1 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
5 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
8 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
11 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
14 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
16 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
17 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
19 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
21 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
22 different rtl trees depending on TARGET_64BIT.
23 (rs6000_gen_lvx): Likewise.
25 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
27 * config/visium/visium.md (nop): Tweak comment.
28 (hazard_nop): Likewise.
30 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
32 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
33 -mspeculate-indirect-jumps.
34 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
35 for -mno-speculate-indirect-jumps.
36 (*call_indirect_elfv2<mode>_nospec): New define_insn.
37 (*call_value_indirect_elfv2<mode>): Disable for
38 -mno-speculate-indirect-jumps.
39 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
40 (indirect_jump): Emit different RTL for
41 -mno-speculate-indirect-jumps.
42 (*indirect_jump<mode>): Disable for
43 -mno-speculate-indirect-jumps.
44 (*indirect_jump<mode>_nospec): New define_insn.
45 (tablejump): Emit different RTL for
46 -mno-speculate-indirect-jumps.
47 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
48 (tablejumpsi_nospec): New define_expand.
49 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
50 (tablejumpdi_nospec): New define_expand.
51 (*tablejump<mode>_internal1): Disable for
52 -mno-speculate-indirect-jumps.
53 (*tablejump<mode>_internal1_nospec): New define_insn.
54 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
57 2018-01-16 Artyom Skrobov tyomitch@gmail.com
59 * caller-save.c (insert_save): Drop unnecessary parameter. All
62 2018-01-16 Jakub Jelinek <jakub@redhat.com>
63 Richard Biener <rguenth@suse.de>
66 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
67 return early, inline manually is_gimple_sizepos. Make sure if we
68 call gimplify_expr we don't end up with a gimple constant.
69 * tree.c (variably_modified_type_p): Don't return true for
70 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
71 * gimplify.h (is_gimple_sizepos): Remove.
73 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
75 PR tree-optimization/83857
76 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
77 vectorizable_live_operation for pure SLP statements.
78 (vectorizable_live_operation): Handle PHIs.
80 2018-01-16 Richard Biener <rguenther@suse.de>
82 PR tree-optimization/83867
83 * tree-vect-stmts.c (vect_transform_stmt): Precompute
84 nested_in_vect_loop_p since the scalar stmt may get invalidated.
86 2018-01-16 Jakub Jelinek <jakub@redhat.com>
89 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
90 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
91 If off is not INTEGER_CST, issue a may not be aligned warning
92 rather than isn't aligned. Use isn%'t rather than isn't.
93 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
95 <case MULT_EXPR>: Improve the case when bottom and one of the
96 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
97 operand, in that case check if the other operand is multiple of
98 bottom divided by the INTEGER_CST operand.
100 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
103 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
104 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
105 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
106 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
107 * config/pa/pa.c (pa_function_arg_advance): Likewise.
108 (pa_function_arg, pa_arg_partial_bytes): Likewise.
109 (pa_function_arg_size): New function.
111 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
113 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
114 in a separate statement.
116 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
118 PR tree-optimization/83847
119 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
120 group gathers and scatters.
122 2018-01-16 Jakub Jelinek <jakub@redhat.com>
124 PR rtl-optimization/86620
125 * params.def (max-sched-ready-insns): Bump minimum value to 1.
127 PR rtl-optimization/83213
128 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
129 to last if both are JUMP_INSNs.
131 PR tree-optimization/83843
132 * gimple-ssa-store-merging.c
133 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
134 store_immediate_info for bswap/nop orig_stores.
136 2018-01-15 Andrew Waterman <andrew@sifive.com>
138 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
140 <UDIV>: Increase cost if !TARGET_DIV.
142 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
144 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
145 (define_attr "cr_logical_3op"): New.
146 (cceq_ior_compare): Adjust.
147 (cceq_ior_compare_complement): Adjust.
148 (*cceq_rev_compare): Adjust.
149 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
150 (is_cracked_insn): Adjust.
151 (insn_must_be_first_in_group): Adjust.
152 * config/rs6000/40x.md: Adjust.
153 * config/rs6000/440.md: Adjust.
154 * config/rs6000/476.md: Adjust.
155 * config/rs6000/601.md: Adjust.
156 * config/rs6000/603.md: Adjust.
157 * config/rs6000/6xx.md: Adjust.
158 * config/rs6000/7450.md: Adjust.
159 * config/rs6000/7xx.md: Adjust.
160 * config/rs6000/8540.md: Adjust.
161 * config/rs6000/cell.md: Adjust.
162 * config/rs6000/e300c2c3.md: Adjust.
163 * config/rs6000/e500mc.md: Adjust.
164 * config/rs6000/e500mc64.md: Adjust.
165 * config/rs6000/e5500.md: Adjust.
166 * config/rs6000/e6500.md: Adjust.
167 * config/rs6000/mpc.md: Adjust.
168 * config/rs6000/power4.md: Adjust.
169 * config/rs6000/power5.md: Adjust.
170 * config/rs6000/power6.md: Adjust.
171 * config/rs6000/power7.md: Adjust.
172 * config/rs6000/power8.md: Adjust.
173 * config/rs6000/power9.md: Adjust.
174 * config/rs6000/rs64.md: Adjust.
175 * config/rs6000/titan.md: Adjust.
177 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
179 * config/i386/predicates.md (indirect_branch_operand): Rewrite
180 ix86_indirect_branch_register logic.
182 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
184 * config/i386/constraints.md (Bs): Update
185 ix86_indirect_branch_register check. Don't check
186 ix86_indirect_branch_register with GOT_memory_operand.
188 * config/i386/predicates.md (GOT_memory_operand): Don't check
189 ix86_indirect_branch_register here.
190 (GOT32_symbol_operand): Likewise.
192 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
194 * config/i386/predicates.md (constant_call_address_operand):
195 Rewrite ix86_indirect_branch_register logic.
196 (sibcall_insn_operand): Likewise.
198 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
200 * config/i386/constraints.md (Bs): Replace
201 ix86_indirect_branch_thunk_register with
202 ix86_indirect_branch_register.
204 * config/i386/i386.md (indirect_jump): Likewise.
205 (tablejump): Likewise.
206 (*sibcall_memory): Likewise.
207 (*sibcall_value_memory): Likewise.
208 Peepholes of indirect call and jump via memory: Likewise.
209 * config/i386/i386.opt: Likewise.
210 * config/i386/predicates.md (indirect_branch_operand): Likewise.
211 (GOT_memory_operand): Likewise.
212 (call_insn_operand): Likewise.
213 (sibcall_insn_operand): Likewise.
214 (GOT32_symbol_operand): Likewise.
216 2018-01-15 Jakub Jelinek <jakub@redhat.com>
219 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
220 type rather than type addr's type points to.
221 (expand_omp_atomic_mutex): Likewise.
222 (expand_omp_atomic): Likewise.
224 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
227 * config/i386/i386.c (output_indirect_thunk_function): Use
228 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
229 for __x86_return_thunk.
231 2018-01-15 Richard Biener <rguenther@suse.de>
234 * expmed.c (extract_bit_field_1): Fix typo.
236 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
239 * config/arm/iterators.md (VF): New mode iterator.
240 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
241 Remove integer-related logic from pattern.
242 (neon_vabd<mode>_3): Likewise.
244 2018-01-15 Jakub Jelinek <jakub@redhat.com>
247 * common.opt (fstrict-overflow): No longer an alias.
248 (fwrapv-pointer): New option.
249 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
250 also for pointer types based on flag_wrapv_pointer.
251 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
252 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
253 opts->x_flag_wrapv got set.
254 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
255 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
256 POINTER_TYPE_OVERFLOW_UNDEFINED.
257 * match.pd: Likewise in address comparison pattern.
258 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
260 2018-01-15 Richard Biener <rguenther@suse.de>
263 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
264 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
265 Reset type names to their identifier if their TYPE_DECL doesn't
266 have linkage (and thus is used for ODR and devirt).
267 (save_debug_info_for_decl): Remove.
268 (save_debug_info_for_type): Likewise.
269 (add_tree_to_fld_list): Adjust.
270 * tree-pretty-print.c (dump_generic_node): Make dumping of
271 type names more robust.
273 2018-01-15 Richard Biener <rguenther@suse.de>
275 * BASE-VER: Bump to 8.0.1.
277 2018-01-14 Martin Sebor <msebor@redhat.com>
280 * builtins.c (check_access): Avoid warning when the no-warning bit
283 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
285 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
286 * ira-color (allocno_hard_regs_compare): Likewise.
288 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
291 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
292 Use .pushsection/.popsection.
294 2018-01-14 Martin Sebor <msebor@redhat.com>
297 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
299 2018-01-14 Jakub Jelinek <jakub@redhat.com>
301 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
302 entry from extra_headers.
303 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
304 extra_headers, make the list bitwise identical to the i?86-*-* one.
306 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
308 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
309 -mcmodel=large with -mindirect-branch=thunk,
310 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
311 -mfunction-return=thunk-extern.
312 * doc/invoke.texi: Document -mcmodel=large is incompatible with
313 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
314 -mfunction-return=thunk and -mfunction-return=thunk-extern.
316 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
318 * config/i386/i386.c (print_reg): Print the name of the full
319 integer register without '%'.
320 (ix86_print_operand): Handle 'V'.
321 * doc/extend.texi: Document 'V' modifier.
323 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
325 * config/i386/constraints.md (Bs): Disallow memory operand for
326 -mindirect-branch-register.
328 * config/i386/predicates.md (indirect_branch_operand): Likewise.
329 (GOT_memory_operand): Likewise.
330 (call_insn_operand): Likewise.
331 (sibcall_insn_operand): Likewise.
332 (GOT32_symbol_operand): Likewise.
333 * config/i386/i386.md (indirect_jump): Call convert_memory_address
334 for -mindirect-branch-register.
335 (tablejump): Likewise.
336 (*sibcall_memory): Likewise.
337 (*sibcall_value_memory): Likewise.
338 Disallow peepholes of indirect call and jump via memory for
339 -mindirect-branch-register.
340 (*call_pop): Replace m with Bw.
341 (*call_value_pop): Likewise.
342 (*sibcall_pop_memory): Replace m with Bs.
343 * config/i386/i386.opt (mindirect-branch-register): New option.
344 * doc/invoke.texi: Document -mindirect-branch-register option.
346 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
348 * config/i386/i386-protos.h (ix86_output_function_return): New.
349 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
350 set function_return_type.
351 (indirect_thunk_name): Add ret_p to indicate thunk for function
353 (output_indirect_thunk_function): Pass false to
355 (ix86_output_indirect_branch_via_reg): Likewise.
356 (ix86_output_indirect_branch_via_push): Likewise.
357 (output_indirect_thunk_function): Create alias for function
358 return thunk if regno < 0.
359 (ix86_output_function_return): New function.
360 (ix86_handle_fndecl_attribute): Handle function_return.
361 (ix86_attribute_table): Add function_return.
362 * config/i386/i386.h (machine_function): Add
363 function_return_type.
364 * config/i386/i386.md (simple_return_internal): Use
365 ix86_output_function_return.
366 (simple_return_internal_long): Likewise.
367 * config/i386/i386.opt (mfunction-return=): New option.
368 (indirect_branch): Mention -mfunction-return=.
369 * doc/extend.texi: Document function_return function attribute.
370 * doc/invoke.texi: Document -mfunction-return= option.
372 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
374 * config/i386/i386-opts.h (indirect_branch): New.
375 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
376 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
377 with local indirect jump when converting indirect call and jump.
378 (ix86_set_indirect_branch_type): New.
379 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
380 (indirectlabelno): New.
381 (indirect_thunk_needed): Likewise.
382 (indirect_thunk_bnd_needed): Likewise.
383 (indirect_thunks_used): Likewise.
384 (indirect_thunks_bnd_used): Likewise.
385 (INDIRECT_LABEL): Likewise.
386 (indirect_thunk_name): Likewise.
387 (output_indirect_thunk): Likewise.
388 (output_indirect_thunk_function): Likewise.
389 (ix86_output_indirect_branch_via_reg): Likewise.
390 (ix86_output_indirect_branch_via_push): Likewise.
391 (ix86_output_indirect_branch): Likewise.
392 (ix86_output_indirect_jmp): Likewise.
393 (ix86_code_end): Call output_indirect_thunk_function if needed.
394 (ix86_output_call_insn): Call ix86_output_indirect_branch if
396 (ix86_handle_fndecl_attribute): Handle indirect_branch.
397 (ix86_attribute_table): Add indirect_branch.
398 * config/i386/i386.h (machine_function): Add indirect_branch_type
399 and has_local_indirect_jump.
400 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
402 (tablejump): Likewise.
403 (*indirect_jump): Use ix86_output_indirect_jmp.
404 (*tablejump_1): Likewise.
405 (simple_return_indirect_internal): Likewise.
406 * config/i386/i386.opt (mindirect-branch=): New option.
407 (indirect_branch): New.
410 (thunk-inline): Likewise.
411 (thunk-extern): Likewise.
412 * doc/extend.texi: Document indirect_branch function attribute.
413 * doc/invoke.texi: Document -mindirect-branch= option.
415 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
418 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
420 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
422 * ipa-inline.c (want_inline_small_function_p): Return false if
423 inlining has already failed with CIF_FINAL_ERROR.
424 (update_caller_keys): Call want_inline_small_function_p before
426 (update_callee_keys): Likewise.
428 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
430 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
432 (rs6000_quadword_masked_address_p): Likewise.
433 (quad_aligned_load_p): Likewise.
434 (quad_aligned_store_p): Likewise.
435 (const_load_sequence_p): Add comment to describe the outer-most loop.
436 (mimic_memory_attributes_and_flags): New function.
437 (rs6000_gen_stvx): Likewise.
438 (replace_swapped_aligned_store): Likewise.
439 (rs6000_gen_lvx): Likewise.
440 (replace_swapped_aligned_load): Likewise.
441 (replace_swapped_load_constant): Capitalize argument name in
442 comment describing this function.
443 (rs6000_analyze_swaps): Add a third pass to search for vector loads
444 and stores that access quad-word aligned addresses and replace
445 with stvx or lvx instructions when appropriate.
446 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
447 New function prototype.
448 (rs6000_quadword_masked_address_p): Likewise.
449 (rs6000_gen_lvx): Likewise.
450 (rs6000_gen_stvx): Likewise.
451 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
452 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
453 when memory address is aligned.
454 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
455 this split to select lvx instruction when memory address is aligned.
456 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
457 instruction when memory address is aligned.
458 (*vsx_le_perm_load_v16qi): Likewise.
459 (four unnamed splitters): Modify to select the stvx instruction
460 when memory is aligned.
462 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
464 * predict.c (determine_unlikely_bbs): Handle correctly BBs
465 which appears in the queue multiple times.
467 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
468 Alan Hayward <alan.hayward@arm.com>
469 David Sherwood <david.sherwood@arm.com>
471 * tree-vectorizer.h (vec_lower_bound): New structure.
472 (_loop_vec_info): Add check_nonzero and lower_bounds.
473 (LOOP_VINFO_CHECK_NONZERO): New macro.
474 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
475 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
476 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
477 fields. Make seg_len the distance travelled, not including the
479 (dr_direction_indicator): Declare.
480 (dr_zero_step_indicator): Likewise.
481 (dr_known_forward_stride_p): Likewise.
482 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
484 (runtime_alias_check_p): Allow runtime alias checks with
486 (operator ==): Compare access_size and align.
487 (prune_runtime_alias_test_list): Rework for new distinction between
488 the access_size and seg_len.
489 (create_intersect_range_checks_index): Likewise. Cope with polynomial
491 (get_segment_min_max): New function.
492 (create_intersect_range_checks): Use it.
493 (dr_step_indicator): New function.
494 (dr_direction_indicator): Likewise.
495 (dr_zero_step_indicator): Likewise.
496 (dr_known_forward_stride_p): Likewise.
497 * tree-loop-distribution.c (data_ref_segment_size): Return
498 DR_STEP * (niters - 1).
499 (compute_alias_check_pairs): Update call to the dr_with_seg_len
501 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
502 (vect_preserves_scalar_order_p): New function, split out from...
503 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
504 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
505 (vect_vfa_access_size): New function.
506 (vect_vfa_align): Likewise.
507 (vect_compile_time_alias): Take access_size_a and access_b arguments.
508 (dump_lower_bound): New function.
509 (vect_check_lower_bound): Likewise.
510 (vect_small_gap_p): Likewise.
511 (vectorizable_with_step_bound_p): Likewise.
512 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
513 depencies if the vectorization factor is 1. Convert the checks
514 for nonzero steps into checks on the bounds of DR_STEP. Try using
515 a bunds check for variable steps if the minimum required step is
516 relatively small. Update calls to the dr_with_seg_len
517 constructor and to vect_compile_time_alias.
