracting negative offset from size.
4919 (builtin_access::overlap): Adjust offset bounds of the access to fall
4920 within the size of the object if possible.
4922 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4924 PR rtl-optimization/83699
4925 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4926 extract_bit_field_as_subreg to cases in which the extracted
4927 value is also a vector.
4929 * lra-constraints.c (process_alt_operands): Test for the equivalence
4930 substitutions when detecting a possible reload cycle.
4932 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4935 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4936 by default if flag_selective_schedling{,2}. Formatting fixes.
4938 PR rtl-optimization/83682
4939 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4940 if it has non-VECTOR_MODE element mode.
4941 (vec_duplicate_p): Likewise.
4944 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4945 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4947 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4950 * config/i386/i386-builtin.def
4951 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4952 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4953 Require also OPTION_MASK_ISA_AVX512F in addition to
4954 OPTION_MASK_ISA_GFNI.
4955 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4956 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4957 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4958 to OPTION_MASK_ISA_GFNI.
4959 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4960 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4961 OPTION_MASK_ISA_AVX512BW.
4962 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4963 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4964 addition to OPTION_MASK_ISA_GFNI.
4965 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4966 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4967 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4968 to OPTION_MASK_ISA_GFNI.
4969 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4970 a requirement for all ISAs rather than any of them with a few
4972 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4974 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4975 bitmasks to be enabled with 3 exceptions, instead of requiring any
4976 enabled ISA with lots of exceptions.
4977 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4978 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4979 Change avx512bw in isa attribute to avx512f.
4980 * config/i386/sgxintrin.h: Add license boilerplate.
4981 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4982 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4983 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4984 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4986 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4987 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4988 temporarily sse2 rather than sse if not enabled already.
4991 * config/i386/sse.md (VI248_VLBW): Rename to ...
4992 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4993 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4994 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4995 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4996 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4997 mode iterator instead of VI248_VLBW.
4999 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5001 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5002 (record_modified): Skip clobbers; add debug output.
5003 (param_change_prob): Use sreal frequencies.
5005 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5007 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5008 punt for user-aligned variables.
5010 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5012 * tree-chrec.c (chrec_contains_symbols): Return true for
5015 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5018 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5019 of (x|y) == x for BICS pattern.
5021 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5023 PR tree-optimization/83605
5024 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5025 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5028 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5030 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5031 * config/epiphany/rtems.h: New file.
5033 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5034 Uros Bizjak <ubizjak@gmail.com>
5037 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5038 QIreg_operand instead of register_operand predicate.
5039 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5040 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5041 comments instead of -fmitigate[-_]rop.
5043 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5046 * cgraphunit.c (symbol_table::compile): Switch to text_section
5047 before calling assembly_start debug hook.
5048 * run-rtl-passes.c (run_rtl_passes): Likewise.
5051 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5053 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5054 range_int_cst_p rather than !symbolic_range_p before calling
5055 extract_range_from_multiplicative_op_1.
5057 2018-01-04 Jeff Law <law@redhat.com>
5059 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5060 redundant test in assertion.
5062 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5064 * doc/rtl.texi: Document machine_mode wrapper classes.
5066 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5068 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5071 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5073 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5074 the VEC_PERM_EXPR fold to fail.
5076 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5079 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5080 to switched_sections.
5082 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5085 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5088 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5091 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5092 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5094 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5097 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5098 is BLKmode and bitpos not zero or mode change is needed.
5100 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5103 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5106 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5109 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5110 instead of MULT rtx. Update all corresponding splitters.
5112 (*ssub<modesuffix>): Ditto.
5114 (*cmp_sadd_di): Update split patterns.
5115 (*cmp_sadd_si): Ditto.
5116 (*cmp_sadd_sidi): Ditto.
5117 (*cmp_ssub_di): Ditto.
5118 (*cmp_ssub_si): Ditto.
5119 (*cmp_ssub_sidi): Ditto.
5120 * config/alpha/predicates.md (const23_operand): New predicate.
5121 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5122 Look for ASHIFT, not MULT inner operand.
5123 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5125 2018-01-04 Martin Liska <mliska@suse.cz>
5127 PR gcov-profile/83669
5128 * gcov.c (output_intermediate_file): Add version to intermediate
5130 * doc/gcov.texi: Document new field 'version' in intermediate
5131 file format. Fix location of '-k' option of gcov command.
5133 2018-01-04 Martin Liska <mliska@suse.cz>
5136 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5138 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5140 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5142 2018-01-03 Martin Sebor <msebor@redhat.com>
5144 PR tree-optimization/83655
5145 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5146 checking calls with invalid arguments.
5148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5150 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5151 (vectorizable_mask_load_store): Delete.
5152 (vectorizable_call): Return false for masked loads and stores.
5153 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5154 instead of gimple_assign_rhs1.
5155 (vectorizable_load): Handle IFN_MASK_LOAD.
5156 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5160 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5162 (vectorizable_mask_load_store): ...here.
5163 (vectorizable_load): ...and here.
5165 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5167 * tree-vect-stmts.c (vect_build_all_ones_mask)
5168 (vect_build_zero_merge_argument): New functions, split out from...
5169 (vectorizable_load): ...here.
5171 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5173 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5175 (vectorizable_mask_load_store): ...here.
5176 (vectorizable_store): ...and here.
5178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5180 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5182 (vectorizable_mask_load_store): ...here.
5184 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5186 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5187 (vect_model_store_cost): Take a vec_load_store_type instead of a
5189 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5190 (vect_model_store_cost): Take a vec_load_store_type instead of a
5192 (vectorizable_mask_load_store): Update accordingly.
5193 (vectorizable_store): Likewise.
5194 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5196 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5198 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5199 IFN_MASK_LOAD calls here rather than...
5200 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5202 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5203 Alan Hayward <alan.hayward@arm.com>
5204 David Sherwood <david.sherwood@arm.com>
5206 * expmed.c (extract_bit_field_1): For vector extracts,
5207 fall back to extract_bit_field_as_subreg if vec_extract
5210 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5211 Alan Hayward <alan.hayward@arm.com>
5212 David Sherwood <david.sherwood@arm.com>
5214 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5215 they are variable or constant sized.
5216 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5217 slots for constant-sized data.
5219 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5220 Alan Hayward <alan.hayward@arm.com>
5221 David Sherwood <david.sherwood@arm.com>
5223 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5224 handling COND_EXPRs with boolean comparisons, try to find a better
5225 basis for the mask type than the boolean itself.
5227 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5229 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5230 is calculated and how it can be overridden.
5231 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5232 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5234 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5238 Alan Hayward <alan.hayward@arm.com>
5239 David Sherwood <david.sherwood@arm.com>
5241 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5242 Remove the mode argument.
5243 (aarch64_simd_valid_immediate): Remove the mode and inverse
5245 * config/aarch64/iterators.md (bitsize): New iterator.
5246 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5247 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5248 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5249 aarch64_simd_valid_immediate.
5250 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5251 (aarch64_reg_or_bic_imm): Likewise.
