1 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
3 * common.opt (gvariable-location-views): New.
4 (gvariable-location-views=incompat5): New.
7 * configure.ac: Test assembler for view support.
8 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
9 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
10 * dwarf2out.c (var_loc_view): New typedef.
11 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
12 (dwarf2out_locviews_in_attribute): New.
13 (dwarf2out_locviews_in_loclist): New.
14 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
15 (enum dw_line_info_opcode): Add LI_adv_address.
16 (struct dw_line_info_table): Add view.
17 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
18 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
19 (zero_view_p): New variable.
20 (ZERO_VIEW_P): New macro.
21 (output_asm_line_debug_info): New.
22 (struct var_loc_node): Add view.
23 (add_AT_view_list, AT_loc_list): New.
24 (add_var_loc_to_decl): Add view param. Test it against last.
25 (new_loc_list): Add view params. Record them.
26 (AT_loc_list_ptr): Handle loc and view lists.
27 (view_list_to_loc_list_val_node): New.
28 (print_dw_val): Handle dw_val_class_view_list.
29 (size_of_die): Likewise.
30 (value_format): Likewise.
31 (loc_list_has_views): New.
32 (gen_llsym): Set vl_symbol too.
33 (maybe_gen_llsym, skip_loc_list_entry): New.
34 (dwarf2out_maybe_output_loclist_view_pair): New.
35 (output_loc_list): Output view list or entries too.
36 (output_view_list_offset): New.
37 (output_die): Handle dw_val_class_view_list.
38 (output_dwarf_version): New.
39 (output_compilation_unit_header): Use it.
40 (output_skeleton_debug_sections): Likewise.
41 (output_rnglists, output_line_info): Likewise.
42 (output_pubnames, output_aranges): Update version comments.
43 (output_one_line_info_table): Output view numbers in asm comments.
44 (dw_loc_list): Determine current endview, pass it to new_loc_list.
46 (loc_list_from_tree_1): Adjust.
47 (add_AT_location_description): Create view list attribute if
48 needed, check it's absent otherwise.
49 (convert_cfa_to_fb_loc_list): Adjust.
50 (maybe_emit_file): Call output_asm_line_debug_info for test.
51 (dwarf2out_var_location): Reset views as needed. Precompute
52 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
54 (new_line_info_table): Reset next view.
55 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
56 (dwarf2out_source_line): Likewise. Output view resets and labels to
57 the assembler, or select appropriate line info opcodes.
58 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
59 (optimize_string_length): Catch it. Adjust.
60 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
61 dw_val_class_view_list, and remove it if no longer needed.
62 (hash_loc_list): Hash view numbers.
63 (loc_list_hasher::equal): Compare them.
64 (optimize_location_lists): Check whether a view list symbol is
65 needed, and whether the locview attribute is present, and
66 whether they match. Remove the locview attribute if no longer
68 (index_location_lists): Call skip_loc_list_entry for test.
69 (dwarf2out_finish): Call output_asm_line_debug_info for test.
70 Use output_dwarf_version.
71 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
72 (struct dw_val_node): Add val_view_list.
73 * final.c (SEEN_NEXT_VIEW): New.
74 (set_next_view_needed): New.
75 (clear_next_view_needed): New.
76 (maybe_output_next_view): New.
77 (final_start_function): Rename to...
78 (final_start_function_1): ... this. Take pointer to FIRST,
79 add SEEN parameter. Emit param bindings in the initial view.
80 (final_start_function): Reintroduce SEEN-less interface.
82 (final_1): ... this. Take SEEN parameter. Output final pending
84 (final): Reintroduce seen-less interface.
85 (final_scan_insn): Output pending next view before switching
86 sections or ending a block. Mark the next view as needed when
87 outputting variable locations. Notify debug backend of section
88 changes, and of location view changes.
89 (rest_of_handle_final): Adjust.
90 * toplev.c (process_options): Autodetect value for debug variable
91 location views option. Warn on incompat5 without -gdwarf-5.
92 * doc/invoke.texi (gvariable-location-views): New.
93 (gvariable-location-views=incompat5): New.
94 (gno-variable-location-views): New.
96 2018-02-08 David Malcolm <dmalcolm@redhat.com>
98 PR tree-optimization/84136
99 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
100 that the result of find_edge is non-NULL.
102 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
105 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
106 storing integer register in SImode. Fix cost of 256 and 512
107 byte aligned SSE register store.
109 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
111 * config/i386/i386.c (ix86_multiplication_cost): Fix
112 multiplication cost for TARGET_AVX512DQ.
114 2018-02-08 Marek Polacek <polacek@redhat.com>
116 PR tree-optimization/84238
117 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
120 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
122 PR tree-optimization/84265
123 * tree-vect-stmts.c (vectorizable_store): Don't treat
124 VMAT_CONTIGUOUS accesses as grouped.
125 (vectorizable_load): Likewise.
127 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
129 PR tree-optimization/81635
130 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
131 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
132 (test_round_for_mask): New functions.
133 (wide_int_cc_tests): Call test_round_for_mask.
134 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
135 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
136 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
137 range returned by get_range_info.
139 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
142 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
143 * symtab.c: Include builtins.h
144 (symtab_node::output_to_lto_symbol_table_p): Move here
145 from lto-streamer-out.c:output_symbol_p.
146 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
147 (output_symbol_p): Move all logic to symtab.c
148 (produce_symtab): Update.
150 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
152 * config/s390/s390-opts.h (enum indirect_branch): Define.
153 * config/s390/s390-protos.h (s390_return_addr_from_memory)
154 (s390_indirect_branch_via_thunk)
155 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
156 (enum s390_indirect_branch_type): Define.
157 * config/s390/s390.c (struct s390_frame_layout, struct
158 machine_function): Remove.
159 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
160 (indirect_branch_table_label_no, indirect_branch_table_name):
162 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
163 (enum s390_indirect_branch_option): Define.
164 (s390_return_addr_from_memory): New function.
165 (s390_handle_string_attribute): New function.
166 (s390_attribute_table): Add new attribute handler.
167 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
168 (s390_indirect_branch_via_thunk): New function.
169 (s390_indirect_branch_via_inline_thunk): New function.
170 (s390_function_ok_for_sibcall): When jumping via thunk disallow
171 sibling call optimization for non z10 compiles.
172 (s390_emit_call): Force indirect branch target to be a single
173 register. Add r1 clobber for non-z10 compiles.
174 (s390_emit_epilogue): Emit return jump via return_use expander.
175 (s390_reorg): Handle JUMP_INSNs as execute targets.
176 (s390_option_override_internal): Perform validity checks for the
177 new command line options.
178 (s390_indirect_branch_attrvalue): New function.
179 (s390_indirect_branch_settings): New function.
180 (s390_set_current_function): Invoke s390_indirect_branch_settings.
181 (s390_output_indirect_thunk_function): New function.
182 (s390_code_end): Implement target hook.
183 (s390_case_values_threshold): Implement target hook.
184 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
186 * config/s390/s390.h (struct s390_frame_layout)
187 (struct machine_function): Move here from s390.c.
188 (TARGET_INDIRECT_BRANCH_NOBP_RET)
189 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
190 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
191 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
192 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
193 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
194 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
195 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
196 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
197 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
198 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
199 (mnemonic attribute): Add values which aren't recognized
201 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
202 pattern for branch conversion. Fix mnemonic attribute.
203 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
204 indirect branch via thunk if requested.
205 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
206 ("*indirect_jump"): Disable for branch conversion using out of
208 ("indirect_jump_via_thunk<mode>_z10")
209 ("indirect_jump_via_thunk<mode>")
210 ("indirect_jump_via_inlinethunk<mode>_z10")
211 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
212 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
213 ("casesi_jump_via_inlinethunk<mode>_z10")
214 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
215 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
216 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
217 ("*indirect2_jump"): Disable for branch conversion.
218 ("casesi_jump"): Turn into expander and expand patterns for branch
220 ("return_use"): New expander.
221 ("*return"): Emit return via thunk and rename it to ...
222 ("*return<mode>"): ... this one.
223 * config/s390/s390.opt: Add new options and and enum for the
226 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
228 * lra-constraints.c (match_reload): Unconditionally use
229 gen_lowpart_SUBREG, rather than selecting between that
230 and equivalent gen_rtx_SUBREG code.
232 2018-02-08 Richard Biener <rguenther@suse.de>
234 PR tree-optimization/84233
235 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
236 changed flag instead of boguously re-using phi_inserted.
238 2018-02-08 Martin Jambor <mjambor@suse.cz>
240 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
241 static local variables.
243 2018-02-08 Richard Biener <rguenther@suse.de>
245 PR tree-optimization/84278
246 * tree-vect-stmts.c (vectorizable_store): When looking for
247 smaller vector types to perform grouped strided loads/stores
248 make sure the mode is supported by the target.
249 (vectorizable_load): Likewise.
251 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
253 * config/aarch64/aarch64.c (aarch64_components_for_bb):
254 Increase LDP/STP opportunities by adding adjacent callee-saves.
256 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
258 PR rtl-optimization/84068
259 PR rtl-optimization/83459
260 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
262 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
264 PR tree-optimization/84224
265 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
266 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
269 2018-02-07 Iain Sandoe <iain@codesourcery.com>
272 * config/rs6000/altivec.md (*restore_world): Remove LR use.
273 * config/rs6000/predicates.md (restore_world_operation): Adjust op
274 count, remove one USE.
276 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
278 * doc/install.texi (Configuration): Document the
279 --with-long-double-format={ibm,ieee} PowerPC configuration
283 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
284 Convert from define_expand to be define_insn_and_split. Rework
285 float/double/_Float128 conversions to QI/HI/SImode to work with
286 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
287 conversions to QI/HImode types did a store and then a load to
288 truncate the value. For conversions to VSX registers, don't split
289 the insn, instead emit the code directly. Use the code iterator
290 any_fix to combine signed and unsigned conversions.
291 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
292 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
293 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
294 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
295 (fix_<mode>di2_hw): Likewise.
296 (fixuns_<mode>di2_hw): Likewise.
297 (fix_<mode>si2_hw): Likewise.
298 (fixuns_<mode>si2_hw): Likewise.
299 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
300 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
301 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
302 fix<uns>_trunc<SFDF:mode>si2_p8.
303 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
305 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
306 (fix<uns>_<mode>_mem): Likewise.
307 (fctiw<u>z_<mode>_mem): Likewise.
308 (fix<uns>_<mode>_mem): Likewise.
309 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
310 the register allocator from doing a direct move to the GPRs to do
311 a store, and instead use the ISA 3.0 store byte/half-word from
312 vector register instruction. For IEEE 128-bit floating point,
313 also optimize stores of 32-bit ints.
314 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
316 2018-02-07 Alan Hayward <alan.hayward@arm.com>
318 * genextract.c (push_pathstr_operand): New function to support
320 (walk_rtx): Call push_pathstr_operand.
321 (print_path): Support [a-zA-Z].
323 2018-02-07 Richard Biener <rguenther@suse.de>
325 PR tree-optimization/84037
326 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
327 (cse_and_gimplify_to_preheader): Declare.
328 (vect_get_place_in_interleaving_chain): Likewise.
329 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
331 (_loop_vec_info::~_loop_vec_info): Delete it.
332 (cse_and_gimplify_to_preheader): New function.
333 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
334 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
335 (vectorizable_load): Likewise. For grouped stores always base
336 the IV on the first element.
337 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
338 condition before gimplifying.
340 2018-02-07 Jakub Jelinek <jakub@redhat.com>
342 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
343 *DIV_EXPR and *MOD_EXPR.
345 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
348 * config/i386/i386.c (ix86_option_override_internal): Mask out
349 the CF_SET bit when checking -fcf-protection.
351 2018-02-07 Tom de Vries <tom@codesourcery.com>
354 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
357 2018-02-07 Richard Biener <rguenther@suse.de>
359 PR tree-optimization/84204
360 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
363 PR tree-optimization/84205
364 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
365 special-case isl_ast_op_zdiv_r.
367 PR tree-optimization/84223
368 * graphite-scop-detection.c (gather_bbs::before_dom_children):
369 Only add conditions from within the region.
370 (gather_bbs::after_dom_children): Adjust.
372 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
375 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
376 * config/avr/avr.md: Only post-reload split REG-REG moves if
377 either register is REGERAL_REG_P.
379 2018-02-07 Jakub Jelinek <jakub@redhat.com>
381 PR tree-optimization/84235
382 * tree-ssa-scopedtables.c
383 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
384 if the subtraction is performed in floating point type where NaNs are
385 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
386 build 1. Formatting fix.
388 2018-02-06 Jakub Jelinek <jakub@redhat.com>
391 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
392 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
393 and skip it regardless of bb boundaries. Use CALL_P macro,
394 don't test INSN_P (insn) together with CALL_P or JUMP_P check
395 unnecessarily, formatting fix.
397 2018-02-06 Michael Collison <michael.collison@arm.com>
399 * config/arm/thumb2.md:
400 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
401 (*thumb_mov_notscc): Ditto.
403 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
406 * config/rs6000/rs6000.md (su code attribute): Use "u" for
407 unsigned_fix, not "s".
409 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
411 * configure.ac (gcc_fn_eh_frame_ro): New function.
412 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
413 correct .eh_frame permissions.
414 * configure: Regenerate.
416 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
418 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
421 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
423 * config/rs6000/rs6000.c (rs6000_option_override_internal):
424 Display warning message for -mno-speculate-indirect-jumps.
426 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
428 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
430 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
432 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
434 PR tree-optimization/84225
435 * tree-eh.c (find_trapping_overflow): Only call
436 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
438 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
441 * config/i386/i386.c: Reimplement the check of possible options
442 -mibt/-mshstk conbination. Change error messages.
443 * doc/invoke.texi: Fix a typo: remove extra '='.
445 2018-02-06 Marek Polacek <polacek@redhat.com>
447 PR tree-optimization/84228
448 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
450 2018-02-06 Tamar Christina <tamar.christina@arm.com>
453 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
454 emitted arch directives.
455 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
456 __ARM_FEATURE_COPROC before changing architectures.
458 2018-02-06 Richard Biener <rguenther@suse.de>
460 * config/i386/i386.c (print_reg): Fix typo.
461 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
463 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
465 * configure: Regenerate.
467 2018-02-05 Martin Sebor <msebor@redhat.com>
469 PR tree-optimization/83369
470 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
473 2018-02-05 Martin Liska <mliska@suse.cz>
475 * doc/invoke.texi: Cherry-pick upstream r323995.
477 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
479 * ira.c (ira_init_register_move_cost): Adjust comment.
481 2018-02-05 Martin Liska <mliska@suse.cz>
483 PR gcov-profile/84137
484 * doc/gcov.texi: Fix typo in documentation.
486 2018-02-05 Martin Liska <mliska@suse.cz>
488 PR gcov-profile/83879
489 * doc/gcov.texi: Document necessity of --dynamic-list-data when
490 using dlopen functionality.
492 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
494 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
495 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
496 _mm_maskz_range_ss, _mm_mask_range_round_ss,
497 _mm_maskz_range_round_ss): New intrinsics.
498 (__builtin_ia32_rangesd128_round)
499 (__builtin_ia32_rangess128_round): Remove.
500 (__builtin_ia32_rangesd128_mask_round,
501 __builtin_ia32_rangess128_mask_round): New builtins.
502 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
503 __builtin_ia32_rangess128_round): Remove.
504 (__builtin_ia32_rangesd128_mask_round,
505 __builtin_ia32_rangess128_mask_round): New builtins.
506 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
507 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
508 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
509 "<round_saeonly_constraint>")): Changed to ...
510 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
511 "<round_saeonly_scalar_constraint>")): ... this.
512 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
513 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
514 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
515 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
516 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
518 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
520 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
522 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
523 Remove all values except native, 8540 and 8548.
525 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
527 * config/i386/i386.c (ix86_output_function_return): Pass
528 INVALID_REGNUM, instead of -1, as invalid register number to
529 indirect_thunk_name and output_indirect_thunk.
531 2018-02-02 Julia Koval <julia.koval@intel.com>
533 * config.gcc: Add -march=icelake.
534 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
535 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
536 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
537 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
538 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
539 (processor_target_table): Add icelake.
540 (ix86_option_override_internal): Handle new PTAs.
541 (get_builtin_code_for_version): Handle icelake.
542 (M_INTEL_COREI7_ICELAKE): New.
543 (fold_builtin_cpu): Handle icelake.
544 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
545 * doc/invoke.texi: Add -march=icelake.
547 2018-02-02 Julia Koval <julia.koval@intel.com>
549 * config/i386/i386.c (ix86_option_override_internal): Change flags type
551 * wide-int-bitmask.h: New.
553 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
556 * config/i386/i386.md: Replace Pmode with word_mode in
557 builtin_setjmp_setup and builtin_longjmp to support x32.
559 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
563 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
565 (rs6000_supported_cpu_names): New static variable.
566 (linux_cpu_translation_table): Likewise.
567 (elf_platform) <cpu>: Define new static variable and use it.
568 Translate kernel AT_PLATFORM name to canonical name if needed.
569 Error if platform name is unknown.
571 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
574 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
576 2018-02-01 Jeff Law <law@redhat.com>
579 * config/i386/i386.c (release_scratch_register_on_entry): Add new
580 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
581 the scratch if RELEASE_VIA_POP is false.
582 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
583 If we have to save a temporary register, decrement SIZE appropriately.
584 Pass new arguments to release_scratch_register_on_entry.
585 (ix86_adjust_stack_and_probe): Likewise.
586 (ix86_emit_probe_stack_range): Pass new arguments to
587 release_scratch_register_on_entry.
589 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
591 PR rtl-optimization/84157
592 * combine.c (change_zero_ext): Use REG_P predicate in
593 front of HARD_REGISTER_P predicate.
595 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
597 * config/avr/avr.c (avr_option_override): Move disabling of
598 -fdelete-null-pointer-checks to...
599 * common/config/avr/avr-common.c (avr_option_optimization_table):
602 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
604 PR tree-optimization/81635
605 * tree-data-ref.c (split_constant_offset_1): For types that
606 wrap on overflow, try to use range info to prove that wrapping
609 2018-02-01 Renlin Li <renlin.li@arm.com>
612 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
614 (aarch64_register_move_cost): Likewise.
615 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
617 (REG_CLASS_NAMES): Likewise.
618 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
619 TAILCALL_ADDR_REGS. Remove IP registers.
620 * config/aarch64/aarch64.md (Ucs): Update register constraint.
622 2018-02-01 Richard Biener <rguenther@suse.de>
624 * domwalk.h (dom_walker::dom_walker): Add additional constructor
625 for specifying RPO order and allow NULL for that.
626 * domwalk.c (dom_walker::dom_walker): Likewise.
627 (dom_walker::walk): Handle NULL RPO order.
628 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
630 (rewrite_update_dom_walker): Likewise.
631 (mark_def_dom_walker): Likewise.
633 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
635 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
636 (aarch64_maybe_expand_sve_subreg_move): Declare.
637 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
638 * config/aarch64/predicates.md (aarch64_any_register_operand): New
640 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
641 that are semantically a reverse operation.
642 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
643 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
644 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
646 (aarch64_can_change_mode_class): For big-endian, forbid changes
647 between two SVE modes if they have different element sizes.
649 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
651 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
652 the TImode handling for big-endian targets.
654 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
656 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
657 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
659 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
660 Remove BSWAP handing for big-endian targets and use the form of
661 LD1RQ appropariate for the mode.
663 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
665 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
666 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
669 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
672 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
673 check for operands that need to go through aarch64_sve_reload_be.
675 2018-02-01 Jakub Jelinek <jakub@redhat.com>
677 PR tree-optimization/81661
678 PR tree-optimization/84117
679 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
680 * tree-eh.c: Include gimplify.h.
681 (find_trapping_overflow, replace_trapping_overflow,
682 rewrite_to_non_trapping_overflow): New functions.
683 * tree-vect-loop.c: Include tree-eh.h.
684 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
685 * tree-data-ref.c: Include tree-eh.h.
686 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
688 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
690 PR rtl-optimization/84123
691 * combine.c (change_zero_ext): Check if hard register satisfies
692 can_change_dest_mode before calling gen_lowpart_SUBREG.
694 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
697 * ira.c (ira_init_register_move_cost): Remove assert.
699 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
701 PR rtl-optimization/84071
702 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
703 * doc/tm.texi: Regenerate.
705 2018-01-31 Richard Biener <rguenther@suse.de>
707 PR tree-optimization/84132
708 * tree-data-ref.c (analyze_miv_subscript): Properly
709 check whether evolution_function_is_affine_multivariate_p
710 before calling gcd_of_steps_may_divide_p.
712 2018-01-31 Julia Koval <julia.koval@intel.com>
715 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
716 * config/i386/i386.md (rdpid_rex64) New.
717 (rdpid): Make 32bit only.
719 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
722 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
723 an IDENTIFIER_NODE for FUNCTION_TYPE's.
725 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
728 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
730 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
732 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
734 PR rtl-optimization/84071
735 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
736 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
738 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
740 * config/arc/arc.c (arc_handle_aux_attribute): New function.
741 (arc_attribute_table): Add 'aux' attribute.
742 (arc_in_small_data_p): Consider aux like variables.
743 (arc_is_aux_reg_p): New function.
744 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
745 (arc_get_aux_arg): New function.
746 (prepare_move_operands): Handle aux-register access.
747 (arc_handle_aux_attribute): New function.
748 * doc/extend.texi (ARC Variable attributes): Add subsection.
750 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
752 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
753 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
754 (arc_attribute_table): Add 'uncached' attribute.
755 (arc_print_operand): Print '.di' flag for uncached memory
757 (arc_in_small_data_p): Do not consider for small data the uncached
759 (arc_is_uncached_mem_p): New function.
760 * config/arc/predicates.md (compact_store_memory_operand): Check
761 for uncached memory accesses.
762 (nonvol_nonimm_operand): Likewise.
763 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
765 2018-01-31 Jakub Jelinek <jakub@redhat.com>
768 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
769 falign-loops=): Add Optimization flag.
771 2018-01-30 Jeff Law <law@redhat.com>
774 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
775 INT_REGISTERS_SAVED. Check it prior to calling
776 get_scratch_register_on_entry.
777 (ix86_adjust_stack_and_probe): Similarly.
778 (ix86_emit_probe_stack_range): Similarly.
779 (ix86_expand_prologue): Corresponding changes.
781 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
784 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
785 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
787 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
790 * lra-constraints.c (curr_insn_transform): Process AND in the
793 2018-01-30 Jakub Jelinek <jakub@redhat.com>
795 PR rtl-optimization/83986
796 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
797 dependence against last_pending_memory_flush in addition to
800 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
802 PR tree-optimization/81611
803 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
806 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
808 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
811 2018-01-30 Richard Biener <rguenther@suse.de>
812 Jakub Jelinek <jakub@redhat.com>
814 PR tree-optimization/84111
815 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
816 inner loops added during recursion, as they don't have up-to-date
819 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
822 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
823 (can_inline_edge_by_limits_p): ... here.
824 (can_early_inline_edge_p, check_callers,
825 update_caller_keys, update_callee_keys, recursive_inlining,
826 add_new_edges_to_heap, speculation_useful_p,
827 inline_small_functions,
828 inline_small_functions, flatten_function,
829 inline_to_all_callers_1): Update.