518 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
520 (vect_loop_versioning): Call it.
521 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
523 (vect_estimate_min_profitable_iters): Account for any bounds checks.
525 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
526 Alan Hayward <alan.hayward@arm.com>
527 David Sherwood <david.sherwood@arm.com>
529 * doc/sourcebuild.texi (vect_scatter_store): Document.
530 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
532 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
534 * genopinit.c (main): Add supports_vec_scatter_store and
535 supports_vec_scatter_store_cached to target_optabs.
536 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
537 IFN_MASK_SCATTER_STORE.
538 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
540 * internal-fn.h (internal_store_fn_p): Declare.
541 (internal_fn_stored_value_index): Likewise.
542 * internal-fn.c (scatter_store_direct): New macro.
543 (expand_scatter_store_optab_fn): New function.
544 (direct_scatter_store_optab_supported_p): New macro.
545 (internal_store_fn_p): New function.
546 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
547 IFN_MASK_SCATTER_STORE.
548 (internal_fn_mask_index): Likewise.
549 (internal_fn_stored_value_index): New function.
550 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
552 * optabs-query.h (supports_vec_scatter_store_p): Declare.
553 * optabs-query.c (supports_vec_scatter_store_p): New function.
554 * tree-vectorizer.h (vect_get_store_rhs): Declare.
555 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
556 true for scatter stores.
557 (vect_gather_scatter_fn_p): Handle scatter stores too.
558 (vect_check_gather_scatter): Consider using scatter stores if
559 supports_vec_scatter_store_p.
560 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
562 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
563 internal_fn_stored_value_index.
564 (check_load_store_masking): Handle scatter stores too.
565 (vect_get_store_rhs): Make public.
566 (vectorizable_call): Use internal_store_fn_p.
567 (vectorizable_store): Handle scatter store internal functions.
568 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
569 when deciding whether the end of the group has been reached.
570 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
571 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
572 (mask_scatter_store<mode>): New insns.
574 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
575 Alan Hayward <alan.hayward@arm.com>
576 David Sherwood <david.sherwood@arm.com>
578 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
579 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
580 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
582 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
583 Use vect_truncate_gather_scatter_offset if we can't treat the
584 operation as a normal gather load or scatter store.
585 (get_group_load_store_type): Take the gather_scatter_info
586 as argument. Try using a gather load or scatter store for
587 single-element groups.
588 (get_load_store_type): Update calls to get_group_load_store_type
589 and vect_use_strided_gather_scatters_p.
591 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
592 Alan Hayward <alan.hayward@arm.com>
593 David Sherwood <david.sherwood@arm.com>
595 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
596 optional tree argument.
597 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
599 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
600 but continue to use the current value as a fallback.
601 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
602 to compare the updates.
603 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
604 (get_load_store_type): Use it when handling a strided access.
605 (vect_get_strided_load_store_ops): New function.
606 (vect_get_data_ptr_increment): Likewise.
607 (vectorizable_load): Handle strided gather loads. Always pass
608 a step to vect_create_data_ref_ptr and bump_vector_ptr.
610 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
611 Alan Hayward <alan.hayward@arm.com>
612 David Sherwood <david.sherwood@arm.com>
614 * doc/md.texi (gather_load@var{m}): Document.
615 (mask_gather_load@var{m}): Likewise.
616 * genopinit.c (main): Add supports_vec_gather_load and
617 supports_vec_gather_load_cached to target_optabs.
618 * optabs-tree.c (init_tree_optimization_optabs): Use
619 ggc_cleared_alloc to allocate target_optabs.
620 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
621 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
623 * internal-fn.h (internal_load_fn_p): Declare.
624 (internal_gather_scatter_fn_p): Likewise.
625 (internal_fn_mask_index): Likewise.
626 (internal_gather_scatter_fn_supported_p): Likewise.
627 * internal-fn.c (gather_load_direct): New macro.
628 (expand_gather_load_optab_fn): New function.
629 (direct_gather_load_optab_supported_p): New macro.
630 (direct_internal_fn_optab): New function.
631 (internal_load_fn_p): Likewise.
632 (internal_gather_scatter_fn_p): Likewise.
633 (internal_fn_mask_index): Likewise.
634 (internal_gather_scatter_fn_supported_p): Likewise.
635 * optabs-query.c (supports_at_least_one_mode_p): New function.
636 (supports_vec_gather_load_p): Likewise.
637 * optabs-query.h (supports_vec_gather_load_p): Declare.
638 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
639 and memory_type field.
640 (NUM_PATTERNS): Bump to 15.
641 * tree-vect-data-refs.c: Include internal-fn.h.
642 (vect_gather_scatter_fn_p): New function.
643 (vect_describe_gather_scatter_call): Likewise.
644 (vect_check_gather_scatter): Try using internal functions for
645 gather loads. Recognize existing calls to a gather load function.
646 (vect_analyze_data_refs): Consider using gather loads if
647 supports_vec_gather_load_p.
648 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
649 (vect_get_gather_scatter_offset_type): Likewise.
650 (vect_convert_mask_for_vectype): Likewise.
651 (vect_add_conversion_to_patterm): Likewise.
652 (vect_try_gather_scatter_pattern): Likewise.
653 (vect_recog_gather_scatter_pattern): New pattern recognizer.
654 (vect_vect_recog_func_ptrs): Add it.
655 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
656 internal_fn_mask_index and internal_gather_scatter_fn_p.
657 (check_load_store_masking): Take the gather_scatter_info as an
658 argument and handle gather loads.
659 (vect_get_gather_scatter_ops): New function.
660 (vectorizable_call): Check internal_load_fn_p.
661 (vectorizable_load): Likewise. Handle gather load internal
663 (vectorizable_store): Update call to check_load_store_masking.
664 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
665 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
666 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
667 (aarch64_gather_scale_operand_d): New predicates.
668 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
669 (mask_gather_load<mode>): New insns.
671 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
672 Alan Hayward <alan.hayward@arm.com>
673 David Sherwood <david.sherwood@arm.com>
675 * optabs.def (fold_left_plus_optab): New optab.
676 * doc/md.texi (fold_left_plus_@var{m}): Document.
677 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
678 * internal-fn.c (fold_left_direct): Define.
679 (expand_fold_left_optab_fn): Likewise.
680 (direct_fold_left_optab_supported_p): Likewise.
681 * fold-const-call.c (fold_const_fold_left): New function.
682 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
683 * tree-parloops.c (valid_reduction_p): New function.
684 (gather_scalar_reductions): Use it.
685 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
686 (vect_finish_replace_stmt): Declare.
687 * tree-vect-loop.c (fold_left_reduction_fn): New function.
688 (needs_fold_left_reduction_p): New function, split out from...
689 (vect_is_simple_reduction): ...here. Accept reductions that
690 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
691 (vect_force_simple_reduction): Also store the reduction type in
692 the assignment's STMT_VINFO_REDUC_TYPE.
693 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
694 (merge_with_identity): New function.
695 (vect_expand_fold_left): Likewise.
696 (vectorize_fold_left_reduction): Likewise.
697 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
698 scalar phi in place for it. Check for target support and reject
699 cases that would reassociate the operation. Defer the transform
700 phase to vectorize_fold_left_reduction.
701 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
702 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
703 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
705 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
707 * tree-if-conv.c (predicate_mem_writes): Remove redundant
708 call to ifc_temp_var.
710 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
711 Alan Hayward <alan.hayward@arm.com>
712 David Sherwood <david.sherwood@arm.com>
714 * target.def (legitimize_address_displacement): Take the original
715 offset as a poly_int.
716 * targhooks.h (default_legitimize_address_displacement): Update
718 * targhooks.c (default_legitimize_address_displacement): Likewise.
719 * doc/tm.texi: Regenerate.
720 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
721 as an argument, moving assert of ad->disp == ad->disp_term to...
722 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
723 Try calling targetm.legitimize_address_displacement before expanding
724 the address rather than afterwards, and adjust for the new interface.
725 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
726 Match the new hook interface. Handle SVE addresses.
727 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
730 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
732 * Makefile.in (OBJS): Add early-remat.o.
733 * target.def (select_early_remat_modes): New hook.
734 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
735 * doc/tm.texi: Regenerate.
736 * targhooks.h (default_select_early_remat_modes): Declare.
737 * targhooks.c (default_select_early_remat_modes): New function.
738 * timevar.def (TV_EARLY_REMAT): New timevar.
739 * passes.def (pass_early_remat): New pass.
740 * tree-pass.h (make_pass_early_remat): Declare.
741 * early-remat.c: New file.
742 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
744 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
746 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
747 Alan Hayward <alan.hayward@arm.com>
748 David Sherwood <david.sherwood@arm.com>
750 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
751 vfm1 with a bound_epilog parameter.
752 (vect_do_peeling): Update calls accordingly, and move the prologue
753 call earlier in the function. Treat the base bound_epilog as 0 for
754 fully-masked loops and retain vf - 1 for other loops. Add 1 to
755 this base when peeling for gaps.
756 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
757 with fully-masked loops.
758 (vect_estimate_min_profitable_iters): Handle the single peeled
759 iteration in that case.
761 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
762 Alan Hayward <alan.hayward@arm.com>
763 David Sherwood <david.sherwood@arm.com>
765 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
766 single-element interleaving even if the size is not a power of 2.
767 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
768 accesses for single-element interleaving if the group size is
771 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
772 Alan Hayward <alan.hayward@arm.com>
773 David Sherwood <david.sherwood@arm.com>
775 * doc/md.texi (fold_extract_last_@var{m}): Document.
776 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
777 * optabs.def (fold_extract_last_optab): New optab.
778 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
779 * internal-fn.c (fold_extract_direct): New macro.
780 (expand_fold_extract_optab_fn): Likewise.
781 (direct_fold_extract_optab_supported_p): Likewise.
782 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
783 * tree-vect-loop.c (vect_model_reduction_cost): Handle
784 EXTRACT_LAST_REDUCTION.
785 (get_initial_def_for_reduction): Do not create an initial vector
786 for EXTRACT_LAST_REDUCTION reductions.
787 (vectorizable_reduction): Leave the scalar phi in place for
788 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
789 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
790 epilogue code for EXTRACT_LAST_REDUCTION and defer the
791 transform phase to vectorizable_condition.
792 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
794 (vect_finish_stmt_generation): ...here.
795 (vect_finish_replace_stmt): New function.
796 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
797 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
799 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
801 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
802 Alan Hayward <alan.hayward@arm.com>
803 David Sherwood <david.sherwood@arm.com>
805 * doc/md.texi (extract_last_@var{m}): Document.
806 * optabs.def (extract_last_optab): New optab.
807 * internal-fn.def (EXTRACT_LAST): New internal function.
808 * internal-fn.c (cond_unary_direct): New macro.
809 (expand_cond_unary_optab_fn): Likewise.
810 (direct_cond_unary_optab_supported_p): Likewise.
811 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
812 loops using EXTRACT_LAST.
813 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
814 (extract_last_<mode>): ...this optab.
815 (vec_extract<mode><Vel>): Update accordingly.
817 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
818 Alan Hayward <alan.hayward@arm.com>
819 David Sherwood <david.sherwood@arm.com>
821 * target.def (empty_mask_is_expensive): New hook.
822 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
823 * doc/tm.texi: Regenerate.
824 * targhooks.h (default_empty_mask_is_expensive): Declare.
825 * targhooks.c (default_empty_mask_is_expensive): New function.
826 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
827 if the target says that empty masks are expensive.
828 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
830 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
832 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
833 Alan Hayward <alan.hayward@arm.com>
834 David Sherwood <david.sherwood@arm.com>
836 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
837 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
838 (vect_use_loop_mask_for_alignment_p): New function.
839 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
840 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
841 niters_skip argument. Make sure that the first niters_skip elements
842 of the first iteration are inactive.
843 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
844 Update call to vect_set_loop_masks_directly.
845 (get_misalign_in_elems): New function, split out from...
846 (vect_gen_prolog_loop_niters): ...here.
847 (vect_update_init_of_dr): Take a code argument that specifies whether
848 the adjustment should be added or subtracted.
849 (vect_update_init_of_drs): Likewise.
850 (vect_prepare_for_masked_peels): New function.
851 (vect_do_peeling): Skip prologue peeling if we're using a mask
852 instead. Update call to vect_update_inits_of_drs.
853 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
855 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
856 alignment. Do not include the number of peeled iterations in
857 the minimum threshold in that case.
858 (vectorizable_induction): Adjust the start value down by
859 LOOP_VINFO_MASK_SKIP_NITERS iterations.
860 (vect_transform_loop): Call vect_prepare_for_masked_peels.
861 Take the number of skipped iterations into account when calculating
863 * tree-vect-stmts.c (vect_gen_while_not): New function.
865 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
866 Alan Hayward <alan.hayward@arm.com>
867 David Sherwood <david.sherwood@arm.com>
869 * doc/sourcebuild.texi (vect_fully_masked): Document.
870 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
872 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
874 (vect_analyze_loop_2): ...here. Don't check the vectorization
875 factor against the number of loop iterations if the loop is
878 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
879 Alan Hayward <alan.hayward@arm.com>
880 David Sherwood <david.sherwood@arm.com>
882 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
883 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
884 (dump_groups): Update accordingly.
885 (iv_use::mem_type): New member variable.
886 (address_p): New function.
887 (record_use): Add a mem_type argument and initialize the new
889 (record_group_use): Add a mem_type argument. Use address_p.
890 Remove obsolete null checks of base_object. Update call to record_use.
891 (find_interesting_uses_op): Update call to record_group_use.
892 (find_interesting_uses_cond): Likewise.
893 (find_interesting_uses_address): Likewise.
894 (get_mem_type_for_internal_fn): New function.
895 (find_address_like_use): Likewise.
896 (find_interesting_uses_stmt): Try find_address_like_use before
897 calling find_interesting_uses_op.
898 (addr_offset_valid_p): Use the iv mem_type field as the type
899 of the addressed memory.
900 (add_autoinc_candidates): Likewise.
901 (get_address_cost): Likewise.
902 (split_small_address_groups_p): Use address_p.
903 (split_address_groups): Likewise.
904 (add_iv_candidate_for_use): Likewise.
905 (autoinc_possible_for_pair): Likewise.
906 (rewrite_groups): Likewise.
907 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
908 (determine_group_iv_cost): Update after split of USE_ADDRESS.
909 (get_alias_ptr_type_for_ptr_address): New function.
910 (rewrite_use_address): Rewrite address uses in calls that were
911 identified by find_address_like_use.
913 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
914 Alan Hayward <alan.hayward@arm.com>
915 David Sherwood <david.sherwood@arm.com>
917 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
919 * gimple-expr.h (is_gimple_addressable: Likewise.
920 * gimple-expr.c (is_gimple_address): Likewise.
921 * internal-fn.c (expand_call_mem_ref): New function.
922 (expand_mask_load_optab_fn): Use it.
923 (expand_mask_store_optab_fn): Likewise.
925 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
926 Alan Hayward <alan.hayward@arm.com>
927 David Sherwood <david.sherwood@arm.com>
929 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
930 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
931 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
932 (cond_umax@var{mode}): Document.
933 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
934 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
935 (cond_umin_optab, cond_umax_optab): New optabs.
936 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
937 (COND_IOR, COND_XOR): New internal functions.
938 * internal-fn.h (get_conditional_internal_fn): Declare.
939 * internal-fn.c (cond_binary_direct): New macro.
940 (expand_cond_binary_optab_fn): Likewise.
941 (direct_cond_binary_optab_supported_p): Likewise.
942 (get_conditional_internal_fn): New function.
943 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
944 Cope with reduction statements that are vectorized as calls rather
946 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
947 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
948 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
949 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
950 (UNSPEC_COND_EOR): New unspecs.
951 (optab): Add mappings for them.
952 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
953 (sve_int_op, sve_fp_op): New int attributes.
955 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
956 Alan Hayward <alan.hayward@arm.com>
957 David Sherwood <david.sherwood@arm.com>
959 * optabs.def (while_ult_optab): New optab.
960 * doc/md.texi (while_ult@var{m}@var{n}): Document.
961 * internal-fn.def (WHILE_ULT): New internal function.
962 * internal-fn.h (direct_internal_fn_supported_p): New override
963 that takes two types as argument.
964 * internal-fn.c (while_direct): New macro.
965 (expand_while_optab_fn): New function.
966 (convert_optab_supported_p): Likewise.
967 (direct_while_optab_supported_p): New macro.
968 * wide-int.h (wi::udiv_ceil): New function.
969 * tree-vectorizer.h (rgroup_masks): New structure.
970 (vec_loop_masks): New typedef.
971 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
973 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
974 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
975 (vect_max_vf): New function.
976 (slpeel_make_loop_iterate_ntimes): Delete.
977 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
978 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
979 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
980 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
981 internal-fn.h, stor-layout.h and optabs-query.h.