5252 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5253 with an insn_type enum and msl with a modifier_type enum.
5254 Replace element_width with a scalar_mode. Change the shift
5255 to unsigned int. Add constructors for scalar_float_mode and
5256 scalar_int_mode elements.
5257 (aarch64_vect_float_const_representable_p): Delete.
5258 (aarch64_can_const_movi_rtx_p)
5259 (aarch64_simd_scalar_immediate_valid_for_move)
5260 (aarch64_simd_make_constant): Update call to
5261 aarch64_simd_valid_immediate.
5262 (aarch64_advsimd_valid_immediate_hs): New function.
5263 (aarch64_advsimd_valid_immediate): Likewise.
5264 (aarch64_simd_valid_immediate): Remove mode and inverse
5265 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5266 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5267 and aarch64_float_const_representable_p on the result.
5268 (aarch64_output_simd_mov_immediate): Remove mode argument.
5269 Update call to aarch64_simd_valid_immediate and use of
5270 simd_immediate_info.
5271 (aarch64_output_scalar_simd_mov_immediate): Update call
5274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5275 Alan Hayward <alan.hayward@arm.com>
5276 David Sherwood <david.sherwood@arm.com>
5278 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5279 (mode_nunits): Likewise CONST_MODE_NUNITS.
5280 * machmode.def (ADJUST_NUNITS): Document.
5281 * genmodes.c (mode_data::need_nunits_adj): New field.
5282 (blank_mode): Update accordingly.
5283 (adj_nunits): New variable.
5284 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5286 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5287 listed in adj_nunits.
5288 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5289 listed in adj_nunits. Don't emit case statements for such modes.
5290 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5291 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5292 nothing if adj_nunits is nonnull.
5293 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5294 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5295 (emit_mode_fbit): Update use of print_maybe_const_decl.
5296 (emit_move_size): Likewise. Treat the array as non-const
5298 (emit_mode_adjustments): Handle adj_nunits.
5300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5302 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5303 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5304 (VECTOR_MODES): Use it.
5305 (make_vector_modes): Take the prefix as an argument.
5307 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5308 Alan Hayward <alan.hayward@arm.com>
5309 David Sherwood <david.sherwood@arm.com>
5311 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5312 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5313 for MODE_VECTOR_BOOL.
5314 * machmode.def (VECTOR_BOOL_MODE): Document.
5315 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5316 (make_vector_bool_mode): New function.
5317 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5319 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5320 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5322 * stor-layout.c (int_mode_for_mode): Likewise.
5323 * tree.c (build_vector_type_for_mode): Likewise.
5324 * varasm.c (output_constant_pool_2): Likewise.
5325 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5326 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5327 for MODE_VECTOR_BOOL.
5328 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5329 of mode class checks.
5330 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5331 instead of a list of mode class checks.
5332 (expand_vector_scalar_condition): Likewise.
5333 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5336 Alan Hayward <alan.hayward@arm.com>
5337 David Sherwood <david.sherwood@arm.com>
5339 * machmode.h (mode_size): Change from unsigned short to
5341 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5342 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5343 or if measurement_type is not polynomial.
5344 (fixed_size_mode::includes_p): Check for constant-sized modes.
5345 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5346 return a poly_uint16 rather than an unsigned short.
5347 (emit_mode_size): Change the type of mode_size from unsigned short
5348 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5349 (emit_mode_adjustments): Cope with polynomial vector sizes.
5350 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5352 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5354 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5355 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5356 * caller-save.c (setup_save_areas): Likewise.
5357 (replace_reg_with_saved_mem): Likewise.
5358 * calls.c (emit_library_call_value_1): Likewise.
5359 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5360 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5361 (gen_lowpart_for_combine): Likewise.
5362 * convert.c (convert_to_integer_1): Likewise.
5363 * cse.c (equiv_constant, cse_insn): Likewise.
5364 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5365 (cselib_subst_to_values): Likewise.
5366 * dce.c (word_dce_process_block): Likewise.
5367 * df-problems.c (df_word_lr_mark_ref): Likewise.
5368 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5369 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5370 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5371 (rtl_for_decl_location): Likewise.
5372 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5373 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5374 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5375 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5376 (expand_expr_real_1): Likewise.
5377 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5378 (pad_below): Likewise.
5379 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5380 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5381 * ira.c (get_subreg_tracking_sizes): Likewise.
5382 * ira-build.c (ira_create_allocno_objects): Likewise.
5383 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5384 (ira_sort_regnos_for_alter_reg): Likewise.
5385 * ira-costs.c (record_operand_costs): Likewise.
5386 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5387 (resolve_simple_move): Likewise.
5388 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5389 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5390 (lra_constraints): Likewise.
5391 (CONST_POOL_OK_P): Reject variable-sized modes.
5392 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5393 (add_pseudo_to_slot, lra_spill): Likewise.
5394 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5395 * optabs-query.c (get_best_extraction_insn): Likewise.
5396 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5397 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5398 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5399 * recog.c (offsettable_address_addr_space_p): Likewise.
5400 * regcprop.c (maybe_mode_change): Likewise.
5401 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5402 * regrename.c (build_def_use): Likewise.
5403 * regstat.c (dump_reg_info): Likewise.
5404 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5405 (find_reloads, find_reloads_subreg_address): Likewise.
5406 * reload1.c (eliminate_regs_1): Likewise.
5407 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5408 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5409 (simplify_binary_operation_1, simplify_subreg): Likewise.
5410 * targhooks.c (default_function_arg_padding): Likewise.
5411 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5412 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5413 (verify_gimple_assign_ternary): Likewise.
5414 * tree-inline.c (estimate_move_cost): Likewise.
5415 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5416 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5417 (get_address_cost_ainc): Likewise.
5418 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5419 (vect_supportable_dr_alignment): Likewise.
5420 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5421 (vectorizable_reduction): Likewise.
5422 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5423 (vectorizable_operation, vectorizable_load): Likewise.
5424 * tree.c (build_same_sized_truth_vector_type): Likewise.
5425 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5426 * var-tracking.c (emit_note_insn_var_location): Likewise.
5427 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5428 (ADDR_VEC_ALIGN): Likewise.
5430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5431 Alan Hayward <alan.hayward@arm.com>
5432 David Sherwood <david.sherwood@arm.com>
5434 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5436 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5437 or if measurement_type is polynomial.
5438 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5439 * combine.c (make_extraction): Likewise.
5440 * dse.c (find_shift_sequence): Likewise.
5441 * dwarf2out.c (mem_loc_descriptor): Likewise.
5442 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5443 (extract_bit_field, extract_low_bits): Likewise.
5444 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5445 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5446 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5447 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5448 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5449 * reload.c (find_reloads): Likewise.
5450 * reload1.c (alter_reg): Likewise.
5451 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5452 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5453 * tree-if-conv.c (predicate_mem_writes): Likewise.
5454 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5455 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5456 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5457 * valtrack.c (dead_debug_insert_temp): Likewise.