831 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
833 * profile-count.c (profile_count::combine_with_ipa_count): Handle
836 2018-01-30 Richard Biener <rguenther@suse.de>
838 PR tree-optimization/83008
839 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
840 invariant and constant vector uses in stmts when they need
843 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
846 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
847 * configure: Regenerate.
849 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
851 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
853 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
854 Use gen_rtx_REG rather than gen_lowpart.
856 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
858 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
859 rather than 0 when creating partial subregs.
861 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
863 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
866 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
869 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
870 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
871 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
872 flags. This restores the settings used before the 2017-07-24.
873 Turning off pre increment/decrement/modify allows IVOPTS to
874 optimize DF/SF loops where the index is an int.
876 2018-01-29 Richard Biener <rguenther@suse.de>
877 Kelvin Nilsen <kelvin@gcc.gnu.org>
880 * tree-vect-stmts.c (vectorizable_call): Don't call
881 targetm.vectorize_builtin_md_vectorized_function if callee is
884 2018-01-22 Carl Love <cel@us.ibm.com>
886 * doc/extend.tex: Fix typo in second arg in
887 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
889 2018-01-29 Richard Biener <rguenther@suse.de>
891 PR tree-optimization/84086
892 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
893 (flush_ssaname_freelist): When SSA names were released reset
896 2018-01-29 Richard Biener <rguenther@suse.de>
898 PR tree-optimization/84057
899 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
900 removed paths when removing edges.
902 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
904 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
905 -mfunction-return=@var{choice}.
907 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
910 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
912 (layout::print_source_line): Likewise.
913 (test_get_line_width_without_trailing_whitespace): Add test cases.
915 2018-01-27 Jakub Jelinek <jakub@redhat.com>
918 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
921 2018-01-26 Jim Wilson <jimw@sifive.com>
923 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
925 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
928 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
930 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
931 and CMP + SUB-immediate -> SUBS.
933 2018-01-26 Martin Sebor <msebor@redhat.com>
935 PR tree-optimization/83896
936 * tree-ssa-strlen.c (get_string_len): Rename...
937 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
938 Avoid assuming length is constant.
939 (handle_char_store): Use HOST_WIDE_INT for string length.
941 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
944 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
945 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
947 2018-01-26 Richard Biener <rguenther@suse.de>
949 PR rtl-optimization/84003
950 * dse.c (record_store): Only record redundant stores when
951 the earlier store aliases at least all accesses the later one does.
953 2018-01-26 Jakub Jelinek <jakub@redhat.com>
955 PR rtl-optimization/83985
956 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
957 REG_CFA_RESTORE insns.
958 (delete_unmarked_insns): Don't ignore separate shrink wrapping
959 REG_CFA_RESTORE insns here.
962 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
963 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
965 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
967 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
968 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
969 (arc_init): Likewise.
970 (arc_override_options): Likewise.
971 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
973 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
975 * config/arc/arc.h (TARGET_DBNZ): Define.
976 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
977 properly set the tune attribute.
978 (dbnz): Use TARGET_DBNZ guard.
979 * config/arc/arc.opt (mtune): Add core3 option.
981 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
983 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
984 recognize new pic like addresses.
985 (arc_delegitimize_address): Clean up.
987 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
989 * config/arc/arc-arches.def: Option mrf16 valid for all
991 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
992 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
993 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
994 * config/arc/arc-tables.opt: Regenerate.
995 * config/arc/arc.c (arc_conditional_register_usage): Handle
996 reduced register file case.
997 (arc_file_start): Set must have build attributes.
998 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1000 * config/arc/arc.opt (mrf16): Add new option.
1001 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1002 * config/arc/genmultilib.awk: Handle new mrf16 option.
1003 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1004 * config/arc/t-multilib: Regenerate.
1005 * doc/invoke.texi (ARC Options): Document mrf16 option.
1007 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1009 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1010 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1011 (arc_attribute_table): Add 'secure_call' attribute.
1012 (arc_print_operand): Print secure call operand.
1013 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1015 (arc_is_secure_call_p): New function. * config/arc/arc.md
1016 (call_i): Add support for sjli instruction.
1017 (call_value_i): Likewise.
1018 * config/arc/constraints.md (Csc): New constraint.
1020 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1021 John Eric Martin <John.Martin@emmicro-us.com>
1023 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1024 * config/arc/arc.c (_arc_jli_section): New struct.
1025 (arc_jli_section): New type.
1026 (rc_jli_sections): New static variable.
1027 (arc_handle_jli_attribute): New function.
1028 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1029 (arc_file_end): New function.
1030 (TARGET_ASM_FILE_END): Define.
1031 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1032 (arc_add_jli_section): New function.
1033 (jli_call_scan): Likewise.
1034 (arc_reorg): Call jli_call_scan.
1035 (arc_output_addsi): Remove 'S' from printing asm operand.
1036 (arc_is_jli_call_p): New function.
1037 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1039 (movhi_insn): Likewise.
1040 (movsi_insn): Likewise.
1041 (movsi_set_cc_insn): Likewise.
1042 (loadqi_update): Likewise.
1043 (load_zeroextendqisi_update): Likewise.
1044 (load_signextendqisi_update): Likewise.
1045 (loadhi_update): Likewise.
1046 (load_zeroextendhisi_update): Likewise.
1047 (load_signextendhisi_update): Likewise.
1048 (loadsi_update): Likewise.
1049 (loadsf_update): Likewise.
1050 (movsicc_insn): Likewise.
1051 (bset_insn): Likewise.
1052 (bxor_insn): Likewise.
1053 (bclr_insn): Likewise.
1054 (bmsk_insn): Likewise.
1055 (bicsi3_insn): Likewise.
1056 (cmpsi_cc_c_insn): Likewise.
1057 (movsi_ne): Likewise.
1058 (movsi_cond_exec): Likewise.
1059 (clrsbsi2): Likewise.
1070 (call_i): Remove 'S' asm letter, add jli instruction.
1071 (call_value_i): Likewise.
1072 * config/arc/arc.op (mjli-always): New option.
1073 * config/arc/constraints.md (Cji): New constraint.
1074 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1076 (subsf3_fpx): Likewise.
1077 (mulsf3_fpx): Likewise.
1078 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1080 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1082 * doc/invoke.texi (ARC): Document mjli-always option.
1084 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1086 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
1087 and use incw and decw where possible
1088 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
1090 2018-01-26 Richard Biener <rguenther@suse.de>
1092 PR tree-optimization/81082
1093 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1094 association if it requires casting to unsigned.
1095 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1096 from fold_plusminus_mult_expr to catch important cases late when
1097 range info is available.
1099 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1101 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1102 * configure.ac (hidden_linkonce): New test.
1103 * configure: Regenerate.
1104 * config.in: Regenerate.
1106 2018-01-26 Julia Koval <julia.koval@intel.com>
1108 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1109 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1110 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1111 _mm_mask_bitshuffle_epi64_mask): Fix type.
1112 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1113 USI_FTYPE_V4DI_V4DI_USI): Remove.
1114 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1115 __builtin_ia32_vpshufbitqmb256_mask,
1116 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1117 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1118 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1120 2018-01-26 Alan Modra <amodra@gmail.com>
1123 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1124 UNSPEC_VBPERMQ. Sort other unspecs.
1126 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1128 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1130 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1133 * predict.c (drop_profile): Do not push/pop cfun; update also
1135 (handle_missing_profiles): Fix logic looking for zero profiles.
1137 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1140 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1141 on functions with #pragma omp declare simd or functions with simd
1143 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1144 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1145 Remove trailing \n from warning_at calls.
1147 2018-01-25 Tom de Vries <tom@codesourcery.com>
1150 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1151 for neutered workers.
1153 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1156 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1157 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1159 2018-01-24 Jeff Law <law@redhat.com>
1162 * i386.c (get_probe_interval): Move to earlier point.
1163 (ix86_compute_frame_layout): If -fstack-clash-protection and
1164 the frame is larger than the probe interval, then use pushes
1165 to save registers rather than reg->mem moves.
1166 (ix86_expand_prologue): Remove conditional for int_registers_saved
1169 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1172 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1173 min/max for never referenced object.
1175 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1178 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1180 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1181 attributes from DECL_ATTRIBUTES (decl) without affecting
1182 DECL_ATTRIBUTES (current_function_decl).
1183 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1184 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1186 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1188 PR tree-optimization/83979
1189 * fold-const.c (fold_comparison): Use constant_boolean_node
1190 instead of boolean_{true,false}_node.
1192 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1194 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1197 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1199 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1200 Simplify the clause that sets the length attribute.
1201 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1202 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1203 clause that sets the length attribute.
1204 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1206 2018-01-24 Tom de Vries <tom@codesourcery.com>
1209 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1210 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1211 Add strict parameter.
1212 (prevent_branch_around_nothing): Insert dummy insn between branch to
1213 label and label with no ptx insn inbetween.
1214 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1216 2018-01-24 Tom de Vries <tom@codesourcery.com>
1219 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1220 for neutered threads in warp.
1221 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1223 2018-01-24 Richard Biener <rguenther@suse.de>
1225 PR tree-optimization/83176
1226 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1229 2018-01-24 Richard Biener <rguenther@suse.de>
1231 PR tree-optimization/82819
1232 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1233 code generating pluses that are no-ops in the target precision.
1235 2018-01-24 Richard Biener <rguenther@suse.de>
1238 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1240 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1242 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1243 to merge probabilities.
1244 * predict.c (probably_never_executed): Also mark as cold functions
1245 with global 0 profile and guessed local profile.
1246 * profile-count.c (profile_probability::combine_with_count): New
1248 * profile-count.h (profile_probability::operator*,
1249 profile_probability::operator*=, profile_probability::operator/,
1250 profile_probability::operator/=): Reduce precision to adjusted
1251 and set value to guessed on contradictory divisions.
1252 (profile_probability::combine_with_freq): Remove.
1253 (profile_probability::combine_wiht_count): Declare.
1254 (profile_count::force_nonzero):: Set to adjusted.
1255 (profile_count::probability_in):: Set quality to adjusted.
1256 * tree-ssa-tail-merge.c (replace_block_by): Use
1259 2018-01-23 Andrew Waterman <andrew@sifive.com>
1260 Jim Wilson <jimw@sifive.com>
1262 * config/riscv/riscv.c (riscv_stack_boundary): New.
1263 (riscv_option_override): Set riscv_stack_boundary. Handle
1264 riscv_preferred_stack_boundary_arg.
1265 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1266 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1267 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1268 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1269 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1270 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1272 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1275 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1276 of struct ix86_frame.
1277 (ix86_expand_epilogue): Likewise. Add a local variable for
1278 the reg_save_offset field in struct ix86_frame.
1280 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1282 PR tree-optimization/82604
1283 * tree-loop-distribution.c (enum partition_kind): New enum item
1284 PKIND_PARTIAL_MEMSET.
1285 (partition_builtin_p): Support above new enum item.
1286 (generate_code_for_partition): Ditto.
1287 (compute_access_range): Differentiate cases that equality can be
1288 proven at all loops, the innermost loops or no loops.
1289 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1290 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1291 (finalize_partitions, distribute_loop): Don't fuse partition of
1292 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1293 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1296 2018-01-23 Martin Liska <mliska@suse.cz>
1298 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1299 order to ignore the predictor.
1300 (PRED_POLYMORPHIC_CALL): Likewise.
1301 (PRED_RECURSIVE_CALL): Likewise.
1303 2018-01-23 Martin Liska <mliska@suse.cz>
1305 * tree-profile.c (tree_profiling): Print function header to
1306 aware reader which function we are working on.
1307 * value-prof.c (gimple_find_values_to_profile): Do not print
1308 not interesting value histograms.
1310 2018-01-23 Martin Liska <mliska@suse.cz>
1312 * profile-count.h (enum profile_quality): Add
1313 profile_uninitialized as the first value. Do not number values
1314 as they are zero based.
1315 (profile_count::verify): Update sanity check.
1316 (profile_probability::verify): Likewise.
1318 2018-01-23 Nathan Sidwell <nathan@acm.org>
1320 * doc/invoke.texi (ffor-scope): Deprecate.
1322 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1324 PR tree-optimization/83510
1325 * domwalk.c (set_all_edges_as_executable): New function.
1326 (dom_walker::dom_walker): Convert bool param
1327 "skip_unreachable_blocks" to enum reachability. Move setup of
1328 edge flags to set_all_edges_as_executable and only do it when
1329 reachability is REACHABLE_BLOCKS.
1330 * domwalk.h (enum dom_walker::reachability): New enum.
1331 (dom_walker::dom_walker): Convert bool param
1332 "skip_unreachable_blocks" to enum reachability.
1333 (set_all_edges_as_executable): New decl.
1334 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1335 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1337 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1338 but converting true to REACHABLE_BLOCKS.
1339 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1341 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1342 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1343 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1345 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1346 if check_all_array_refs will be called.
1348 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1350 * tree.c (selftest::test_location_wrappers): Add more test
1353 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1355 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1356 (selftest::test_bit_in_range): Likewise.
1358 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1361 * doc/sourcebuild.texi (vect_float): Say that the selector
1362 only describes the situation when -funsafe-math-optimizations is on.
1363 (vect_float_strict): Document.
1365 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1367 PR tree-optimization/83965
1368 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1369 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1370 instead of checking only for a reduction.
1371 (vect_recog_widen_sum_pattern): Likewise.
1373 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1375 * predict.c (probably_never_executed): Only use precise profile info.
1376 (compute_function_frequency): Skip after inlining hack since we now
1377 have quality checking.
1379 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1381 * profile-count.h (profile_probability::very_unlikely,
1382 profile_probability::unlikely, profile_probability::even): Set
1383 precision to guessed.
1385 2018-01-23 Richard Biener <rguenther@suse.de>
1387 PR tree-optimization/83963
1388 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1389 Properly terminate dominator walk when crossing the exit edge not
1390 when visiting its source block.
1392 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1395 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1396 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1398 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1400 PR tree-optimization/83957
1401 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1402 semicolon after for body surrounded by braces.
1404 PR tree-optimization/83081
1405 * profile-count.h (profile_probability::split): New method.
1406 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1407 Use profile_probability::split.
1408 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1409 when splitting a single conditional jump into 2.
1411 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1413 PR tree-optimization/69452
1414 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1417 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1419 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1420 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1421 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1423 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1425 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1426 * config/rl78/rl78.md: New define_expand "movdi"
1427 * config/rl78/rl78.c: New function definition rl78_split_movdi
1429 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1432 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1434 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1435 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1436 128-bit to produce an UNSPEC move to get the double word with the
1437 signbit and then a shift directly to do signbit.
1438 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1439 implementation with a new version that just does either a direct
1440 move or a regular move. Move memory interface to separate insns.
1441 Move insns so they are next to the expander.
1442 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1443 with signbit move. Split big and little endian case.
1444 (signbit<mode>2_dm_mem_le): Likewise.
1445 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1446 (signbit<mode>2_dm2): Likewise.
1448 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1450 * config/rl78/rl78.md: New define_expand "anddi3".
1452 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1454 * config/rl78/rl78.md: New define_expand "umindi3".
1456 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1458 * config/rl78/rl78.md: New define_expand "smindi3".
1460 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1462 * config/rl78/rl78.md: New define_expand "smaxdi3".
1464 2018-01-22 Carl Love <cel@us.ibm.com>
1466 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1467 LVX_V1TI): Add macro expansion.
1468 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1469 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1470 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1471 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1472 Change check to determine if the instruction is a byte reversing
1473 entry. Fix typo in comment.
1474 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1475 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1476 Add def_builtin calls for new builtins.
1477 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1478 Add define_insn expansion.
1480 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1482 * config/rl78/rl78.md: New define_expand "umaxdi3".
1484 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1486 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1487 for non-QImode registers
1489 2018-01-22 Richard Biener <rguenther@suse.de>
1491 PR tree-optimization/83963
1492 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1493 including the loop exit block.
1494 (scop_detection::merge_sese): Likewise.
1495 (scop_detection::add_scop): Do it here instead.
1497 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1499 * doc/sourcebuild.texi (arm_softfloat): Document.
1501 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1504 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1505 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1506 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1508 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1509 David Edelsohn <dje.gcc@gmail.com>
1512 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1513 Change "crset eq" to "crset 2".
1514 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1515 (*call_indirect_aix<mode>_nospec): Likewise.
1516 (*call_value_indirect_aix<mode>_nospec): Likewise.
1517 (*call_indirect_elfv2<mode>_nospec): Likewise.
1518 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1519 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1520 change assembly output from . to $.
1521 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1522 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1523 (*tablejump<mode>_internal1_nospec): Likewise.
1525 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1528 * config/sh/sh_optimize_sett_clrt.cc:
1529 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1531 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1533 PR tree-optimization/83940
1534 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1535 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1536 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1537 use it to pass back the definition type.
1538 (vect_check_store_rhs): Likewise rhs_dt_out.
1539 (vect_build_gather_load_calls): Add a mask_dt argument and use
1540 it instead of a call to vect_is_simple_use.
1541 (vectorizable_store): Update calls to vect_check_load_store_mask
1542 and vect_check_store_rhs. Use the dt returned by the latter instead
1543 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1544 instead of calls to vect_is_simple_use. Pass the scalar rather
1545 than the vector operand to vect_is_simple_use when handling
1546 second and subsequent copies of an rhs value.
1547 (vectorizable_load): Update calls to vect_check_load_store_mask
1548 and vect_build_gather_load_calls. Use the cached mask_dt and
1549 gs_info.offset_dt instead of calls to vect_is_simple_use.
1551 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1554 * tree-emutls.c: Include gimplify.h.
1555 (lower_emutls_2): New function.
1556 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1557 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1558 it before further processing.
1561 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1562 UINTVAL (trueop1) instead of INTVAL (op1).
1564 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1568 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1569 INCOMING_FRAME_SP_OFFSET if not defined.
1570 (scan_trace): Add ENTRY argument. If true and
1571 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1572 emit a note to adjust the CFA offset.
1573 (create_cfi_notes): Adjust scan_trace callers.
1574 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1575 INCOMING_FRAME_SP_OFFSET in the CIE.
1576 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1577 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1579 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1580 * doc/tm.texi: Regenerated.
1582 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1584 PR rtl-optimization/83147
1585 * lra-constraints.c (remove_inheritance_pseudos): Use
1586 lra_substitute_pseudo_within_insn.
1588 2018-01-19 Tom de Vries <tom@codesourcery.com>
1589 Cesar Philippidis <cesar@codesourcery.com>
1592 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1594 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1597 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1598 spaces for function labels.
1600 2018-01-19 Martin Liska <mliska@suse.cz>
1602 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1603 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1604 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1605 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1606 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1607 (PRED_CONST_RETURN): Change from 69 to 65.
1608 (PRED_NULL_RETURN): Change from 91 to 71.
1609 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1610 (PRED_LOOP_GUARD): Change from 66 to 73.
1612 2018-01-19 Martin Liska <mliska@suse.cz>
1614 * predict.c (predict_insn_def): Add new assert.
1615 (struct branch_predictor): Change type to signed integer.
1616 (test_prediction_value_range): Amend test to cover
1618 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1619 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1620 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1621 (PRED_LOOP_IV_COMPARE): Likewise.
1622 * predict.h (PROB_UNINITIALIZED): Define new constant.
1624 2018-01-19 Martin Liska <mliska@suse.cz>
1626 * predict.c (dump_prediction): Add new format for
1627 analyze_brprob.py script which is enabled with -details
1629 * profile-count.h (precise_p): New function.
1631 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1633 PR tree-optimization/83922
1634 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1635 there are no statements that need masking.
1636 (vect_active_double_reduction_p): New function.
1637 (vect_analyze_loop_operations): Use it when handling phis that
1638 are not in the loop header.
1640 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1642 PR tree-optimization/83914
1643 * tree-vect-loop.c (vectorizable_induction): Don't convert
1644 init_expr or apply the peeling adjustment for inductions
1645 that are nested within the vectorized loop.
1647 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1649 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1652 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1656 * function.h (gimplify_parameters): Add gimple_seq * argument.
1657 * function.c: Include gimple.h and options.h.
1658 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1659 for the added local temporaries if needed.
1660 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1661 if there are any parameter cleanups, wrap whole body into a
1662 try/finally with the cleanups.
1664 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1667 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1668 Use GET_MODE_CLASS for scalar floating point.
1670 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1674 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1675 Fix call of call_cgraph_insertion_hooks.
1677 2018-01-18 Martin Sebor <msebor@redhat.com>
1679 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1681 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1684 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1687 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1690 * common.opt: (-ffile-prefix-map): New option.
1691 * opts.c (common_handle_option): Defer it.
1692 * opts-global.c (handle_common_deferred_options): Handle it.
1693 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1694 * file-prefix-map.h: New file.
1695 (remap_debug_filename, add_debug_prefix_map): ...here.
1696 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1697 * final.c (debug_prefix_map, add_debug_prefix_map
1698 remap_debug_filename): Move to...
1699 * file-prefix-map.c: New file.
1700 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1701 generalize, get rid of alloca(), use strrchr() instead of strchr().
1702 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1703 Implement in terms of add_prefix_map().
1704 (remap_macro_filename, remap_debug_filename): Implement in term of
1706 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1707 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1708 * dbxout.c: Include file-prefix-map.h.
1709 * varasm.c: Likewise.
1710 * vmsdbgout.c: Likewise.
1711 * xcoffout.c: Likewise.
1712 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1713 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1714 * doc/invoke.texi (-ffile-prefix-map): Document.
1715 (-fdebug-prefix-map): Update description.
1717 2018-01-18 Martin Liska <mliska@suse.cz>
1719 * config/i386/i386.c (indirect_thunk_name): Document that also
1721 (output_indirect_thunk): Document why both instructions
1722 (pause and lfence) are generated.
1724 2018-01-18 Richard Biener <rguenther@suse.de>
1726 PR tree-optimization/83887
1727 * graphite-scop-detection.c
1728 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1729 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1730 (scop_detection::merge_sese): Re-implement with a flood-fill
1731 algorithm that properly finds a SESE region if it exists.
1733 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1736 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1737 pointer_diff optimizations use view_convert instead of convert.
1739 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1741 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1742 Generate different code for -mno-speculate-indirect-jumps.
1743 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1744 (*call_indirect_aix<mode>): Disable for
1745 -mno-speculate-indirect-jumps.
1746 (*call_indirect_aix<mode>_nospec): New define_insn.
1747 (*call_value_indirect_aix<mode>): Disable for
1748 -mno-speculate-indirect-jumps.
1749 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1750 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1751 -mno-speculate-indirect-jumps.
1752 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1754 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1756 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1757 long double type, set the flags for noting the default long double
1758 type, even if we don't pass or return a long double type.
1760 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1763 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1766 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1768 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1769 support for merge[hl].
1770 (fold_mergehl_helper): New helper function.
1771 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1772 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1773 (altivec_vmrglw_direct): Add xxmrglw insn.
1775 2018-01-17 Andrew Waterman <andrew@sifive.com>
1777 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1778 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1780 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1783 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1784 call the lto_location_cache before reading the
1785 DECL_SOURCE_LOCATION of the types.
1787 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1788 Richard Sandiford <richard.sandiford@linaro.org>
1790 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1791 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1792 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1793 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1794 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1796 * config/aarch64/constraints.md (aarch64_movti_operand):
1798 * config/aarch64/predicates.md (Uti): Add new constraint.