982 (vect_set_loop_mask): New function.
983 (add_preheader_seq): Likewise.
984 (add_header_seq): Likewise.
985 (interleave_supported_p): Likewise.
986 (vect_maybe_permute_loop_masks): Likewise.
987 (vect_set_loop_masks_directly): Likewise.
988 (vect_set_loop_condition_masked): Likewise.
989 (vect_set_loop_condition_unmasked): New function, split out from
990 slpeel_make_loop_iterate_ntimes.
991 (slpeel_make_loop_iterate_ntimes): Rename to..
992 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
993 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
994 (vect_do_peeling): Update call accordingly.
995 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
997 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
998 mask_compare_type, can_fully_mask_p and fully_masked_p.
999 (release_vec_loop_masks): New function.
1000 (_loop_vec_info): Use it to free the loop masks.
1001 (can_produce_all_loop_masks_p): New function.
1002 (vect_get_max_nscalars_per_iter): Likewise.
1003 (vect_verify_full_masking): Likewise.
1004 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1005 retries, and free the mask rgroups before retrying. Check loop-wide
1006 reasons for disallowing fully-masked loops. Make the final decision
1007 about whether use a fully-masked loop or not.
1008 (vect_estimate_min_profitable_iters): Do not assume that peeling
1009 for the number of iterations will be needed for fully-masked loops.
1010 (vectorizable_reduction): Disable fully-masked loops.
1011 (vectorizable_live_operation): Likewise.
1012 (vect_halve_mask_nunits): New function.
1013 (vect_double_mask_nunits): Likewise.
1014 (vect_record_loop_mask): Likewise.
1015 (vect_get_loop_mask): Likewise.
1016 (vect_transform_loop): Handle the case in which the final loop
1017 iteration might handle a partial vector. Call vect_set_loop_condition
1018 instead of slpeel_make_loop_iterate_ntimes.
1019 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1020 (check_load_store_masking): New function.
1021 (prepare_load_store_mask): Likewise.
1022 (vectorizable_store): Handle fully-masked loops.
1023 (vectorizable_load): Likewise.
1024 (supportable_widening_operation): Use vect_halve_mask_nunits for
1026 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1027 (vect_gen_while): New function.
1028 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1029 (aarch64_uqdec<mode>): New insn.
1031 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1032 Alan Hayward <alan.hayward@arm.com>
1033 David Sherwood <david.sherwood@arm.com>
1035 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1036 (reduc_xor_scal_optab): New optabs.
1037 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1038 (reduc_xor_scal_@var{m}): Document.
1039 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1040 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1042 * fold-const-call.c (fold_const_call): Handle them.
1043 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
1044 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
1045 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
1046 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
1047 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
1048 (UNSPEC_XORV): New unspecs.
1049 (optab): Add entries for them.
1050 (BITWISEV): New int iterator.
1051 (bit_reduc_op): New int attributes.
1053 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1054 Alan Hayward <alan.hayward@arm.com>
1055 David Sherwood <david.sherwood@arm.com>
1057 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
1058 * internal-fn.def (VEC_SHL_INSERT): New internal function.
1059 * optabs.def (vec_shl_insert_optab): New optab.
1060 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
1061 (duplicate_and_interleave): Likewise.
1062 * tree-vect-loop.c: Include internal-fn.h.
1063 (neutral_op_for_slp_reduction): New function, split out from
1064 get_initial_defs_for_reduction.
1065 (get_initial_def_for_reduction): Handle option 2 for variable-length
1066 vectors by loading the neutral value into a vector and then shifting
1067 the initial value into element 0.
1068 (get_initial_defs_for_reduction): Replace the code argument with
1069 the neutral value calculated by neutral_op_for_slp_reduction.
1070 Use gimple_build_vector for constant-length vectors.
1071 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
1072 but the first group_size elements have a neutral value.
1073 Use duplicate_and_interleave otherwise.
1074 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
1075 Update call to get_initial_defs_for_reduction. Handle SLP
1076 reductions for variable-length vectors by creating one vector
1077 result for each scalar result, with the elements associated
1078 with other scalar results stubbed out with the neutral value.
1079 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
1080 Require IFN_VEC_SHL_INSERT for double reductions on
1081 variable-length vectors, or SLP reductions that have
1082 a neutral value. Require can_duplicate_and_interleave_p
1083 support for variable-length unchained SLP reductions if there
1084 is no neutral value, such as for MIN/MAX reductions. Also require
1085 the number of vector elements to be a multiple of the number of
1086 SLP statements when doing variable-length unchained SLP reductions.
1087 Update call to vect_create_epilog_for_reduction.
1088 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
1089 and remove initial values.
1090 (duplicate_and_interleave): Make public.
1091 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
1092 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
1094 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1095 Alan Hayward <alan.hayward@arm.com>
1096 David Sherwood <david.sherwood@arm.com>
1098 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
1099 (can_duplicate_and_interleave_p): New function.
1100 (vect_get_and_check_slp_defs): Take the vector of statements
1101 rather than just the current one. Remove excess parentheses.
1102 Restriction rejectinon of vect_constant_def and vect_external_def
1103 for variable-length vectors to boolean types, or types for which
1104 can_duplicate_and_interleave_p is false.
1105 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
1106 (duplicate_and_interleave): New function.
1107 (vect_get_constant_vectors): Use gimple_build_vector for
1108 constant-length vectors and suitable variable-length constant
1109 vectors. Use duplicate_and_interleave for other variable-length
1110 vectors. Don't defer the update when inserting new statements.
1112 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1113 Alan Hayward <alan.hayward@arm.com>
1114 David Sherwood <david.sherwood@arm.com>
1116 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
1117 min_profitable_iters doesn't go negative.
1119 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1120 Alan Hayward <alan.hayward@arm.com>
1121 David Sherwood <david.sherwood@arm.com>
1123 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
1124 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
1125 * optabs.def (vec_mask_load_lanes_optab): New optab.
1126 (vec_mask_store_lanes_optab): Likewise.
1127 * internal-fn.def (MASK_LOAD_LANES): New internal function.
1128 (MASK_STORE_LANES): Likewise.
1129 * internal-fn.c (mask_load_lanes_direct): New macro.
1130 (mask_store_lanes_direct): Likewise.
1131 (expand_mask_load_optab_fn): Handle masked operations.
1132 (expand_mask_load_lanes_optab_fn): New macro.
1133 (expand_mask_store_optab_fn): Handle masked operations.
1134 (expand_mask_store_lanes_optab_fn): New macro.
1135 (direct_mask_load_lanes_optab_supported_p): Likewise.
1136 (direct_mask_store_lanes_optab_supported_p): Likewise.
1137 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
1139 (vect_load_lanes_supported): Likewise.
1140 * tree-vect-data-refs.c (strip_conversion): New function.
1141 (can_group_stmts_p): Likewise.
1142 (vect_analyze_data_ref_accesses): Use it instead of checking
1143 for a pair of assignments.
1144 (vect_store_lanes_supported): Take a masked_p parameter.
1145 (vect_load_lanes_supported): Likewise.
1146 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
1147 vect_store_lanes_supported and vect_load_lanes_supported.
1148 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
1149 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
1150 parameter. Don't allow gaps for masked accesses.
1151 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
1152 and vect_load_lanes_supported.
1153 (get_load_store_type): Take a masked_p parameter and update
1154 call to get_group_load_store_type.
1155 (vectorizable_store): Update call to get_load_store_type.
1156 Handle IFN_MASK_STORE_LANES.
1157 (vectorizable_load): Update call to get_load_store_type.
1158 Handle IFN_MASK_LOAD_LANES.
1160 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1161 Alan Hayward <alan.hayward@arm.com>
1162 David Sherwood <david.sherwood@arm.com>
1164 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
1166 * config/aarch64/aarch64-protos.h
1167 (aarch64_sve_struct_memory_operand_p): Declare.
1168 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
1169 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
1170 (VPRED, vpred): Handle SVE structure modes.
1171 * config/aarch64/constraints.md (Utx): New constraint.
1172 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
1173 (aarch64_sve_struct_nonimmediate_operand): New predicates.
1174 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
1175 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
1176 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
1177 structure modes. Split into pieces after RA.
1178 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
1179 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
1181 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
1182 SVE structure modes.
1183 (aarch64_classify_address): Likewise.
1184 (sizetochar): Move earlier in file.
1185 (aarch64_print_operand): Handle SVE register lists.
1186 (aarch64_array_mode): New function.
1187 (aarch64_sve_struct_memory_operand_p): Likewise.
1188 (TARGET_ARRAY_MODE): Redefine.
1190 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1191 Alan Hayward <alan.hayward@arm.com>
1192 David Sherwood <david.sherwood@arm.com>
1194 * target.def (array_mode): New target hook.
1195 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
1196 * doc/tm.texi: Regenerate.
1197 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
1198 * hooks.c (hook_optmode_mode_uhwi_none): New function.
1199 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
1201 * stor-layout.c (mode_for_array): Likewise. Support polynomial
1204 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1205 Alan Hayward <alan.hayward@arm.com>
1206 David Sherwood <david.sherwood@arm.com>
1208 * fold-const.c (fold_binary_loc): Check the argument types
1209 rather than the result type when testing for a vector operation.
1211 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1213 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
1214 * doc/tm.texi: Regenerate.
1216 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1217 Alan Hayward <alan.hayward@arm.com>
1218 David Sherwood <david.sherwood@arm.com>
1220 * doc/invoke.texi (-msve-vector-bits=): Document new option.
1221 (sve): Document new AArch64 extension.
1222 * doc/md.texi (w): Extend the description of the AArch64
1223 constraint to include SVE vectors.
1224 (Upl, Upa): Document new AArch64 predicate constraints.
1225 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
1227 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
1228 (msve-vector-bits=): New option.
1229 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
1230 SVE when these are disabled.
1231 (sve): New extension.
1232 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
1233 modes. Adjust their number of units based on aarch64_sve_vg.
1234 (MAX_BITSIZE_MODE_ANY_MODE): Define.
1235 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
1236 aarch64_addr_query_type.
1237 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
1238 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
1239 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
1240 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
1241 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
1242 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
1243 (aarch64_simd_imm_zero_p): Delete.
1244 (aarch64_check_zero_based_sve_index_immediate): Declare.
1245 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
1246 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
1247 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
1248 (aarch64_sve_float_mul_immediate_p): Likewise.
1249 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
1251 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
1252 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
1253 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
1254 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
1255 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
1256 (aarch64_regmode_natural_size): Likewise.
1257 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
1258 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
1260 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
1261 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
1262 for VG and the SVE predicate registers.
1263 (V_ALIASES): Add a "z"-prefixed alias.
1264 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
1265 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
1266 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
1267 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
1268 (REG_CLASS_NAMES): Add entries for them.
1269 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
1270 and the predicate registers.
1271 (aarch64_sve_vg): Declare.
1272 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
1273 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
1274 (REGMODE_NATURAL_SIZE): Define.
1275 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
1277 * config/aarch64/aarch64.c: Include cfgrtl.h.
1278 (simd_immediate_info): Add a constructor for series vectors,
1279 and an associated step field.
1280 (aarch64_sve_vg): New variable.
1281 (aarch64_dbx_register_number): Handle VG and the predicate registers.
1282 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
1283 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
1284 (VEC_ANY_DATA, VEC_STRUCT): New constants.
1285 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
1286 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
1287 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
1288 (aarch64_get_mask_mode): New functions.
1289 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
1290 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
1291 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
1292 predicate modes and predicate registers. Explicitly restrict
1293 GPRs to modes of 16 bytes or smaller. Only allow FP registers
1294 to store a vector mode if it is recognized by
1295 aarch64_classify_vector_mode.
1296 (aarch64_regmode_natural_size): New function.
1297 (aarch64_hard_regno_caller_save_mode): Return the original mode
1299 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
1300 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
1301 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
1302 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
1304 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
1305 does not overlap dest if the function is frame-related. Handle
1307 (aarch64_split_add_offset): New function.
1308 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
1309 them aarch64_add_offset.
1310 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
1311 and update call to aarch64_sub_sp.
1312 (aarch64_add_cfa_expression): New function.
1313 (aarch64_expand_prologue): Pass extra temporary registers to the
1314 functions above. Handle the case in which we need to emit new
1315 DW_CFA_expressions for registers that were originally saved
1316 relative to the stack pointer, but now have to be expressed
1317 relative to the frame pointer.
1318 (aarch64_output_mi_thunk): Pass extra temporary registers to the
1320 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
1321 IP0 and IP1 values for SVE frames.
1322 (aarch64_expand_vec_series): New function.
1323 (aarch64_expand_sve_widened_duplicate): Likewise.
1324 (aarch64_expand_sve_const_vector): Likewise.
1325 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
1326 Handle SVE constants. Use emit_move_insn to move a force_const_mem
1327 into the register, rather than emitting a SET directly.
1328 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
1329 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
1330 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
1331 (offset_9bit_signed_scaled_p): New functions.
1332 (aarch64_replicate_bitmask_imm): New function.
1333 (aarch64_bitmask_imm): Use it.
1334 (aarch64_cannot_force_const_mem): Reject expressions involving
1335 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
1336 (aarch64_classify_index): Handle SVE indices, by requiring
1337 a plain register index with a scale that matches the element size.
1338 (aarch64_classify_address): Handle SVE addresses. Assert that
1339 the mode of the address is VOIDmode or an integer mode.
1340 Update call to aarch64_classify_symbol.
1341 (aarch64_classify_symbolic_expression): Update call to
1342 aarch64_classify_symbol.
1343 (aarch64_const_vec_all_in_range_p): New function.
1344 (aarch64_print_vector_float_operand): Likewise.
1345 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
1346 "vN" for FP registers with SVE modes. Handle (const ...) vectors
1347 and the FP immediates 1.0 and 0.5.
1348 (aarch64_print_address_internal): Handle SVE addresses.
1349 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
1350 (aarch64_regno_regclass): Handle predicate registers.
1351 (aarch64_secondary_reload): Handle big-endian reloads of SVE
1353 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
1354 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
1355 (aarch64_convert_sve_vector_bits): New function.
1356 (aarch64_override_options): Use it to handle -msve-vector-bits=.
1357 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
1359 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
1360 Handle SVE vector and predicate modes. Accept VL-based constants
1361 that need only one temporary register, and VL offsets that require
1362 no temporary registers.
1363 (aarch64_conditional_register_usage): Mark the predicate registers
1364 as fixed if SVE isn't available.
1365 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
1366 Return true for SVE vector and predicate modes.
1367 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
1368 rather than an unsigned int. Handle SVE modes.
1369 (aarch64_preferred_simd_mode): Update call accordingly. Handle
1371 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
1373 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
1374 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
1375 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
1376 (aarch64_sve_float_mul_immediate_p): New functions.
1377 (aarch64_sve_valid_immediate): New function.
1378 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
1379 Explicitly reject structure modes. Check for INDEX constants.
1380 Handle PTRUE and PFALSE constants.
1381 (aarch64_check_zero_based_sve_index_immediate): New function.
1382 (aarch64_simd_imm_zero_p): Delete.
1383 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
1384 vector modes. Accept constants in the range of CNT[BHWD].
1385 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
1386 ask for an Advanced SIMD mode.
1387 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
1388 (aarch64_simd_vector_alignment): Handle SVE predicates.
1389 (aarch64_vectorize_preferred_vector_alignment): New function.
1390 (aarch64_simd_vector_alignment_reachable): Use it instead of
1392 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
1393 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
1395 (MAX_VECT_LEN): Delete.
1396 (expand_vec_perm_d): Add a vec_flags field.
1397 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
1398 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
1399 (aarch64_evpc_ext): Don't apply a big-endian lane correction
1401 (aarch64_evpc_rev): Rename to...
1402 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
1403 (aarch64_evpc_rev_global): New function.
1404 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
1405 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
1407 (aarch64_evpc_sve_tbl): New function.
1408 (aarch64_expand_vec_perm_const_1): Update after rename of
1409 aarch64_evpc_rev. Handle SVE permutes too, trying
1410 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
1411 than aarch64_evpc_tbl.
1412 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
1413 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
1414 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
1415 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
1416 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
1417 (aarch64_expand_sve_vcond): New functions.
1418 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
1419 of aarch64_vector_mode_p.
1420 (aarch64_dwarf_poly_indeterminate_value): New function.
1421 (aarch64_compute_pressure_classes): Likewise.
1422 (aarch64_can_change_mode_class): Likewise.
1423 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
1424 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
1425 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
1426 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
1427 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
1428 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
1429 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
1430 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
1432 (Dn, Dl, Dr): Accept const as well as const_vector.
1433 (Dz): Likewise. Compare against CONST0_RTX.
1434 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
1435 of "vector" where appropriate.