5458 * varasm.c (mergeable_constant_section): Likewise.
5459 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5461 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5462 Alan Hayward <alan.hayward@arm.com>
5463 David Sherwood <david.sherwood@arm.com>
5465 * expr.c (expand_assignment): Cope with polynomial mode sizes
5466 when assigning to a CONCAT.
5468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5469 Alan Hayward <alan.hayward@arm.com>
5470 David Sherwood <david.sherwood@arm.com>
5472 * machmode.h (mode_precision): Change from unsigned short to
5474 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5476 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5477 or if measurement_type is not polynomial.
5478 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5479 in which the mode is already known to be a scalar_int_mode.
5480 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5481 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5483 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5484 for GET_MODE_PRECISION.
5485 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5486 for GET_MODE_PRECISION.
5487 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5489 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5490 (expand_field_assignment, make_extraction): Likewise.
5491 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5492 (get_last_value): Likewise.
5493 * convert.c (convert_to_integer_1): Likewise.
5494 * cse.c (cse_insn): Likewise.
5495 * expr.c (expand_expr_real_1): Likewise.
5496 * lra-constraints.c (simplify_operand_subreg): Likewise.
5497 * optabs-query.c (can_atomic_load_p): Likewise.
5498 * optabs.c (expand_atomic_load): Likewise.
5499 (expand_atomic_store): Likewise.
5500 * ree.c (combine_reaching_defs): Likewise.
5501 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5502 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5503 * tree.h (type_has_mode_precision_p): Likewise.
5504 * ubsan.c (instrument_si_overflow): Likewise.
5506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5507 Alan Hayward <alan.hayward@arm.com>
5508 David Sherwood <david.sherwood@arm.com>
5510 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5511 polynomial numbers of units.
5512 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5513 (valid_vector_subparts_p): New function.
5514 (build_vector_type): Remove temporary shim and take the number
5515 of units as a poly_uint64 rather than an int.
5516 (build_opaque_vector_type): Take the number of units as a
5517 poly_uint64 rather than an int.
5518 * tree.c (build_vector_from_ctor): Handle polynomial
5519 TYPE_VECTOR_SUBPARTS.
5520 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5521 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5522 (build_vector_from_val): If the number of units is variable,
5523 use build_vec_duplicate_cst for constant operands and
5524 VEC_DUPLICATE_EXPR otherwise.
5525 (make_vector_type): Remove temporary is_constant ().
5526 (build_vector_type, build_opaque_vector_type): Take the number of
5527 units as a poly_uint64 rather than an int.
5528 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5530 * cfgexpand.c (expand_debug_expr): Likewise.
5531 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5532 (store_constructor, expand_expr_real_1): Likewise.
5533 (const_scalar_mask_from_tree): Likewise.
5534 * fold-const-call.c (fold_const_reduction): Likewise.
5535 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5536 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5537 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5538 (fold_relational_const): Likewise.
5539 (native_interpret_vector): Likewise. Change the size from an
5540 int to an unsigned int.
5541 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5542 TYPE_VECTOR_SUBPARTS.
5543 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5544 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5545 duplicating a non-constant operand into a variable-length vector.
5546 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5547 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5548 * ipa-icf.c (sem_variable::equals): Likewise.
5549 * match.pd: Likewise.
5550 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5551 * print-tree.c (print_node): Likewise.
5552 * stor-layout.c (layout_type): Likewise.
5553 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5554 * tree-cfg.c (verify_gimple_comparison): Likewise.
5555 (verify_gimple_assign_binary): Likewise.
5556 (verify_gimple_assign_ternary): Likewise.
5557 (verify_gimple_assign_single): Likewise.
5558 * tree-pretty-print.c (dump_generic_node): Likewise.
5559 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5560 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5561 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5562 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5563 (vect_shift_permute_load_chain): Likewise.
5564 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5565 (expand_vector_condition, optimize_vector_constructor): Likewise.
5566 (lower_vec_perm, get_compute_type): Likewise.
5567 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5568 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5569 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5570 (vect_recog_mask_conversion_pattern): Likewise.
5571 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5572 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5573 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5574 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5575 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5576 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5577 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5578 (supportable_widening_operation): Likewise.
5579 (supportable_narrowing_operation): Likewise.
5580 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5582 * varasm.c (output_constant): Likewise.
5584 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5585 Alan Hayward <alan.hayward@arm.com>
5586 David Sherwood <david.sherwood@arm.com>
5588 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5589 so that both the length == 3 and length != 3 cases set up their
5590 own permute vectors. Add comments explaining why we know the
5591 number of elements is constant.
5592 (vect_permute_load_chain): Likewise.
5594 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5595 Alan Hayward <alan.hayward@arm.com>
5596 David Sherwood <david.sherwood@arm.com>
5598 * machmode.h (mode_nunits): Change from unsigned char to
5600 (ONLY_FIXED_SIZE_MODES): New macro.
5601 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5602 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5603 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5605 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5606 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5607 or if measurement_type is not polynomial.
5608 * genmodes.c (ZERO_COEFFS): New macro.
5609 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5611 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5612 Use ZERO_COEFFS when emitting initializers.
5613 * data-streamer.h (bp_pack_poly_value): New function.
5614 (bp_unpack_poly_value): Likewise.
5615 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5616 for GET_MODE_NUNITS.
5617 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5618 for GET_MODE_NUNITS.
5619 * tree.c (make_vector_type): Remove temporary shim and make
5620 the real function take the number of units as a poly_uint64
5622 (build_vector_type_for_mode): Handle polynomial nunits.
5623 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5624 * emit-rtl.c (const_vec_series_p_1): Likewise.
5625 (gen_rtx_CONST_VECTOR): Likewise.
5626 * fold-const.c (test_vec_duplicate_folding): Likewise.
5627 * genrecog.c (validate_pattern): Likewise.
5628 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5629 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5630 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5631 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5632 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5633 * rtlanal.c (subreg_get_info): Likewise.
5634 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5635 (vect_grouped_load_supported): Likewise.
5636 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5637 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5638 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5639 (simplify_const_unary_operation, simplify_binary_operation_1)
5640 (simplify_const_binary_operation, simplify_ternary_operation)
5641 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5642 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5643 instead of CONST_VECTOR_NUNITS.
5644 * varasm.c (output_constant_pool_2): Likewise.
5645 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5646 explicit-encoded elements in the XVEC for variable-length vectors.
5648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5650 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5652 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5653 Alan Hayward <alan.hayward@arm.com>
5654 David Sherwood <david.sherwood@arm.com>
5656 * coretypes.h (fixed_size_mode): Declare.
5657 (fixed_size_mode_pod): New typedef.
5658 * builtins.h (target_builtins::x_apply_args_mode)
5659 (target_builtins::x_apply_result_mode): Change type to
5660 fixed_size_mode_pod.
5661 * builtins.c (apply_args_size, apply_result_size, result_vector)
5662 (expand_builtin_apply_args_1, expand_builtin_apply)
5663 (expand_builtin_return): Update accordingly.