1800 2018-01-17 Carl Love <cel@us.ibm.com>
1801 * config/rs6000/vsx.md (define_expand xl_len_r,
1802 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1803 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1805 (define_expand, define_insn): Move the shift left from the
1806 define_insn to the define_expand for lxvl and stxvl instructions.
1807 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1808 and XL_LEN_R definitions to PURE.
1810 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1812 * config/i386/i386.c (indirect_thunk_name): Declare regno
1813 as unsigned int. Compare regno with INVALID_REGNUM.
1814 (output_indirect_thunk): Ditto.
1815 (output_indirect_thunk_function): Ditto.
1816 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1817 in the call to output_indirect_thunk_function.
1819 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1822 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1823 rather than the size of inner_type to determine the stack slot size
1824 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1826 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1829 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1832 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1834 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1835 endian Linux systems to optionally enable multilibs for selecting
1836 the long double type if the user configured an explicit type.
1837 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1838 have no long double multilibs if not defined.
1839 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1840 warn if the user used -mabi={ieee,ibm}longdouble and we built
1841 multilibs for long double.
1842 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1843 appropriate multilib option.
1844 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1846 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1847 for building long double multilibs.
1848 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1850 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1852 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1855 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1857 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1860 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1863 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1866 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1868 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1869 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1871 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1873 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1874 different rtl trees depending on TARGET_64BIT.
1875 (rs6000_gen_lvx): Likewise.
1877 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1879 * config/visium/visium.md (nop): Tweak comment.
1880 (hazard_nop): Likewise.
1882 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1884 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1885 -mspeculate-indirect-jumps.
1886 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1887 for -mno-speculate-indirect-jumps.
1888 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1889 (*call_value_indirect_elfv2<mode>): Disable for
1890 -mno-speculate-indirect-jumps.
1891 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1892 (indirect_jump): Emit different RTL for
1893 -mno-speculate-indirect-jumps.
1894 (*indirect_jump<mode>): Disable for
1895 -mno-speculate-indirect-jumps.
1896 (*indirect_jump<mode>_nospec): New define_insn.
1897 (tablejump): Emit different RTL for
1898 -mno-speculate-indirect-jumps.
1899 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1900 (tablejumpsi_nospec): New define_expand.
1901 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1902 (tablejumpdi_nospec): New define_expand.
1903 (*tablejump<mode>_internal1): Disable for
1904 -mno-speculate-indirect-jumps.
1905 (*tablejump<mode>_internal1_nospec): New define_insn.
1906 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1909 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1911 * caller-save.c (insert_save): Drop unnecessary parameter. All
1914 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1915 Richard Biener <rguenth@suse.de>
1918 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1919 return early, inline manually is_gimple_sizepos. Make sure if we
1920 call gimplify_expr we don't end up with a gimple constant.
1921 * tree.c (variably_modified_type_p): Don't return true for
1922 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1923 * gimplify.h (is_gimple_sizepos): Remove.
1925 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1927 PR tree-optimization/83857
1928 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1929 vectorizable_live_operation for pure SLP statements.
1930 (vectorizable_live_operation): Handle PHIs.
1932 2018-01-16 Richard Biener <rguenther@suse.de>
1934 PR tree-optimization/83867
1935 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1936 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1938 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1941 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1942 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1943 If off is not INTEGER_CST, issue a may not be aligned warning
1944 rather than isn't aligned. Use isn%'t rather than isn't.
1945 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1947 <case MULT_EXPR>: Improve the case when bottom and one of the
1948 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1949 operand, in that case check if the other operand is multiple of
1950 bottom divided by the INTEGER_CST operand.
1952 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1955 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1956 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1957 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1958 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1959 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1960 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1961 (pa_function_arg_size): New function.
1963 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1965 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1966 in a separate statement.
1968 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1970 PR tree-optimization/83847
1971 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1972 group gathers and scatters.
1974 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1976 PR rtl-optimization/86620
1977 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1979 PR rtl-optimization/83213
1980 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1981 to last if both are JUMP_INSNs.
1983 PR tree-optimization/83843
1984 * gimple-ssa-store-merging.c
1985 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1986 store_immediate_info for bswap/nop orig_stores.
1988 2018-01-15 Andrew Waterman <andrew@sifive.com>
1990 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1992 <UDIV>: Increase cost if !TARGET_DIV.
1994 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1996 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1997 (define_attr "cr_logical_3op"): New.
1998 (cceq_ior_compare): Adjust.
1999 (cceq_ior_compare_complement): Adjust.
2000 (*cceq_rev_compare): Adjust.
2001 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2002 (is_cracked_insn): Adjust.
2003 (insn_must_be_first_in_group): Adjust.
2004 * config/rs6000/40x.md: Adjust.
2005 * config/rs6000/440.md: Adjust.
2006 * config/rs6000/476.md: Adjust.
2007 * config/rs6000/601.md: Adjust.
2008 * config/rs6000/603.md: Adjust.
2009 * config/rs6000/6xx.md: Adjust.
2010 * config/rs6000/7450.md: Adjust.
2011 * config/rs6000/7xx.md: Adjust.
2012 * config/rs6000/8540.md: Adjust.
2013 * config/rs6000/cell.md: Adjust.
2014 * config/rs6000/e300c2c3.md: Adjust.
2015 * config/rs6000/e500mc.md: Adjust.
2016 * config/rs6000/e500mc64.md: Adjust.
2017 * config/rs6000/e5500.md: Adjust.
2018 * config/rs6000/e6500.md: Adjust.
2019 * config/rs6000/mpc.md: Adjust.
2020 * config/rs6000/power4.md: Adjust.
2021 * config/rs6000/power5.md: Adjust.
2022 * config/rs6000/power6.md: Adjust.
2023 * config/rs6000/power7.md: Adjust.
2024 * config/rs6000/power8.md: Adjust.
2025 * config/rs6000/power9.md: Adjust.
2026 * config/rs6000/rs64.md: Adjust.
2027 * config/rs6000/titan.md: Adjust.
2029 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2031 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2032 ix86_indirect_branch_register logic.
2034 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2036 * config/i386/constraints.md (Bs): Update
2037 ix86_indirect_branch_register check. Don't check
2038 ix86_indirect_branch_register with GOT_memory_operand.
2040 * config/i386/predicates.md (GOT_memory_operand): Don't check
2041 ix86_indirect_branch_register here.
2042 (GOT32_symbol_operand): Likewise.
2044 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2046 * config/i386/predicates.md (constant_call_address_operand):
2047 Rewrite ix86_indirect_branch_register logic.
2048 (sibcall_insn_operand): Likewise.
2050 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2052 * config/i386/constraints.md (Bs): Replace
2053 ix86_indirect_branch_thunk_register with
2054 ix86_indirect_branch_register.
2056 * config/i386/i386.md (indirect_jump): Likewise.
2057 (tablejump): Likewise.
2058 (*sibcall_memory): Likewise.
2059 (*sibcall_value_memory): Likewise.
2060 Peepholes of indirect call and jump via memory: Likewise.
2061 * config/i386/i386.opt: Likewise.
2062 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2063 (GOT_memory_operand): Likewise.
2064 (call_insn_operand): Likewise.
2065 (sibcall_insn_operand): Likewise.
2066 (GOT32_symbol_operand): Likewise.
2068 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2071 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2072 type rather than type addr's type points to.
2073 (expand_omp_atomic_mutex): Likewise.
2074 (expand_omp_atomic): Likewise.
2076 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2079 * config/i386/i386.c (output_indirect_thunk_function): Use
2080 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2081 for __x86_return_thunk.
2083 2018-01-15 Richard Biener <rguenther@suse.de>
2086 * expmed.c (extract_bit_field_1): Fix typo.
2088 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2091 * config/arm/iterators.md (VF): New mode iterator.
2092 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2093 Remove integer-related logic from pattern.
2094 (neon_vabd<mode>_3): Likewise.
2096 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2099 * common.opt (fstrict-overflow): No longer an alias.
2100 (fwrapv-pointer): New option.
2101 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2102 also for pointer types based on flag_wrapv_pointer.
2103 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2104 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2105 opts->x_flag_wrapv got set.
2106 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2107 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2108 POINTER_TYPE_OVERFLOW_UNDEFINED.
2109 * match.pd: Likewise in address comparison pattern.
2110 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2112 2018-01-15 Richard Biener <rguenther@suse.de>
2115 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2116 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2117 Reset type names to their identifier if their TYPE_DECL doesn't
2118 have linkage (and thus is used for ODR and devirt).
2119 (save_debug_info_for_decl): Remove.
2120 (save_debug_info_for_type): Likewise.
2121 (add_tree_to_fld_list): Adjust.
2122 * tree-pretty-print.c (dump_generic_node): Make dumping of
2123 type names more robust.
2125 2018-01-15 Richard Biener <rguenther@suse.de>
2127 * BASE-VER: Bump to 8.0.1.
2129 2018-01-14 Martin Sebor <msebor@redhat.com>
2132 * builtins.c (check_access): Avoid warning when the no-warning bit
2135 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2137 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2138 * ira-color (allocno_hard_regs_compare): Likewise.
2140 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2143 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2144 Use .pushsection/.popsection.
2146 2018-01-14 Martin Sebor <msebor@redhat.com>
2149 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2151 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2153 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2154 entry from extra_headers.
2155 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2156 extra_headers, make the list bitwise identical to the i?86-*-* one.
2158 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2160 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2161 -mcmodel=large with -mindirect-branch=thunk,
2162 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2163 -mfunction-return=thunk-extern.
2164 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2165 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2166 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2168 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2170 * config/i386/i386.c (print_reg): Print the name of the full
2171 integer register without '%'.
2172 (ix86_print_operand): Handle 'V'.
2173 * doc/extend.texi: Document 'V' modifier.
2175 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2177 * config/i386/constraints.md (Bs): Disallow memory operand for
2178 -mindirect-branch-register.
2180 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2181 (GOT_memory_operand): Likewise.
2182 (call_insn_operand): Likewise.
2183 (sibcall_insn_operand): Likewise.
2184 (GOT32_symbol_operand): Likewise.
2185 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2186 for -mindirect-branch-register.
2187 (tablejump): Likewise.
2188 (*sibcall_memory): Likewise.
2189 (*sibcall_value_memory): Likewise.
2190 Disallow peepholes of indirect call and jump via memory for
2191 -mindirect-branch-register.
2192 (*call_pop): Replace m with Bw.
2193 (*call_value_pop): Likewise.
2194 (*sibcall_pop_memory): Replace m with Bs.
2195 * config/i386/i386.opt (mindirect-branch-register): New option.
2196 * doc/invoke.texi: Document -mindirect-branch-register option.
2198 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2200 * config/i386/i386-protos.h (ix86_output_function_return): New.
2201 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2202 set function_return_type.
2203 (indirect_thunk_name): Add ret_p to indicate thunk for function
2205 (output_indirect_thunk_function): Pass false to
2206 indirect_thunk_name.
2207 (ix86_output_indirect_branch_via_reg): Likewise.
2208 (ix86_output_indirect_branch_via_push): Likewise.
2209 (output_indirect_thunk_function): Create alias for function
2210 return thunk if regno < 0.
2211 (ix86_output_function_return): New function.
2212 (ix86_handle_fndecl_attribute): Handle function_return.
2213 (ix86_attribute_table): Add function_return.
2214 * config/i386/i386.h (machine_function): Add
2215 function_return_type.
2216 * config/i386/i386.md (simple_return_internal): Use
2217 ix86_output_function_return.
2218 (simple_return_internal_long): Likewise.
2219 * config/i386/i386.opt (mfunction-return=): New option.
2220 (indirect_branch): Mention -mfunction-return=.
2221 * doc/extend.texi: Document function_return function attribute.
2222 * doc/invoke.texi: Document -mfunction-return= option.
2224 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2226 * config/i386/i386-opts.h (indirect_branch): New.
2227 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2228 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2229 with local indirect jump when converting indirect call and jump.
2230 (ix86_set_indirect_branch_type): New.
2231 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2232 (indirectlabelno): New.
2233 (indirect_thunk_needed): Likewise.
2234 (indirect_thunk_bnd_needed): Likewise.
2235 (indirect_thunks_used): Likewise.
2236 (indirect_thunks_bnd_used): Likewise.
2237 (INDIRECT_LABEL): Likewise.
2238 (indirect_thunk_name): Likewise.
2239 (output_indirect_thunk): Likewise.
2240 (output_indirect_thunk_function): Likewise.
2241 (ix86_output_indirect_branch_via_reg): Likewise.
2242 (ix86_output_indirect_branch_via_push): Likewise.
2243 (ix86_output_indirect_branch): Likewise.
2244 (ix86_output_indirect_jmp): Likewise.
2245 (ix86_code_end): Call output_indirect_thunk_function if needed.
2246 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2248 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2249 (ix86_attribute_table): Add indirect_branch.
2250 * config/i386/i386.h (machine_function): Add indirect_branch_type
2251 and has_local_indirect_jump.
2252 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2254 (tablejump): Likewise.
2255 (*indirect_jump): Use ix86_output_indirect_jmp.
2256 (*tablejump_1): Likewise.
2257 (simple_return_indirect_internal): Likewise.
2258 * config/i386/i386.opt (mindirect-branch=): New option.
2259 (indirect_branch): New.
2262 (thunk-inline): Likewise.
2263 (thunk-extern): Likewise.
2264 * doc/extend.texi: Document indirect_branch function attribute.
2265 * doc/invoke.texi: Document -mindirect-branch= option.
2267 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2270 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2272 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2274 * ipa-inline.c (want_inline_small_function_p): Return false if
2275 inlining has already failed with CIF_FINAL_ERROR.
2276 (update_caller_keys): Call want_inline_small_function_p before
2278 (update_callee_keys): Likewise.
2280 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2282 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2284 (rs6000_quadword_masked_address_p): Likewise.
2285 (quad_aligned_load_p): Likewise.
2286 (quad_aligned_store_p): Likewise.
2287 (const_load_sequence_p): Add comment to describe the outer-most loop.
2288 (mimic_memory_attributes_and_flags): New function.
2289 (rs6000_gen_stvx): Likewise.
2290 (replace_swapped_aligned_store): Likewise.
2291 (rs6000_gen_lvx): Likewise.
2292 (replace_swapped_aligned_load): Likewise.
2293 (replace_swapped_load_constant): Capitalize argument name in
2294 comment describing this function.
2295 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2296 and stores that access quad-word aligned addresses and replace
2297 with stvx or lvx instructions when appropriate.
2298 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2299 New function prototype.
2300 (rs6000_quadword_masked_address_p): Likewise.
2301 (rs6000_gen_lvx): Likewise.
2302 (rs6000_gen_stvx): Likewise.
2303 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2304 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2305 when memory address is aligned.
2306 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2307 this split to select lvx instruction when memory address is aligned.
2308 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2309 instruction when memory address is aligned.
2310 (*vsx_le_perm_load_v16qi): Likewise.
2311 (four unnamed splitters): Modify to select the stvx instruction
2312 when memory is aligned.
2314 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2316 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2317 which appears in the queue multiple times.
2319 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2320 Alan Hayward <alan.hayward@arm.com>
2321 David Sherwood <david.sherwood@arm.com>
2323 * tree-vectorizer.h (vec_lower_bound): New structure.
2324 (_loop_vec_info): Add check_nonzero and lower_bounds.
2325 (LOOP_VINFO_CHECK_NONZERO): New macro.
2326 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2327 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2328 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2329 fields. Make seg_len the distance travelled, not including the
2331 (dr_direction_indicator): Declare.
2332 (dr_zero_step_indicator): Likewise.
2333 (dr_known_forward_stride_p): Likewise.
2334 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2336 (runtime_alias_check_p): Allow runtime alias checks with
2338 (operator ==): Compare access_size and align.
2339 (prune_runtime_alias_test_list): Rework for new distinction between
2340 the access_size and seg_len.
2341 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2343 (get_segment_min_max): New function.
2344 (create_intersect_range_checks): Use it.
2345 (dr_step_indicator): New function.
2346 (dr_direction_indicator): Likewise.
2347 (dr_zero_step_indicator): Likewise.
2348 (dr_known_forward_stride_p): Likewise.
2349 * tree-loop-distribution.c (data_ref_segment_size): Return
2350 DR_STEP * (niters - 1).
2351 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2353 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2354 (vect_preserves_scalar_order_p): New function, split out from...
2355 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2356 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2357 (vect_vfa_access_size): New function.
2358 (vect_vfa_align): Likewise.
2359 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2360 (dump_lower_bound): New function.
2361 (vect_check_lower_bound): Likewise.
2362 (vect_small_gap_p): Likewise.
2363 (vectorizable_with_step_bound_p): Likewise.
2364 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2365 depencies if the vectorization factor is 1. Convert the checks
2366 for nonzero steps into checks on the bounds of DR_STEP. Try using
2367 a bunds check for variable steps if the minimum required step is
2368 relatively small. Update calls to the dr_with_seg_len
2369 constructor and to vect_compile_time_alias.
2370 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2372 (vect_loop_versioning): Call it.
2373 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2375 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2377 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2378 Alan Hayward <alan.hayward@arm.com>
2379 David Sherwood <david.sherwood@arm.com>
2381 * doc/sourcebuild.texi (vect_scatter_store): Document.
2382 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2384 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2386 * genopinit.c (main): Add supports_vec_scatter_store and
2387 supports_vec_scatter_store_cached to target_optabs.
2388 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2389 IFN_MASK_SCATTER_STORE.
2390 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2392 * internal-fn.h (internal_store_fn_p): Declare.
2393 (internal_fn_stored_value_index): Likewise.
2394 * internal-fn.c (scatter_store_direct): New macro.
2395 (expand_scatter_store_optab_fn): New function.
2396 (direct_scatter_store_optab_supported_p): New macro.
2397 (internal_store_fn_p): New function.
2398 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2399 IFN_MASK_SCATTER_STORE.
2400 (internal_fn_mask_index): Likewise.
2401 (internal_fn_stored_value_index): New function.
2402 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2404 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2405 * optabs-query.c (supports_vec_scatter_store_p): New function.
2406 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2407 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2408 true for scatter stores.
2409 (vect_gather_scatter_fn_p): Handle scatter stores too.
2410 (vect_check_gather_scatter): Consider using scatter stores if
2411 supports_vec_scatter_store_p.
2412 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2414 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2415 internal_fn_stored_value_index.
2416 (check_load_store_masking): Handle scatter stores too.
2417 (vect_get_store_rhs): Make public.
2418 (vectorizable_call): Use internal_store_fn_p.
2419 (vectorizable_store): Handle scatter store internal functions.
2420 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2421 when deciding whether the end of the group has been reached.
2422 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2423 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2424 (mask_scatter_store<mode>): New insns.
2426 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2427 Alan Hayward <alan.hayward@arm.com>
2428 David Sherwood <david.sherwood@arm.com>
2430 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2431 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2432 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2434 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2435 Use vect_truncate_gather_scatter_offset if we can't treat the
2436 operation as a normal gather load or scatter store.
2437 (get_group_load_store_type): Take the gather_scatter_info
2438 as argument. Try using a gather load or scatter store for
2439 single-element groups.
2440 (get_load_store_type): Update calls to get_group_load_store_type
2441 and vect_use_strided_gather_scatters_p.
2443 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2444 Alan Hayward <alan.hayward@arm.com>
2445 David Sherwood <david.sherwood@arm.com>
2447 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2448 optional tree argument.
2449 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2451 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2452 but continue to use the current value as a fallback.
2453 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2454 to compare the updates.
2455 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2456 (get_load_store_type): Use it when handling a strided access.
2457 (vect_get_strided_load_store_ops): New function.
2458 (vect_get_data_ptr_increment): Likewise.
2459 (vectorizable_load): Handle strided gather loads. Always pass
2460 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2462 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2463 Alan Hayward <alan.hayward@arm.com>
2464 David Sherwood <david.sherwood@arm.com>
2466 * doc/md.texi (gather_load@var{m}): Document.
2467 (mask_gather_load@var{m}): Likewise.
2468 * genopinit.c (main): Add supports_vec_gather_load and
2469 supports_vec_gather_load_cached to target_optabs.
2470 * optabs-tree.c (init_tree_optimization_optabs): Use
2471 ggc_cleared_alloc to allocate target_optabs.
2472 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2473 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2475 * internal-fn.h (internal_load_fn_p): Declare.
2476 (internal_gather_scatter_fn_p): Likewise.
2477 (internal_fn_mask_index): Likewise.
2478 (internal_gather_scatter_fn_supported_p): Likewise.
2479 * internal-fn.c (gather_load_direct): New macro.
2480 (expand_gather_load_optab_fn): New function.
2481 (direct_gather_load_optab_supported_p): New macro.
2482 (direct_internal_fn_optab): New function.
2483 (internal_load_fn_p): Likewise.
2484 (internal_gather_scatter_fn_p): Likewise.
2485 (internal_fn_mask_index): Likewise.
2486 (internal_gather_scatter_fn_supported_p): Likewise.
2487 * optabs-query.c (supports_at_least_one_mode_p): New function.
2488 (supports_vec_gather_load_p): Likewise.
2489 * optabs-query.h (supports_vec_gather_load_p): Declare.
2490 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2491 and memory_type field.
2492 (NUM_PATTERNS): Bump to 15.
2493 * tree-vect-data-refs.c: Include internal-fn.h.
2494 (vect_gather_scatter_fn_p): New function.
2495 (vect_describe_gather_scatter_call): Likewise.
2496 (vect_check_gather_scatter): Try using internal functions for
2497 gather loads. Recognize existing calls to a gather load function.
2498 (vect_analyze_data_refs): Consider using gather loads if
2499 supports_vec_gather_load_p.
2500 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2501 (vect_get_gather_scatter_offset_type): Likewise.
2502 (vect_convert_mask_for_vectype): Likewise.
2503 (vect_add_conversion_to_patterm): Likewise.
2504 (vect_try_gather_scatter_pattern): Likewise.
2505 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2506 (vect_vect_recog_func_ptrs): Add it.
2507 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2508 internal_fn_mask_index and internal_gather_scatter_fn_p.
2509 (check_load_store_masking): Take the gather_scatter_info as an
2510 argument and handle gather loads.
2511 (vect_get_gather_scatter_ops): New function.
2512 (vectorizable_call): Check internal_load_fn_p.
2513 (vectorizable_load): Likewise. Handle gather load internal
2515 (vectorizable_store): Update call to check_load_store_masking.
2516 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2517 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2518 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2519 (aarch64_gather_scale_operand_d): New predicates.
2520 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2521 (mask_gather_load<mode>): New insns.
2523 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2524 Alan Hayward <alan.hayward@arm.com>
2525 David Sherwood <david.sherwood@arm.com>
2527 * optabs.def (fold_left_plus_optab): New optab.
2528 * doc/md.texi (fold_left_plus_@var{m}): Document.
2529 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2530 * internal-fn.c (fold_left_direct): Define.
2531 (expand_fold_left_optab_fn): Likewise.
2532 (direct_fold_left_optab_supported_p): Likewise.
2533 * fold-const-call.c (fold_const_fold_left): New function.
2534 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2535 * tree-parloops.c (valid_reduction_p): New function.
2536 (gather_scalar_reductions): Use it.
2537 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2538 (vect_finish_replace_stmt): Declare.
2539 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2540 (needs_fold_left_reduction_p): New function, split out from...