1436 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
1437 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
1438 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
1439 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
1440 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
1441 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
1442 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
1443 (v_int_equiv): Extend to SVE modes.
1444 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
1446 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
1447 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
1448 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
1449 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
1450 (SVE_COND_FP_CMP): New int iterators.
1451 (perm_hilo): Handle the new unpack unspecs.
1452 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
1454 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
1455 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
1456 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
1457 (aarch64_equality_operator, aarch64_constant_vector_operand)
1458 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
1459 (aarch64_sve_nonimmediate_operand): Likewise.
1460 (aarch64_sve_general_operand): Likewise.
1461 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
1462 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
1463 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
1464 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
1465 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
1466 (aarch64_sve_float_arith_immediate): Likewise.
1467 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
1468 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
1469 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
1470 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
1471 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
1472 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
1473 (aarch64_sve_float_arith_operand): Likewise.
1474 (aarch64_sve_float_arith_with_sub_operand): Likewise.
1475 (aarch64_sve_float_mul_operand): Likewise.
1476 (aarch64_sve_vec_perm_operand): Likewise.
1477 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
1478 (aarch64_mov_operand): Accept const_poly_int and const_vector.
1479 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
1480 as well as const_vector.
1481 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
1482 in file. Use CONST0_RTX and CONSTM1_RTX.
1483 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
1484 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
1485 Use aarch64_simd_imm_zero.
1486 * config/aarch64/aarch64-sve.md: New file.
1487 * config/aarch64/aarch64.md: Include it.
1488 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
1489 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
1490 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
1491 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
1492 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
1493 (sve): New attribute.
1494 (enabled): Disable instructions with the sve attribute unless
1496 (movqi, movhi): Pass CONST_POLY_INT operaneds through
1497 aarch64_expand_mov_immediate.
1498 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
1499 CNT[BHSD] immediates.
1500 (movti): Split CONST_POLY_INT moves into two halves.
1501 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
1502 Split additions that need a temporary here if the destination
1503 is the stack pointer.
1504 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
1505 (*add<mode>3_poly_1): New instruction.
1506 (set_clobber_cc): New expander.
1508 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1510 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
1511 parameter and use it instead of GET_MODE_SIZE (innermode). Use
1512 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
1513 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
1514 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
1515 Change innermode from fixed_mode_size to machine_mode.
1516 (simplify_subreg): Update call accordingly. Handle a constant-sized
1517 subreg of a variable-length CONST_VECTOR.
1519 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1520 Alan Hayward <alan.hayward@arm.com>
1521 David Sherwood <david.sherwood@arm.com>
1523 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
1524 (add_offset_to_base): New function, split out from...
1525 (create_mem_ref): ...here. When handling a scale other than 1,
1526 check first whether the address is valid without the offset.
1527 Add it into the base if so, leaving the index and scale as-is.
1529 2018-01-12 Jakub Jelinek <jakub@redhat.com>
1532 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
1533 fold_for_warn before checking if arg2 is INTEGER_CST.
1535 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
1537 * config/rs6000/predicates.md (load_multiple_operation): Delete.
1538 (store_multiple_operation): Delete.
1539 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
1540 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
1541 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
1542 guarded by TARGET_STRING.
1543 (rs6000_output_load_multiple): Delete.
1544 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
1545 OPTION_MASK_STRING / TARGET_STRING handling.
1546 (print_operand) <'N', 'O'>: Add comment that these are unused now.
1547 (const rs6000_opt_masks) <"string">: Change mask to 0.
1548 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
1549 (MASK_STRING): Delete.
1550 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
1552 (load_multiple): Delete.
1559 (store_multiple): Delete.
1566 (movmemsi_8reg): Delete.
1567 (corresponding unnamed define_insn): Delete.
1568 (movmemsi_6reg): Delete.
1569 (corresponding unnamed define_insn): Delete.
1570 (movmemsi_4reg): Delete.
1571 (corresponding unnamed define_insn): Delete.
1572 (movmemsi_2reg): Delete.
1573 (corresponding unnamed define_insn): Delete.
1574 (movmemsi_1reg): Delete.
1575 (corresponding unnamed define_insn): Delete.
1576 * config/rs6000/rs6000.opt (mno-string): New.
1577 (mstring): Replace by deprecation warning stub.
1578 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
1580 2018-01-12 Jakub Jelinek <jakub@redhat.com>
1582 * regrename.c (regrename_do_replace): If replacing the same
1583 reg multiple times, try to reuse last created gen_raw_REG.
1586 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
1587 main to workaround a bug in GDB.
1589 2018-01-12 Tom de Vries <tom@codesourcery.com>
1592 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
1594 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
1596 PR rtl-optimization/80481
1597 * ira-color.c (get_cap_member): New function.
1598 (allocnos_conflict_by_live_ranges_p): Use it.
1599 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
1600 (setup_slot_coalesced_allocno_live_ranges): Ditto.
1602 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
1605 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
1606 (*saddl_se_1): Ditto.
1608 (*saddl_se_1): Ditto.
1610 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
1612 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
1613 rather than wi::to_widest for DR_INITs.
1614 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
1615 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
1616 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
1618 (vect_analyze_group_access_1): Note that here.
1620 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
1622 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
1623 polynomial type sizes.
1625 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
1627 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
1628 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
1629 (gimple_add_tmp_var): Likewise.
1631 2018-01-12 Martin Liska <mliska@suse.cz>
1633 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
1634 (gimple_alloc_sizes): Likewise.
1635 (dump_gimple_statistics): Use PRIu64 in printf format.
1636 * gimple.h: Change uint64_t to int.
1638 2018-01-12 Martin Liska <mliska@suse.cz>
1640 * tree-core.h: Use uint64_t instead of int.
1641 * tree.c (tree_node_counts): Likewise.
1642 (tree_node_sizes): Likewise.
1643 (dump_tree_statistics): Use PRIu64 in printf format.
1645 2018-01-12 Martin Liska <mliska@suse.cz>
1647 * Makefile.in: As qsort_chk is implemented in vec.c, add
1648 vec.o to linkage of gencfn-macros.
1649 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
1650 passing the info to record_node_allocation_statistics.
1651 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
1653 * ggc-common.c (struct ggc_usage): Add operator== and use
1654 it in operator< and compare function.
1655 * mem-stats.h (struct mem_usage): Likewise.
1656 * vec.c (struct vec_usage): Remove operator< and compare
1657 function. Can be simply inherited.
1659 2018-01-12 Martin Jambor <mjambor@suse.cz>
1662 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
1663 * tree-ssa-math-opts.c: Include domwalk.h.
1664 (convert_mult_to_fma_1): New function.
1665 (fma_transformation_info): New type.
1666 (fma_deferring_state): Likewise.
1667 (cancel_fma_deferring): New function.
1668 (result_of_phi): Likewise.
1669 (last_fma_candidate_feeds_initial_phi): Likewise.
1670 (convert_mult_to_fma): Added deferring logic, split actual
1671 transformation to convert_mult_to_fma_1.
1672 (math_opts_dom_walker): New type.
1673 (math_opts_dom_walker::after_dom_children): New method, body moved
1674 here from pass_optimize_widening_mul::execute, added deferring logic
1676 (pass_optimize_widening_mul::execute): Moved most of code to
1677 math_opts_dom_walker::after_dom_children.
1678 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
1679 * config/i386/i386.c (ix86_option_override_internal): Added
1680 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
1682 2018-01-12 Richard Biener <rguenther@suse.de>
1685 * dwarf2out.c (gen_variable_die): Do not reset old_die for
1686 inline instance vars.
1688 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
1691 * config/rx/rx.c (rx_is_restricted_memory_address):
1694 2018-01-12 Richard Biener <rguenther@suse.de>
1696 PR tree-optimization/80846
1697 * target.def (split_reduction): New target hook.
1698 * targhooks.c (default_split_reduction): New function.
1699 * targhooks.h (default_split_reduction): Declare.
1700 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
1701 target requests first reduce vectors by combining low and high
1703 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
1704 (get_vectype_for_scalar_type_and_size): Export.
1705 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
1706 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
1707 * doc/tm.texi: Regenerate.
1708 * config/i386/i386.c (ix86_split_reduction): Implement
1709 TARGET_VECTORIZE_SPLIT_REDUCTION.
1711 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1714 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
1715 in PIC mode except for TARGET_VXWORKS_RTP.
1716 * config/sparc/sparc.c: Include cfgrtl.h.
1717 (TARGET_INIT_PIC_REG): Define.
1718 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
1719 (sparc_pic_register_p): New predicate.
1720 (sparc_legitimate_address_p): Use it.
1721 (sparc_legitimize_pic_address): Likewise.
1722 (sparc_delegitimize_address): Likewise.
1723 (sparc_mode_dependent_address_p): Likewise.
1724 (gen_load_pcrel_sym): Remove 4th parameter.
1725 (load_got_register): Adjust call to above. Remove obsolete stuff.
1726 (sparc_expand_prologue): Do not call load_got_register here.
1727 (sparc_flat_expand_prologue): Likewise.
1728 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
1729 (sparc_use_pseudo_pic_reg): New function.
1730 (sparc_init_pic_reg): Likewise.
1731 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1732 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
1734 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
1736 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
1737 Add item for branch_cost.
1739 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1741 PR rtl-optimization/83565
1742 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
1743 not extend the result to a larger mode for rotate operations.
1744 (num_sign_bit_copies1): Likewise.
1746 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1749 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
1751 Use values-Xc.o for -pedantic.
1752 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
1754 2018-01-12 Martin Liska <mliska@suse.cz>
1757 * ipa-devirt.c (final_warning_record::grow_type_warnings):
1759 (possible_polymorphic_call_targets): Use it.
1760 (ipa_devirt): Likewise.
1762 2018-01-12 Martin Liska <mliska@suse.cz>
1764 * profile-count.h (enum profile_quality): Use 0 as invalid
1765 enum value of profile_quality.
1767 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
1769 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
1770 -mext-string options.
1772 2018-01-12 Richard Biener <rguenther@suse.de>
1774 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
1775 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
1776 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
1778 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
1780 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
1782 * configure.ac (--with-long-double-format): Add support for the
1783 configuration option to change the default long double format on
1785 * config.gcc (powerpc*-linux*-*): Likewise.
1786 * configure: Regenerate.
1787 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
1788 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
1789 used without modification.
1791 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1793 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
1794 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
1795 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
1796 MISC_BUILTIN_SPEC_BARRIER.
1797 (rs6000_init_builtins): Likewise.
1798 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
1800 (speculation_barrier): New define_insn.
1801 * doc/extend.texi: Document __builtin_speculation_barrier.
1803 2018-01-11 Jakub Jelinek <jakub@redhat.com>
1806 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
1807 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
1808 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
1810 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
1811 integral modes instead of "ss" and "sd".
1812 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
1813 vectors with 32-bit and 64-bit elements.
1814 (vecdupssescalarmodesuffix): New mode attribute.
1815 (vec_dup<mode>): Use it.
1817 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
1820 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
1821 frame if argument is passed on stack.
1823 2018-01-11 Jakub Jelinek <jakub@redhat.com>
1826 * ree.c (combine_reaching_defs): Optimize also
1827 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
1828 reg2=any_extend(exp); reg1=reg2;, formatting fix.
1830 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
1833 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
1835 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
1838 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
1839 after they are computed.
1841 2018-01-11 Bin Cheng <bin.cheng@arm.com>
1843 PR tree-optimization/83695
1844 * gimple-loop-linterchange.cc
1845 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
1846 reset cached scev information after interchange.
1847 (pass_linterchange::execute): Remove call to scev_reset_htab.
1849 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1851 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
1852 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
1853 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
1854 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
1855 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
1856 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
1857 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
1858 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
1859 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
1860 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
1861 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
1862 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
1863 (V_lane_reg): Likewise.
1864 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
1866 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
1867 (vfmal_lane_low<mode>_intrinsic,
1868 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
1869 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
1870 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
1871 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
1872 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
1873 vfmsl_lane_high<mode>_intrinsic): New define_insns.
1875 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1877 * config/arm/arm-cpus.in (fp16fml): New feature.
1878 (ALL_SIMD): Add fp16fml.
1879 (armv8.2-a): Add fp16fml as an option.
1880 (armv8.3-a): Likewise.
1881 (armv8.4-a): Add fp16fml as part of fp16.
1882 * config/arm/arm.h (TARGET_FP16FML): Define.
1883 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
1885 * config/arm/arm-modes.def (V2HF): Define.
1886 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
1887 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
1888 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
1889 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
1890 vfmsl_low, vfmsl_high): New set of builtins.
1891 * config/arm/iterators.md (PLUSMINUS): New code iterator.
1892 (vfml_op): New code attribute.
1893 (VFMLHALVES): New int iterator.
1894 (VFML, VFMLSEL): New mode attributes.
1895 (V_reg): Define mapping for V2HF.
1896 (V_hi, V_lo): New mode attributes.
1897 (VF_constraint): Likewise.
1898 (vfml_half, vfml_half_selector): New int attributes.
1899 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
1901 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
1902 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
1904 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
1905 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
1906 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
1907 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
1909 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
1910 Document new effective target and option set.
1912 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1914 * config/arm/arm-cpus.in (armv8_4): New feature.
1915 (ARMv8_4a): New fgroup.
1916 (armv8.4-a): New arch.
1917 * config/arm/arm-tables.opt: Regenerate.
1918 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
1919 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
1920 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
1921 Add matching rules for -march=armv8.4-a and extensions.
1922 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
1924 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
1927 * config/rx/rx.md (BW): New mode attribute.
1928 (sync_lock_test_and_setsi): Add mode suffix to insn output.
1930 2018-01-11 Richard Biener <rguenther@suse.de>
1932 PR tree-optimization/83435
1933 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
1934 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
1935 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
1937 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1938 Alan Hayward <alan.hayward@arm.com>
1939 David Sherwood <david.sherwood@arm.com>
1941 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
1943 (aarch64_classify_address): Initialize it. Track polynomial offsets.
1944 (aarch64_print_address_internal): Use it to check for a zero offset.
1946 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
1947 Alan Hayward <alan.hayward@arm.com>
1948 David Sherwood <david.sherwood@arm.com>
1950 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
1951 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
1952 Return a poly_int64 rather than a HOST_WIDE_INT.
1953 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
1954 rather than a HOST_WIDE_INT.
1955 * config/aarch64/aarch64.h (aarch64_frame): Protect with
1956 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
1957 hard_fp_offset, frame_size, initial_adjust, callee_offset and
1958 final_offset from HOST_WIDE_INT to poly_int64.
1959 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
1960 to_constant when getting the number of units in an Advanced SIMD
1962 (aarch64_builtin_vectorized_function): Check for a constant number
1964 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
1966 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
1967 attribute instead of GET_MODE_NUNITS.
1968 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
1969 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
1970 GET_MODE_SIZE for fixed-size registers.
1971 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
1972 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
1973 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
1974 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
1975 (aarch64_print_operand, aarch64_print_address_internal)
1976 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
1977 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
1978 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
1979 Handle polynomial GET_MODE_SIZE.
1980 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
1981 wider than SImode without modification.
1982 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
1983 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
1984 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
1985 passing and returning SVE modes.
1986 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
1987 rather than GEN_INT.
1988 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
1989 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
1990 (aarch64_allocate_and_probe_stack_space): Likewise.
1991 (aarch64_layout_frame): Cope with polynomial offsets.
1992 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
1993 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
1995 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
1996 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
1997 poly_int64 rather than a HOST_WIDE_INT.
1998 (aarch64_get_separate_components, aarch64_process_components)
1999 (aarch64_expand_prologue, aarch64_expand_epilogue)
2000 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2001 (aarch64_anchor_offset): New function, split out from...
2002 (aarch64_legitimize_address): ...here.
2003 (aarch64_builtin_vectorization_cost): Handle polynomial
2004 TYPE_VECTOR_SUBPARTS.
2005 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2007 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2008 number of elements from the PARALLEL rather than the mode.
2009 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2010 rather than GET_MODE_BITSIZE.
2011 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2012 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2013 (aarch64_expand_vec_perm_const_1): Handle polynomial
2014 d->perm.length () and d->perm elements.
2015 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2016 Apply to_constant to d->perm elements.
2017 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2018 polynomial CONST_VECTOR_NUNITS.
2019 (aarch64_move_pointer): Take amount as a poly_int64 rather
2021 (aarch64_progress_pointer): Avoid temporary variable.
2022 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2023 the mode attribute instead of GET_MODE.
2025 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2026 Alan Hayward <alan.hayward@arm.com>
2027 David Sherwood <david.sherwood@arm.com>
2029 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2030 x exists before using it.
2031 (aarch64_add_constant_internal): Rename to...
2032 (aarch64_add_offset_1): ...this. Replace regnum with separate
2033 src and dest rtxes. Handle the case in which they're different,
2034 including when the offset is zero. Replace scratchreg with an rtx.