5665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5667 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5668 * cselib.c (cselib_hash_rtx): Likewise.
5669 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5670 CONST_VECTOR encoding.
5672 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5673 Jeff Law <law@redhat.com>
5676 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5677 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5678 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5679 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5682 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5683 explicitly probe *sp in a noreturn function if there were any callee
5684 register saves or frame pointer is needed.
5686 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5689 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5690 BLKmode for ternary, binary or unary expressions.
5693 * var-tracking.c (delete_vta_debug_insn): New inline function.
5694 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5695 insns from get_insns () to NULL instead of each bb separately.
5696 Use delete_vta_debug_insn. No longer static.
5697 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5698 delete_vta_debug_insns callers.
5699 * rtl.h (delete_vta_debug_insns): Declare.
5700 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5701 instead of variable_tracking_main.
5703 2018-01-03 Martin Sebor <msebor@redhat.com>
5705 PR tree-optimization/83603
5706 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5707 arguments past the endof the argument list in functions declared
5708 without a prototype.
5709 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5710 Avoid checking when arguments are null.
5712 2018-01-03 Martin Sebor <msebor@redhat.com>
5715 * doc/extend.texi (attribute const): Fix a typo.
5716 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5717 issuing -Wsuggest-attribute for void functions.
5719 2018-01-03 Martin Sebor <msebor@redhat.com>
5721 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5722 offset_int::from instead of wide_int::to_shwi.
5723 (maybe_diag_overlap): Remove assertion.
5724 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5725 * gimple-ssa-sprintf.c (format_directive): Same.
5726 (parse_directive): Same.
5727 (sprintf_dom_walker::compute_format_length): Same.
5728 (try_substitute_return_value): Same.
5730 2018-01-03 Jeff Law <law@redhat.com>
5733 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5734 non-constant residual for zero at runtime and avoid probing in
5735 that case. Reorganize code for trailing problem to mirror handling
5738 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5740 PR tree-optimization/83501
5741 * tree-ssa-strlen.c (get_string_cst): New.
5742 (handle_char_store): Call get_string_cst.
5744 2018-01-03 Martin Liska <mliska@suse.cz>
5746 PR tree-optimization/83593
5747 * tree-ssa-strlen.c: Include tree-cfg.h.
5748 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5749 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5750 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5752 (strlen_dom_walker::before_dom_children): Call
5753 gimple_purge_dead_eh_edges. Dump tranformation with details
5755 (strlen_dom_walker::before_dom_children): Update call by adding
5756 new argument cleanup_eh.
5757 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5759 2018-01-03 Martin Liska <mliska@suse.cz>
5762 * cif-code.def (VARIADIC_THUNK): New enum value.
5763 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5766 2018-01-03 Jan Beulich <jbeulich@suse.com>
5768 * sse.md (mov<mode>_internal): Tighten condition for when to use
5769 vmovdqu<ssescalarsize> for TI and OI modes.
5771 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5773 Update copyright years.
5775 2018-01-03 Martin Liska <mliska@suse.cz>
5778 * ipa-visibility.c (function_and_variable_visibility): Skip
5779 functions with noipa attribure.
5781 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5783 * gcc.c (process_command): Update copyright notice dates.
5784 * gcov-dump.c (print_version): Ditto.
5785 * gcov.c (print_version): Ditto.
5786 * gcov-tool.c (print_version): Ditto.
5787 * gengtype.c (create_file): Ditto.
5788 * doc/cpp.texi: Bump @copying's copyright year.
5789 * doc/cppinternals.texi: Ditto.
5790 * doc/gcc.texi: Ditto.
5791 * doc/gccint.texi: Ditto.
5792 * doc/gcov.texi: Ditto.
5793 * doc/install.texi: Ditto.
5794 * doc/invoke.texi: Ditto.
5796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5798 * vector-builder.h (vector_builder::m_full_nelts): Change from
5799 unsigned int to poly_uint64.
5800 (vector_builder::full_nelts): Update prototype accordingly.
5801 (vector_builder::new_vector): Likewise.
5802 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5803 (vector_builder::operator ==): Likewise.
5804 (vector_builder::finalize): Likewise.
5805 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5806 Take the number of elements as a poly_uint64 rather than an
5808 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5809 from unsigned int to poly_uint64.
5810 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5811 (vec_perm_indices::new_vector): Likewise.
5812 (vec_perm_indices::length): Likewise.
5813 (vec_perm_indices::nelts_per_input): Likewise.
5814 (vec_perm_indices::input_nelts): Likewise.
5815 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5816 number of elements per input as a poly_uint64 rather than an
5817 unsigned int. Use the original encoding for variable-length
5818 vectors, rather than clamping each individual element.
5819 For the second and subsequent elements in each pattern,
5820 clamp the step and base before clamping their sum.
5821 (vec_perm_indices::series_p): Handle polynomial element counts.
5822 (vec_perm_indices::all_in_range_p): Likewise.
5823 (vec_perm_indices_to_tree): Likewise.
5824 (vec_perm_indices_to_rtx): Likewise.
5825 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5826 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5827 (tree_vector_builder::new_binary_operation): Handle polynomial
5828 element counts. Return false if we need to know the number
5829 of elements at compile time.
5830 * fold-const.c (fold_vec_perm): Punt if the number of elements
5831 isn't known at compile time.
5833 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5835 * vec-perm-indices.h (vec_perm_builder): Change element type
5836 from HOST_WIDE_INT to poly_int64.
5837 (vec_perm_indices::element_type): Update accordingly.
5838 (vec_perm_indices::clamp): Handle polynomial element_types.
5839 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5840 (vec_perm_indices::all_in_range_p): Likewise.
5841 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5843 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5844 polynomial vec_perm_indices element types.
5845 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5846 * fold-const.c (fold_vec_perm): Likewise.
5847 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5848 * tree-vect-generic.c (lower_vec_perm): Likewise.
5849 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5850 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5851 element type to HOST_WIDE_INT.
5853 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5854 Alan Hayward <alan.hayward@arm.com>
5855 David Sherwood <david.sherwood@arm.com>
5857 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5858 rather than an int. Use plus_constant.
5859 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5860 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5862 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5863 Alan Hayward <alan.hayward@arm.com>
5864 David Sherwood <david.sherwood@arm.com>
5866 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5867 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
5874 mode sizes. Require a constant size for BLKmode parameters
5875 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5876 forces a parameter to be padded at the lsb end in order to
5877 fill a complete number of words, require the parameter size
5878 to be ordered wrt UNITS_PER_WORD.
5880 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5881 Alan Hayward <alan.hayward@arm.com>
5882 David Sherwood <david.sherwood@arm.com>
5884 * reload1.c (spill_stack_slot_width): Change element type
5885 from unsigned int to poly_uint64_pod.
5886 (alter_reg): Treat mode sizes as polynomial.
5888 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5889 Alan Hayward <alan.hayward@arm.com>
5890 David Sherwood <david.sherwood@arm.com>
5892 * reload.c (complex_word_subreg_p): New function.