2541 (vect_is_simple_reduction): ...here. Accept reductions that
2542 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2543 (vect_force_simple_reduction): Also store the reduction type in
2544 the assignment's STMT_VINFO_REDUC_TYPE.
2545 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2546 (merge_with_identity): New function.
2547 (vect_expand_fold_left): Likewise.
2548 (vectorize_fold_left_reduction): Likewise.
2549 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2550 scalar phi in place for it. Check for target support and reject
2551 cases that would reassociate the operation. Defer the transform
2552 phase to vectorize_fold_left_reduction.
2553 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2554 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2555 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2557 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2559 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2560 call to ifc_temp_var.
2562 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2563 Alan Hayward <alan.hayward@arm.com>
2564 David Sherwood <david.sherwood@arm.com>
2566 * target.def (legitimize_address_displacement): Take the original
2567 offset as a poly_int.
2568 * targhooks.h (default_legitimize_address_displacement): Update
2570 * targhooks.c (default_legitimize_address_displacement): Likewise.
2571 * doc/tm.texi: Regenerate.
2572 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2573 as an argument, moving assert of ad->disp == ad->disp_term to...
2574 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2575 Try calling targetm.legitimize_address_displacement before expanding
2576 the address rather than afterwards, and adjust for the new interface.
2577 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2578 Match the new hook interface. Handle SVE addresses.
2579 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2582 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2584 * Makefile.in (OBJS): Add early-remat.o.
2585 * target.def (select_early_remat_modes): New hook.
2586 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2587 * doc/tm.texi: Regenerate.
2588 * targhooks.h (default_select_early_remat_modes): Declare.
2589 * targhooks.c (default_select_early_remat_modes): New function.
2590 * timevar.def (TV_EARLY_REMAT): New timevar.
2591 * passes.def (pass_early_remat): New pass.
2592 * tree-pass.h (make_pass_early_remat): Declare.
2593 * early-remat.c: New file.
2594 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2596 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2598 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2599 Alan Hayward <alan.hayward@arm.com>
2600 David Sherwood <david.sherwood@arm.com>
2602 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2603 vfm1 with a bound_epilog parameter.
2604 (vect_do_peeling): Update calls accordingly, and move the prologue
2605 call earlier in the function. Treat the base bound_epilog as 0 for
2606 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2607 this base when peeling for gaps.
2608 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2609 with fully-masked loops.
2610 (vect_estimate_min_profitable_iters): Handle the single peeled
2611 iteration in that case.
2613 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2614 Alan Hayward <alan.hayward@arm.com>
2615 David Sherwood <david.sherwood@arm.com>
2617 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2618 single-element interleaving even if the size is not a power of 2.
2619 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2620 accesses for single-element interleaving if the group size is
2623 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2624 Alan Hayward <alan.hayward@arm.com>
2625 David Sherwood <david.sherwood@arm.com>
2627 * doc/md.texi (fold_extract_last_@var{m}): Document.
2628 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2629 * optabs.def (fold_extract_last_optab): New optab.
2630 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2631 * internal-fn.c (fold_extract_direct): New macro.
2632 (expand_fold_extract_optab_fn): Likewise.
2633 (direct_fold_extract_optab_supported_p): Likewise.
2634 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2635 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2636 EXTRACT_LAST_REDUCTION.
2637 (get_initial_def_for_reduction): Do not create an initial vector
2638 for EXTRACT_LAST_REDUCTION reductions.
2639 (vectorizable_reduction): Leave the scalar phi in place for
2640 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2641 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2642 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2643 transform phase to vectorizable_condition.
2644 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2646 (vect_finish_stmt_generation): ...here.
2647 (vect_finish_replace_stmt): New function.
2648 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2649 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2651 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2653 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2654 Alan Hayward <alan.hayward@arm.com>
2655 David Sherwood <david.sherwood@arm.com>
2657 * doc/md.texi (extract_last_@var{m}): Document.
2658 * optabs.def (extract_last_optab): New optab.
2659 * internal-fn.def (EXTRACT_LAST): New internal function.
2660 * internal-fn.c (cond_unary_direct): New macro.
2661 (expand_cond_unary_optab_fn): Likewise.
2662 (direct_cond_unary_optab_supported_p): Likewise.
2663 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2664 loops using EXTRACT_LAST.
2665 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2666 (extract_last_<mode>): ...this optab.
2667 (vec_extract<mode><Vel>): Update accordingly.
2669 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2670 Alan Hayward <alan.hayward@arm.com>
2671 David Sherwood <david.sherwood@arm.com>
2673 * target.def (empty_mask_is_expensive): New hook.
2674 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2675 * doc/tm.texi: Regenerate.
2676 * targhooks.h (default_empty_mask_is_expensive): Declare.
2677 * targhooks.c (default_empty_mask_is_expensive): New function.
2678 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2679 if the target says that empty masks are expensive.
2680 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2682 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2684 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2685 Alan Hayward <alan.hayward@arm.com>
2686 David Sherwood <david.sherwood@arm.com>
2688 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2689 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2690 (vect_use_loop_mask_for_alignment_p): New function.
2691 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2692 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2693 niters_skip argument. Make sure that the first niters_skip elements
2694 of the first iteration are inactive.
2695 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2696 Update call to vect_set_loop_masks_directly.
2697 (get_misalign_in_elems): New function, split out from...
2698 (vect_gen_prolog_loop_niters): ...here.
2699 (vect_update_init_of_dr): Take a code argument that specifies whether
2700 the adjustment should be added or subtracted.
2701 (vect_update_init_of_drs): Likewise.
2702 (vect_prepare_for_masked_peels): New function.
2703 (vect_do_peeling): Skip prologue peeling if we're using a mask
2704 instead. Update call to vect_update_inits_of_drs.
2705 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2707 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2708 alignment. Do not include the number of peeled iterations in
2709 the minimum threshold in that case.
2710 (vectorizable_induction): Adjust the start value down by
2711 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2712 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2713 Take the number of skipped iterations into account when calculating
2715 * tree-vect-stmts.c (vect_gen_while_not): New function.
2717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2718 Alan Hayward <alan.hayward@arm.com>
2719 David Sherwood <david.sherwood@arm.com>
2721 * doc/sourcebuild.texi (vect_fully_masked): Document.
2722 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2724 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2726 (vect_analyze_loop_2): ...here. Don't check the vectorization
2727 factor against the number of loop iterations if the loop is
2730 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2731 Alan Hayward <alan.hayward@arm.com>
2732 David Sherwood <david.sherwood@arm.com>
2734 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2735 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2736 (dump_groups): Update accordingly.
2737 (iv_use::mem_type): New member variable.
2738 (address_p): New function.
2739 (record_use): Add a mem_type argument and initialize the new
2741 (record_group_use): Add a mem_type argument. Use address_p.
2742 Remove obsolete null checks of base_object. Update call to record_use.
2743 (find_interesting_uses_op): Update call to record_group_use.
2744 (find_interesting_uses_cond): Likewise.
2745 (find_interesting_uses_address): Likewise.
2746 (get_mem_type_for_internal_fn): New function.
2747 (find_address_like_use): Likewise.
2748 (find_interesting_uses_stmt): Try find_address_like_use before
2749 calling find_interesting_uses_op.
2750 (addr_offset_valid_p): Use the iv mem_type field as the type
2751 of the addressed memory.
2752 (add_autoinc_candidates): Likewise.
2753 (get_address_cost): Likewise.
2754 (split_small_address_groups_p): Use address_p.
2755 (split_address_groups): Likewise.
2756 (add_iv_candidate_for_use): Likewise.
2757 (autoinc_possible_for_pair): Likewise.
2758 (rewrite_groups): Likewise.
2759 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2760 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2761 (get_alias_ptr_type_for_ptr_address): New function.
2762 (rewrite_use_address): Rewrite address uses in calls that were
2763 identified by find_address_like_use.
2765 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2766 Alan Hayward <alan.hayward@arm.com>
2767 David Sherwood <david.sherwood@arm.com>
2769 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2771 * gimple-expr.h (is_gimple_addressable: Likewise.
2772 * gimple-expr.c (is_gimple_address): Likewise.
2773 * internal-fn.c (expand_call_mem_ref): New function.
2774 (expand_mask_load_optab_fn): Use it.
2775 (expand_mask_store_optab_fn): Likewise.
2777 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2778 Alan Hayward <alan.hayward@arm.com>
2779 David Sherwood <david.sherwood@arm.com>
2781 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2782 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2783 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2784 (cond_umax@var{mode}): Document.
2785 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2786 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2787 (cond_umin_optab, cond_umax_optab): New optabs.
2788 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2789 (COND_IOR, COND_XOR): New internal functions.
2790 * internal-fn.h (get_conditional_internal_fn): Declare.
2791 * internal-fn.c (cond_binary_direct): New macro.
2792 (expand_cond_binary_optab_fn): Likewise.
2793 (direct_cond_binary_optab_supported_p): Likewise.
2794 (get_conditional_internal_fn): New function.
2795 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2796 Cope with reduction statements that are vectorized as calls rather
2798 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2799 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2800 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2801 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2802 (UNSPEC_COND_EOR): New unspecs.
2803 (optab): Add mappings for them.
2804 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2805 (sve_int_op, sve_fp_op): New int attributes.
2807 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2808 Alan Hayward <alan.hayward@arm.com>
2809 David Sherwood <david.sherwood@arm.com>
2811 * optabs.def (while_ult_optab): New optab.
2812 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2813 * internal-fn.def (WHILE_ULT): New internal function.
2814 * internal-fn.h (direct_internal_fn_supported_p): New override
2815 that takes two types as argument.
2816 * internal-fn.c (while_direct): New macro.
2817 (expand_while_optab_fn): New function.
2818 (convert_optab_supported_p): Likewise.
2819 (direct_while_optab_supported_p): New macro.
2820 * wide-int.h (wi::udiv_ceil): New function.
2821 * tree-vectorizer.h (rgroup_masks): New structure.
2822 (vec_loop_masks): New typedef.
2823 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2825 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2826 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2827 (vect_max_vf): New function.
2828 (slpeel_make_loop_iterate_ntimes): Delete.
2829 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2830 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2831 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2832 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2833 internal-fn.h, stor-layout.h and optabs-query.h.
2834 (vect_set_loop_mask): New function.
2835 (add_preheader_seq): Likewise.
2836 (add_header_seq): Likewise.
2837 (interleave_supported_p): Likewise.
2838 (vect_maybe_permute_loop_masks): Likewise.
2839 (vect_set_loop_masks_directly): Likewise.
2840 (vect_set_loop_condition_masked): Likewise.
2841 (vect_set_loop_condition_unmasked): New function, split out from
2842 slpeel_make_loop_iterate_ntimes.
2843 (slpeel_make_loop_iterate_ntimes): Rename to..
2844 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2845 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2846 (vect_do_peeling): Update call accordingly.
2847 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2849 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2850 mask_compare_type, can_fully_mask_p and fully_masked_p.
2851 (release_vec_loop_masks): New function.
2852 (_loop_vec_info): Use it to free the loop masks.
2853 (can_produce_all_loop_masks_p): New function.
2854 (vect_get_max_nscalars_per_iter): Likewise.
2855 (vect_verify_full_masking): Likewise.
2856 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2857 retries, and free the mask rgroups before retrying. Check loop-wide
2858 reasons for disallowing fully-masked loops. Make the final decision
2859 about whether use a fully-masked loop or not.
2860 (vect_estimate_min_profitable_iters): Do not assume that peeling
2861 for the number of iterations will be needed for fully-masked loops.
2862 (vectorizable_reduction): Disable fully-masked loops.
2863 (vectorizable_live_operation): Likewise.
2864 (vect_halve_mask_nunits): New function.
2865 (vect_double_mask_nunits): Likewise.
2866 (vect_record_loop_mask): Likewise.
2867 (vect_get_loop_mask): Likewise.
2868 (vect_transform_loop): Handle the case in which the final loop
2869 iteration might handle a partial vector. Call vect_set_loop_condition
2870 instead of slpeel_make_loop_iterate_ntimes.
2871 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2872 (check_load_store_masking): New function.
2873 (prepare_load_store_mask): Likewise.
2874 (vectorizable_store): Handle fully-masked loops.
2875 (vectorizable_load): Likewise.
2876 (supportable_widening_operation): Use vect_halve_mask_nunits for
2878 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2879 (vect_gen_while): New function.
2880 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2881 (aarch64_uqdec<mode>): New insn.
2883 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2884 Alan Hayward <alan.hayward@arm.com>
2885 David Sherwood <david.sherwood@arm.com>
2887 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2888 (reduc_xor_scal_optab): New optabs.
2889 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2890 (reduc_xor_scal_@var{m}): Document.
2891 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2892 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2894 * fold-const-call.c (fold_const_call): Handle them.
2895 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2896 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2897 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2898 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2899 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2900 (UNSPEC_XORV): New unspecs.
2901 (optab): Add entries for them.
2902 (BITWISEV): New int iterator.
2903 (bit_reduc_op): New int attributes.
2905 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2906 Alan Hayward <alan.hayward@arm.com>
2907 David Sherwood <david.sherwood@arm.com>
2909 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2910 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2911 * optabs.def (vec_shl_insert_optab): New optab.
2912 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2913 (duplicate_and_interleave): Likewise.
2914 * tree-vect-loop.c: Include internal-fn.h.
2915 (neutral_op_for_slp_reduction): New function, split out from
2916 get_initial_defs_for_reduction.
2917 (get_initial_def_for_reduction): Handle option 2 for variable-length
2918 vectors by loading the neutral value into a vector and then shifting
2919 the initial value into element 0.
2920 (get_initial_defs_for_reduction): Replace the code argument with
2921 the neutral value calculated by neutral_op_for_slp_reduction.
2922 Use gimple_build_vector for constant-length vectors.
2923 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2924 but the first group_size elements have a neutral value.
2925 Use duplicate_and_interleave otherwise.
2926 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2927 Update call to get_initial_defs_for_reduction. Handle SLP
2928 reductions for variable-length vectors by creating one vector
2929 result for each scalar result, with the elements associated
2930 with other scalar results stubbed out with the neutral value.
2931 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2932 Require IFN_VEC_SHL_INSERT for double reductions on
2933 variable-length vectors, or SLP reductions that have
2934 a neutral value. Require can_duplicate_and_interleave_p
2935 support for variable-length unchained SLP reductions if there
2936 is no neutral value, such as for MIN/MAX reductions. Also require
2937 the number of vector elements to be a multiple of the number of
2938 SLP statements when doing variable-length unchained SLP reductions.
2939 Update call to vect_create_epilog_for_reduction.
2940 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2941 and remove initial values.
2942 (duplicate_and_interleave): Make public.
2943 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2944 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2946 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2947 Alan Hayward <alan.hayward@arm.com>
2948 David Sherwood <david.sherwood@arm.com>
2950 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2951 (can_duplicate_and_interleave_p): New function.
2952 (vect_get_and_check_slp_defs): Take the vector of statements
2953 rather than just the current one. Remove excess parentheses.
2954 Restriction rejectinon of vect_constant_def and vect_external_def
2955 for variable-length vectors to boolean types, or types for which
2956 can_duplicate_and_interleave_p is false.
2957 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2958 (duplicate_and_interleave): New function.
2959 (vect_get_constant_vectors): Use gimple_build_vector for
2960 constant-length vectors and suitable variable-length constant
2961 vectors. Use duplicate_and_interleave for other variable-length
2962 vectors. Don't defer the update when inserting new statements.
2964 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2965 Alan Hayward <alan.hayward@arm.com>
2966 David Sherwood <david.sherwood@arm.com>
2968 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2969 min_profitable_iters doesn't go negative.
2971 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2972 Alan Hayward <alan.hayward@arm.com>
2973 David Sherwood <david.sherwood@arm.com>
2975 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2976 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2977 * optabs.def (vec_mask_load_lanes_optab): New optab.
2978 (vec_mask_store_lanes_optab): Likewise.
2979 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2980 (MASK_STORE_LANES): Likewise.
2981 * internal-fn.c (mask_load_lanes_direct): New macro.
2982 (mask_store_lanes_direct): Likewise.
2983 (expand_mask_load_optab_fn): Handle masked operations.
2984 (expand_mask_load_lanes_optab_fn): New macro.
2985 (expand_mask_store_optab_fn): Handle masked operations.
2986 (expand_mask_store_lanes_optab_fn): New macro.
2987 (direct_mask_load_lanes_optab_supported_p): Likewise.
2988 (direct_mask_store_lanes_optab_supported_p): Likewise.
2989 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2991 (vect_load_lanes_supported): Likewise.
2992 * tree-vect-data-refs.c (strip_conversion): New function.
2993 (can_group_stmts_p): Likewise.
2994 (vect_analyze_data_ref_accesses): Use it instead of checking
2995 for a pair of assignments.
2996 (vect_store_lanes_supported): Take a masked_p parameter.
2997 (vect_load_lanes_supported): Likewise.
2998 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2999 vect_store_lanes_supported and vect_load_lanes_supported.
3000 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3001 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3002 parameter. Don't allow gaps for masked accesses.
3003 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3004 and vect_load_lanes_supported.
3005 (get_load_store_type): Take a masked_p parameter and update
3006 call to get_group_load_store_type.
3007 (vectorizable_store): Update call to get_load_store_type.
3008 Handle IFN_MASK_STORE_LANES.
3009 (vectorizable_load): Update call to get_load_store_type.
3010 Handle IFN_MASK_LOAD_LANES.
3012 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3013 Alan Hayward <alan.hayward@arm.com>
3014 David Sherwood <david.sherwood@arm.com>
3016 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3018 * config/aarch64/aarch64-protos.h
3019 (aarch64_sve_struct_memory_operand_p): Declare.
3020 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3021 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3022 (VPRED, vpred): Handle SVE structure modes.
3023 * config/aarch64/constraints.md (Utx): New constraint.
3024 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3025 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3026 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3027 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3028 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3029 structure modes. Split into pieces after RA.
3030 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3031 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3033 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3034 SVE structure modes.
3035 (aarch64_classify_address): Likewise.
3036 (sizetochar): Move earlier in file.
3037 (aarch64_print_operand): Handle SVE register lists.
3038 (aarch64_array_mode): New function.
3039 (aarch64_sve_struct_memory_operand_p): Likewise.
3040 (TARGET_ARRAY_MODE): Redefine.
3042 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3043 Alan Hayward <alan.hayward@arm.com>
3044 David Sherwood <david.sherwood@arm.com>
3046 * target.def (array_mode): New target hook.
3047 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3048 * doc/tm.texi: Regenerate.
3049 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3050 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3051 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3053 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3056 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3057 Alan Hayward <alan.hayward@arm.com>
3058 David Sherwood <david.sherwood@arm.com>
3060 * fold-const.c (fold_binary_loc): Check the argument types
3061 rather than the result type when testing for a vector operation.
3063 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3065 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3066 * doc/tm.texi: Regenerate.
3068 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3069 Alan Hayward <alan.hayward@arm.com>
3070 David Sherwood <david.sherwood@arm.com>
3072 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3073 (sve): Document new AArch64 extension.
3074 * doc/md.texi (w): Extend the description of the AArch64
3075 constraint to include SVE vectors.
3076 (Upl, Upa): Document new AArch64 predicate constraints.
3077 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3079 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3080 (msve-vector-bits=): New option.
3081 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3082 SVE when these are disabled.
3083 (sve): New extension.
3084 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3085 modes. Adjust their number of units based on aarch64_sve_vg.
3086 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3087 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3088 aarch64_addr_query_type.
3089 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3090 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3091 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3092 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3093 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3094 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3095 (aarch64_simd_imm_zero_p): Delete.
3096 (aarch64_check_zero_based_sve_index_immediate): Declare.
3097 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3098 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3099 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3100 (aarch64_sve_float_mul_immediate_p): Likewise.
3101 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3103 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3104 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3105 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3106 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3107 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3108 (aarch64_regmode_natural_size): Likewise.
3109 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3110 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3112 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3113 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3114 for VG and the SVE predicate registers.
3115 (V_ALIASES): Add a "z"-prefixed alias.
3116 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3117 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3118 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3119 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3120 (REG_CLASS_NAMES): Add entries for them.
3121 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3122 and the predicate registers.
3123 (aarch64_sve_vg): Declare.
3124 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3125 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3126 (REGMODE_NATURAL_SIZE): Define.
3127 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3129 * config/aarch64/aarch64.c: Include cfgrtl.h.
3130 (simd_immediate_info): Add a constructor for series vectors,
3131 and an associated step field.
3132 (aarch64_sve_vg): New variable.
3133 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3134 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3135 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3136 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3137 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3138 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3139 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3140 (aarch64_get_mask_mode): New functions.
3141 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3142 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3143 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3144 predicate modes and predicate registers. Explicitly restrict
3145 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3146 to store a vector mode if it is recognized by
3147 aarch64_classify_vector_mode.
3148 (aarch64_regmode_natural_size): New function.
3149 (aarch64_hard_regno_caller_save_mode): Return the original mode
3151 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3152 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3153 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3154 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3156 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3157 does not overlap dest if the function is frame-related. Handle
3159 (aarch64_split_add_offset): New function.
3160 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3161 them aarch64_add_offset.
3162 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3163 and update call to aarch64_sub_sp.
3164 (aarch64_add_cfa_expression): New function.
3165 (aarch64_expand_prologue): Pass extra temporary registers to the
3166 functions above. Handle the case in which we need to emit new
3167 DW_CFA_expressions for registers that were originally saved
3168 relative to the stack pointer, but now have to be expressed
3169 relative to the frame pointer.
3170 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3172 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3173 IP0 and IP1 values for SVE frames.
3174 (aarch64_expand_vec_series): New function.
3175 (aarch64_expand_sve_widened_duplicate): Likewise.
3176 (aarch64_expand_sve_const_vector): Likewise.
3177 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3178 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3179 into the register, rather than emitting a SET directly.
3180 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3181 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3182 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3183 (offset_9bit_signed_scaled_p): New functions.
3184 (aarch64_replicate_bitmask_imm): New function.
3185 (aarch64_bitmask_imm): Use it.
3186 (aarch64_cannot_force_const_mem): Reject expressions involving
3187 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3188 (aarch64_classify_index): Handle SVE indices, by requiring
3189 a plain register index with a scale that matches the element size.
3190 (aarch64_classify_address): Handle SVE addresses. Assert that
3191 the mode of the address is VOIDmode or an integer mode.
3192 Update call to aarch64_classify_symbol.
3193 (aarch64_classify_symbolic_expression): Update call to
3194 aarch64_classify_symbol.
3195 (aarch64_const_vec_all_in_range_p): New function.
3196 (aarch64_print_vector_float_operand): Likewise.
3197 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3198 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3199 and the FP immediates 1.0 and 0.5.
3200 (aarch64_print_address_internal): Handle SVE addresses.
3201 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3202 (aarch64_regno_regclass): Handle predicate registers.
3203 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3205 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3206 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3207 (aarch64_convert_sve_vector_bits): New function.
3208 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3209 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3211 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3212 Handle SVE vector and predicate modes. Accept VL-based constants
3213 that need only one temporary register, and VL offsets that require
3214 no temporary registers.
3215 (aarch64_conditional_register_usage): Mark the predicate registers
3216 as fixed if SVE isn't available.
3217 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3218 Return true for SVE vector and predicate modes.