2035 Use 2 additions if there is no spare register into which we can
2036 move a 16-bit constant.
2037 (aarch64_add_constant): Delete.
2038 (aarch64_add_offset): Replace reg with separate src and dest
2039 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2040 Use aarch64_add_offset_1.
2041 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2042 an rtx rather than an int. Take the delta as a poly_int64
2043 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
2044 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
2045 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
2046 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
2047 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
2049 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
2050 aarch64_add_constant.
2052 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2054 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
2055 Use scalar_float_mode.
2057 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2059 * config/aarch64/aarch64-simd.md
2060 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
2061 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
2062 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
2063 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
2064 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
2065 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
2066 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
2067 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
2068 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
2069 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
2071 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2074 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
2075 targ_options->x_arm_arch_string is non NULL.
2077 2018-01-11 Tamar Christina <tamar.christina@arm.com>
2079 * config/aarch64/aarch64.h
2080 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
2082 2018-01-11 Sudakshina Das <sudi.das@arm.com>
2085 * expmed.c (emit_store_flag_force): Swap if const op0
2086 and change VOIDmode to mode of op0.
2088 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2090 PR rtl-optimization/83761
2091 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
2092 than bytes to mode_for_size.
2094 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
2097 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
2098 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
2101 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
2104 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
2105 when in layout mode.
2106 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
2107 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
2110 2018-01-10 Michael Collison <michael.collison@arm.com>
2112 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
2113 * config/aarch64/aarch64-option-extension.def: Add
2114 AARCH64_OPT_EXTENSION of 'fp16fml'.
2115 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2116 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
2117 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
2118 * config/aarch64/constraints.md (Ui7): New constraint.
2119 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
2120 (VFMLA_SEL_W): Ditto.
2123 (VFMLA16_LOW): New int iterator.
2124 (VFMLA16_HIGH): Ditto.
2125 (UNSPEC_FMLAL): New unspec.
2126 (UNSPEC_FMLSL): Ditto.
2127 (UNSPEC_FMLAL2): Ditto.
2128 (UNSPEC_FMLSL2): Ditto.
2129 (f16mac): New code attribute.
2130 * config/aarch64/aarch64-simd-builtins.def
2131 (aarch64_fmlal_lowv2sf): Ditto.
2132 (aarch64_fmlsl_lowv2sf): Ditto.
2133 (aarch64_fmlalq_lowv4sf): Ditto.
2134 (aarch64_fmlslq_lowv4sf): Ditto.
2135 (aarch64_fmlal_highv2sf): Ditto.
2136 (aarch64_fmlsl_highv2sf): Ditto.
2137 (aarch64_fmlalq_highv4sf): Ditto.
2138 (aarch64_fmlslq_highv4sf): Ditto.
2139 (aarch64_fmlal_lane_lowv2sf): Ditto.
2140 (aarch64_fmlsl_lane_lowv2sf): Ditto.
2141 (aarch64_fmlal_laneq_lowv2sf): Ditto.
2142 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
2143 (aarch64_fmlalq_lane_lowv4sf): Ditto.
2144 (aarch64_fmlsl_lane_lowv4sf): Ditto.
2145 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
2146 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
2147 (aarch64_fmlal_lane_highv2sf): Ditto.
2148 (aarch64_fmlsl_lane_highv2sf): Ditto.
2149 (aarch64_fmlal_laneq_highv2sf): Ditto.
2150 (aarch64_fmlsl_laneq_highv2sf): Ditto.
2151 (aarch64_fmlalq_lane_highv4sf): Ditto.
2152 (aarch64_fmlsl_lane_highv4sf): Ditto.
2153 (aarch64_fmlalq_laneq_highv4sf): Ditto.
2154 (aarch64_fmlsl_laneq_highv4sf): Ditto.
2155 * config/aarch64/aarch64-simd.md:
2156 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
2157 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
2158 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
2159 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
2160 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
2161 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
2162 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
2163 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
2164 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
2165 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
2166 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
2167 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
2168 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
2169 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
2170 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
2171 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
2172 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
2173 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
2174 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
2175 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
2176 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
2177 (vfmlsl_low_u32): Ditto.
2178 (vfmlalq_low_u32): Ditto.
2179 (vfmlslq_low_u32): Ditto.
2180 (vfmlal_high_u32): Ditto.
2181 (vfmlsl_high_u32): Ditto.
2182 (vfmlalq_high_u32): Ditto.
2183 (vfmlslq_high_u32): Ditto.
2184 (vfmlal_lane_low_u32): Ditto.
2185 (vfmlsl_lane_low_u32): Ditto.
2186 (vfmlal_laneq_low_u32): Ditto.
2187 (vfmlsl_laneq_low_u32): Ditto.
2188 (vfmlalq_lane_low_u32): Ditto.
2189 (vfmlslq_lane_low_u32): Ditto.
2190 (vfmlalq_laneq_low_u32): Ditto.
2191 (vfmlslq_laneq_low_u32): Ditto.
2192 (vfmlal_lane_high_u32): Ditto.
2193 (vfmlsl_lane_high_u32): Ditto.
2194 (vfmlal_laneq_high_u32): Ditto.
2195 (vfmlsl_laneq_high_u32): Ditto.
2196 (vfmlalq_lane_high_u32): Ditto.
2197 (vfmlslq_lane_high_u32): Ditto.
2198 (vfmlalq_laneq_high_u32): Ditto.
2199 (vfmlslq_laneq_high_u32): Ditto.
2200 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
2201 (AARCH64_FL_FOR_ARCH8_4): New.
2202 (AARCH64_ISA_F16FML): New ISA flag.
2203 (TARGET_F16FML): New feature flag for fp16fml.
2204 (doc/invoke.texi): Document new fp16fml option.
2206 2018-01-10 Michael Collison <michael.collison@arm.com>
2208 * config/aarch64/aarch64-builtins.c:
2209 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
2210 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2211 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
2212 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
2213 (AARCH64_ISA_SHA3): New ISA flag.
2214 (TARGET_SHA3): New feature flag for sha3.
2215 * config/aarch64/iterators.md (sha512_op): New int attribute.
2216 (CRYPTO_SHA512): New int iterator.
2217 (UNSPEC_SHA512H): New unspec.
2218 (UNSPEC_SHA512H2): Ditto.
2219 (UNSPEC_SHA512SU0): Ditto.
2220 (UNSPEC_SHA512SU1): Ditto.
2221 * config/aarch64/aarch64-simd-builtins.def
2222 (aarch64_crypto_sha512hqv2di): New builtin.
2223 (aarch64_crypto_sha512h2qv2di): Ditto.
2224 (aarch64_crypto_sha512su0qv2di): Ditto.
2225 (aarch64_crypto_sha512su1qv2di): Ditto.
2226 (aarch64_eor3qv8hi): Ditto.
2227 (aarch64_rax1qv2di): Ditto.
2228 (aarch64_xarqv2di): Ditto.
2229 (aarch64_bcaxqv8hi): Ditto.
2230 * config/aarch64/aarch64-simd.md:
2231 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
2232 (aarch64_crypto_sha512su0qv2di): Ditto.
2233 (aarch64_crypto_sha512su1qv2di): Ditto.
2234 (aarch64_eor3qv8hi): Ditto.
2235 (aarch64_rax1qv2di): Ditto.
2236 (aarch64_xarqv2di): Ditto.
2237 (aarch64_bcaxqv8hi): Ditto.
2238 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
2239 (vsha512h2q_u64): Ditto.
2240 (vsha512su0q_u64): Ditto.
2241 (vsha512su1q_u64): Ditto.
2242 (veor3q_u16): Ditto.
2243 (vrax1q_u64): Ditto.
2245 (vbcaxq_u16): Ditto.
2246 * config/arm/types.md (crypto_sha512): New type attribute.
2247 (crypto_sha3): Ditto.
2248 (doc/invoke.texi): Document new sha3 option.
2250 2018-01-10 Michael Collison <michael.collison@arm.com>
2252 * config/aarch64/aarch64-builtins.c:
2253 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
2254 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2255 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
2256 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
2257 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
2258 (AARCH64_ISA_SM4): New ISA flag.
2259 (TARGET_SM4): New feature flag for sm4.
2260 * config/aarch64/aarch64-simd-builtins.def
2261 (aarch64_sm3ss1qv4si): Ditto.
2262 (aarch64_sm3tt1aq4si): Ditto.
2263 (aarch64_sm3tt1bq4si): Ditto.
2264 (aarch64_sm3tt2aq4si): Ditto.
2265 (aarch64_sm3tt2bq4si): Ditto.
2266 (aarch64_sm3partw1qv4si): Ditto.
2267 (aarch64_sm3partw2qv4si): Ditto.
2268 (aarch64_sm4eqv4si): Ditto.
2269 (aarch64_sm4ekeyqv4si): Ditto.
2270 * config/aarch64/aarch64-simd.md:
2271 (aarch64_sm3ss1qv4si): Ditto.
2272 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
2273 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
2274 (aarch64_sm4eqv4si): Ditto.
2275 (aarch64_sm4ekeyqv4si): Ditto.
2276 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
2277 (sm3part_op): Ditto.
2278 (CRYPTO_SM3TT): Ditto.
2279 (CRYPTO_SM3PART): Ditto.
2280 (UNSPEC_SM3SS1): New unspec.
2281 (UNSPEC_SM3TT1A): Ditto.
2282 (UNSPEC_SM3TT1B): Ditto.
2283 (UNSPEC_SM3TT2A): Ditto.
2284 (UNSPEC_SM3TT2B): Ditto.
2285 (UNSPEC_SM3PARTW1): Ditto.
2286 (UNSPEC_SM3PARTW2): Ditto.
2287 (UNSPEC_SM4E): Ditto.
2288 (UNSPEC_SM4EKEY): Ditto.
2289 * config/aarch64/constraints.md (Ui2): New constraint.
2290 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
2291 * config/arm/types.md (crypto_sm3): New type attribute.
2292 (crypto_sm4): Ditto.
2293 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
2294 (vsm3tt1aq_u32): Ditto.
2295 (vsm3tt1bq_u32): Ditto.
2296 (vsm3tt2aq_u32): Ditto.
2297 (vsm3tt2bq_u32): Ditto.
2298 (vsm3partw1q_u32): Ditto.
2299 (vsm3partw2q_u32): Ditto.
2300 (vsm4eq_u32): Ditto.
2301 (vsm4ekeyq_u32): Ditto.
2302 (doc/invoke.texi): Document new sm4 option.
2304 2018-01-10 Michael Collison <michael.collison@arm.com>
2306 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
2307 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
2308 (AARCH64_FL_FOR_ARCH8_4): New.
2309 (AARCH64_FL_V8_4): New flag.
2310 (doc/invoke.texi): Document new armv8.4-a option.
2312 2018-01-10 Michael Collison <michael.collison@arm.com>
2314 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2315 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
2316 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
2317 * config/aarch64/aarch64-option-extension.def: Add
2318 AARCH64_OPT_EXTENSION of 'sha2'.
2319 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
2320 (crypto): Disable sha2 and aes if crypto disabled.
2321 (crypto): Enable aes and sha2 if enabled.
2322 (simd): Disable sha2 and aes if simd disabled.
2323 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
2325 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
2326 (TARGET_SHA2): New feature flag for sha2.
2327 (TARGET_AES): New feature flag for aes.
2328 * config/aarch64/aarch64-simd.md:
2329 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
2330 conditional on TARGET_AES.
2331 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
2332 (aarch64_crypto_sha1hsi): Make pattern conditional
2334 (aarch64_crypto_sha1hv4si): Ditto.
2335 (aarch64_be_crypto_sha1hv4si): Ditto.
2336 (aarch64_crypto_sha1su1v4si): Ditto.
2337 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
2338 (aarch64_crypto_sha1su0v4si): Ditto.
2339 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
2340 (aarch64_crypto_sha256su0v4si): Ditto.
2341 (aarch64_crypto_sha256su1v4si): Ditto.
2342 (doc/invoke.texi): Document new aes and sha2 options.
2344 2018-01-10 Martin Sebor <msebor@redhat.com>
2346 PR tree-optimization/83781
2347 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
2350 2018-01-11 Martin Sebor <msebor@gmail.com>
2351 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2353 PR tree-optimization/83501
2354 PR tree-optimization/81703
2356 * tree-ssa-strlen.c (get_string_cst): Rename...
2357 (get_string_len): ...to this. Handle global constants.
2358 (handle_char_store): Adjust.
2360 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
2361 Jim Wilson <jimw@sifive.com>
2363 * config/riscv/riscv-protos.h (riscv_output_return): New.
2364 * config/riscv/riscv.c (struct machine_function): New naked_p field.
2365 (riscv_attribute_table, riscv_output_return),
2366 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
2367 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
2368 (riscv_compute_frame_info): Only compute frame->mask if not a naked
2370 (riscv_expand_prologue): Add early return for naked function.
2371 (riscv_expand_epilogue): Likewise.
2372 (riscv_function_ok_for_sibcall): Return false for naked function.
2373 (riscv_set_current_function): New.
2374 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
2375 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
2376 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
2377 * doc/extend.texi (RISC-V Function Attributes): New.
2379 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
2381 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
2382 check for 128-bit long double before checking TCmode.
2383 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
2384 128-bit long doubles before checking TFmode or TCmode.
2385 (FLOAT128_IBM_P): Likewise.
2387 2018-01-10 Martin Sebor <msebor@redhat.com>
2389 PR tree-optimization/83671
2390 * builtins.c (c_strlen): Unconditionally return zero for the empty
2392 Use -Warray-bounds for warnings.
2393 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
2394 for non-constant array indices with COMPONENT_REF, arrays of
2395 arrays, and pointers to arrays.
2396 (gimple_fold_builtin_strlen): Determine and set length range for
2397 non-constant character arrays.
2399 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
2402 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
2405 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
2407 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
2409 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
2412 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
2413 VECTOR_MEM_ALTIVEC_OR_VSX_P.
2414 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
2415 indexed_or_indirect_operand predicate.
2416 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
2417 (*vsx_le_perm_load_v8hi): Likewise.
2418 (*vsx_le_perm_load_v16qi): Likewise.
2419 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
2420 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
2421 (*vsx_le_perm_store_v8hi): Likewise.
2422 (*vsx_le_perm_store_v16qi): Likewise.
2423 (eight unnamed splitters): Likewise.
2425 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
2427 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
2428 * config/rs6000/emmintrin.h: Likewise.
2429 * config/rs6000/mmintrin.h: Likewise.
2430 * config/rs6000/xmmintrin.h: Likewise.
2432 2018-01-10 David Malcolm <dmalcolm@redhat.com>
2435 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
2437 * tree.c (tree_nop_conversion): Return true for location wrapper
2439 (maybe_wrap_with_location): New function.
2440 (selftest::check_strip_nops): New function.
2441 (selftest::test_location_wrappers): New function.
2442 (selftest::tree_c_tests): Call it.
2443 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
2444 (maybe_wrap_with_location): New decl.
2445 (EXPR_LOCATION_WRAPPER_P): New macro.
2446 (location_wrapper_p): New inline function.
2447 (tree_strip_any_location_wrapper): New inline function.
2449 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
2452 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
2453 stack_realign_offset for the largest alignment of stack slot
2455 (ix86_find_max_used_stack_alignment): New function.
2456 (ix86_finalize_stack_frame_flags): Use it. Set
2457 max_used_stack_alignment if we don't realign stack.
2458 * config/i386/i386.h (machine_function): Add
2459 max_used_stack_alignment.
2461 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
2463 * config/arm/arm.opt (-mbranch-cost): New option.
2464 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
2467 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
2470 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
2471 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
2473 2018-01-10 Richard Biener <rguenther@suse.de>
2476 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
2477 early out so it also covers the case where we have a non-NULL
2480 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
2482 PR tree-optimization/83753
2483 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
2484 for non-strided grouped accesses if the number of elements is 1.
2486 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
2489 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
2490 * i386.h (TARGET_USE_GATHER): Define.
2491 * x86-tune.def (X86_TUNE_USE_GATHER): New.
2493 2018-01-10 Martin Liska <mliska@suse.cz>
2496 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
2497 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
2499 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
2500 CLEANUP_NO_PARTITIONING is not set.
2502 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
2504 * doc/rtl.texi: Remove documentation of (const ...) wrappers
2505 for vectors, as a partial revert of r254296.
2506 * rtl.h (const_vec_p): Delete.
2507 (const_vec_duplicate_p): Don't test for vector CONSTs.
2508 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
2509 * expmed.c (make_tree): Likewise.
2512 * common.md (E, F): Use CONSTANT_P instead of checking for
2514 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
2515 checking for CONST_VECTOR.