5893 (reload_inner_reg_of_subreg, push_reload): Use it.
5895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5896 Alan Hayward <alan.hayward@arm.com>
5897 David Sherwood <david.sherwood@arm.com>
5899 * lra-constraints.c (process_alt_operands): Reject matched
5900 operands whose sizes aren't ordered.
5901 (match_reload): Refer to this check here.
5903 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5904 Alan Hayward <alan.hayward@arm.com>
5905 David Sherwood <david.sherwood@arm.com>
5907 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5908 that the mode size is in the set {1, 2, 4, 8, 16}.
5910 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5911 Alan Hayward <alan.hayward@arm.com>
5912 David Sherwood <david.sherwood@arm.com>
5914 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5915 Use plus_constant instead of gen_rtx_PLUS.
5917 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5918 Alan Hayward <alan.hayward@arm.com>
5919 David Sherwood <david.sherwood@arm.com>
5921 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5922 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5923 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5924 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5925 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5926 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5927 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5928 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5929 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5930 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5932 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5933 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5934 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5935 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5936 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5937 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5938 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5939 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5940 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5941 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5943 * expr.c (emit_move_resolve_push): Treat the input and result
5944 of PUSH_ROUNDING as a poly_int64.
5945 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5946 (emit_push_insn): Likewise.
5947 * lra-eliminations.c (mark_not_eliminable): Likewise.
5948 * recog.c (push_operand): Likewise.
5949 * reload1.c (elimination_effects): Likewise.
5950 * rtlanal.c (nonzero_bits1): Likewise.
5951 * calls.c (store_one_arg): Likewise. Require the padding to be
5952 known at compile time.
5954 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5955 Alan Hayward <alan.hayward@arm.com>
5956 David Sherwood <david.sherwood@arm.com>
5958 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5959 Use plus_constant instead of gen_rtx_PLUS.
5961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5962 Alan Hayward <alan.hayward@arm.com>
5963 David Sherwood <david.sherwood@arm.com>
5965 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5969 Alan Hayward <alan.hayward@arm.com>
5970 David Sherwood <david.sherwood@arm.com>
5972 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5973 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5974 via stack temporaries. Treat the mode size as polynomial too.
5976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5977 Alan Hayward <alan.hayward@arm.com>
5978 David Sherwood <david.sherwood@arm.com>
5980 * expr.c (expand_expr_real_2): When handling conversions involving
5981 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5982 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5983 as a poly_uint64 too.
5985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5986 Alan Hayward <alan.hayward@arm.com>
5987 David Sherwood <david.sherwood@arm.com>
5989 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5992 Alan Hayward <alan.hayward@arm.com>
5993 David Sherwood <david.sherwood@arm.com>
5995 * combine.c (can_change_dest_mode): Handle polynomial
5996 REGMODE_NATURAL_SIZE.
5997 * expmed.c (store_bit_field_1): Likewise.
5998 * expr.c (store_constructor): Likewise.
5999 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6000 and polynomial REGMODE_NATURAL_SIZE.
6001 (gen_lowpart_common): Likewise.
6002 * reginfo.c (record_subregs_of_mode): Likewise.
6003 * rtlanal.c (read_modify_subreg_p): Likewise.
6005 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6006 Alan Hayward <alan.hayward@arm.com>
6007 David Sherwood <david.sherwood@arm.com>
6009 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6010 numbers of elements.
6012 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6013 Alan Hayward <alan.hayward@arm.com>
6014 David Sherwood <david.sherwood@arm.com>
6016 * match.pd: Cope with polynomial numbers of vector elements.
6018 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6019 Alan Hayward <alan.hayward@arm.com>
6020 David Sherwood <david.sherwood@arm.com>
6022 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6023 in a POINTER_PLUS_EXPR.
6025 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6026 Alan Hayward <alan.hayward@arm.com>
6027 David Sherwood <david.sherwood@arm.com>
6029 * omp-simd-clone.c (simd_clone_subparts): New function.
6030 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6031 (ipa_simd_modify_function_body): Likewise.
6033 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6034 Alan Hayward <alan.hayward@arm.com>
6035 David Sherwood <david.sherwood@arm.com>
6037 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6038 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6039 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6040 (expand_vector_condition, vector_element): Likewise.
6041 (subparts_gt): New function.
6042 (get_compute_type): Use subparts_gt.
6043 (count_type_subparts): Delete.
6044 (expand_vector_operations_1): Use subparts_gt instead of
6045 count_type_subparts.
6047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6048 Alan Hayward <alan.hayward@arm.com>
6049 David Sherwood <david.sherwood@arm.com>
6051 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6052 (vect_compile_time_alias): ...this new function. Do the calculation
6053 on poly_ints rather than trees.
6054 (vect_prune_runtime_alias_test_list): Update call accordingly.
6056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6057 Alan Hayward <alan.hayward@arm.com>
6058 David Sherwood <david.sherwood@arm.com>
6060 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6062 (vect_schedule_slp_instance): Likewise.
6064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6065 Alan Hayward <alan.hayward@arm.com>
6066 David Sherwood <david.sherwood@arm.com>
6068 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6069 constant and extern definitions for variable-length vectors.
6070 (vect_get_constant_vectors): Note that the number of units
6071 is known to be constant.
6073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6074 Alan Hayward <alan.hayward@arm.com>
6075 David Sherwood <david.sherwood@arm.com>
6077 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6078 of units as polynomial. Choose between WIDE and NARROW based
6081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6082 Alan Hayward <alan.hayward@arm.com>
6083 David Sherwood <david.sherwood@arm.com>
6085 * tree-vect-stmts.c (simd_clone_subparts): New function.
6086 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6089 Alan Hayward <alan.hayward@arm.com>
6090 David Sherwood <david.sherwood@arm.com>
6092 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6093 vectors as polynomial. Use build_index_vector for
6096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6097 Alan Hayward <alan.hayward@arm.com>
6098 David Sherwood <david.sherwood@arm.com>
6100 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6101 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6102 for variable-length vectors.
6103 (vectorizable_mask_load_store): Treat the number of units as
6104 polynomial, asserting that it is constant if the condition has
6105 already been enforced.
6106 (vectorizable_store, vectorizable_load): Likewise.
6108 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6109 Alan Hayward <alan.hayward@arm.com>
6110 David Sherwood <david.sherwood@arm.com>
6112 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6113 of units as polynomial. Punt if we can't tell at compile time
6114 which vector contains the final result.
6116 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6117 Alan Hayward <alan.hayward@arm.com>
6118 David Sherwood <david.sherwood@arm.com>
6120 * tree-vect-loop.c (vectorizable_induction): Treat the number
6121 of units as polynomial. Punt on SLP inductions. Use an integer
6122 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6123 cast of such a series for variable-length floating-point
6126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6127 Alan Hayward <alan.hayward@arm.com>
6128 David Sherwood <david.sherwood@arm.com>
6130 * tree.h (build_index_vector): Declare.