3219 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3220 rather than an unsigned int. Handle SVE modes.
3221 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3223 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3225 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3226 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3227 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3228 (aarch64_sve_float_mul_immediate_p): New functions.
3229 (aarch64_sve_valid_immediate): New function.
3230 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3231 Explicitly reject structure modes. Check for INDEX constants.
3232 Handle PTRUE and PFALSE constants.
3233 (aarch64_check_zero_based_sve_index_immediate): New function.
3234 (aarch64_simd_imm_zero_p): Delete.
3235 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3236 vector modes. Accept constants in the range of CNT[BHWD].
3237 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3238 ask for an Advanced SIMD mode.
3239 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3240 (aarch64_simd_vector_alignment): Handle SVE predicates.
3241 (aarch64_vectorize_preferred_vector_alignment): New function.
3242 (aarch64_simd_vector_alignment_reachable): Use it instead of
3244 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3245 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3247 (MAX_VECT_LEN): Delete.
3248 (expand_vec_perm_d): Add a vec_flags field.
3249 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3250 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3251 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3253 (aarch64_evpc_rev): Rename to...
3254 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3255 (aarch64_evpc_rev_global): New function.
3256 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3257 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3259 (aarch64_evpc_sve_tbl): New function.
3260 (aarch64_expand_vec_perm_const_1): Update after rename of
3261 aarch64_evpc_rev. Handle SVE permutes too, trying
3262 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3263 than aarch64_evpc_tbl.
3264 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3265 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3266 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3267 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3268 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3269 (aarch64_expand_sve_vcond): New functions.
3270 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3271 of aarch64_vector_mode_p.
3272 (aarch64_dwarf_poly_indeterminate_value): New function.
3273 (aarch64_compute_pressure_classes): Likewise.
3274 (aarch64_can_change_mode_class): Likewise.
3275 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3276 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3277 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3278 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3279 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3280 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3281 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3282 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3284 (Dn, Dl, Dr): Accept const as well as const_vector.
3285 (Dz): Likewise. Compare against CONST0_RTX.
3286 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3287 of "vector" where appropriate.
3288 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3289 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3290 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3291 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3292 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3293 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3294 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3295 (v_int_equiv): Extend to SVE modes.
3296 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3298 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3299 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3300 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3301 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3302 (SVE_COND_FP_CMP): New int iterators.
3303 (perm_hilo): Handle the new unpack unspecs.
3304 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3306 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3307 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3308 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3309 (aarch64_equality_operator, aarch64_constant_vector_operand)
3310 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3311 (aarch64_sve_nonimmediate_operand): Likewise.
3312 (aarch64_sve_general_operand): Likewise.
3313 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3314 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3315 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3316 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3317 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3318 (aarch64_sve_float_arith_immediate): Likewise.
3319 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3320 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3321 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3322 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3323 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3324 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3325 (aarch64_sve_float_arith_operand): Likewise.
3326 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3327 (aarch64_sve_float_mul_operand): Likewise.
3328 (aarch64_sve_vec_perm_operand): Likewise.
3329 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3330 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3331 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3332 as well as const_vector.
3333 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3334 in file. Use CONST0_RTX and CONSTM1_RTX.
3335 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3336 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3337 Use aarch64_simd_imm_zero.
3338 * config/aarch64/aarch64-sve.md: New file.
3339 * config/aarch64/aarch64.md: Include it.
3340 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3341 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3342 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3343 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3344 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3345 (sve): New attribute.
3346 (enabled): Disable instructions with the sve attribute unless
3348 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3349 aarch64_expand_mov_immediate.
3350 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3351 CNT[BHSD] immediates.
3352 (movti): Split CONST_POLY_INT moves into two halves.
3353 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3354 Split additions that need a temporary here if the destination
3355 is the stack pointer.
3356 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3357 (*add<mode>3_poly_1): New instruction.
3358 (set_clobber_cc): New expander.
3360 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3362 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3363 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3364 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3365 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3366 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3367 Change innermode from fixed_mode_size to machine_mode.
3368 (simplify_subreg): Update call accordingly. Handle a constant-sized
3369 subreg of a variable-length CONST_VECTOR.
3371 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3372 Alan Hayward <alan.hayward@arm.com>
3373 David Sherwood <david.sherwood@arm.com>
3375 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3376 (add_offset_to_base): New function, split out from...
3377 (create_mem_ref): ...here. When handling a scale other than 1,
3378 check first whether the address is valid without the offset.
3379 Add it into the base if so, leaving the index and scale as-is.
3381 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3384 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3385 fold_for_warn before checking if arg2 is INTEGER_CST.
3387 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3389 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3390 (store_multiple_operation): Delete.
3391 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3392 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3393 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3394 guarded by TARGET_STRING.
3395 (rs6000_output_load_multiple): Delete.
3396 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3397 OPTION_MASK_STRING / TARGET_STRING handling.
3398 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3399 (const rs6000_opt_masks) <"string">: Change mask to 0.
3400 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3401 (MASK_STRING): Delete.
3402 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3404 (load_multiple): Delete.
3411 (store_multiple): Delete.
3418 (movmemsi_8reg): Delete.
3419 (corresponding unnamed define_insn): Delete.
3420 (movmemsi_6reg): Delete.
3421 (corresponding unnamed define_insn): Delete.
3422 (movmemsi_4reg): Delete.
3423 (corresponding unnamed define_insn): Delete.
3424 (movmemsi_2reg): Delete.
3425 (corresponding unnamed define_insn): Delete.
3426 (movmemsi_1reg): Delete.
3427 (corresponding unnamed define_insn): Delete.
3428 * config/rs6000/rs6000.opt (mno-string): New.
3429 (mstring): Replace by deprecation warning stub.
3430 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3432 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3434 * regrename.c (regrename_do_replace): If replacing the same
3435 reg multiple times, try to reuse last created gen_raw_REG.
3438 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3439 main to workaround a bug in GDB.
3441 2018-01-12 Tom de Vries <tom@codesourcery.com>
3444 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3446 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3448 PR rtl-optimization/80481
3449 * ira-color.c (get_cap_member): New function.
3450 (allocnos_conflict_by_live_ranges_p): Use it.
3451 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3452 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3454 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3457 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3458 (*saddl_se_1): Ditto.
3460 (*ssubl_se_1): Ditto.
3462 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3464 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3465 rather than wi::to_widest for DR_INITs.
3466 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3467 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3468 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3470 (vect_analyze_group_access_1): Note that here.
3472 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3474 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3475 polynomial type sizes.
3477 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3479 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3480 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3481 (gimple_add_tmp_var): Likewise.
3483 2018-01-12 Martin Liska <mliska@suse.cz>
3485 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3486 (gimple_alloc_sizes): Likewise.
3487 (dump_gimple_statistics): Use PRIu64 in printf format.
3488 * gimple.h: Change uint64_t to int.
3490 2018-01-12 Martin Liska <mliska@suse.cz>
3492 * tree-core.h: Use uint64_t instead of int.
3493 * tree.c (tree_node_counts): Likewise.
3494 (tree_node_sizes): Likewise.
3495 (dump_tree_statistics): Use PRIu64 in printf format.
3497 2018-01-12 Martin Liska <mliska@suse.cz>
3499 * Makefile.in: As qsort_chk is implemented in vec.c, add
3500 vec.o to linkage of gencfn-macros.
3501 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3502 passing the info to record_node_allocation_statistics.
3503 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3505 * ggc-common.c (struct ggc_usage): Add operator== and use
3506 it in operator< and compare function.
3507 * mem-stats.h (struct mem_usage): Likewise.
3508 * vec.c (struct vec_usage): Remove operator< and compare
3509 function. Can be simply inherited.
3511 2018-01-12 Martin Jambor <mjambor@suse.cz>
3514 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3515 * tree-ssa-math-opts.c: Include domwalk.h.
3516 (convert_mult_to_fma_1): New function.
3517 (fma_transformation_info): New type.
3518 (fma_deferring_state): Likewise.
3519 (cancel_fma_deferring): New function.
3520 (result_of_phi): Likewise.
3521 (last_fma_candidate_feeds_initial_phi): Likewise.
3522 (convert_mult_to_fma): Added deferring logic, split actual
3523 transformation to convert_mult_to_fma_1.
3524 (math_opts_dom_walker): New type.
3525 (math_opts_dom_walker::after_dom_children): New method, body moved
3526 here from pass_optimize_widening_mul::execute, added deferring logic
3528 (pass_optimize_widening_mul::execute): Moved most of code to
3529 math_opts_dom_walker::after_dom_children.
3530 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3531 * config/i386/i386.c (ix86_option_override_internal): Added
3532 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3534 2018-01-12 Richard Biener <rguenther@suse.de>
3537 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3538 inline instance vars.
3540 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3543 * config/rx/rx.c (rx_is_restricted_memory_address):
3546 2018-01-12 Richard Biener <rguenther@suse.de>
3548 PR tree-optimization/80846
3549 * target.def (split_reduction): New target hook.
3550 * targhooks.c (default_split_reduction): New function.
3551 * targhooks.h (default_split_reduction): Declare.
3552 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3553 target requests first reduce vectors by combining low and high
3555 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3556 (get_vectype_for_scalar_type_and_size): Export.
3557 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3558 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3559 * doc/tm.texi: Regenerate.
3560 * config/i386/i386.c (ix86_split_reduction): Implement
3561 TARGET_VECTORIZE_SPLIT_REDUCTION.
3563 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3566 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3567 in PIC mode except for TARGET_VXWORKS_RTP.
3568 * config/sparc/sparc.c: Include cfgrtl.h.
3569 (TARGET_INIT_PIC_REG): Define.
3570 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3571 (sparc_pic_register_p): New predicate.
3572 (sparc_legitimate_address_p): Use it.
3573 (sparc_legitimize_pic_address): Likewise.
3574 (sparc_delegitimize_address): Likewise.
3575 (sparc_mode_dependent_address_p): Likewise.
3576 (gen_load_pcrel_sym): Remove 4th parameter.
3577 (load_got_register): Adjust call to above. Remove obsolete stuff.
3578 (sparc_expand_prologue): Do not call load_got_register here.
3579 (sparc_flat_expand_prologue): Likewise.
3580 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3581 (sparc_use_pseudo_pic_reg): New function.
3582 (sparc_init_pic_reg): Likewise.
3583 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3584 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3586 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3588 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3589 Add item for branch_cost.
3591 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3593 PR rtl-optimization/83565
3594 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3595 not extend the result to a larger mode for rotate operations.
3596 (num_sign_bit_copies1): Likewise.
3598 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3601 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3603 Use values-Xc.o for -pedantic.
3604 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3606 2018-01-12 Martin Liska <mliska@suse.cz>
3609 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3611 (possible_polymorphic_call_targets): Use it.
3612 (ipa_devirt): Likewise.
3614 2018-01-12 Martin Liska <mliska@suse.cz>
3616 * profile-count.h (enum profile_quality): Use 0 as invalid
3617 enum value of profile_quality.
3619 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3621 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3622 -mext-string options.
3624 2018-01-12 Richard Biener <rguenther@suse.de>
3626 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3627 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3628 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3630 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3632 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3634 * configure.ac (--with-long-double-format): Add support for the
3635 configuration option to change the default long double format on
3637 * config.gcc (powerpc*-linux*-*): Likewise.
3638 * configure: Regenerate.
3639 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3640 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3641 used without modification.
3643 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3645 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3646 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3647 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3648 MISC_BUILTIN_SPEC_BARRIER.
3649 (rs6000_init_builtins): Likewise.
3650 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3652 (speculation_barrier): New define_insn.
3653 * doc/extend.texi: Document __builtin_speculation_barrier.
3655 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3658 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3659 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3660 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3662 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3663 integral modes instead of "ss" and "sd".
3664 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3665 vectors with 32-bit and 64-bit elements.
3666 (vecdupssescalarmodesuffix): New mode attribute.
3667 (vec_dup<mode>): Use it.
3669 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3672 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3673 frame if argument is passed on stack.
3675 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3678 * ree.c (combine_reaching_defs): Optimize also
3679 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3680 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3682 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3685 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3687 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3690 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3691 after they are computed.
3693 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3695 PR tree-optimization/83695
3696 * gimple-loop-linterchange.cc
3697 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3698 reset cached scev information after interchange.
3699 (pass_linterchange::execute): Remove call to scev_reset_htab.
3701 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3703 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3704 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3705 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3706 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3707 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3708 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3709 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3710 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3711 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3712 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3713 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3714 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3715 (V_lane_reg): Likewise.
3716 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3718 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3719 (vfmal_lane_low<mode>_intrinsic,
3720 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3721 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3722 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3723 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3724 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3725 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3727 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3729 * config/arm/arm-cpus.in (fp16fml): New feature.
3730 (ALL_SIMD): Add fp16fml.
3731 (armv8.2-a): Add fp16fml as an option.
3732 (armv8.3-a): Likewise.
3733 (armv8.4-a): Add fp16fml as part of fp16.
3734 * config/arm/arm.h (TARGET_FP16FML): Define.
3735 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3737 * config/arm/arm-modes.def (V2HF): Define.
3738 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3739 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3740 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3741 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3742 vfmsl_low, vfmsl_high): New set of builtins.
3743 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3744 (vfml_op): New code attribute.
3745 (VFMLHALVES): New int iterator.
3746 (VFML, VFMLSEL): New mode attributes.
3747 (V_reg): Define mapping for V2HF.
3748 (V_hi, V_lo): New mode attributes.
3749 (VF_constraint): Likewise.
3750 (vfml_half, vfml_half_selector): New int attributes.
3751 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3753 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3754 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3756 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3757 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3758 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3759 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3761 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3762 Document new effective target and option set.
3764 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3766 * config/arm/arm-cpus.in (armv8_4): New feature.
3767 (ARMv8_4a): New fgroup.
3768 (armv8.4-a): New arch.
3769 * config/arm/arm-tables.opt: Regenerate.
3770 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3771 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3772 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3773 Add matching rules for -march=armv8.4-a and extensions.
3774 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3776 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3779 * config/rx/rx.md (BW): New mode attribute.
3780 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3782 2018-01-11 Richard Biener <rguenther@suse.de>
3784 PR tree-optimization/83435
3785 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3786 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3787 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3789 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3790 Alan Hayward <alan.hayward@arm.com>
3791 David Sherwood <david.sherwood@arm.com>
3793 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3795 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3796 (aarch64_print_address_internal): Use it to check for a zero offset.
3798 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3799 Alan Hayward <alan.hayward@arm.com>
3800 David Sherwood <david.sherwood@arm.com>
3802 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3803 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3804 Return a poly_int64 rather than a HOST_WIDE_INT.
3805 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3806 rather than a HOST_WIDE_INT.
3807 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3808 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3809 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3810 final_offset from HOST_WIDE_INT to poly_int64.
3811 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3812 to_constant when getting the number of units in an Advanced SIMD
3814 (aarch64_builtin_vectorized_function): Check for a constant number
3816 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3818 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3819 attribute instead of GET_MODE_NUNITS.
3820 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3821 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3822 GET_MODE_SIZE for fixed-size registers.
3823 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3824 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3825 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3826 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3827 (aarch64_print_operand, aarch64_print_address_internal)
3828 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3829 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3830 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3831 Handle polynomial GET_MODE_SIZE.
3832 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3833 wider than SImode without modification.
3834 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3835 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3836 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3837 passing and returning SVE modes.
3838 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3839 rather than GEN_INT.
3840 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3841 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3842 (aarch64_allocate_and_probe_stack_space): Likewise.
3843 (aarch64_layout_frame): Cope with polynomial offsets.
3844 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3845 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3847 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3848 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3849 poly_int64 rather than a HOST_WIDE_INT.
3850 (aarch64_get_separate_components, aarch64_process_components)
3851 (aarch64_expand_prologue, aarch64_expand_epilogue)
3852 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3853 (aarch64_anchor_offset): New function, split out from...
3854 (aarch64_legitimize_address): ...here.
3855 (aarch64_builtin_vectorization_cost): Handle polynomial
3856 TYPE_VECTOR_SUBPARTS.
3857 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3859 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3860 number of elements from the PARALLEL rather than the mode.
3861 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3862 rather than GET_MODE_BITSIZE.
3863 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3864 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3865 (aarch64_expand_vec_perm_const_1): Handle polynomial
3866 d->perm.length () and d->perm elements.
3867 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3868 Apply to_constant to d->perm elements.
3869 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3870 polynomial CONST_VECTOR_NUNITS.
3871 (aarch64_move_pointer): Take amount as a poly_int64 rather
3873 (aarch64_progress_pointer): Avoid temporary variable.
3874 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3875 the mode attribute instead of GET_MODE.
3877 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3878 Alan Hayward <alan.hayward@arm.com>
3879 David Sherwood <david.sherwood@arm.com>
3881 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3882 x exists before using it.
3883 (aarch64_add_constant_internal): Rename to...
3884 (aarch64_add_offset_1): ...this. Replace regnum with separate
3885 src and dest rtxes. Handle the case in which they're different,
3886 including when the offset is zero. Replace scratchreg with an rtx.
3887 Use 2 additions if there is no spare register into which we can
3888 move a 16-bit constant.
3889 (aarch64_add_constant): Delete.
3890 (aarch64_add_offset): Replace reg with separate src and dest
3891 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3892 Use aarch64_add_offset_1.
3893 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3894 an rtx rather than an int. Take the delta as a poly_int64
3895 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3896 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3897 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3898 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3899 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3901 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3902 aarch64_add_constant.
3904 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3906 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3907 Use scalar_float_mode.
3909 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3911 * config/aarch64/aarch64-simd.md
3912 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3913 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3914 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3915 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3916 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3917 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3918 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3919 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3920 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3921 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3923 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3926 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3927 targ_options->x_arm_arch_string is non NULL.
3929 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3931 * config/aarch64/aarch64.h
3932 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3934 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3937 * expmed.c (emit_store_flag_force): Swap if const op0
3938 and change VOIDmode to mode of op0.
3940 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3942 PR rtl-optimization/83761
3943 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3944 than bytes to mode_for_size.
3946 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3949 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3950 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3953 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3956 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3957 when in layout mode.
3958 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3959 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3962 2018-01-10 Michael Collison <michael.collison@arm.com>
3964 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3965 * config/aarch64/aarch64-option-extension.def: Add
3966 AARCH64_OPT_EXTENSION of 'fp16fml'.
3967 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3968 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3969 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3970 * config/aarch64/constraints.md (Ui7): New constraint.
3971 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3972 (VFMLA_SEL_W): Ditto.
3975 (VFMLA16_LOW): New int iterator.
3976 (VFMLA16_HIGH): Ditto.
3977 (UNSPEC_FMLAL): New unspec.
3978 (UNSPEC_FMLSL): Ditto.
3979 (UNSPEC_FMLAL2): Ditto.
3980 (UNSPEC_FMLSL2): Ditto.
3981 (f16mac): New code attribute.
3982 * config/aarch64/aarch64-simd-builtins.def
3983 (aarch64_fmlal_lowv2sf): Ditto.
3984 (aarch64_fmlsl_lowv2sf): Ditto.
3985 (aarch64_fmlalq_lowv4sf): Ditto.
3986 (aarch64_fmlslq_lowv4sf): Ditto.
3987 (aarch64_fmlal_highv2sf): Ditto.
3988 (aarch64_fmlsl_highv2sf): Ditto.
3989 (aarch64_fmlalq_highv4sf): Ditto.
3990 (aarch64_fmlslq_highv4sf): Ditto.
3991 (aarch64_fmlal_lane_lowv2sf): Ditto.
3992 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3993 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3994 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3995 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3996 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3997 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3998 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3999 (aarch64_fmlal_lane_highv2sf): Ditto.
4000 (aarch64_fmlsl_lane_highv2sf): Ditto.
4001 (aarch64_fmlal_laneq_highv2sf): Ditto.
4002 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4003 (aarch64_fmlalq_lane_highv4sf): Ditto.
4004 (aarch64_fmlsl_lane_highv4sf): Ditto.
4005 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4006 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4007 * config/aarch64/aarch64-simd.md:
4008 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4009 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4010 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4011 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4012 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4013 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4014 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4015 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4016 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4017 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4018 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4019 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4020 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4021 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4022 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4023 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4024 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4025 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4026 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4027 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4028 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4029 (vfmlsl_low_u32): Ditto.
4030 (vfmlalq_low_u32): Ditto.
4031 (vfmlslq_low_u32): Ditto.
4032 (vfmlal_high_u32): Ditto.
4033 (vfmlsl_high_u32): Ditto.
4034 (vfmlalq_high_u32): Ditto.
4035 (vfmlslq_high_u32): Ditto.
4036 (vfmlal_lane_low_u32): Ditto.
4037 (vfmlsl_lane_low_u32): Ditto.
4038 (vfmlal_laneq_low_u32): Ditto.
4039 (vfmlsl_laneq_low_u32): Ditto.
4040 (vfmlalq_lane_low_u32): Ditto.
4041 (vfmlslq_lane_low_u32): Ditto.
4042 (vfmlalq_laneq_low_u32): Ditto.
4043 (vfmlslq_laneq_low_u32): Ditto.
4044 (vfmlal_lane_high_u32): Ditto.
4045 (vfmlsl_lane_high_u32): Ditto.
4046 (vfmlal_laneq_high_u32): Ditto.
4047 (vfmlsl_laneq_high_u32): Ditto.
4048 (vfmlalq_lane_high_u32): Ditto.
4049 (vfmlslq_lane_high_u32): Ditto.
4050 (vfmlalq_laneq_high_u32): Ditto.
4051 (vfmlslq_laneq_high_u32): Ditto.
4052 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4053 (AARCH64_FL_FOR_ARCH8_4): New.
4054 (AARCH64_ISA_F16FML): New ISA flag.
4055 (TARGET_F16FML): New feature flag for fp16fml.
4056 (doc/invoke.texi): Document new fp16fml option.
4058 2018-01-10 Michael Collison <michael.collison@arm.com>
4060 * config/aarch64/aarch64-builtins.c:
4061 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4062 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4063 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4064 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4065 (AARCH64_ISA_SHA3): New ISA flag.
4066 (TARGET_SHA3): New feature flag for sha3.
4067 * config/aarch64/iterators.md (sha512_op): New int attribute.
4068 (CRYPTO_SHA512): New int iterator.
4069 (UNSPEC_SHA512H): New unspec.
4070 (UNSPEC_SHA512H2): Ditto.
4071 (UNSPEC_SHA512SU0): Ditto.
4072 (UNSPEC_SHA512SU1): Ditto.
4073 * config/aarch64/aarch64-simd-builtins.def
4074 (aarch64_crypto_sha512hqv2di): New builtin.
4075 (aarch64_crypto_sha512h2qv2di): Ditto.
4076 (aarch64_crypto_sha512su0qv2di): Ditto.
4077 (aarch64_crypto_sha512su1qv2di): Ditto.
4078 (aarch64_eor3qv8hi): Ditto.
4079 (aarch64_rax1qv2di): Ditto.
4080 (aarch64_xarqv2di): Ditto.
4081 (aarch64_bcaxqv8hi): Ditto.
4082 * config/aarch64/aarch64-simd.md:
4083 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4084 (aarch64_crypto_sha512su0qv2di): Ditto.
4085 (aarch64_crypto_sha512su1qv2di): Ditto.
4086 (aarch64_eor3qv8hi): Ditto.