2517 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
2520 * predict.c (force_edge_cold): Handle in more sane way edges
2523 2018-01-09 Carl Love <cel@us.ibm.com>
2525 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
2527 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
2528 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
2529 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
2530 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
2531 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
2532 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
2533 * config/rs6000/rs6000-protos.h: Add extern defition for
2534 rs6000_generate_float2_double_code.
2535 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
2537 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
2538 (float2_v2df): Add define_expand.
2540 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
2543 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
2544 op_mode in the force_to_mode call.
2546 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
2548 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
2549 instead of checking each element individually.
2550 (aarch64_evpc_uzp): Likewise.
2551 (aarch64_evpc_zip): Likewise.
2552 (aarch64_evpc_ext): Likewise.
2553 (aarch64_evpc_rev): Likewise.
2554 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
2555 instead of checking each element individually. Return true without
2557 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
2558 whether all selected elements come from the same input, instead of
2559 checking each element individually. Remove calls to gen_rtx_REG,
2560 start_sequence and end_sequence and instead assert that no rtl is
2563 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
2565 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
2566 order of HIGH and CONST checks.
2568 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
2570 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
2571 if the destination isn't an SSA_NAME.
2573 2018-01-09 Richard Biener <rguenther@suse.de>
2575 PR tree-optimization/83668
2576 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
2578 (canonicalize_loop_form): ... here, renamed from ...
2579 (canonicalize_loop_closed_ssa_form): ... this and amended to
2580 swap successor edges for loop exit blocks to make us use
2581 the RPO order we need for initial schedule generation.
2583 2018-01-09 Joseph Myers <joseph@codesourcery.com>
2585 PR tree-optimization/64811
2586 * match.pd: When optimizing comparisons with Inf, avoid
2587 introducing or losing exceptions from comparisons with NaN.
2589 2018-01-09 Martin Liska <mliska@suse.cz>
2592 * asan.c (shadow_mem_size): Add gcc_assert.
2594 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
2596 Don't save registers in main().
2599 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
2600 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
2601 * config/avr/avr.c (avr_set_current_function): Don't error if
2602 naked, OS_task or OS_main are specified at the same time.
2603 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
2605 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
2607 * common/config/avr/avr-common.c (avr_option_optimization_table):
2608 Switch on -mmain-is-OS_task for optimizing compilations.
2610 2018-01-09 Richard Biener <rguenther@suse.de>
2612 PR tree-optimization/83572
2613 * graphite.c: Include cfganal.h.
2614 (graphite_transform_loops): Connect infinite loops to exit
2615 and remove fake edges at the end.
2617 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
2619 * ipa-inline.c (edge_badness): Revert accidental checkin.
2621 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
2624 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
2625 symbols; not inline clones.
2627 2018-01-09 Jakub Jelinek <jakub@redhat.com>
2630 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
2631 hard registers. Formatting fixes.
2633 PR preprocessor/83722
2634 * gcc.c (try_generate_repro): Pass
2635 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
2636 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
2639 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
2640 Kito Cheng <kito.cheng@gmail.com>
2642 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
2643 (riscv_leaf_function_p): Delete.
2644 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
2646 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2648 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
2650 (do_ifelse): New function.
2651 (do_isel): New function.
2652 (do_sub3): New function.
2653 (do_add3): New function.
2654 (do_load_mask_compare): New function.
2655 (do_overlap_load_compare): New function.
2656 (expand_compare_loop): New function.
2657 (expand_block_compare): Call expand_compare_loop() when appropriate.
2658 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
2660 (-mblock-compare-inline-loop-limit): New option.
2662 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2665 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
2666 Reverse order of second and third operands in first alternative.
2667 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
2668 of first and second elements in UNSPEC_VPERMR vector.
2669 (altivec_expand_vec_perm_le): Likewise.
2671 2017-01-08 Jeff Law <law@redhat.com>
2673 PR rtl-optimizatin/81308
2674 * tree-switch-conversion.c (cfg_altered): New file scoped static.
2675 (process_switch): If group_case_labels makes a change, then set
2677 (pass_convert_switch::execute): If a switch is converted, then
2678 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
2680 PR rtl-optimization/81308
2681 * recog.c (split_all_insns): Conditionally cleanup the CFG after
2684 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
2686 PR target/83663 - Revert r255946
2687 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
2688 generation for cases where splatting a value is not useful.
2689 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
2690 across a vec_duplicate and a paradoxical subreg forming a vector
2691 mode to a vec_concat.
2693 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2695 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
2696 -march=armv8.3-a variants.
2697 * config/arm/t-multilib: Likewise.
2698 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
2700 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2702 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
2704 (cceq_ior_compare_complement): Give it a name so I can use it, and
2705 change boolean_or_operator predicate to boolean_operator so it can
2706 be used to generate a crand.
2707 (eqne): New code iterator.
2708 (bd/bd_neg): New code_attrs.
2709 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
2710 a single define_insn.
2711 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
2712 decrement (bdnzt/bdnzf/bdzt/bdzf).
2713 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
2714 with the new names of the branch decrement patterns, and added the
2715 names of the branch decrement conditional patterns.
2717 2018-01-08 Richard Biener <rguenther@suse.de>
2719 PR tree-optimization/83563
2720 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
2723 2018-01-08 Richard Biener <rguenther@suse.de>
2726 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
2728 2018-01-08 Richard Biener <rguenther@suse.de>
2730 PR tree-optimization/83685
2731 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
2732 references to abnormals.
2734 2018-01-08 Richard Biener <rguenther@suse.de>
2737 * dwarf2out.c (output_indirect_strings): Handle empty
2738 skeleton_debug_str_hash.
2739 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
2741 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
2743 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
2744 (emit_store_direct): Likewise.
2745 (arc_trampoline_adjust_address): Likewise.
2746 (arc_asm_trampoline_template): New function.
2747 (arc_initialize_trampoline): Use asm_trampoline_template.
2748 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
2749 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
2750 * config/arc/arc.md (flush_icache): Delete pattern.
2752 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
2754 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
2755 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
2758 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
2761 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
2762 by not USED_FOR_TARGET.
2763 (make_pass_resolve_sw_modes): Likewise.
2765 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
2767 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
2770 2018-01-08 Richard Biener <rguenther@suse.de>
2773 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
2775 2018-01-08 Richard Biener <rguenther@suse.de>
2778 * match.pd ((t * 2) / 2) -> t): Add missing :c.
2780 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
2783 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
2784 basic blocks with a small number of successors.
2785 (convert_control_dep_chain_into_preds): Improve handling of
2787 (dump_predicates): Split apart into...
2788 (dump_pred_chain): ...here...
2789 (dump_pred_info): ...and here.
2790 (can_one_predicate_be_invalidated_p): Add debugging printfs.
2791 (can_chain_union_be_invalidated_p): Improve check for invalidation
2793 (uninit_uses_cannot_happen): Avoid unnecessary if
2794 convert_control_dep_chain_into_preds yielded nothing.
2796 2018-01-06 Martin Sebor <msebor@redhat.com>
2798 PR tree-optimization/83640
2799 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
2800 subtracting negative offset from size.
2801 (builtin_access::overlap): Adjust offset bounds of the access to fall
2802 within the size of the object if possible.
2804 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
2806 PR rtl-optimization/83699
2807 * expmed.c (extract_bit_field_1): Restrict the vector usage of
2808 extract_bit_field_as_subreg to cases in which the extracted
2809 value is also a vector.
2811 * lra-constraints.c (process_alt_operands): Test for the equivalence
2812 substitutions when detecting a possible reload cycle.
2814 2018-01-06 Jakub Jelinek <jakub@redhat.com>
2817 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
2818 by default if flag_selective_schedling{,2}. Formatting fixes.
2820 PR rtl-optimization/83682
2821 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
2822 if it has non-VECTOR_MODE element mode.
2823 (vec_duplicate_p): Likewise.
2826 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
2827 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
2829 2018-01-05 Jakub Jelinek <jakub@redhat.com>
2832 * config/i386/i386-builtin.def
2833 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
2834 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
2835 Require also OPTION_MASK_ISA_AVX512F in addition to
2836 OPTION_MASK_ISA_GFNI.
2837 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
2838 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
2839 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
2840 to OPTION_MASK_ISA_GFNI.
2841 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
2842 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
2843 OPTION_MASK_ISA_AVX512BW.
2844 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
2845 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
2846 addition to OPTION_MASK_ISA_GFNI.
2847 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
2848 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
2849 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
2850 to OPTION_MASK_ISA_GFNI.
2851 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
2852 a requirement for all ISAs rather than any of them with a few
2854 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
2856 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
2857 bitmasks to be enabled with 3 exceptions, instead of requiring any
2858 enabled ISA with lots of exceptions.
2859 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
2860 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
2861 Change avx512bw in isa attribute to avx512f.
2862 * config/i386/sgxintrin.h: Add license boilerplate.
2863 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
2864 to __AVX512F__ and __AVX512VL to __AVX512VL__.
2865 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
2866 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
2868 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
2869 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
2870 temporarily sse2 rather than sse if not enabled already.
2873 * config/i386/sse.md (VI248_VLBW): Rename to ...
2874 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
2875 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
2876 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
2877 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
2878 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
2879 mode iterator instead of VI248_VLBW.
2881 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
2883 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
2884 (record_modified): Skip clobbers; add debug output.
2885 (param_change_prob): Use sreal frequencies.
2887 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
2889 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
2890 punt for user-aligned variables.
2892 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
2894 * tree-chrec.c (chrec_contains_symbols): Return true for
2897 2018-01-05 Sudakshina Das <sudi.das@arm.com>
2900 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
2901 of (x|y) == x for BICS pattern.
2903 2018-01-05 Jakub Jelinek <jakub@redhat.com>
2905 PR tree-optimization/83605
2906 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
2907 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
2910 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
2912 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
2913 * config/epiphany/rtems.h: New file.
2915 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2916 Uros Bizjak <ubizjak@gmail.com>
2919 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
2920 QIreg_operand instead of register_operand predicate.
2921 * config/i386/i386.c (ix86_rop_should_change_byte_p,
2922 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
2923 comments instead of -fmitigate[-_]rop.
2925 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2928 * cgraphunit.c (symbol_table::compile): Switch to text_section
2929 before calling assembly_start debug hook.
2930 * run-rtl-passes.c (run_rtl_passes): Likewise.
2933 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2935 * tree-vrp.c (extract_range_from_binary_expr_1): Check
2936 range_int_cst_p rather than !symbolic_range_p before calling
2937 extract_range_from_multiplicative_op_1.
2939 2017-01-04 Jeff Law <law@redhat.com>
2941 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
2942 redundant test in assertion.
2944 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2946 * doc/rtl.texi: Document machine_mode wrapper classes.
2948 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2950 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
2953 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2955 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
2956 the VEC_PERM_EXPR fold to fail.
2958 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2961 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
2962 to switched_sections.
2964 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2967 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
2970 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
2973 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
2974 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
2976 2018-01-04 Jakub Jelinek <jakub@redhat.com>
2979 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
2980 is BLKmode and bitpos not zero or mode change is needed.
2982 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
2985 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
2988 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
2991 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
2992 instead of MULT rtx. Update all corresponding splitters.
2994 (*ssub<modesuffix>): Ditto.
2996 (*cmp_sadd_di): Update split patterns.
2997 (*cmp_sadd_si): Ditto.
2998 (*cmp_sadd_sidi): Ditto.
2999 (*cmp_ssub_di): Ditto.
3000 (*cmp_ssub_si): Ditto.
3001 (*cmp_ssub_sidi): Ditto.
3002 * config/alpha/predicates.md (const23_operand): New predicate.
3003 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3004 Look for ASHIFT, not MULT inner operand.
3005 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3007 2018-01-04 Martin Liska <mliska@suse.cz>
3009 PR gcov-profile/83669
3010 * gcov.c (output_intermediate_file): Add version to intermediate
3012 * doc/gcov.texi: Document new field 'version' in intermediate
3013 file format. Fix location of '-k' option of gcov command.
3015 2018-01-04 Martin Liska <mliska@suse.cz>
3018 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3020 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3022 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3024 2018-01-03 Martin Sebor <msebor@redhat.com>
3026 PR tree-optimization/83655
3027 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3028 checking calls with invalid arguments.
3030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3032 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3033 (vectorizable_mask_load_store): Delete.
3034 (vectorizable_call): Return false for masked loads and stores.
3035 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3036 instead of gimple_assign_rhs1.
3037 (vectorizable_load): Handle IFN_MASK_LOAD.
3038 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3040 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3042 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
3044 (vectorizable_mask_load_store): ...here.
3045 (vectorizable_load): ...and here.
3047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3049 * tree-vect-stmts.c (vect_build_all_ones_mask)
3050 (vect_build_zero_merge_argument): New functions, split out from...
3051 (vectorizable_load): ...here.
3053 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3055 * tree-vect-stmts.c (vect_check_store_rhs): New function,
3057 (vectorizable_mask_load_store): ...here.
3058 (vectorizable_store): ...and here.
3060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3062 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
3064 (vectorizable_mask_load_store): ...here.
3066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3068 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
3069 (vect_model_store_cost): Take a vec_load_store_type instead of a
3071 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
3072 (vect_model_store_cost): Take a vec_load_store_type instead of a
3074 (vectorizable_mask_load_store): Update accordingly.
3075 (vectorizable_store): Likewise.
3076 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
3078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3080 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
3081 IFN_MASK_LOAD calls here rather than...
3082 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
3084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3085 Alan Hayward <alan.hayward@arm.com>
3086 David Sherwood <david.sherwood@arm.com>
3088 * expmed.c (extract_bit_field_1): For vector extracts,
3089 fall back to extract_bit_field_as_subreg if vec_extract
3092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3093 Alan Hayward <alan.hayward@arm.com>
3094 David Sherwood <david.sherwood@arm.com>
3096 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
3097 they are variable or constant sized.
3098 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
3099 slots for constant-sized data.
3101 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3102 Alan Hayward <alan.hayward@arm.com>
3103 David Sherwood <david.sherwood@arm.com>
3105 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
3106 handling COND_EXPRs with boolean comparisons, try to find a better
3107 basis for the mask type than the boolean itself.
3109 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3111 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
3112 is calculated and how it can be overridden.
3113 * genmodes.c (max_bitsize_mode_any_mode): New variable.
3114 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
3116 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
3119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3120 Alan Hayward <alan.hayward@arm.com>
3121 David Sherwood <david.sherwood@arm.com>
3123 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
3124 Remove the mode argument.
3125 (aarch64_simd_valid_immediate): Remove the mode and inverse
3127 * config/aarch64/iterators.md (bitsize): New iterator.
3128 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
3129 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
3130 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
3131 aarch64_simd_valid_immediate.
3132 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
3133 (aarch64_reg_or_bic_imm): Likewise.
3134 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
3135 with an insn_type enum and msl with a modifier_type enum.
3136 Replace element_width with a scalar_mode. Change the shift
3137 to unsigned int. Add constructors for scalar_float_mode and
3138 scalar_int_mode elements.
3139 (aarch64_vect_float_const_representable_p): Delete.
3140 (aarch64_can_const_movi_rtx_p)
3141 (aarch64_simd_scalar_immediate_valid_for_move)
3142 (aarch64_simd_make_constant): Update call to
3143 aarch64_simd_valid_immediate.
3144 (aarch64_advsimd_valid_immediate_hs): New function.
3145 (aarch64_advsimd_valid_immediate): Likewise.
3146 (aarch64_simd_valid_immediate): Remove mode and inverse
3147 arguments. Rewrite to use the above. Use const_vec_duplicate_p
3148 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
3149 and aarch64_float_const_representable_p on the result.
3150 (aarch64_output_simd_mov_immediate): Remove mode argument.
3151 Update call to aarch64_simd_valid_immediate and use of
3152 simd_immediate_info.
3153 (aarch64_output_scalar_simd_mov_immediate): Update call
3156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3157 Alan Hayward <alan.hayward@arm.com>
3158 David Sherwood <david.sherwood@arm.com>
3160 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
3161 (mode_nunits): Likewise CONST_MODE_NUNITS.
3162 * machmode.def (ADJUST_NUNITS): Document.
3163 * genmodes.c (mode_data::need_nunits_adj): New field.
3164 (blank_mode): Update accordingly.
3165 (adj_nunits): New variable.
3166 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
3168 (emit_mode_size_inline): Set need_bytesize_adj for all modes
3169 listed in adj_nunits.
3170 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
3171 listed in adj_nunits. Don't emit case statements for such modes.
3172 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
3173 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
3174 nothing if adj_nunits is nonnull.
3175 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
3176 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
3177 (emit_mode_fbit): Update use of print_maybe_const_decl.
3178 (emit_move_size): Likewise. Treat the array as non-const
3180 (emit_mode_adjustments): Handle adj_nunits.
3182 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3184 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
3185 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
3186 (VECTOR_MODES): Use it.