6131 * tree.c (build_index_vector): New function.
6132 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6133 of units as polynomial, forcibly converting it to a constant if
6134 vectorizable_reduction has already enforced the condition.
6135 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6136 to create a {1,2,3,...} vector.
6137 (vectorizable_reduction): Treat the number of units as polynomial.
6138 Choose vectype_in based on the largest scalar element size rather
6139 than the smallest number of units. Enforce the restrictions
6142 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6143 Alan Hayward <alan.hayward@arm.com>
6144 David Sherwood <david.sherwood@arm.com>
6146 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6147 number of units as polynomial.
6149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6150 Alan Hayward <alan.hayward@arm.com>
6151 David Sherwood <david.sherwood@arm.com>
6153 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6154 * target.def (autovectorize_vector_sizes): Return the vector sizes
6155 by pointer, using vector_sizes rather than a bitmask.
6156 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6157 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6158 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6160 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6161 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6162 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6163 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6164 * omp-general.c (omp_max_vf): Likewise.
6165 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6166 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6167 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6168 * tree-vect-slp.c (vect_slp_bb): Likewise.
6169 * doc/tm.texi: Regenerate.
6170 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6172 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6173 the vector size as a poly_uint64 rather than an unsigned int.
6174 (current_vector_size): Change from an unsigned int to a poly_uint64.
6175 (get_vectype_for_scalar_type): Update accordingly.
6176 * tree.h (build_truth_vector_type): Take the size and number of
6177 units as a poly_uint64 rather than an unsigned int.
6178 (build_vector_type): Add a temporary overload that takes
6179 the number of units as a poly_uint64 rather than an unsigned int.
6180 * tree.c (make_vector_type): Likewise.
6181 (build_truth_vector_type): Take the number of units as a poly_uint64
6182 rather than an unsigned int.
6184 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6185 Alan Hayward <alan.hayward@arm.com>
6186 David Sherwood <david.sherwood@arm.com>
6188 * target.def (get_mask_mode): Take the number of units and length
6189 as poly_uint64s rather than unsigned ints.
6190 * targhooks.h (default_get_mask_mode): Update accordingly.
6191 * targhooks.c (default_get_mask_mode): Likewise.
6192 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6193 * doc/tm.texi: Regenerate.
6195 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6196 Alan Hayward <alan.hayward@arm.com>
6197 David Sherwood <david.sherwood@arm.com>
6199 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6200 * omp-general.c (omp_max_vf): Likewise.
6201 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6202 (expand_omp_simd): Handle polynomial safelen.
6203 * omp-low.c (omplow_simd_context): Add a default constructor.
6204 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6205 (lower_rec_simd_input_clauses): Update accordingly.
6206 (lower_rec_input_clauses): Likewise.
6208 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6209 Alan Hayward <alan.hayward@arm.com>
6210 David Sherwood <david.sherwood@arm.com>
6212 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6213 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6214 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6215 (vect_analyze_slp_cost): Likewise.
6216 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6217 (vect_model_load_cost): Likewise.
6219 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6220 Alan Hayward <alan.hayward@arm.com>
6221 David Sherwood <david.sherwood@arm.com>
6223 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6224 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6225 from an unsigned int * to a poly_uint64_pod *.
6226 (calculate_unrolling_factor): New function.
6227 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6230 Alan Hayward <alan.hayward@arm.com>
6231 David Sherwood <david.sherwood@arm.com>
6233 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6234 from an unsigned int to a poly_uint64.
6235 (_loop_vec_info::slp_unrolling_factor): Likewise.
6236 (_loop_vec_info::vectorization_factor): Change from an int
6238 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6239 (vect_get_num_vectors): New function.
6240 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6241 (vect_get_num_copies): Use vect_get_num_vectors.
6242 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6243 to an unsigned int *.
6244 (vect_analyze_data_refs): Change min_vf from an int * to a
6246 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6247 than an unsigned HOST_WIDE_INT.
6248 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6249 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6250 to an unsigned int *.
6251 (vect_analyze_data_ref_dependences): Likewise.
6252 (vect_compute_data_ref_alignment): Handle polynomial vf.
6253 (vect_enhance_data_refs_alignment): Likewise.
6254 (vect_prune_runtime_alias_test_list): Likewise.
6255 (vect_shift_permute_load_chain): Likewise.
6256 (vect_supportable_dr_alignment): Likewise.
6257 (dependence_distance_ge_vf): Take the vectorization factor as a
6258 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6259 (vect_analyze_data_refs): Change min_vf from an int * to a
6261 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6262 vfm1 as a poly_uint64 rather than an int. Make the same change
6263 for the returned bound_scalar.
6264 (vect_gen_vector_loop_niters): Handle polynomial vf.
6265 (vect_do_peeling): Likewise. Update call to
6266 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6267 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6269 * tree-vect-loop.c (vect_determine_vectorization_factor)
6270 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6271 (vect_get_known_peeling_cost): Likewise.
6272 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6273 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6274 (vect_transform_loop): Likewise. Use the lowest possible VF when
6275 updating the upper bounds of the loop.
6276 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6278 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6279 polynomial unroll factors.
6280 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6281 (vect_make_slp_decision): Likewise.
6282 (vect_supported_load_permutation_p): Likewise, and polynomial
6284 (vect_analyze_slp_cost): Handle polynomial vf.
6285 (vect_slp_analyze_node_operations): Likewise.
6286 (vect_slp_analyze_bb_1): Likewise.
6287 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6288 than an unsigned HOST_WIDE_INT.
6289 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6290 (vectorizable_load): Handle polynomial vf.
6291 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6293 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6295 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6296 Alan Hayward <alan.hayward@arm.com>
6297 David Sherwood <david.sherwood@arm.com>
6299 * match.pd: Handle bit operations involving three constants
6300 and try to fold one pair.
6302 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6304 * tree-vect-loop-manip.c: Include gimple-fold.h.
6305 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6306 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6307 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6308 Add a path that uses a step of VF instead of 1, but disable it
6310 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6311 and niters_no_overflow parameters. Update calls to
6312 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6313 Create a new SSA name if the latter choses to use a ste other
6314 than zero, and return it via niters_vector_mult_vf_var.
6315 * tree-vect-loop.c (vect_transform_loop): Update calls to
6316 vect_do_peeling, vect_gen_vector_loop_niters and
6317 slpeel_make_loop_iterate_ntimes.
6318 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6319 (vect_gen_vector_loop_niters): Update declarations after above changes.
6321 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6323 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6324 128-bit round to integer instructions.
6325 (ceil<mode>2): Likewise.
6326 (btrunc<mode>2): Likewise.
6327 (round<mode>2): Likewise.
6329 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6331 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6332 unaligned VSX load/store on P8/P9.
6333 (expand_block_clear): Allow the use of unaligned VSX
6334 load/store on P8/P9.
6336 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6338 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6340 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6341 swap associated with both a load and a store.