4087 (aarch64_rax1qv2di): Ditto.
4088 (aarch64_xarqv2di): Ditto.
4089 (aarch64_bcaxqv8hi): Ditto.
4090 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4091 (vsha512h2q_u64): Ditto.
4092 (vsha512su0q_u64): Ditto.
4093 (vsha512su1q_u64): Ditto.
4094 (veor3q_u16): Ditto.
4095 (vrax1q_u64): Ditto.
4097 (vbcaxq_u16): Ditto.
4098 * config/arm/types.md (crypto_sha512): New type attribute.
4099 (crypto_sha3): Ditto.
4100 (doc/invoke.texi): Document new sha3 option.
4102 2018-01-10 Michael Collison <michael.collison@arm.com>
4104 * config/aarch64/aarch64-builtins.c:
4105 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4106 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4107 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4108 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4109 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4110 (AARCH64_ISA_SM4): New ISA flag.
4111 (TARGET_SM4): New feature flag for sm4.
4112 * config/aarch64/aarch64-simd-builtins.def
4113 (aarch64_sm3ss1qv4si): Ditto.
4114 (aarch64_sm3tt1aq4si): Ditto.
4115 (aarch64_sm3tt1bq4si): Ditto.
4116 (aarch64_sm3tt2aq4si): Ditto.
4117 (aarch64_sm3tt2bq4si): Ditto.
4118 (aarch64_sm3partw1qv4si): Ditto.
4119 (aarch64_sm3partw2qv4si): Ditto.
4120 (aarch64_sm4eqv4si): Ditto.
4121 (aarch64_sm4ekeyqv4si): Ditto.
4122 * config/aarch64/aarch64-simd.md:
4123 (aarch64_sm3ss1qv4si): Ditto.
4124 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4125 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4126 (aarch64_sm4eqv4si): Ditto.
4127 (aarch64_sm4ekeyqv4si): Ditto.
4128 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4129 (sm3part_op): Ditto.
4130 (CRYPTO_SM3TT): Ditto.
4131 (CRYPTO_SM3PART): Ditto.
4132 (UNSPEC_SM3SS1): New unspec.
4133 (UNSPEC_SM3TT1A): Ditto.
4134 (UNSPEC_SM3TT1B): Ditto.
4135 (UNSPEC_SM3TT2A): Ditto.
4136 (UNSPEC_SM3TT2B): Ditto.
4137 (UNSPEC_SM3PARTW1): Ditto.
4138 (UNSPEC_SM3PARTW2): Ditto.
4139 (UNSPEC_SM4E): Ditto.
4140 (UNSPEC_SM4EKEY): Ditto.
4141 * config/aarch64/constraints.md (Ui2): New constraint.
4142 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4143 * config/arm/types.md (crypto_sm3): New type attribute.
4144 (crypto_sm4): Ditto.
4145 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4146 (vsm3tt1aq_u32): Ditto.
4147 (vsm3tt1bq_u32): Ditto.
4148 (vsm3tt2aq_u32): Ditto.
4149 (vsm3tt2bq_u32): Ditto.
4150 (vsm3partw1q_u32): Ditto.
4151 (vsm3partw2q_u32): Ditto.
4152 (vsm4eq_u32): Ditto.
4153 (vsm4ekeyq_u32): Ditto.
4154 (doc/invoke.texi): Document new sm4 option.
4156 2018-01-10 Michael Collison <michael.collison@arm.com>
4158 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4159 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4160 (AARCH64_FL_FOR_ARCH8_4): New.
4161 (AARCH64_FL_V8_4): New flag.
4162 (doc/invoke.texi): Document new armv8.4-a option.
4164 2018-01-10 Michael Collison <michael.collison@arm.com>
4166 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4167 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4168 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4169 * config/aarch64/aarch64-option-extension.def: Add
4170 AARCH64_OPT_EXTENSION of 'sha2'.
4171 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4172 (crypto): Disable sha2 and aes if crypto disabled.
4173 (crypto): Enable aes and sha2 if enabled.
4174 (simd): Disable sha2 and aes if simd disabled.
4175 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4177 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4178 (TARGET_SHA2): New feature flag for sha2.
4179 (TARGET_AES): New feature flag for aes.
4180 * config/aarch64/aarch64-simd.md:
4181 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4182 conditional on TARGET_AES.
4183 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4184 (aarch64_crypto_sha1hsi): Make pattern conditional
4186 (aarch64_crypto_sha1hv4si): Ditto.
4187 (aarch64_be_crypto_sha1hv4si): Ditto.
4188 (aarch64_crypto_sha1su1v4si): Ditto.
4189 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4190 (aarch64_crypto_sha1su0v4si): Ditto.
4191 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4192 (aarch64_crypto_sha256su0v4si): Ditto.
4193 (aarch64_crypto_sha256su1v4si): Ditto.
4194 (doc/invoke.texi): Document new aes and sha2 options.
4196 2018-01-10 Martin Sebor <msebor@redhat.com>
4198 PR tree-optimization/83781
4199 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4202 2018-01-11 Martin Sebor <msebor@gmail.com>
4203 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4205 PR tree-optimization/83501
4206 PR tree-optimization/81703
4208 * tree-ssa-strlen.c (get_string_cst): Rename...
4209 (get_string_len): ...to this. Handle global constants.
4210 (handle_char_store): Adjust.
4212 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4213 Jim Wilson <jimw@sifive.com>
4215 * config/riscv/riscv-protos.h (riscv_output_return): New.
4216 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4217 (riscv_attribute_table, riscv_output_return),
4218 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4219 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4220 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4222 (riscv_expand_prologue): Add early return for naked function.
4223 (riscv_expand_epilogue): Likewise.
4224 (riscv_function_ok_for_sibcall): Return false for naked function.
4225 (riscv_set_current_function): New.
4226 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4227 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4228 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4229 * doc/extend.texi (RISC-V Function Attributes): New.
4231 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4233 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4234 check for 128-bit long double before checking TCmode.
4235 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4236 128-bit long doubles before checking TFmode or TCmode.
4237 (FLOAT128_IBM_P): Likewise.
4239 2018-01-10 Martin Sebor <msebor@redhat.com>
4241 PR tree-optimization/83671
4242 * builtins.c (c_strlen): Unconditionally return zero for the empty
4244 Use -Warray-bounds for warnings.
4245 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4246 for non-constant array indices with COMPONENT_REF, arrays of
4247 arrays, and pointers to arrays.
4248 (gimple_fold_builtin_strlen): Determine and set length range for
4249 non-constant character arrays.
4251 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4254 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4257 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4259 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4261 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4264 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4265 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4266 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4267 indexed_or_indirect_operand predicate.
4268 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4269 (*vsx_le_perm_load_v8hi): Likewise.
4270 (*vsx_le_perm_load_v16qi): Likewise.
4271 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4272 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4273 (*vsx_le_perm_store_v8hi): Likewise.
4274 (*vsx_le_perm_store_v16qi): Likewise.
4275 (eight unnamed splitters): Likewise.
4277 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4279 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4280 * config/rs6000/emmintrin.h: Likewise.
4281 * config/rs6000/mmintrin.h: Likewise.
4282 * config/rs6000/xmmintrin.h: Likewise.
4284 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4287 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4289 * tree.c (tree_nop_conversion): Return true for location wrapper
4291 (maybe_wrap_with_location): New function.
4292 (selftest::check_strip_nops): New function.
4293 (selftest::test_location_wrappers): New function.
4294 (selftest::tree_c_tests): Call it.
4295 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4296 (maybe_wrap_with_location): New decl.
4297 (EXPR_LOCATION_WRAPPER_P): New macro.
4298 (location_wrapper_p): New inline function.
4299 (tree_strip_any_location_wrapper): New inline function.
4301 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4304 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4305 stack_realign_offset for the largest alignment of stack slot
4307 (ix86_find_max_used_stack_alignment): New function.
4308 (ix86_finalize_stack_frame_flags): Use it. Set
4309 max_used_stack_alignment if we don't realign stack.
4310 * config/i386/i386.h (machine_function): Add
4311 max_used_stack_alignment.
4313 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4315 * config/arm/arm.opt (-mbranch-cost): New option.
4316 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4319 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4322 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4323 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4325 2018-01-10 Richard Biener <rguenther@suse.de>
4328 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4329 early out so it also covers the case where we have a non-NULL
4332 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4334 PR tree-optimization/83753
4335 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4336 for non-strided grouped accesses if the number of elements is 1.
4338 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4341 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4342 * i386.h (TARGET_USE_GATHER): Define.
4343 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4345 2018-01-10 Martin Liska <mliska@suse.cz>
4348 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4349 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4351 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4352 CLEANUP_NO_PARTITIONING is not set.
4354 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4356 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4357 for vectors, as a partial revert of r254296.
4358 * rtl.h (const_vec_p): Delete.
4359 (const_vec_duplicate_p): Don't test for vector CONSTs.
4360 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4361 * expmed.c (make_tree): Likewise.
4364 * common.md (E, F): Use CONSTANT_P instead of checking for
4366 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4367 checking for CONST_VECTOR.
4369 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4372 * predict.c (force_edge_cold): Handle in more sane way edges
4375 2018-01-09 Carl Love <cel@us.ibm.com>
4377 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4379 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4380 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4381 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4382 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4383 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4384 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4385 * config/rs6000/rs6000-protos.h: Add extern defition for
4386 rs6000_generate_float2_double_code.
4387 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4389 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4390 (float2_v2df): Add define_expand.
4392 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4395 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4396 op_mode in the force_to_mode call.
4398 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4400 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4401 instead of checking each element individually.
4402 (aarch64_evpc_uzp): Likewise.
4403 (aarch64_evpc_zip): Likewise.
4404 (aarch64_evpc_ext): Likewise.
4405 (aarch64_evpc_rev): Likewise.
4406 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4407 instead of checking each element individually. Return true without
4409 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4410 whether all selected elements come from the same input, instead of
4411 checking each element individually. Remove calls to gen_rtx_REG,
4412 start_sequence and end_sequence and instead assert that no rtl is
4415 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4417 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4418 order of HIGH and CONST checks.
4420 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4422 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4423 if the destination isn't an SSA_NAME.
4425 2018-01-09 Richard Biener <rguenther@suse.de>
4427 PR tree-optimization/83668
4428 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4430 (canonicalize_loop_form): ... here, renamed from ...
4431 (canonicalize_loop_closed_ssa_form): ... this and amended to
4432 swap successor edges for loop exit blocks to make us use
4433 the RPO order we need for initial schedule generation.
4435 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4437 PR tree-optimization/64811
4438 * match.pd: When optimizing comparisons with Inf, avoid
4439 introducing or losing exceptions from comparisons with NaN.
4441 2018-01-09 Martin Liska <mliska@suse.cz>
4444 * asan.c (shadow_mem_size): Add gcc_assert.
4446 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4448 Don't save registers in main().
4451 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4452 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4453 * config/avr/avr.c (avr_set_current_function): Don't error if
4454 naked, OS_task or OS_main are specified at the same time.
4455 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4457 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4459 * common/config/avr/avr-common.c (avr_option_optimization_table):
4460 Switch on -mmain-is-OS_task for optimizing compilations.
4462 2018-01-09 Richard Biener <rguenther@suse.de>
4464 PR tree-optimization/83572
4465 * graphite.c: Include cfganal.h.
4466 (graphite_transform_loops): Connect infinite loops to exit
4467 and remove fake edges at the end.
4469 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4471 * ipa-inline.c (edge_badness): Revert accidental checkin.
4473 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4476 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4477 symbols; not inline clones.
4479 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4482 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4483 hard registers. Formatting fixes.
4485 PR preprocessor/83722
4486 * gcc.c (try_generate_repro): Pass
4487 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4488 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4491 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4492 Kito Cheng <kito.cheng@gmail.com>
4494 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4495 (riscv_leaf_function_p): Delete.
4496 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4498 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4500 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4502 (do_ifelse): New function.
4503 (do_isel): New function.
4504 (do_sub3): New function.
4505 (do_add3): New function.
4506 (do_load_mask_compare): New function.
4507 (do_overlap_load_compare): New function.
4508 (expand_compare_loop): New function.
4509 (expand_block_compare): Call expand_compare_loop() when appropriate.
4510 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4512 (-mblock-compare-inline-loop-limit): New option.
4514 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4517 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4518 Reverse order of second and third operands in first alternative.
4519 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4520 of first and second elements in UNSPEC_VPERMR vector.
4521 (altivec_expand_vec_perm_le): Likewise.
4523 2018-01-08 Jeff Law <law@redhat.com>
4525 PR rtl-optimizatin/81308
4526 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4527 (process_switch): If group_case_labels makes a change, then set
4529 (pass_convert_switch::execute): If a switch is converted, then
4530 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4532 PR rtl-optimization/81308
4533 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4536 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4538 PR target/83663 - Revert r255946
4539 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4540 generation for cases where splatting a value is not useful.
4541 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4542 across a vec_duplicate and a paradoxical subreg forming a vector
4543 mode to a vec_concat.
4545 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4547 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4548 -march=armv8.3-a variants.
4549 * config/arm/t-multilib: Likewise.
4550 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4552 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4554 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4556 (cceq_ior_compare_complement): Give it a name so I can use it, and
4557 change boolean_or_operator predicate to boolean_operator so it can
4558 be used to generate a crand.
4559 (eqne): New code iterator.
4560 (bd/bd_neg): New code_attrs.
4561 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4562 a single define_insn.
4563 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4564 decrement (bdnzt/bdnzf/bdzt/bdzf).
4565 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4566 with the new names of the branch decrement patterns, and added the
4567 names of the branch decrement conditional patterns.
4569 2018-01-08 Richard Biener <rguenther@suse.de>
4571 PR tree-optimization/83563
4572 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4575 2018-01-08 Richard Biener <rguenther@suse.de>
4578 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4580 2018-01-08 Richard Biener <rguenther@suse.de>
4582 PR tree-optimization/83685
4583 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4584 references to abnormals.
4586 2018-01-08 Richard Biener <rguenther@suse.de>
4589 * dwarf2out.c (output_indirect_strings): Handle empty
4590 skeleton_debug_str_hash.
4591 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4593 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4595 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4596 (emit_store_direct): Likewise.
4597 (arc_trampoline_adjust_address): Likewise.
4598 (arc_asm_trampoline_template): New function.
4599 (arc_initialize_trampoline): Use asm_trampoline_template.
4600 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4601 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4602 * config/arc/arc.md (flush_icache): Delete pattern.
4604 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4606 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4607 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4610 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4613 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4614 by not USED_FOR_TARGET.
4615 (make_pass_resolve_sw_modes): Likewise.
4617 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4619 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4622 2018-01-08 Richard Biener <rguenther@suse.de>
4625 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4627 2018-01-08 Richard Biener <rguenther@suse.de>
4630 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4632 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4635 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4636 basic blocks with a small number of successors.
4637 (convert_control_dep_chain_into_preds): Improve handling of
4639 (dump_predicates): Split apart into...
4640 (dump_pred_chain): ...here...
4641 (dump_pred_info): ...and here.
4642 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4643 (can_chain_union_be_invalidated_p): Improve check for invalidation
4645 (uninit_uses_cannot_happen): Avoid unnecessary if
4646 convert_control_dep_chain_into_preds yielded nothing.
4648 2018-01-06 Martin Sebor <msebor@redhat.com>
4650 PR tree-optimization/83640
4651 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4652 subtracting negative offset from size.
4653 (builtin_access::overlap): Adjust offset bounds of the access to fall
4654 within the size of the object if possible.
4656 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4658 PR rtl-optimization/83699
4659 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4660 extract_bit_field_as_subreg to cases in which the extracted
4661 value is also a vector.
4663 * lra-constraints.c (process_alt_operands): Test for the equivalence
4664 substitutions when detecting a possible reload cycle.
4666 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4669 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4670 by default if flag_selective_schedling{,2}. Formatting fixes.
4672 PR rtl-optimization/83682
4673 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4674 if it has non-VECTOR_MODE element mode.
4675 (vec_duplicate_p): Likewise.
4678 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4679 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4681 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4684 * config/i386/i386-builtin.def
4685 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4686 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4687 Require also OPTION_MASK_ISA_AVX512F in addition to
4688 OPTION_MASK_ISA_GFNI.
4689 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4690 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4691 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4692 to OPTION_MASK_ISA_GFNI.
4693 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4694 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4695 OPTION_MASK_ISA_AVX512BW.
4696 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4697 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4698 addition to OPTION_MASK_ISA_GFNI.
4699 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4700 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4701 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4702 to OPTION_MASK_ISA_GFNI.
4703 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4704 a requirement for all ISAs rather than any of them with a few
4706 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4708 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4709 bitmasks to be enabled with 3 exceptions, instead of requiring any
4710 enabled ISA with lots of exceptions.
4711 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4712 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4713 Change avx512bw in isa attribute to avx512f.
4714 * config/i386/sgxintrin.h: Add license boilerplate.
4715 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4716 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4717 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4718 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4720 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4721 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4722 temporarily sse2 rather than sse if not enabled already.
4725 * config/i386/sse.md (VI248_VLBW): Rename to ...
4726 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4727 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4728 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4729 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4730 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4731 mode iterator instead of VI248_VLBW.
4733 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4735 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4736 (record_modified): Skip clobbers; add debug output.
4737 (param_change_prob): Use sreal frequencies.
4739 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4741 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4742 punt for user-aligned variables.
4744 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4746 * tree-chrec.c (chrec_contains_symbols): Return true for
4749 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4752 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4753 of (x|y) == x for BICS pattern.
4755 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4757 PR tree-optimization/83605
4758 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4759 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4762 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4764 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4765 * config/epiphany/rtems.h: New file.
4767 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4768 Uros Bizjak <ubizjak@gmail.com>
4771 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4772 QIreg_operand instead of register_operand predicate.
4773 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4774 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4775 comments instead of -fmitigate[-_]rop.
4777 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4780 * cgraphunit.c (symbol_table::compile): Switch to text_section
4781 before calling assembly_start debug hook.
4782 * run-rtl-passes.c (run_rtl_passes): Likewise.
4785 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4787 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4788 range_int_cst_p rather than !symbolic_range_p before calling
4789 extract_range_from_multiplicative_op_1.
4791 2018-01-04 Jeff Law <law@redhat.com>
4793 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4794 redundant test in assertion.
4796 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4798 * doc/rtl.texi: Document machine_mode wrapper classes.
4800 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4802 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4805 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4807 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4808 the VEC_PERM_EXPR fold to fail.
4810 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4813 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4814 to switched_sections.
4816 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4819 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4822 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4825 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4826 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4828 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4831 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4832 is BLKmode and bitpos not zero or mode change is needed.
4834 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4837 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4840 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4843 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4844 instead of MULT rtx. Update all corresponding splitters.
4846 (*ssub<modesuffix>): Ditto.
4848 (*cmp_sadd_di): Update split patterns.
4849 (*cmp_sadd_si): Ditto.
4850 (*cmp_sadd_sidi): Ditto.
4851 (*cmp_ssub_di): Ditto.
4852 (*cmp_ssub_si): Ditto.
4853 (*cmp_ssub_sidi): Ditto.
4854 * config/alpha/predicates.md (const23_operand): New predicate.
4855 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4856 Look for ASHIFT, not MULT inner operand.
4857 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4859 2018-01-04 Martin Liska <mliska@suse.cz>
4861 PR gcov-profile/83669
4862 * gcov.c (output_intermediate_file): Add version to intermediate
4864 * doc/gcov.texi: Document new field 'version' in intermediate
4865 file format. Fix location of '-k' option of gcov command.
4867 2018-01-04 Martin Liska <mliska@suse.cz>
4870 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4872 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4874 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4876 2018-01-03 Martin Sebor <msebor@redhat.com>
4878 PR tree-optimization/83655
4879 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4880 checking calls with invalid arguments.
4882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4884 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4885 (vectorizable_mask_load_store): Delete.
4886 (vectorizable_call): Return false for masked loads and stores.
4887 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4888 instead of gimple_assign_rhs1.
4889 (vectorizable_load): Handle IFN_MASK_LOAD.
4890 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4892 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4894 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4896 (vectorizable_mask_load_store): ...here.
4897 (vectorizable_load): ...and here.
4899 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4901 * tree-vect-stmts.c (vect_build_all_ones_mask)
4902 (vect_build_zero_merge_argument): New functions, split out from...
4903 (vectorizable_load): ...here.
4905 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4907 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4909 (vectorizable_mask_load_store): ...here.
4910 (vectorizable_store): ...and here.
4912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4914 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4916 (vectorizable_mask_load_store): ...here.
4918 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4920 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4921 (vect_model_store_cost): Take a vec_load_store_type instead of a
4923 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4924 (vect_model_store_cost): Take a vec_load_store_type instead of a
4926 (vectorizable_mask_load_store): Update accordingly.
4927 (vectorizable_store): Likewise.
4928 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4930 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4932 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4933 IFN_MASK_LOAD calls here rather than...
4934 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4936 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4937 Alan Hayward <alan.hayward@arm.com>
4938 David Sherwood <david.sherwood@arm.com>
4940 * expmed.c (extract_bit_field_1): For vector extracts,
4941 fall back to extract_bit_field_as_subreg if vec_extract
4944 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4945 Alan Hayward <alan.hayward@arm.com>
4946 David Sherwood <david.sherwood@arm.com>
4948 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4949 they are variable or constant sized.
4950 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4951 slots for constant-sized data.
4953 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4954 Alan Hayward <alan.hayward@arm.com>
4955 David Sherwood <david.sherwood@arm.com>
4957 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4958 handling COND_EXPRs with boolean comparisons, try to find a better
4959 basis for the mask type than the boolean itself.
4961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4963 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4964 is calculated and how it can be overridden.
4965 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4966 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4968 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4971 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4972 Alan Hayward <alan.hayward@arm.com>
4973 David Sherwood <david.sherwood@arm.com>
4975 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4976 Remove the mode argument.
4977 (aarch64_simd_valid_immediate): Remove the mode and inverse
4979 * config/aarch64/iterators.md (bitsize): New iterator.
4980 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4981 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4982 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4983 aarch64_simd_valid_immediate.
4984 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4985 (aarch64_reg_or_bic_imm): Likewise.
4986 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4987 with an insn_type enum and msl with a modifier_type enum.
4988 Replace element_width with a scalar_mode. Change the shift
4989 to unsigned int. Add constructors for scalar_float_mode and
4990 scalar_int_mode elements.
4991 (aarch64_vect_float_const_representable_p): Delete.
4992 (aarch64_can_const_movi_rtx_p)
4993 (aarch64_simd_scalar_immediate_valid_for_move)
4994 (aarch64_simd_make_constant): Update call to
4995 aarch64_simd_valid_immediate.
4996 (aarch64_advsimd_valid_immediate_hs): New function.
4997 (aarch64_advsimd_valid_immediate): Likewise.
4998 (aarch64_simd_valid_immediate): Remove mode and inverse
4999 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5000 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5001 and aarch64_float_const_representable_p on the result.
5002 (aarch64_output_simd_mov_immediate): Remove mode argument.
5003 Update call to aarch64_simd_valid_immediate and use of
5004 simd_immediate_info.
5005 (aarch64_output_scalar_simd_mov_immediate): Update call
5008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5009 Alan Hayward <alan.hayward@arm.com>
5010 David Sherwood <david.sherwood@arm.com>
5012 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5013 (mode_nunits): Likewise CONST_MODE_NUNITS.