3187 (make_vector_modes): Take the prefix as an argument.
3189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3190 Alan Hayward <alan.hayward@arm.com>
3191 David Sherwood <david.sherwood@arm.com>
3193 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
3194 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
3195 for MODE_VECTOR_BOOL.
3196 * machmode.def (VECTOR_BOOL_MODE): Document.
3197 * genmodes.c (VECTOR_BOOL_MODE): New macro.
3198 (make_vector_bool_mode): New function.
3199 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
3201 * lto-streamer-in.c (lto_input_mode_table): Likewise.
3202 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
3204 * stor-layout.c (int_mode_for_mode): Likewise.
3205 * tree.c (build_vector_type_for_mode): Likewise.
3206 * varasm.c (output_constant_pool_2): Likewise.
3207 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
3208 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
3209 for MODE_VECTOR_BOOL.
3210 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
3211 of mode class checks.
3212 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
3213 instead of a list of mode class checks.
3214 (expand_vector_scalar_condition): Likewise.
3215 (type_for_widest_vector_mode): Handle BImode as an inner mode.
3217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3218 Alan Hayward <alan.hayward@arm.com>
3219 David Sherwood <david.sherwood@arm.com>
3221 * machmode.h (mode_size): Change from unsigned short to
3223 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
3224 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
3225 or if measurement_type is not polynomial.
3226 (fixed_size_mode::includes_p): Check for constant-sized modes.
3227 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
3228 return a poly_uint16 rather than an unsigned short.
3229 (emit_mode_size): Change the type of mode_size from unsigned short
3230 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
3231 (emit_mode_adjustments): Cope with polynomial vector sizes.
3232 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
3234 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
3236 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
3237 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
3238 * caller-save.c (setup_save_areas): Likewise.
3239 (replace_reg_with_saved_mem): Likewise.
3240 * calls.c (emit_library_call_value_1): Likewise.
3241 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
3242 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
3243 (gen_lowpart_for_combine): Likewise.
3244 * convert.c (convert_to_integer_1): Likewise.
3245 * cse.c (equiv_constant, cse_insn): Likewise.
3246 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
3247 (cselib_subst_to_values): Likewise.
3248 * dce.c (word_dce_process_block): Likewise.
3249 * df-problems.c (df_word_lr_mark_ref): Likewise.
3250 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
3251 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
3252 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
3253 (rtl_for_decl_location): Likewise.
3254 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
3255 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
3256 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
3257 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
3258 (expand_expr_real_1): Likewise.
3259 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
3260 (pad_below): Likewise.
3261 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
3262 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
3263 * ira.c (get_subreg_tracking_sizes): Likewise.
3264 * ira-build.c (ira_create_allocno_objects): Likewise.
3265 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
3266 (ira_sort_regnos_for_alter_reg): Likewise.
3267 * ira-costs.c (record_operand_costs): Likewise.
3268 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
3269 (resolve_simple_move): Likewise.
3270 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
3271 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
3272 (lra_constraints): Likewise.
3273 (CONST_POOL_OK_P): Reject variable-sized modes.
3274 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
3275 (add_pseudo_to_slot, lra_spill): Likewise.
3276 * omp-low.c (omp_clause_aligned_alignment): Likewise.
3277 * optabs-query.c (get_best_extraction_insn): Likewise.
3278 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
3279 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
3280 (expand_mult_highpart, valid_multiword_target_p): Likewise.
3281 * recog.c (offsettable_address_addr_space_p): Likewise.
3282 * regcprop.c (maybe_mode_change): Likewise.
3283 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
3284 * regrename.c (build_def_use): Likewise.
3285 * regstat.c (dump_reg_info): Likewise.
3286 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
3287 (find_reloads, find_reloads_subreg_address): Likewise.
3288 * reload1.c (eliminate_regs_1): Likewise.
3289 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
3290 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
3291 (simplify_binary_operation_1, simplify_subreg): Likewise.
3292 * targhooks.c (default_function_arg_padding): Likewise.
3293 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
3294 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
3295 (verify_gimple_assign_ternary): Likewise.
3296 * tree-inline.c (estimate_move_cost): Likewise.
3297 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3298 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
3299 (get_address_cost_ainc): Likewise.
3300 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
3301 (vect_supportable_dr_alignment): Likewise.
3302 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
3303 (vectorizable_reduction): Likewise.
3304 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
3305 (vectorizable_operation, vectorizable_load): Likewise.
3306 * tree.c (build_same_sized_truth_vector_type): Likewise.
3307 * valtrack.c (cleanup_auto_inc_dec): Likewise.
3308 * var-tracking.c (emit_note_insn_var_location): Likewise.
3309 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
3310 (ADDR_VEC_ALIGN): Likewise.
3312 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3313 Alan Hayward <alan.hayward@arm.com>
3314 David Sherwood <david.sherwood@arm.com>
3316 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
3318 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
3319 or if measurement_type is polynomial.
3320 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
3321 * combine.c (make_extraction): Likewise.
3322 * dse.c (find_shift_sequence): Likewise.
3323 * dwarf2out.c (mem_loc_descriptor): Likewise.
3324 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
3325 (extract_bit_field, extract_low_bits): Likewise.
3326 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
3327 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
3328 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
3329 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
3330 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
3331 * reload.c (find_reloads): Likewise.
3332 * reload1.c (alter_reg): Likewise.
3333 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
3334 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
3335 * tree-if-conv.c (predicate_mem_writes): Likewise.
3336 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
3337 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
3338 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
3339 * valtrack.c (dead_debug_insert_temp): Likewise.
3340 * varasm.c (mergeable_constant_section): Likewise.
3341 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
3343 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3344 Alan Hayward <alan.hayward@arm.com>
3345 David Sherwood <david.sherwood@arm.com>
3347 * expr.c (expand_assignment): Cope with polynomial mode sizes
3348 when assigning to a CONCAT.
3350 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3351 Alan Hayward <alan.hayward@arm.com>
3352 David Sherwood <david.sherwood@arm.com>
3354 * machmode.h (mode_precision): Change from unsigned short to
3356 (mode_to_precision): Return a poly_uint16 rather than an unsigned
3358 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
3359 or if measurement_type is not polynomial.
3360 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
3361 in which the mode is already known to be a scalar_int_mode.
3362 * genmodes.c (emit_mode_precision): Change the type of mode_precision
3363 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
3365 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
3366 for GET_MODE_PRECISION.
3367 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
3368 for GET_MODE_PRECISION.
3369 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
3371 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
3372 (expand_field_assignment, make_extraction): Likewise.
3373 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
3374 (get_last_value): Likewise.
3375 * convert.c (convert_to_integer_1): Likewise.
3376 * cse.c (cse_insn): Likewise.
3377 * expr.c (expand_expr_real_1): Likewise.
3378 * lra-constraints.c (simplify_operand_subreg): Likewise.
3379 * optabs-query.c (can_atomic_load_p): Likewise.
3380 * optabs.c (expand_atomic_load): Likewise.
3381 (expand_atomic_store): Likewise.
3382 * ree.c (combine_reaching_defs): Likewise.
3383 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
3384 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
3385 * tree.h (type_has_mode_precision_p): Likewise.
3386 * ubsan.c (instrument_si_overflow): Likewise.
3388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3389 Alan Hayward <alan.hayward@arm.com>
3390 David Sherwood <david.sherwood@arm.com>
3392 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
3393 polynomial numbers of units.
3394 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
3395 (valid_vector_subparts_p): New function.
3396 (build_vector_type): Remove temporary shim and take the number
3397 of units as a poly_uint64 rather than an int.
3398 (build_opaque_vector_type): Take the number of units as a
3399 poly_uint64 rather than an int.
3400 * tree.c (build_vector_from_ctor): Handle polynomial
3401 TYPE_VECTOR_SUBPARTS.
3402 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
3403 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
3404 (build_vector_from_val): If the number of units is variable,
3405 use build_vec_duplicate_cst for constant operands and
3406 VEC_DUPLICATE_EXPR otherwise.
3407 (make_vector_type): Remove temporary is_constant ().
3408 (build_vector_type, build_opaque_vector_type): Take the number of
3409 units as a poly_uint64 rather than an int.
3410 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
3412 * cfgexpand.c (expand_debug_expr): Likewise.
3413 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
3414 (store_constructor, expand_expr_real_1): Likewise.
3415 (const_scalar_mask_from_tree): Likewise.
3416 * fold-const-call.c (fold_const_reduction): Likewise.
3417 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
3418 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
3419 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
3420 (fold_relational_const): Likewise.
3421 (native_interpret_vector): Likewise. Change the size from an
3422 int to an unsigned int.
3423 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
3424 TYPE_VECTOR_SUBPARTS.
3425 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
3426 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
3427 duplicating a non-constant operand into a variable-length vector.
3428 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
3429 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
3430 * ipa-icf.c (sem_variable::equals): Likewise.
3431 * match.pd: Likewise.
3432 * omp-simd-clone.c (simd_clone_subparts): Likewise.
3433 * print-tree.c (print_node): Likewise.
3434 * stor-layout.c (layout_type): Likewise.
3435 * targhooks.c (default_builtin_vectorization_cost): Likewise.
3436 * tree-cfg.c (verify_gimple_comparison): Likewise.
3437 (verify_gimple_assign_binary): Likewise.
3438 (verify_gimple_assign_ternary): Likewise.
3439 (verify_gimple_assign_single): Likewise.
3440 * tree-pretty-print.c (dump_generic_node): Likewise.
3441 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3442 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
3443 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
3444 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
3445 (vect_shift_permute_load_chain): Likewise.
3446 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
3447 (expand_vector_condition, optimize_vector_constructor): Likewise.
3448 (lower_vec_perm, get_compute_type): Likewise.
3449 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
3450 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
3451 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
3452 (vect_recog_mask_conversion_pattern): Likewise.
3453 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
3454 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
3455 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3456 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
3457 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
3458 (vectorizable_shift, vectorizable_operation, vectorizable_store)
3459 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
3460 (supportable_widening_operation): Likewise.
3461 (supportable_narrowing_operation): Likewise.
3462 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
3464 * varasm.c (output_constant): Likewise.
3466 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3467 Alan Hayward <alan.hayward@arm.com>
3468 David Sherwood <david.sherwood@arm.com>
3470 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
3471 so that both the length == 3 and length != 3 cases set up their
3472 own permute vectors. Add comments explaining why we know the
3473 number of elements is constant.
3474 (vect_permute_load_chain): Likewise.
3476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3477 Alan Hayward <alan.hayward@arm.com>
3478 David Sherwood <david.sherwood@arm.com>
3480 * machmode.h (mode_nunits): Change from unsigned char to
3482 (ONLY_FIXED_SIZE_MODES): New macro.
3483 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
3484 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
3485 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
3487 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
3488 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
3489 or if measurement_type is not polynomial.
3490 * genmodes.c (ZERO_COEFFS): New macro.
3491 (emit_mode_nunits_inline): Make mode_nunits_inline return a
3493 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
3494 Use ZERO_COEFFS when emitting initializers.
3495 * data-streamer.h (bp_pack_poly_value): New function.
3496 (bp_unpack_poly_value): Likewise.
3497 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
3498 for GET_MODE_NUNITS.
3499 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
3500 for GET_MODE_NUNITS.
3501 * tree.c (make_vector_type): Remove temporary shim and make
3502 the real function take the number of units as a poly_uint64
3504 (build_vector_type_for_mode): Handle polynomial nunits.
3505 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
3506 * emit-rtl.c (const_vec_series_p_1): Likewise.
3507 (gen_rtx_CONST_VECTOR): Likewise.
3508 * fold-const.c (test_vec_duplicate_folding): Likewise.
3509 * genrecog.c (validate_pattern): Likewise.
3510 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
3511 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
3512 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
3513 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
3514 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
3515 * rtlanal.c (subreg_get_info): Likewise.
3516 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
3517 (vect_grouped_load_supported): Likewise.
3518 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
3519 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
3520 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
3521 (simplify_const_unary_operation, simplify_binary_operation_1)
3522 (simplify_const_binary_operation, simplify_ternary_operation)
3523 (test_vector_ops_duplicate, test_vector_ops): Likewise.
3524 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
3525 instead of CONST_VECTOR_NUNITS.
3526 * varasm.c (output_constant_pool_2): Likewise.
3527 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
3528 explicit-encoded elements in the XVEC for variable-length vectors.
3530 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3532 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
3534 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3535 Alan Hayward <alan.hayward@arm.com>
3536 David Sherwood <david.sherwood@arm.com>
3538 * coretypes.h (fixed_size_mode): Declare.
3539 (fixed_size_mode_pod): New typedef.
3540 * builtins.h (target_builtins::x_apply_args_mode)
3541 (target_builtins::x_apply_result_mode): Change type to
3542 fixed_size_mode_pod.
3543 * builtins.c (apply_args_size, apply_result_size, result_vector)
3544 (expand_builtin_apply_args_1, expand_builtin_apply)
3545 (expand_builtin_return): Update accordingly.
3547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3549 * cse.c (hash_rtx_cb): Hash only the encoded elements.
3550 * cselib.c (cselib_hash_rtx): Likewise.
3551 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
3552 CONST_VECTOR encoding.
3554 2017-01-03 Jakub Jelinek <jakub@redhat.com>
3555 Jeff Law <law@redhat.com>
3558 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
3559 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
3560 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
3561 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
3564 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
3565 explicitly probe *sp in a noreturn function if there were any callee
3566 register saves or frame pointer is needed.
3568 2018-01-03 Jakub Jelinek <jakub@redhat.com>
3571 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
3572 BLKmode for ternary, binary or unary expressions.
3575 * var-tracking.c (delete_vta_debug_insn): New inline function.
3576 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
3577 insns from get_insns () to NULL instead of each bb separately.
3578 Use delete_vta_debug_insn. No longer static.
3579 (vt_debug_insns_local, variable_tracking_main_1): Adjust
3580 delete_vta_debug_insns callers.
3581 * rtl.h (delete_vta_debug_insns): Declare.
3582 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
3583 instead of variable_tracking_main.
3585 2018-01-03 Martin Sebor <msebor@redhat.com>
3587 PR tree-optimization/83603
3588 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
3589 arguments past the endof the argument list in functions declared
3590 without a prototype.
3591 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
3592 Avoid checking when arguments are null.
3594 2018-01-03 Martin Sebor <msebor@redhat.com>
3597 * doc/extend.texi (attribute const): Fix a typo.
3598 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
3599 issuing -Wsuggest-attribute for void functions.
3601 2018-01-03 Martin Sebor <msebor@redhat.com>
3603 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
3604 offset_int::from instead of wide_int::to_shwi.
3605 (maybe_diag_overlap): Remove assertion.
3606 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
3607 * gimple-ssa-sprintf.c (format_directive): Same.
3608 (parse_directive): Same.
3609 (sprintf_dom_walker::compute_format_length): Same.
3610 (try_substitute_return_value): Same.
3612 2017-01-03 Jeff Law <law@redhat.com>
3615 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
3616 non-constant residual for zero at runtime and avoid probing in
3617 that case. Reorganize code for trailing problem to mirror handling
3620 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3622 PR tree-optimization/83501
3623 * tree-ssa-strlen.c (get_string_cst): New.
3624 (handle_char_store): Call get_string_cst.
3626 2018-01-03 Martin Liska <mliska@suse.cz>
3628 PR tree-optimization/83593
3629 * tree-ssa-strlen.c: Include tree-cfg.h.
3630 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
3631 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
3632 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
3634 (strlen_dom_walker::before_dom_children): Call
3635 gimple_purge_dead_eh_edges. Dump tranformation with details
3637 (strlen_dom_walker::before_dom_children): Update call by adding
3638 new argument cleanup_eh.
3639 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
3641 2018-01-03 Martin Liska <mliska@suse.cz>
3644 * cif-code.def (VARIADIC_THUNK): New enum value.
3645 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
3648 2018-01-03 Jan Beulich <jbeulich@suse.com>
3650 * sse.md (mov<mode>_internal): Tighten condition for when to use
3651 vmovdqu<ssescalarsize> for TI and OI modes.
3653 2018-01-03 Jakub Jelinek <jakub@redhat.com>
3655 Update copyright years.
3657 2018-01-03 Martin Liska <mliska@suse.cz>
3660 * ipa-visibility.c (function_and_variable_visibility): Skip
3661 functions with noipa attribure.
3663 2018-01-03 Jakub Jelinek <jakub@redhat.com>
3665 * gcc.c (process_command): Update copyright notice dates.
3666 * gcov-dump.c (print_version): Ditto.
3667 * gcov.c (print_version): Ditto.
3668 * gcov-tool.c (print_version): Ditto.
3669 * gengtype.c (create_file): Ditto.
3670 * doc/cpp.texi: Bump @copying's copyright year.