6343 2018-01-02 Andrew Waterman <andrew@sifive.com>
6345 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6346 * config/riscv/riscv.md (clear_cache): Use it.
6348 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6350 * web.c: Remove out-of-date comment.
6352 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6354 * expr.c (fixup_args_size_notes): Check that any existing
6355 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6356 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6357 (emit_single_push_insn): ...here.
6359 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6361 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6362 (const_vector_encoded_nelts): New function.
6363 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6364 (const_vector_int_elt, const_vector_elt): Declare.
6365 * emit-rtl.c (const_vector_int_elt_1): New function.
6366 (const_vector_elt): Likewise.
6367 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6368 of CONST_VECTOR_ELT.
6370 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6372 * expr.c: Include rtx-vector-builder.h.
6373 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6374 directly on the tree encoding.
6375 (const_vector_from_tree): Likewise.
6376 * optabs.c: Include rtx-vector-builder.h.
6377 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6378 sequence of "u" values.
6379 * vec-perm-indices.c: Include rtx-vector-builder.h.
6380 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6381 directly on the vec_perm_indices encoding.
6383 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6385 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6386 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6387 * rtx-vector-builder.h: New file.
6388 * rtx-vector-builder.c: Likewise.
6389 * rtl.h (rtx_def::u2): Add a const_vector field.
6390 (CONST_VECTOR_NPATTERNS): New macro.
6391 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6392 (CONST_VECTOR_DUPLICATE_P): Likewise.
6393 (CONST_VECTOR_STEPPED_P): Likewise.
6394 (CONST_VECTOR_ENCODED_ELT): Likewise.
6395 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6396 (unwrap_const_vec_duplicate): Likewise.
6397 (const_vec_series_p): Check for a non-duplicated vector encoding.
6398 Say that the function only returns true for integer vectors.
6399 * emit-rtl.c: Include rtx-vector-builder.h.
6400 (gen_const_vec_duplicate_1): Delete.
6401 (gen_const_vector): Call gen_const_vec_duplicate instead of
6402 gen_const_vec_duplicate_1.
6403 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6404 (gen_const_vec_duplicate): Use rtx_vector_builder.
6405 (gen_const_vec_series): Likewise.
6406 (gen_rtx_CONST_VECTOR): Likewise.
6407 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6408 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6409 Build a new vector rather than modifying a CONST_VECTOR in-place.
6410 (handle_special_swappables): Update call accordingly.
6411 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6412 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6413 Build a new vector rather than modifying a CONST_VECTOR in-place.
6414 (handle_special_swappables): Update call accordingly.
6416 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6418 * simplify-rtx.c (simplify_const_binary_operation): Use
6419 CONST_VECTOR_ELT instead of XVECEXP.
6421 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6423 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6424 the selector elements to be different from the data elements
6425 if the selector is a VECTOR_CST.
6426 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6427 ssizetype for the selector.
6429 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6431 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6432 before testing each element individually.
6433 * tree-vect-generic.c (lower_vec_perm): Likewise.
6435 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6437 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6438 * selftest-run-tests.c (selftest::run_tests): Call it.
6439 * vector-builder.h (vector_builder::operator ==): New function.
6440 (vector_builder::operator !=): Likewise.
6441 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6442 (vec_perm_indices::all_from_input_p): New function.
6443 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6444 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6445 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6446 instead of reading the VECTOR_CST directly. Detect whether both
6447 vector inputs are the same before constructing the vec_perm_indices,
6448 and update the number of inputs argument accordingly. Use the
6449 utility functions added above. Only construct sel2 if we need to.
6451 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6453 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6454 the broadcast of the low byte.
6455 (expand_mult_highpart): Use an explicit encoding for the permutes.
6456 * optabs-query.c (can_mult_highpart_p): Likewise.
6457 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6458 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6459 (vectorizable_bswap): Likewise.
6460 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6461 explicit encoding for the power-of-2 permutes.
6462 (vect_permute_store_chain): Likewise.
6463 (vect_grouped_load_supported): Likewise.
6464 (vect_permute_load_chain): Likewise.
6466 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6468 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6469 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6470 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6471 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6472 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6473 (vect_gen_perm_mask_any): Likewise.
6475 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6477 * int-vector-builder.h: New file.
6478 * vec-perm-indices.h: Include int-vector-builder.h.
6479 (vec_perm_indices): Redefine as an int_vector_builder.
6480 (auto_vec_perm_indices): Delete.
6481 (vec_perm_builder): Redefine as a stand-alone class.
6482 (vec_perm_indices::vec_perm_indices): New function.
6483 (vec_perm_indices::clamp): Likewise.
6484 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6485 (vec_perm_indices::new_vector): New function.
6486 (vec_perm_indices::new_expanded_vector): Update for new
6487 vec_perm_indices class.
6488 (vec_perm_indices::rotate_inputs): New function.
6489 (vec_perm_indices::all_in_range_p): Operate directly on the
6490 encoded form, without computing elided elements.
6491 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6492 encoding. Update for new vec_perm_indices class.
6493 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6494 the given vec_perm_builder.
6495 (expand_vec_perm_var): Update vec_perm_builder constructor.
6496 (expand_mult_highpart): Use vec_perm_builder instead of
6497 auto_vec_perm_indices.
6498 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6499 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6500 or double series encoding as appropriate.
6501 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6502 vec_perm_indices instead of auto_vec_perm_indices.
6503 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6504 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6505 (vect_permute_store_chain): Likewise.
6506 (vect_grouped_load_supported): Likewise.
6507 (vect_permute_load_chain): Likewise.
6508 (vect_shift_permute_load_chain): Likewise.
6509 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6510 (vect_transform_slp_perm_load): Likewise.
6511 (vect_schedule_slp_instance): Likewise.
6512 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6513 (vectorizable_mask_load_store): Likewise.
6514 (vectorizable_bswap): Likewise.
6515 (vectorizable_store): Likewise.
6516 (vectorizable_load): Likewise.
6517 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6518 vec_perm_indices instead of auto_vec_perm_indices. Use
6519 tree_to_vec_perm_builder to read the vector from a tree.
6520 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6521 vec_perm_builder instead of a vec_perm_indices.
6522 (have_whole_vector_shift): Use vec_perm_builder and
6523 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6524 truncation to calc_vec_perm_mask_for_shift.
6525 (vect_create_epilog_for_reduction): Likewise.
6526 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6527 from auto_vec_perm_indices to vec_perm_indices.
6528 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6529 instead of changing individual elements.
6530 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6531 the vector in d.perm.
6532 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6533 from auto_vec_perm_indices to vec_perm_indices.
6534 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6535 instead of changing individual elements.
6536 (arm_vectorize_vec_perm_const): Use new_vector to install
6537 the vector in d.perm.
6538 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6539 Update vec_perm_builder constructor.
6540 (rs6000_expand_interleave): Likewise.
6541 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6542 (rs6000_expand_interleave): Likewise.
6544 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6546 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6547 to qimode could truncate the indices.
6548 * optabs.c (expand_vec_perm_var): Likewise.