5014 * machmode.def (ADJUST_NUNITS): Document.
5015 * genmodes.c (mode_data::need_nunits_adj): New field.
5016 (blank_mode): Update accordingly.
5017 (adj_nunits): New variable.
5018 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5020 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5021 listed in adj_nunits.
5022 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5023 listed in adj_nunits. Don't emit case statements for such modes.
5024 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5025 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5026 nothing if adj_nunits is nonnull.
5027 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5028 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5029 (emit_mode_fbit): Update use of print_maybe_const_decl.
5030 (emit_move_size): Likewise. Treat the array as non-const
5032 (emit_mode_adjustments): Handle adj_nunits.
5034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5036 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5037 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5038 (VECTOR_MODES): Use it.
5039 (make_vector_modes): Take the prefix as an argument.
5041 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5042 Alan Hayward <alan.hayward@arm.com>
5043 David Sherwood <david.sherwood@arm.com>
5045 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5046 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5047 for MODE_VECTOR_BOOL.
5048 * machmode.def (VECTOR_BOOL_MODE): Document.
5049 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5050 (make_vector_bool_mode): New function.
5051 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5053 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5054 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5056 * stor-layout.c (int_mode_for_mode): Likewise.
5057 * tree.c (build_vector_type_for_mode): Likewise.
5058 * varasm.c (output_constant_pool_2): Likewise.
5059 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5060 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5061 for MODE_VECTOR_BOOL.
5062 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5063 of mode class checks.
5064 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5065 instead of a list of mode class checks.
5066 (expand_vector_scalar_condition): Likewise.
5067 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5070 Alan Hayward <alan.hayward@arm.com>
5071 David Sherwood <david.sherwood@arm.com>
5073 * machmode.h (mode_size): Change from unsigned short to
5075 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5076 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5077 or if measurement_type is not polynomial.
5078 (fixed_size_mode::includes_p): Check for constant-sized modes.
5079 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5080 return a poly_uint16 rather than an unsigned short.
5081 (emit_mode_size): Change the type of mode_size from unsigned short
5082 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5083 (emit_mode_adjustments): Cope with polynomial vector sizes.
5084 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5086 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5088 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5089 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5090 * caller-save.c (setup_save_areas): Likewise.
5091 (replace_reg_with_saved_mem): Likewise.
5092 * calls.c (emit_library_call_value_1): Likewise.
5093 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5094 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5095 (gen_lowpart_for_combine): Likewise.
5096 * convert.c (convert_to_integer_1): Likewise.
5097 * cse.c (equiv_constant, cse_insn): Likewise.
5098 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5099 (cselib_subst_to_values): Likewise.
5100 * dce.c (word_dce_process_block): Likewise.
5101 * df-problems.c (df_word_lr_mark_ref): Likewise.
5102 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5103 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5104 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5105 (rtl_for_decl_location): Likewise.
5106 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5107 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5108 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5109 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5110 (expand_expr_real_1): Likewise.
5111 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5112 (pad_below): Likewise.
5113 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5114 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5115 * ira.c (get_subreg_tracking_sizes): Likewise.
5116 * ira-build.c (ira_create_allocno_objects): Likewise.
5117 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5118 (ira_sort_regnos_for_alter_reg): Likewise.
5119 * ira-costs.c (record_operand_costs): Likewise.
5120 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5121 (resolve_simple_move): Likewise.
5122 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5123 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5124 (lra_constraints): Likewise.
5125 (CONST_POOL_OK_P): Reject variable-sized modes.
5126 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5127 (add_pseudo_to_slot, lra_spill): Likewise.
5128 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5129 * optabs-query.c (get_best_extraction_insn): Likewise.
5130 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5131 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5132 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5133 * recog.c (offsettable_address_addr_space_p): Likewise.
5134 * regcprop.c (maybe_mode_change): Likewise.
5135 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5136 * regrename.c (build_def_use): Likewise.
5137 * regstat.c (dump_reg_info): Likewise.
5138 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5139 (find_reloads, find_reloads_subreg_address): Likewise.
5140 * reload1.c (eliminate_regs_1): Likewise.
5141 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5142 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5143 (simplify_binary_operation_1, simplify_subreg): Likewise.
5144 * targhooks.c (default_function_arg_padding): Likewise.
5145 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5146 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5147 (verify_gimple_assign_ternary): Likewise.
5148 * tree-inline.c (estimate_move_cost): Likewise.
5149 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5150 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5151 (get_address_cost_ainc): Likewise.
5152 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5153 (vect_supportable_dr_alignment): Likewise.
5154 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5155 (vectorizable_reduction): Likewise.
5156 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5157 (vectorizable_operation, vectorizable_load): Likewise.
5158 * tree.c (build_same_sized_truth_vector_type): Likewise.
5159 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5160 * var-tracking.c (emit_note_insn_var_location): Likewise.
5161 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5162 (ADDR_VEC_ALIGN): Likewise.
5164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5165 Alan Hayward <alan.hayward@arm.com>
5166 David Sherwood <david.sherwood@arm.com>
5168 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5170 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5171 or if measurement_type is polynomial.
5172 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5173 * combine.c (make_extraction): Likewise.
5174 * dse.c (find_shift_sequence): Likewise.
5175 * dwarf2out.c (mem_loc_descriptor): Likewise.
5176 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5177 (extract_bit_field, extract_low_bits): Likewise.
5178 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5179 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5180 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5181 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5182 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5183 * reload.c (find_reloads): Likewise.
5184 * reload1.c (alter_reg): Likewise.
5185 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5186 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5187 * tree-if-conv.c (predicate_mem_writes): Likewise.
5188 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5189 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5190 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5191 * valtrack.c (dead_debug_insert_temp): Likewise.
5192 * varasm.c (mergeable_constant_section): Likewise.
5193 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5195 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5196 Alan Hayward <alan.hayward@arm.com>
5197 David Sherwood <david.sherwood@arm.com>
5199 * expr.c (expand_assignment): Cope with polynomial mode sizes
5200 when assigning to a CONCAT.
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 * machmode.h (mode_precision): Change from unsigned short to
5208 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5210 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5211 or if measurement_type is not polynomial.
5212 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5213 in which the mode is already known to be a scalar_int_mode.
5214 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5215 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5217 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5218 for GET_MODE_PRECISION.
5219 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5220 for GET_MODE_PRECISION.
5221 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5223 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5224 (expand_field_assignment, make_extraction): Likewise.
5225 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5226 (get_last_value): Likewise.
5227 * convert.c (convert_to_integer_1): Likewise.
5228 * cse.c (cse_insn): Likewise.
5229 * expr.c (expand_expr_real_1): Likewise.
5230 * lra-constraints.c (simplify_operand_subreg): Likewise.
5231 * optabs-query.c (can_atomic_load_p): Likewise.
5232 * optabs.c (expand_atomic_load): Likewise.
5233 (expand_atomic_store): Likewise.
5234 * ree.c (combine_reaching_defs): Likewise.
5235 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5236 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5237 * tree.h (type_has_mode_precision_p): Likewise.
5238 * ubsan.c (instrument_si_overflow): Likewise.
5240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5241 Alan Hayward <alan.hayward@arm.com>
5242 David Sherwood <david.sherwood@arm.com>
5244 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5245 polynomial numbers of units.
5246 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5247 (valid_vector_subparts_p): New function.
5248 (build_vector_type): Remove temporary shim and take the number
5249 of units as a poly_uint64 rather than an int.
5250 (build_opaque_vector_type): Take the number of units as a
5251 poly_uint64 rather than an int.
5252 * tree.c (build_vector_from_ctor): Handle polynomial
5253 TYPE_VECTOR_SUBPARTS.
5254 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5255 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5256 (build_vector_from_val): If the number of units is variable,
5257 use build_vec_duplicate_cst for constant operands and
5258 VEC_DUPLICATE_EXPR otherwise.
5259 (make_vector_type): Remove temporary is_constant ().
5260 (build_vector_type, build_opaque_vector_type): Take the number of
5261 units as a poly_uint64 rather than an int.
5262 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5264 * cfgexpand.c (expand_debug_expr): Likewise.
5265 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5266 (store_constructor, expand_expr_real_1): Likewise.
5267 (const_scalar_mask_from_tree): Likewise.
5268 * fold-const-call.c (fold_const_reduction): Likewise.
5269 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5270 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5271 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5272 (fold_relational_const): Likewise.
5273 (native_interpret_vector): Likewise. Change the size from an
5274 int to an unsigned int.
5275 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5276 TYPE_VECTOR_SUBPARTS.
5277 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5278 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5279 duplicating a non-constant operand into a variable-length vector.
5280 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5281 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5282 * ipa-icf.c (sem_variable::equals): Likewise.
5283 * match.pd: Likewise.
5284 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5285 * print-tree.c (print_node): Likewise.
5286 * stor-layout.c (layout_type): Likewise.
5287 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5288 * tree-cfg.c (verify_gimple_comparison): Likewise.
5289 (verify_gimple_assign_binary): Likewise.
5290 (verify_gimple_assign_ternary): Likewise.
5291 (verify_gimple_assign_single): Likewise.
5292 * tree-pretty-print.c (dump_generic_node): Likewise.
5293 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5294 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5295 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5296 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5297 (vect_shift_permute_load_chain): Likewise.
5298 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5299 (expand_vector_condition, optimize_vector_constructor): Likewise.
5300 (lower_vec_perm, get_compute_type): Likewise.
5301 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5302 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5303 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5304 (vect_recog_mask_conversion_pattern): Likewise.
5305 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5306 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5307 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5308 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5309 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5310 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5311 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5312 (supportable_widening_operation): Likewise.
5313 (supportable_narrowing_operation): Likewise.
5314 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5316 * varasm.c (output_constant): Likewise.
5318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5319 Alan Hayward <alan.hayward@arm.com>
5320 David Sherwood <david.sherwood@arm.com>
5322 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5323 so that both the length == 3 and length != 3 cases set up their
5324 own permute vectors. Add comments explaining why we know the
5325 number of elements is constant.
5326 (vect_permute_load_chain): Likewise.
5328 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5329 Alan Hayward <alan.hayward@arm.com>
5330 David Sherwood <david.sherwood@arm.com>
5332 * machmode.h (mode_nunits): Change from unsigned char to
5334 (ONLY_FIXED_SIZE_MODES): New macro.
5335 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5336 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5337 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5339 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5340 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5341 or if measurement_type is not polynomial.
5342 * genmodes.c (ZERO_COEFFS): New macro.
5343 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5345 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5346 Use ZERO_COEFFS when emitting initializers.
5347 * data-streamer.h (bp_pack_poly_value): New function.
5348 (bp_unpack_poly_value): Likewise.
5349 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5350 for GET_MODE_NUNITS.
5351 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5352 for GET_MODE_NUNITS.
5353 * tree.c (make_vector_type): Remove temporary shim and make
5354 the real function take the number of units as a poly_uint64
5356 (build_vector_type_for_mode): Handle polynomial nunits.
5357 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5358 * emit-rtl.c (const_vec_series_p_1): Likewise.
5359 (gen_rtx_CONST_VECTOR): Likewise.
5360 * fold-const.c (test_vec_duplicate_folding): Likewise.
5361 * genrecog.c (validate_pattern): Likewise.
5362 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5363 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5364 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5365 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5366 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5367 * rtlanal.c (subreg_get_info): Likewise.
5368 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5369 (vect_grouped_load_supported): Likewise.
5370 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5371 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5372 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5373 (simplify_const_unary_operation, simplify_binary_operation_1)
5374 (simplify_const_binary_operation, simplify_ternary_operation)
5375 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5376 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5377 instead of CONST_VECTOR_NUNITS.
5378 * varasm.c (output_constant_pool_2): Likewise.
5379 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5380 explicit-encoded elements in the XVEC for variable-length vectors.
5382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5384 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5386 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5387 Alan Hayward <alan.hayward@arm.com>
5388 David Sherwood <david.sherwood@arm.com>
5390 * coretypes.h (fixed_size_mode): Declare.
5391 (fixed_size_mode_pod): New typedef.
5392 * builtins.h (target_builtins::x_apply_args_mode)
5393 (target_builtins::x_apply_result_mode): Change type to
5394 fixed_size_mode_pod.
5395 * builtins.c (apply_args_size, apply_result_size, result_vector)
5396 (expand_builtin_apply_args_1, expand_builtin_apply)
5397 (expand_builtin_return): Update accordingly.
5399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5401 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5402 * cselib.c (cselib_hash_rtx): Likewise.
5403 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5404 CONST_VECTOR encoding.
5406 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5407 Jeff Law <law@redhat.com>
5410 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5411 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5412 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5413 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5416 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5417 explicitly probe *sp in a noreturn function if there were any callee
5418 register saves or frame pointer is needed.
5420 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5423 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5424 BLKmode for ternary, binary or unary expressions.
5427 * var-tracking.c (delete_vta_debug_insn): New inline function.
5428 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5429 insns from get_insns () to NULL instead of each bb separately.
5430 Use delete_vta_debug_insn. No longer static.
5431 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5432 delete_vta_debug_insns callers.
5433 * rtl.h (delete_vta_debug_insns): Declare.
5434 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5435 instead of variable_tracking_main.
5437 2018-01-03 Martin Sebor <msebor@redhat.com>
5439 PR tree-optimization/83603
5440 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5441 arguments past the endof the argument list in functions declared
5442 without a prototype.
5443 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5444 Avoid checking when arguments are null.
5446 2018-01-03 Martin Sebor <msebor@redhat.com>
5449 * doc/extend.texi (attribute const): Fix a typo.
5450 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5451 issuing -Wsuggest-attribute for void functions.
5453 2018-01-03 Martin Sebor <msebor@redhat.com>
5455 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5456 offset_int::from instead of wide_int::to_shwi.
5457 (maybe_diag_overlap): Remove assertion.
5458 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5459 * gimple-ssa-sprintf.c (format_directive): Same.
5460 (parse_directive): Same.
5461 (sprintf_dom_walker::compute_format_length): Same.
5462 (try_substitute_return_value): Same.
5464 2018-01-03 Jeff Law <law@redhat.com>
5467 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5468 non-constant residual for zero at runtime and avoid probing in
5469 that case. Reorganize code for trailing problem to mirror handling
5472 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5474 PR tree-optimization/83501
5475 * tree-ssa-strlen.c (get_string_cst): New.
5476 (handle_char_store): Call get_string_cst.
5478 2018-01-03 Martin Liska <mliska@suse.cz>
5480 PR tree-optimization/83593
5481 * tree-ssa-strlen.c: Include tree-cfg.h.
5482 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5483 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5484 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5486 (strlen_dom_walker::before_dom_children): Call
5487 gimple_purge_dead_eh_edges. Dump tranformation with details
5489 (strlen_dom_walker::before_dom_children): Update call by adding
5490 new argument cleanup_eh.
5491 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5493 2018-01-03 Martin Liska <mliska@suse.cz>
5496 * cif-code.def (VARIADIC_THUNK): New enum value.
5497 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5500 2018-01-03 Jan Beulich <jbeulich@suse.com>
5502 * sse.md (mov<mode>_internal): Tighten condition for when to use
5503 vmovdqu<ssescalarsize> for TI and OI modes.
5505 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5507 Update copyright years.
5509 2018-01-03 Martin Liska <mliska@suse.cz>
5512 * ipa-visibility.c (function_and_variable_visibility): Skip
5513 functions with noipa attribure.
5515 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5517 * gcc.c (process_command): Update copyright notice dates.
5518 * gcov-dump.c (print_version): Ditto.
5519 * gcov.c (print_version): Ditto.
5520 * gcov-tool.c (print_version): Ditto.
5521 * gengtype.c (create_file): Ditto.
5522 * doc/cpp.texi: Bump @copying's copyright year.
5523 * doc/cppinternals.texi: Ditto.
5524 * doc/gcc.texi: Ditto.
5525 * doc/gccint.texi: Ditto.
5526 * doc/gcov.texi: Ditto.
5527 * doc/install.texi: Ditto.
5528 * doc/invoke.texi: Ditto.
5530 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5532 * vector-builder.h (vector_builder::m_full_nelts): Change from
5533 unsigned int to poly_uint64.
5534 (vector_builder::full_nelts): Update prototype accordingly.
5535 (vector_builder::new_vector): Likewise.
5536 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5537 (vector_builder::operator ==): Likewise.
5538 (vector_builder::finalize): Likewise.
5539 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5540 Take the number of elements as a poly_uint64 rather than an
5542 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5543 from unsigned int to poly_uint64.
5544 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5545 (vec_perm_indices::new_vector): Likewise.
5546 (vec_perm_indices::length): Likewise.
5547 (vec_perm_indices::nelts_per_input): Likewise.
5548 (vec_perm_indices::input_nelts): Likewise.
5549 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5550 number of elements per input as a poly_uint64 rather than an
5551 unsigned int. Use the original encoding for variable-length
5552 vectors, rather than clamping each individual element.
5553 For the second and subsequent elements in each pattern,
5554 clamp the step and base before clamping their sum.
5555 (vec_perm_indices::series_p): Handle polynomial element counts.
5556 (vec_perm_indices::all_in_range_p): Likewise.
5557 (vec_perm_indices_to_tree): Likewise.
5558 (vec_perm_indices_to_rtx): Likewise.
5559 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5560 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5561 (tree_vector_builder::new_binary_operation): Handle polynomial
5562 element counts. Return false if we need to know the number
5563 of elements at compile time.
5564 * fold-const.c (fold_vec_perm): Punt if the number of elements
5565 isn't known at compile time.
5567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5569 * vec-perm-indices.h (vec_perm_builder): Change element type
5570 from HOST_WIDE_INT to poly_int64.
5571 (vec_perm_indices::element_type): Update accordingly.
5572 (vec_perm_indices::clamp): Handle polynomial element_types.
5573 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5574 (vec_perm_indices::all_in_range_p): Likewise.
5575 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5577 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5578 polynomial vec_perm_indices element types.
5579 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5580 * fold-const.c (fold_vec_perm): Likewise.
5581 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5582 * tree-vect-generic.c (lower_vec_perm): Likewise.
5583 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5584 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5585 element type to HOST_WIDE_INT.
5587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5588 Alan Hayward <alan.hayward@arm.com>
5589 David Sherwood <david.sherwood@arm.com>
5591 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5592 rather than an int. Use plus_constant.
5593 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5594 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5596 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5597 Alan Hayward <alan.hayward@arm.com>
5598 David Sherwood <david.sherwood@arm.com>
5600 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5601 a HOST_WIDE_INT to a poly_int64.
5603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5604 Alan Hayward <alan.hayward@arm.com>
5605 David Sherwood <david.sherwood@arm.com>
5607 * calls.c (load_register_parameters): Cope with polynomial
5608 mode sizes. Require a constant size for BLKmode parameters
5609 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5610 forces a parameter to be padded at the lsb end in order to
5611 fill a complete number of words, require the parameter size
5612 to be ordered wrt UNITS_PER_WORD.
5614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5615 Alan Hayward <alan.hayward@arm.com>
5616 David Sherwood <david.sherwood@arm.com>
5618 * reload1.c (spill_stack_slot_width): Change element type
5619 from unsigned int to poly_uint64_pod.
5620 (alter_reg): Treat mode sizes as polynomial.
5622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5623 Alan Hayward <alan.hayward@arm.com>
5624 David Sherwood <david.sherwood@arm.com>
5626 * reload.c (complex_word_subreg_p): New function.
5627 (reload_inner_reg_of_subreg, push_reload): Use it.
5629 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5630 Alan Hayward <alan.hayward@arm.com>
5631 David Sherwood <david.sherwood@arm.com>
5633 * lra-constraints.c (process_alt_operands): Reject matched
5634 operands whose sizes aren't ordered.
5635 (match_reload): Refer to this check here.
5637 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5638 Alan Hayward <alan.hayward@arm.com>
5639 David Sherwood <david.sherwood@arm.com>
5641 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5642 that the mode size is in the set {1, 2, 4, 8, 16}.
5644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5645 Alan Hayward <alan.hayward@arm.com>
5646 David Sherwood <david.sherwood@arm.com>
5648 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5649 Use plus_constant instead of gen_rtx_PLUS.
5651 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5652 Alan Hayward <alan.hayward@arm.com>
5653 David Sherwood <david.sherwood@arm.com>
5655 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5656 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5657 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5658 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5659 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5660 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5661 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5662 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5663 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5664 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5666 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5667 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5668 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5669 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5670 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5671 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5672 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5673 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5674 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5675 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5677 * expr.c (emit_move_resolve_push): Treat the input and result
5678 of PUSH_ROUNDING as a poly_int64.
5679 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5680 (emit_push_insn): Likewise.
5681 * lra-eliminations.c (mark_not_eliminable): Likewise.
5682 * recog.c (push_operand): Likewise.
5683 * reload1.c (elimination_effects): Likewise.
5684 * rtlanal.c (nonzero_bits1): Likewise.
5685 * calls.c (store_one_arg): Likewise. Require the padding to be
5686 known at compile time.
5688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5689 Alan Hayward <alan.hayward@arm.com>
5690 David Sherwood <david.sherwood@arm.com>
5692 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5693 Use plus_constant instead of gen_rtx_PLUS.
5695 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5696 Alan Hayward <alan.hayward@arm.com>
5697 David Sherwood <david.sherwood@arm.com>
5699 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5703 Alan Hayward <alan.hayward@arm.com>
5704 David Sherwood <david.sherwood@arm.com>
5706 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5707 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5708 via stack temporaries. Treat the mode size as polynomial too.
5710 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5711 Alan Hayward <alan.hayward@arm.com>
5712 David Sherwood <david.sherwood@arm.com>
5714 * expr.c (expand_expr_real_2): When handling conversions involving
5715 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5716 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5717 as a poly_uint64 too.
5719 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5720 Alan Hayward <alan.hayward@arm.com>
5721 David Sherwood <david.sherwood@arm.com>
5723 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5726 Alan Hayward <alan.hayward@arm.com>
5727 David Sherwood <david.sherwood@arm.com>
5729 * combine.c (can_change_dest_mode): Handle polynomial
5730 REGMODE_NATURAL_SIZE.
5731 * expmed.c (store_bit_field_1): Likewise.
5732 * expr.c (store_constructor): Likewise.
5733 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5734 and polynomial REGMODE_NATURAL_SIZE.
5735 (gen_lowpart_common): Likewise.
5736 * reginfo.c (record_subregs_of_mode): Likewise.
5737 * rtlanal.c (read_modify_subreg_p): Likewise.
5739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5740 Alan Hayward <alan.hayward@arm.com>
5741 David Sherwood <david.sherwood@arm.com>
5743 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5744 numbers of elements.
5746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5747 Alan Hayward <alan.hayward@arm.com>
5748 David Sherwood <david.sherwood@arm.com>
5750 * match.pd: Cope with polynomial numbers of vector elements.
5752 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5753 Alan Hayward <alan.hayward@arm.com>
5754 David Sherwood <david.sherwood@arm.com>
5756 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5757 in a POINTER_PLUS_EXPR.
5759 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5760 Alan Hayward <alan.hayward@arm.com>
5761 David Sherwood <david.sherwood@arm.com>
5763 * omp-simd-clone.c (simd_clone_subparts): New function.
5764 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5765 (ipa_simd_modify_function_body): Likewise.
5767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5768 Alan Hayward <alan.hayward@arm.com>
5769 David Sherwood <david.sherwood@arm.com>
5771 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5772 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5773 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5774 (expand_vector_condition, vector_element): Likewise.
5775 (subparts_gt): New function.
5776 (get_compute_type): Use subparts_gt.
5777 (count_type_subparts): Delete.
5778 (expand_vector_operations_1): Use subparts_gt instead of
5779 count_type_subparts.
5781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5782 Alan Hayward <alan.hayward@arm.com>
5783 David Sherwood <david.sherwood@arm.com>
5785 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5786 (vect_compile_time_alias): ...this new function. Do the calculation
5787 on poly_ints rather than trees.
5788 (vect_prune_runtime_alias_test_list): Update call accordingly.
5790 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5791 Alan Hayward <alan.hayward@arm.com>
5792 David Sherwood <david.sherwood@arm.com>
5794 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5796 (vect_schedule_slp_instance): Likewise.
5798 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5799 Alan Hayward <alan.hayward@arm.com>
5800 David Sherwood <david.sherwood@arm.com>
5802 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5803 constant and extern definitions for variable-length vectors.
5804 (vect_get_constant_vectors): Note that the number of units
5805 is known to be constant.
5807 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5808 Alan Hayward <alan.hayward@arm.com>
5809 David Sherwood <david.sherwood@arm.com>
5811 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5812 of units as polynomial. Choose between WIDE and NARROW based
5815 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5816 Alan Hayward <alan.hayward@arm.com>
5817 David Sherwood <david.sherwood@arm.com>
5819 * tree-vect-stmts.c (simd_clone_subparts): New function.
5820 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5822 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5823 Alan Hayward <alan.hayward@arm.com>
5824 David Sherwood <david.sherwood@arm.com>
5826 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5827 vectors as polynomial. Use build_index_vector for
5830 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5831 Alan Hayward <alan.hayward@arm.com>
5832 David Sherwood <david.sherwood@arm.com>
5834 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5835 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5836 for variable-length vectors.
5837 (vectorizable_mask_load_store): Treat the number of units as
5838 polynomial, asserting that it is constant if the condition has
5839 already been enforced.
5840 (vectorizable_store, vectorizable_load): Likewise.
5842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5843 Alan Hayward <alan.hayward@arm.com>
5844 David Sherwood <david.sherwood@arm.com>
5846 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5847 of units as polynomial. Punt if we can't tell at compile time
5848 which vector contains the final result.
5850 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5851 Alan Hayward <alan.hayward@arm.com>
5852 David Sherwood <david.sherwood@arm.com>
5854 * tree-vect-loop.c (vectorizable_induction): Treat the number
5855 of units as polynomial. Punt on SLP inductions. Use an integer
5856 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5857 cast of such a series for variable-length floating-point
5860 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5861 Alan Hayward <alan.hayward@arm.com>
5862 David Sherwood <david.sherwood@arm.com>
5864 * tree.h (build_index_vector): Declare.
5865 * tree.c (build_index_vector): New function.
5866 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5867 of units as polynomial, forcibly converting it to a constant if
5868 vectorizable_reduction has already enforced the condition.
5869 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5870 to create a {1,2,3,...} vector.
5871 (vectorizable_reduction): Treat the number of units as polynomial.
5872 Choose vectype_in based on the largest scalar element size rather
5873 than the smallest number of units. Enforce the restrictions
5876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5877 Alan Hayward <alan.hayward@arm.com>
5878 David Sherwood <david.sherwood@arm.com>
5880 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5881 number of units as polynomial.
5883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5884 Alan Hayward <alan.hayward@arm.com>
5885 David Sherwood <david.sherwood@arm.com>
5887 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5888 * target.def (autovectorize_vector_sizes): Return the vector sizes
5889 by pointer, using vector_sizes rather than a bitmask.
5890 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5891 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5892 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5894 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5895 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5896 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5897 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5898 * omp-general.c (omp_max_vf): Likewise.
5899 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5900 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5901 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5902 * tree-vect-slp.c (vect_slp_bb): Likewise.
5903 * doc/tm.texi: Regenerate.
5904 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5906 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5907 the vector size as a poly_uint64 rather than an unsigned int.
5908 (current_vector_size): Change from an unsigned int to a poly_uint64.
5909 (get_vectype_for_scalar_type): Update accordingly.
5910 * tree.h (build_truth_vector_type): Take the size and number of
5911 units as a poly_uint64 rather than an unsigned int.
5912 (build_vector_type): Add a temporary overload that takes
5913 the number of units as a poly_uint64 rather than an unsigned int.
5914 * tree.c (make_vector_type): Likewise.
5915 (build_truth_vector_type): Take the number of units as a poly_uint64
5916 rather than an unsigned int.
5918 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5919 Alan Hayward <alan.hayward@arm.com>
5920 David Sherwood <david.sherwood@arm.com>
5922 * target.def (get_mask_mode): Take the number of units and length
5923 as poly_uint64s rather than unsigned ints.
5924 * targhooks.h (default_get_mask_mode): Update accordingly.
5925 * targhooks.c (default_get_mask_mode): Likewise.
5926 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5927 * doc/tm.texi: Regenerate.
5929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5930 Alan Hayward <alan.hayward@arm.com>
5931 David Sherwood <david.sherwood@arm.com>
5933 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5934 * omp-general.c (omp_max_vf): Likewise.
5935 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5936 (expand_omp_simd): Handle polynomial safelen.
5937 * omp-low.c (omplow_simd_context): Add a default constructor.
5938 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5939 (lower_rec_simd_input_clauses): Update accordingly.
5940 (lower_rec_input_clauses): Likewise.
5942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5943 Alan Hayward <alan.hayward@arm.com>
5944 David Sherwood <david.sherwood@arm.com>
5946 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5947 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5948 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5949 (vect_analyze_slp_cost): Likewise.
5950 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5951 (vect_model_load_cost): Likewise.
5953 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5954 Alan Hayward <alan.hayward@arm.com>
5955 David Sherwood <david.sherwood@arm.com>
5957 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5958 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5959 from an unsigned int * to a poly_uint64_pod *.
5960 (calculate_unrolling_factor): New function.
5961 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5964 Alan Hayward <alan.hayward@arm.com>
5965 David Sherwood <david.sherwood@arm.com>
5967 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5968 from an unsigned int to a poly_uint64.
5969 (_loop_vec_info::slp_unrolling_factor): Likewise.
5970 (_loop_vec_info::vectorization_factor): Change from an int
5972 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5973 (vect_get_num_vectors): New function.
5974 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5975 (vect_get_num_copies): Use vect_get_num_vectors.
5976 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5977 to an unsigned int *.
5978 (vect_analyze_data_refs): Change min_vf from an int * to a
5980 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5981 than an unsigned HOST_WIDE_INT.
5982 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5983 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5984 to an unsigned int *.
5985 (vect_analyze_data_ref_dependences): Likewise.
5986 (vect_compute_data_ref_alignment): Handle polynomial vf.
5987 (vect_enhance_data_refs_alignment): Likewise.
5988 (vect_prune_runtime_alias_test_list): Likewise.
5989 (vect_shift_permute_load_chain): Likewise.
5990 (vect_supportable_dr_alignment): Likewise.
5991 (dependence_distance_ge_vf): Take the vectorization factor as a
5992 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5993 (vect_analyze_data_refs): Change min_vf from an int * to a
5995 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5996 vfm1 as a poly_uint64 rather than an int. Make the same change
5997 for the returned bound_scalar.
5998 (vect_gen_vector_loop_niters): Handle polynomial vf.
5999 (vect_do_peeling): Likewise. Update call to
6000 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6001 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6003 * tree-vect-loop.c (vect_determine_vectorization_factor)
6004 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6005 (vect_get_known_peeling_cost): Likewise.
6006 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6007 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6008 (vect_transform_loop): Likewise. Use the lowest possible VF when
6009 updating the upper bounds of the loop.
6010 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6012 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6013 polynomial unroll factors.
6014 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6015 (vect_make_slp_decision): Likewise.
6016 (vect_supported_load_permutation_p): Likewise, and polynomial
6018 (vect_analyze_slp_cost): Handle polynomial vf.
6019 (vect_slp_analyze_node_operations): Likewise.
6020 (vect_slp_analyze_bb_1): Likewise.
6021 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6022 than an unsigned HOST_WIDE_INT.
6023 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6024 (vectorizable_load): Handle polynomial vf.
6025 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6027 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6030 Alan Hayward <alan.hayward@arm.com>
6031 David Sherwood <david.sherwood@arm.com>
6033 * match.pd: Handle bit operations involving three constants
6034 and try to fold one pair.
6036 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6038 * tree-vect-loop-manip.c: Include gimple-fold.h.
6039 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6040 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6041 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6042 Add a path that uses a step of VF instead of 1, but disable it
6044 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6045 and niters_no_overflow parameters. Update calls to
6046 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6047 Create a new SSA name if the latter choses to use a ste other
6048 than zero, and return it via niters_vector_mult_vf_var.
6049 * tree-vect-loop.c (vect_transform_loop): Update calls to
6050 vect_do_peeling, vect_gen_vector_loop_niters and
6051 slpeel_make_loop_iterate_ntimes.
6052 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6053 (vect_gen_vector_loop_niters): Update declarations after above changes.
6055 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6057 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6058 128-bit round to integer instructions.
6059 (ceil<mode>2): Likewise.
6060 (btrunc<mode>2): Likewise.
6061 (round<mode>2): Likewise.
6063 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6065 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6066 unaligned VSX load/store on P8/P9.
6067 (expand_block_clear): Allow the use of unaligned VSX
6068 load/store on P8/P9.
6070 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6072 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6074 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6075 swap associated with both a load and a store.
6077 2018-01-02 Andrew Waterman <andrew@sifive.com>
6079 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6080 * config/riscv/riscv.md (clear_cache): Use it.
6082 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6084 * web.c: Remove out-of-date comment.
6086 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6088 * expr.c (fixup_args_size_notes): Check that any existing
6089 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6090 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6091 (emit_single_push_insn): ...here.
6093 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6095 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6096 (const_vector_encoded_nelts): New function.
6097 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6098 (const_vector_int_elt, const_vector_elt): Declare.
6099 * emit-rtl.c (const_vector_int_elt_1): New function.
6100 (const_vector_elt): Likewise.
6101 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6102 of CONST_VECTOR_ELT.
6104 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6106 * expr.c: Include rtx-vector-builder.h.
6107 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6108 directly on the tree encoding.
6109 (const_vector_from_tree): Likewise.
6110 * optabs.c: Include rtx-vector-builder.h.
6111 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6112 sequence of "u" values.
6113 * vec-perm-indices.c: Include rtx-vector-builder.h.
6114 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6115 directly on the vec_perm_indices encoding.
6117 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6119 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6120 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6121 * rtx-vector-builder.h: New file.
6122 * rtx-vector-builder.c: Likewise.
6123 * rtl.h (rtx_def::u2): Add a const_vector field.
6124 (CONST_VECTOR_NPATTERNS): New macro.
6125 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6126 (CONST_VECTOR_DUPLICATE_P): Likewise.
6127 (CONST_VECTOR_STEPPED_P): Likewise.
6128 (CONST_VECTOR_ENCODED_ELT): Likewise.
6129 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6130 (unwrap_const_vec_duplicate): Likewise.
6131 (const_vec_series_p): Check for a non-duplicated vector encoding.
6132 Say that the function only returns true for integer vectors.
6133 * emit-rtl.c: Include rtx-vector-builder.h.
6134 (gen_const_vec_duplicate_1): Delete.
6135 (gen_const_vector): Call gen_const_vec_duplicate instead of
6136 gen_const_vec_duplicate_1.
6137 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6138 (gen_const_vec_duplicate): Use rtx_vector_builder.
6139 (gen_const_vec_series): Likewise.
6140 (gen_rtx_CONST_VECTOR): Likewise.
6141 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6142 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6143 Build a new vector rather than modifying a CONST_VECTOR in-place.
6144 (handle_special_swappables): Update call accordingly.
6145 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6146 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6147 Build a new vector rather than modifying a CONST_VECTOR in-place.
6148 (handle_special_swappables): Update call accordingly.
6150 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6152 * simplify-rtx.c (simplify_const_binary_operation): Use
6153 CONST_VECTOR_ELT instead of XVECEXP.
6155 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6157 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6158 the selector elements to be different from the data elements
6159 if the selector is a VECTOR_CST.
6160 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6161 ssizetype for the selector.
6163 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6165 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6166 before testing each element individually.
6167 * tree-vect-generic.c (lower_vec_perm): Likewise.
6169 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6171 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6172 * selftest-run-tests.c (selftest::run_tests): Call it.
6173 * vector-builder.h (vector_builder::operator ==): New function.
6174 (vector_builder::operator !=): Likewise.
6175 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6176 (vec_perm_indices::all_from_input_p): New function.
6177 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6178 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6179 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6180 instead of reading the VECTOR_CST directly. Detect whether both
6181 vector inputs are the same before constructing the vec_perm_indices,
6182 and update the number of inputs argument accordingly. Use the
6183 utility functions added above. Only construct sel2 if we need to.
6185 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6187 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6188 the broadcast of the low byte.
6189 (expand_mult_highpart): Use an explicit encoding for the permutes.
6190 * optabs-query.c (can_mult_highpart_p): Likewise.
6191 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6192 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6193 (vectorizable_bswap): Likewise.
6194 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6195 explicit encoding for the power-of-2 permutes.
6196 (vect_permute_store_chain): Likewise.
6197 (vect_grouped_load_supported): Likewise.
6198 (vect_permute_load_chain): Likewise.
6200 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6202 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6203 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6204 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6205 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6206 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6207 (vect_gen_perm_mask_any): Likewise.
6209 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6211 * int-vector-builder.h: New file.
6212 * vec-perm-indices.h: Include int-vector-builder.h.
6213 (vec_perm_indices): Redefine as an int_vector_builder.
6214 (auto_vec_perm_indices): Delete.
6215 (vec_perm_builder): Redefine as a stand-alone class.
6216 (vec_perm_indices::vec_perm_indices): New function.
6217 (vec_perm_indices::clamp): Likewise.
6218 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6219 (vec_perm_indices::new_vector): New function.
6220 (vec_perm_indices::new_expanded_vector): Update for new
6221 vec_perm_indices class.
6222 (vec_perm_indices::rotate_inputs): New function.
6223 (vec_perm_indices::all_in_range_p): Operate directly on the
6224 encoded form, without computing elided elements.
6225 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6226 encoding. Update for new vec_perm_indices class.
6227 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6228 the given vec_perm_builder.
6229 (expand_vec_perm_var): Update vec_perm_builder constructor.
6230 (expand_mult_highpart): Use vec_perm_builder instead of
6231 auto_vec_perm_indices.
6232 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6233 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6234 or double series encoding as appropriate.
6235 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6236 vec_perm_indices instead of auto_vec_perm_indices.
6237 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6238 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6239 (vect_permute_store_chain): Likewise.
6240 (vect_grouped_load_supported): Likewise.
6241 (vect_permute_load_chain): Likewise.
6242 (vect_shift_permute_load_chain): Likewise.
6243 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6244 (vect_transform_slp_perm_load): Likewise.
6245 (vect_schedule_slp_instance): Likewise.
6246 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6247 (vectorizable_mask_load_store): Likewise.
6248 (vectorizable_bswap): Likewise.
6249 (vectorizable_store): Likewise.
6250 (vectorizable_load): Likewise.
6251 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6252 vec_perm_indices instead of auto_vec_perm_indices. Use
6253 tree_to_vec_perm_builder to read the vector from a tree.
6254 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6255 vec_perm_builder instead of a vec_perm_indices.
6256 (have_whole_vector_shift): Use vec_perm_builder and
6257 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6258 truncation to calc_vec_perm_mask_for_shift.
6259 (vect_create_epilog_for_reduction): Likewise.
6260 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6261 from auto_vec_perm_indices to vec_perm_indices.
6262 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6263 instead of changing individual elements.
6264 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6265 the vector in d.perm.
6266 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6267 from auto_vec_perm_indices to vec_perm_indices.
6268 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6269 instead of changing individual elements.
6270 (arm_vectorize_vec_perm_const): Use new_vector to install
6271 the vector in d.perm.
6272 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6273 Update vec_perm_builder constructor.
6274 (rs6000_expand_interleave): Likewise.
6275 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6276 (rs6000_expand_interleave): Likewise.
6278 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6280 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6281 to qimode could truncate the indices.
6282 * optabs.c (expand_vec_perm_var): Likewise.
6284 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6286 * Makefile.in (OBJS): Add vec-perm-indices.o.
6287 * vec-perm-indices.h: New file.
6288 * vec-perm-indices.c: Likewise.
6289 * target.h (vec_perm_indices): Replace with a forward class
6291 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6292 * optabs.h: Include vec-perm-indices.h.
6293 (expand_vec_perm): Delete.
6294 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6295 (expand_vec_perm_const): Declare.
6296 * target.def (vec_perm_const_ok): Replace with...
6297 (vec_perm_const): ...this new hook.
6298 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6299 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6300 * doc/tm.texi: Regenerate.
6301 * optabs.def (vec_perm_const): Delete.
6302 * doc/md.texi (vec_perm_const): Likewise.
6303 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6304 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6305 expand_vec_perm for constant permutation vectors. Assert that
6306 the mode of variable permutation vectors is the integer equivalent
6307 of the mode that is being permuted.
6308 * optabs-query.h (selector_fits_mode_p): Declare.
6309 * optabs-query.c: Include vec-perm-indices.h.
6310 (selector_fits_mode_p): New function.
6311 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6312 is defined, instead of checking whether the vec_perm_const_optab
6313 exists. Use targetm.vectorize.vec_perm_const instead of
6314 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6315 fit in the vector mode before using a variable permute.
6316 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6317 vec_perm_indices instead of an rtx.
6318 (expand_vec_perm): Replace with...
6319 (expand_vec_perm_const): ...this new function. Take the selector
6320 as a vec_perm_indices rather than an rtx. Also take the mode of
6321 the selector. Update call to shift_amt_for_vec_perm_mask.
6322 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6323 Use vec_perm_indices::new_expanded_vector to expand the original
6324 selector into bytes. Check whether the indices fit in the vector
6325 mode before using a variable permute.
6326 (expand_vec_perm_var): Make global.
6327 (expand_mult_highpart): Use expand_vec_perm_const.
6328 * fold-const.c: Includes vec-perm-indices.h.
6329 * tree-ssa-forwprop.c: Likewise.
6330 * tree-vect-data-refs.c: Likewise.
6331 * tree-vect-generic.c: Likewise.
6332 * tree-vect-loop.c: Likewise.
6333 * tree-vect-slp.c: Likewise.
6334 * tree-vect-stmts.c: Likewise.
6335 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6337 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6338 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6339 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6340 (aarch64_vectorize_vec_perm_const): ...this new function.
6341 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6342 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6343 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6344 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6345 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6346 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6347 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6349 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6350 check for NEON modes.
6351 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6352 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6353 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6354 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6356 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6357 the old VEC_PERM_CONST conditions.
6358 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6359 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6360 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6361 (ia64_vectorize_vec_perm_const_ok): Merge into...
6362 (ia64_vectorize_vec_perm_const): ...this new function.
6363 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6364 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6365 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6366 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6367 * config/mips/mips.c (mips_expand_vec_perm_const)
6368 (mips_vectorize_vec_perm_const_ok): Merge into...
6369 (mips_vectorize_vec_perm_const): ...this new function.
6370 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6371 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6372 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6373 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6374 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6375 (rs6000_expand_vec_perm_const): Delete.
6376 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6378 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6379 (altivec_expand_vec_perm_const_le): Take each operand individually.
6380 Operate on constant selectors rather than rtxes.
6381 (altivec_expand_vec_perm_const): Likewise. Update call to
6382 altivec_expand_vec_perm_const_le.
6383 (rs6000_expand_vec_perm_const): Delete.
6384 (rs6000_vectorize_vec_perm_const_ok): Delete.
6385 (rs6000_vectorize_vec_perm_const): New function.
6386 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6387 an element count and rtx array.
6388 (rs6000_expand_extract_even): Update call accordingly.
6389 (rs6000_expand_interleave): Likewise.
6390 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6391 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6392 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6393 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6394 (rs6000_expand_vec_perm_const): Delete.
6395 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6396 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6397 (altivec_expand_vec_perm_const_le): Take each operand individually.
6398 Operate on constant selectors rather than rtxes.
6399 (altivec_expand_vec_perm_const): Likewise. Update call to
6400 altivec_expand_vec_perm_const_le.
6401 (rs6000_expand_vec_perm_const): Delete.
6402 (rs6000_vectorize_vec_perm_const_ok): Delete.
6403 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6404 reference to the SPE evmerge intructions.
6405 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6406 an element count and rtx array.
6407 (rs6000_expand_extract_even): Update call accordingly.
6408 (rs6000_expand_interleave): Likewise.
6409 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6410 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6412 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6414 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6416 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6417 vector mode and that that mode matches the mode of the data
6419 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6420 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6421 directly using expand_vec_perm_1 when forcing selectors into
6423 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6425 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6427 * optabs-query.h (can_vec_perm_p): Delete.
6428 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6429 * optabs-query.c (can_vec_perm_p): Split into...
6430 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6431 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6432 particular selector is valid.
6433 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6434 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6435 (vect_grouped_load_supported): Likewise.
6436 (vect_shift_permute_load_chain): Likewise.
6437 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6438 (vect_transform_slp_perm_load): Likewise.
6439 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6440 (vectorizable_bswap): Likewise.
6441 (vect_gen_perm_mask_checked): Likewise.
6442 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6443 implementations of variable permutation vectors into account
6444 when deciding which selector to use.
6445 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6446 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6447 with a false third argument.
6448 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6449 to test whether the constant selector is valid and can_vec_perm_var_p
6450 to test whether a variable selector is valid.
6452 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6454 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6455 * optabs-query.c (can_vec_perm_p): Likewise.
6456 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6457 instead of vec_perm_indices.
6458 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6459 (vect_gen_perm_mask_checked): Likewise,
6460 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6461 (vect_gen_perm_mask_checked): Likewise,
6463 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6465 * optabs-query.h (qimode_for_vec_perm): Declare.
6466 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6467 (qimode_for_vec_perm): ...this new function.
6468 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6470 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6472 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6473 does not have a conditional at the top.
6475 2018-01-02 Richard Biener <rguenther@suse.de>
6477 * ipa-inline.c (big_speedup_p): Fix expression.
6479 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6482 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6485 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6489 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6490 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6491 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6492 cond_taken_branch_cost 3->4.
6494 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6496 PR tree-optimization/83581
6497 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6498 TODO_cleanup_cfg if any changes have been made.
6501 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6502 convert_modes if target mode has the right side, but different mode
6506 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6507 last argument when extracting from CONCAT. If either from_real or
6508 from_imag is NULL, use expansion through memory. If result is not
6509 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6510 the parts directly to inner mode, if even that fails, use expansion
6514 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6515 check for bswap in mode rather than HImode and use that in expand_unop
6518 Copyright (C) 2018 Free Software Foundation, Inc.
6520 Copying and distribution of this file, with or without modification,
6521 are permitted in any medium without royalty provided the copyright
6522 notice and this notice are preserved.