6550 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6552 * Makefile.in (OBJS): Add vec-perm-indices.o.
6553 * vec-perm-indices.h: New file.
6554 * vec-perm-indices.c: Likewise.
6555 * target.h (vec_perm_indices): Replace with a forward class
6557 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6558 * optabs.h: Include vec-perm-indices.h.
6559 (expand_vec_perm): Delete.
6560 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6561 (expand_vec_perm_const): Declare.
6562 * target.def (vec_perm_const_ok): Replace with...
6563 (vec_perm_const): ...this new hook.
6564 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6565 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6566 * doc/tm.texi: Regenerate.
6567 * optabs.def (vec_perm_const): Delete.
6568 * doc/md.texi (vec_perm_const): Likewise.
6569 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6570 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6571 expand_vec_perm for constant permutation vectors. Assert that
6572 the mode of variable permutation vectors is the integer equivalent
6573 of the mode that is being permuted.
6574 * optabs-query.h (selector_fits_mode_p): Declare.
6575 * optabs-query.c: Include vec-perm-indices.h.
6576 (selector_fits_mode_p): New function.
6577 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6578 is defined, instead of checking whether the vec_perm_const_optab
6579 exists. Use targetm.vectorize.vec_perm_const instead of
6580 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6581 fit in the vector mode before using a variable permute.
6582 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6583 vec_perm_indices instead of an rtx.
6584 (expand_vec_perm): Replace with...
6585 (expand_vec_perm_const): ...this new function. Take the selector
6586 as a vec_perm_indices rather than an rtx. Also take the mode of
6587 the selector. Update call to shift_amt_for_vec_perm_mask.
6588 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6589 Use vec_perm_indices::new_expanded_vector to expand the original
6590 selector into bytes. Check whether the indices fit in the vector
6591 mode before using a variable permute.
6592 (expand_vec_perm_var): Make global.
6593 (expand_mult_highpart): Use expand_vec_perm_const.
6594 * fold-const.c: Includes vec-perm-indices.h.
6595 * tree-ssa-forwprop.c: Likewise.
6596 * tree-vect-data-refs.c: Likewise.
6597 * tree-vect-generic.c: Likewise.
6598 * tree-vect-loop.c: Likewise.
6599 * tree-vect-slp.c: Likewise.
6600 * tree-vect-stmts.c: Likewise.
6601 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6603 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6604 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6605 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6606 (aarch64_vectorize_vec_perm_const): ...this new function.
6607 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6608 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6609 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6610 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6611 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6612 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6613 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6615 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6616 check for NEON modes.
6617 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6618 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6619 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6620 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6622 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6623 the old VEC_PERM_CONST conditions.
6624 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6625 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6626 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6627 (ia64_vectorize_vec_perm_const_ok): Merge into...
6628 (ia64_vectorize_vec_perm_const): ...this new function.
6629 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6630 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6631 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6632 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6633 * config/mips/mips.c (mips_expand_vec_perm_const)
6634 (mips_vectorize_vec_perm_const_ok): Merge into...
6635 (mips_vectorize_vec_perm_const): ...this new function.
6636 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6637 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6638 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6639 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6640 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6641 (rs6000_expand_vec_perm_const): Delete.
6642 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6644 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6645 (altivec_expand_vec_perm_const_le): Take each operand individually.
6646 Operate on constant selectors rather than rtxes.
6647 (altivec_expand_vec_perm_const): Likewise. Update call to
6648 altivec_expand_vec_perm_const_le.
6649 (rs6000_expand_vec_perm_const): Delete.
6650 (rs6000_vectorize_vec_perm_const_ok): Delete.
6651 (rs6000_vectorize_vec_perm_const): New function.
6652 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6653 an element count and rtx array.
6654 (rs6000_expand_extract_even): Update call accordingly.
6655 (rs6000_expand_interleave): Likewise.
6656 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6657 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6658 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6659 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6660 (rs6000_expand_vec_perm_const): Delete.
6661 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6662 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6663 (altivec_expand_vec_perm_const_le): Take each operand individually.
6664 Operate on constant selectors rather than rtxes.
6665 (altivec_expand_vec_perm_const): Likewise. Update call to
6666 altivec_expand_vec_perm_const_le.
6667 (rs6000_expand_vec_perm_const): Delete.
6668 (rs6000_vectorize_vec_perm_const_ok): Delete.
6669 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6670 reference to the SPE evmerge intructions.
6671 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6672 an element count and rtx array.
6673 (rs6000_expand_extract_even): Update call accordingly.
6674 (rs6000_expand_interleave): Likewise.
6675 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6676 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6678 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6680 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6682 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6683 vector mode and that that mode matches the mode of the data
6685 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6686 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6687 directly using expand_vec_perm_1 when forcing selectors into
6689 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6691 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6693 * optabs-query.h (can_vec_perm_p): Delete.
6694 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6695 * optabs-query.c (can_vec_perm_p): Split into...
6696 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6697 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6698 particular selector is valid.
6699 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6700 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6701 (vect_grouped_load_supported): Likewise.
6702 (vect_shift_permute_load_chain): Likewise.
6703 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6704 (vect_transform_slp_perm_load): Likewise.
6705 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6706 (vectorizable_bswap): Likewise.
6707 (vect_gen_perm_mask_checked): Likewise.
6708 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6709 implementations of variable permutation vectors into account
6710 when deciding which selector to use.
6711 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6712 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6713 with a false third argument.
6714 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6715 to test whether the constant selector is valid and can_vec_perm_var_p
6716 to test whether a variable selector is valid.
6718 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6720 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6721 * optabs-query.c (can_vec_perm_p): Likewise.
6722 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6723 instead of vec_perm_indices.
6724 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6725 (vect_gen_perm_mask_checked): Likewise,
6726 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6727 (vect_gen_perm_mask_checked): Likewise,
6729 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6731 * optabs-query.h (qimode_for_vec_perm): Declare.
6732 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6733 (qimode_for_vec_perm): ...this new function.
6734 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6736 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6738 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6739 does not have a conditional at the top.
6741 2018-01-02 Richard Biener <rguenther@suse.de>
6743 * ipa-inline.c (big_speedup_p): Fix expression.
6745 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6748 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6751 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6755 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6756 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6757 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6758 cond_taken_branch_cost 3->4.
6760 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6762 PR tree-optimization/83581
6763 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6764 TODO_cleanup_cfg if any changes have been made.
6767 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6768 convert_modes if target mode has the right side, but different mode
6772 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6773 last argument when extracting from CONCAT. If either from_real or
6774 from_imag is NULL, use expansion through memory. If result is not
6775 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6776 the parts directly to inner mode, if even that fails, use expansion
6780 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6781 check for bswap in mode rather than HImode and use that in expand_unop
6784 Copyright (C) 2018 Free Software Foundation, Inc.
6786 Copying and distribution of this file, with or without modification,
6787 are permitted in any medium without royalty provided the copyright
6788 notice and this notice are preserved.