3671 * doc/cppinternals.texi: Ditto.
3672 * doc/gcc.texi: Ditto.
3673 * doc/gccint.texi: Ditto.
3674 * doc/gcov.texi: Ditto.
3675 * doc/install.texi: Ditto.
3676 * doc/invoke.texi: Ditto.
3678 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3680 * vector-builder.h (vector_builder::m_full_nelts): Change from
3681 unsigned int to poly_uint64.
3682 (vector_builder::full_nelts): Update prototype accordingly.
3683 (vector_builder::new_vector): Likewise.
3684 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
3685 (vector_builder::operator ==): Likewise.
3686 (vector_builder::finalize): Likewise.
3687 * int-vector-builder.h (int_vector_builder::int_vector_builder):
3688 Take the number of elements as a poly_uint64 rather than an
3690 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
3691 from unsigned int to poly_uint64.
3692 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
3693 (vec_perm_indices::new_vector): Likewise.
3694 (vec_perm_indices::length): Likewise.
3695 (vec_perm_indices::nelts_per_input): Likewise.
3696 (vec_perm_indices::input_nelts): Likewise.
3697 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
3698 number of elements per input as a poly_uint64 rather than an
3699 unsigned int. Use the original encoding for variable-length
3700 vectors, rather than clamping each individual element.
3701 For the second and subsequent elements in each pattern,
3702 clamp the step and base before clamping their sum.
3703 (vec_perm_indices::series_p): Handle polynomial element counts.
3704 (vec_perm_indices::all_in_range_p): Likewise.
3705 (vec_perm_indices_to_tree): Likewise.
3706 (vec_perm_indices_to_rtx): Likewise.
3707 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
3708 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
3709 (tree_vector_builder::new_binary_operation): Handle polynomial
3710 element counts. Return false if we need to know the number
3711 of elements at compile time.
3712 * fold-const.c (fold_vec_perm): Punt if the number of elements
3713 isn't known at compile time.
3715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3717 * vec-perm-indices.h (vec_perm_builder): Change element type
3718 from HOST_WIDE_INT to poly_int64.
3719 (vec_perm_indices::element_type): Update accordingly.
3720 (vec_perm_indices::clamp): Handle polynomial element_types.
3721 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
3722 (vec_perm_indices::all_in_range_p): Likewise.
3723 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
3725 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
3726 polynomial vec_perm_indices element types.
3727 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
3728 * fold-const.c (fold_vec_perm): Likewise.
3729 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
3730 * tree-vect-generic.c (lower_vec_perm): Likewise.
3731 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
3732 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
3733 element type to HOST_WIDE_INT.
3735 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3736 Alan Hayward <alan.hayward@arm.com>
3737 David Sherwood <david.sherwood@arm.com>
3739 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
3740 rather than an int. Use plus_constant.
3741 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
3742 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
3744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3745 Alan Hayward <alan.hayward@arm.com>
3746 David Sherwood <david.sherwood@arm.com>
3748 * calls.c (emit_call_1, expand_call): Change struct_value_size from
3749 a HOST_WIDE_INT to a poly_int64.
3751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3752 Alan Hayward <alan.hayward@arm.com>
3753 David Sherwood <david.sherwood@arm.com>
3755 * calls.c (load_register_parameters): Cope with polynomial
3756 mode sizes. Require a constant size for BLKmode parameters
3757 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
3758 forces a parameter to be padded at the lsb end in order to
3759 fill a complete number of words, require the parameter size
3760 to be ordered wrt UNITS_PER_WORD.
3762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3763 Alan Hayward <alan.hayward@arm.com>
3764 David Sherwood <david.sherwood@arm.com>
3766 * reload1.c (spill_stack_slot_width): Change element type
3767 from unsigned int to poly_uint64_pod.
3768 (alter_reg): Treat mode sizes as polynomial.
3770 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3771 Alan Hayward <alan.hayward@arm.com>
3772 David Sherwood <david.sherwood@arm.com>
3774 * reload.c (complex_word_subreg_p): New function.
3775 (reload_inner_reg_of_subreg, push_reload): Use it.
3777 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3778 Alan Hayward <alan.hayward@arm.com>
3779 David Sherwood <david.sherwood@arm.com>
3781 * lra-constraints.c (process_alt_operands): Reject matched
3782 operands whose sizes aren't ordered.
3783 (match_reload): Refer to this check here.
3785 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3786 Alan Hayward <alan.hayward@arm.com>
3787 David Sherwood <david.sherwood@arm.com>
3789 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
3790 that the mode size is in the set {1, 2, 4, 8, 16}.
3792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3793 Alan Hayward <alan.hayward@arm.com>
3794 David Sherwood <david.sherwood@arm.com>
3796 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
3797 Use plus_constant instead of gen_rtx_PLUS.
3799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3800 Alan Hayward <alan.hayward@arm.com>
3801 David Sherwood <david.sherwood@arm.com>
3803 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
3804 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
3805 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
3806 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
3807 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
3808 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
3809 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
3810 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
3811 * config/i386/i386.c (ix86_push_rounding): ...this new function.
3812 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
3814 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
3815 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
3816 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
3817 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
3818 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
3819 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
3820 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
3821 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
3822 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
3823 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
3825 * expr.c (emit_move_resolve_push): Treat the input and result
3826 of PUSH_ROUNDING as a poly_int64.
3827 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
3828 (emit_push_insn): Likewise.
3829 * lra-eliminations.c (mark_not_eliminable): Likewise.
3830 * recog.c (push_operand): Likewise.
3831 * reload1.c (elimination_effects): Likewise.
3832 * rtlanal.c (nonzero_bits1): Likewise.
3833 * calls.c (store_one_arg): Likewise. Require the padding to be
3834 known at compile time.
3836 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3837 Alan Hayward <alan.hayward@arm.com>
3838 David Sherwood <david.sherwood@arm.com>
3840 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
3841 Use plus_constant instead of gen_rtx_PLUS.
3843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3844 Alan Hayward <alan.hayward@arm.com>
3845 David Sherwood <david.sherwood@arm.com>
3847 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
3850 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3851 Alan Hayward <alan.hayward@arm.com>
3852 David Sherwood <david.sherwood@arm.com>
3854 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
3855 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
3856 via stack temporaries. Treat the mode size as polynomial too.
3858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3859 Alan Hayward <alan.hayward@arm.com>
3860 David Sherwood <david.sherwood@arm.com>
3862 * expr.c (expand_expr_real_2): When handling conversions involving
3863 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
3864 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
3865 as a poly_uint64 too.
3867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3868 Alan Hayward <alan.hayward@arm.com>
3869 David Sherwood <david.sherwood@arm.com>
3871 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
3873 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3874 Alan Hayward <alan.hayward@arm.com>
3875 David Sherwood <david.sherwood@arm.com>
3877 * combine.c (can_change_dest_mode): Handle polynomial
3878 REGMODE_NATURAL_SIZE.
3879 * expmed.c (store_bit_field_1): Likewise.
3880 * expr.c (store_constructor): Likewise.
3881 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
3882 and polynomial REGMODE_NATURAL_SIZE.
3883 (gen_lowpart_common): Likewise.
3884 * reginfo.c (record_subregs_of_mode): Likewise.
3885 * rtlanal.c (read_modify_subreg_p): Likewise.
3887 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3888 Alan Hayward <alan.hayward@arm.com>
3889 David Sherwood <david.sherwood@arm.com>
3891 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
3892 numbers of elements.
3894 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3895 Alan Hayward <alan.hayward@arm.com>
3896 David Sherwood <david.sherwood@arm.com>
3898 * match.pd: Cope with polynomial numbers of vector elements.
3900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3901 Alan Hayward <alan.hayward@arm.com>
3902 David Sherwood <david.sherwood@arm.com>
3904 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
3905 in a POINTER_PLUS_EXPR.
3907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3908 Alan Hayward <alan.hayward@arm.com>
3909 David Sherwood <david.sherwood@arm.com>
3911 * omp-simd-clone.c (simd_clone_subparts): New function.
3912 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
3913 (ipa_simd_modify_function_body): Likewise.
3915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3916 Alan Hayward <alan.hayward@arm.com>
3917 David Sherwood <david.sherwood@arm.com>
3919 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
3920 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
3921 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
3922 (expand_vector_condition, vector_element): Likewise.
3923 (subparts_gt): New function.
3924 (get_compute_type): Use subparts_gt.
3925 (count_type_subparts): Delete.
3926 (expand_vector_operations_1): Use subparts_gt instead of
3927 count_type_subparts.
3929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3930 Alan Hayward <alan.hayward@arm.com>
3931 David Sherwood <david.sherwood@arm.com>
3933 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
3934 (vect_compile_time_alias): ...this new function. Do the calculation
3935 on poly_ints rather than trees.
3936 (vect_prune_runtime_alias_test_list): Update call accordingly.
3938 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3939 Alan Hayward <alan.hayward@arm.com>
3940 David Sherwood <david.sherwood@arm.com>
3942 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
3944 (vect_schedule_slp_instance): Likewise.
3946 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3947 Alan Hayward <alan.hayward@arm.com>
3948 David Sherwood <david.sherwood@arm.com>
3950 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
3951 constant and extern definitions for variable-length vectors.
3952 (vect_get_constant_vectors): Note that the number of units
3953 is known to be constant.
3955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3956 Alan Hayward <alan.hayward@arm.com>
3957 David Sherwood <david.sherwood@arm.com>
3959 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
3960 of units as polynomial. Choose between WIDE and NARROW based
3963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3964 Alan Hayward <alan.hayward@arm.com>
3965 David Sherwood <david.sherwood@arm.com>
3967 * tree-vect-stmts.c (simd_clone_subparts): New function.
3968 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
3970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3971 Alan Hayward <alan.hayward@arm.com>
3972 David Sherwood <david.sherwood@arm.com>
3974 * tree-vect-stmts.c (vectorizable_call): Treat the number of
3975 vectors as polynomial. Use build_index_vector for
3978 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3979 Alan Hayward <alan.hayward@arm.com>
3980 David Sherwood <david.sherwood@arm.com>
3982 * tree-vect-stmts.c (get_load_store_type): Treat the number of
3983 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
3984 for variable-length vectors.
3985 (vectorizable_mask_load_store): Treat the number of units as
3986 polynomial, asserting that it is constant if the condition has
3987 already been enforced.
3988 (vectorizable_store, vectorizable_load): Likewise.
3990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3991 Alan Hayward <alan.hayward@arm.com>
3992 David Sherwood <david.sherwood@arm.com>
3994 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
3995 of units as polynomial. Punt if we can't tell at compile time
3996 which vector contains the final result.
3998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3999 Alan Hayward <alan.hayward@arm.com>
4000 David Sherwood <david.sherwood@arm.com>
4002 * tree-vect-loop.c (vectorizable_induction): Treat the number
4003 of units as polynomial. Punt on SLP inductions. Use an integer
4004 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4005 cast of such a series for variable-length floating-point
4008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4009 Alan Hayward <alan.hayward@arm.com>
4010 David Sherwood <david.sherwood@arm.com>
4012 * tree.h (build_index_vector): Declare.
4013 * tree.c (build_index_vector): New function.
4014 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4015 of units as polynomial, forcibly converting it to a constant if
4016 vectorizable_reduction has already enforced the condition.
4017 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4018 to create a {1,2,3,...} vector.
4019 (vectorizable_reduction): Treat the number of units as polynomial.
4020 Choose vectype_in based on the largest scalar element size rather
4021 than the smallest number of units. Enforce the restrictions
4024 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4025 Alan Hayward <alan.hayward@arm.com>
4026 David Sherwood <david.sherwood@arm.com>
4028 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4029 number of units as polynomial.
4031 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4032 Alan Hayward <alan.hayward@arm.com>
4033 David Sherwood <david.sherwood@arm.com>
4035 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4036 * target.def (autovectorize_vector_sizes): Return the vector sizes
4037 by pointer, using vector_sizes rather than a bitmask.
4038 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4039 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4040 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4042 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
4043 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
4044 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
4045 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
4046 * omp-general.c (omp_max_vf): Likewise.
4047 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4048 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
4049 * tree-vect-loop.c (vect_analyze_loop): Likewise.
4050 * tree-vect-slp.c (vect_slp_bb): Likewise.
4051 * doc/tm.texi: Regenerate.
4052 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
4054 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
4055 the vector size as a poly_uint64 rather than an unsigned int.
4056 (current_vector_size): Change from an unsigned int to a poly_uint64.
4057 (get_vectype_for_scalar_type): Update accordingly.
4058 * tree.h (build_truth_vector_type): Take the size and number of
4059 units as a poly_uint64 rather than an unsigned int.
4060 (build_vector_type): Add a temporary overload that takes
4061 the number of units as a poly_uint64 rather than an unsigned int.
4062 * tree.c (make_vector_type): Likewise.
4063 (build_truth_vector_type): Take the number of units as a poly_uint64
4064 rather than an unsigned int.
4066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4067 Alan Hayward <alan.hayward@arm.com>
4068 David Sherwood <david.sherwood@arm.com>
4070 * target.def (get_mask_mode): Take the number of units and length
4071 as poly_uint64s rather than unsigned ints.
4072 * targhooks.h (default_get_mask_mode): Update accordingly.
4073 * targhooks.c (default_get_mask_mode): Likewise.
4074 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
4075 * doc/tm.texi: Regenerate.
4077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4078 Alan Hayward <alan.hayward@arm.com>
4079 David Sherwood <david.sherwood@arm.com>
4081 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
4082 * omp-general.c (omp_max_vf): Likewise.
4083 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
4084 (expand_omp_simd): Handle polynomial safelen.
4085 * omp-low.c (omplow_simd_context): Add a default constructor.
4086 (omplow_simd_context::max_vf): Change from int to poly_uint64.
4087 (lower_rec_simd_input_clauses): Update accordingly.
4088 (lower_rec_input_clauses): Likewise.
4090 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4091 Alan Hayward <alan.hayward@arm.com>
4092 David Sherwood <david.sherwood@arm.com>
4094 * tree-vectorizer.h (vect_nunits_for_cost): New function.
4095 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
4096 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
4097 (vect_analyze_slp_cost): Likewise.
4098 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
4099 (vect_model_load_cost): Likewise.
4101 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4102 Alan Hayward <alan.hayward@arm.com>
4103 David Sherwood <david.sherwood@arm.com>
4105 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
4106 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
4107 from an unsigned int * to a poly_uint64_pod *.
4108 (calculate_unrolling_factor): New function.
4109 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
4111 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4112 Alan Hayward <alan.hayward@arm.com>
4113 David Sherwood <david.sherwood@arm.com>
4115 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
4116 from an unsigned int to a poly_uint64.
4117 (_loop_vec_info::slp_unrolling_factor): Likewise.
4118 (_loop_vec_info::vectorization_factor): Change from an int
4120 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
4121 (vect_get_num_vectors): New function.
4122 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
4123 (vect_get_num_copies): Use vect_get_num_vectors.
4124 (vect_analyze_data_ref_dependences): Change max_vf from an int *
4125 to an unsigned int *.
4126 (vect_analyze_data_refs): Change min_vf from an int * to a
4128 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
4129 than an unsigned HOST_WIDE_INT.
4130 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
4131 (vect_analyze_data_ref_dependence): Change max_vf from an int *
4132 to an unsigned int *.
4133 (vect_analyze_data_ref_dependences): Likewise.
4134 (vect_compute_data_ref_alignment): Handle polynomial vf.
4135 (vect_enhance_data_refs_alignment): Likewise.
4136 (vect_prune_runtime_alias_test_list): Likewise.
4137 (vect_shift_permute_load_chain): Likewise.
4138 (vect_supportable_dr_alignment): Likewise.
4139 (dependence_distance_ge_vf): Take the vectorization factor as a
4140 poly_uint64 rather than an unsigned HOST_WIDE_INT.
4141 (vect_analyze_data_refs): Change min_vf from an int * to a
4143 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
4144 vfm1 as a poly_uint64 rather than an int. Make the same change
4145 for the returned bound_scalar.
4146 (vect_gen_vector_loop_niters): Handle polynomial vf.
4147 (vect_do_peeling): Likewise. Update call to
4148 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
4149 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
4151 * tree-vect-loop.c (vect_determine_vectorization_factor)
4152 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
4153 (vect_get_known_peeling_cost): Likewise.
4154 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
4155 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
4156 (vect_transform_loop): Likewise. Use the lowest possible VF when
4157 updating the upper bounds of the loop.
4158 (vect_min_worthwhile_factor): Make static. Return an unsigned int
4160 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
4161 polynomial unroll factors.
4162 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
4163 (vect_make_slp_decision): Likewise.
4164 (vect_supported_load_permutation_p): Likewise, and polynomial
4166 (vect_analyze_slp_cost): Handle polynomial vf.
4167 (vect_slp_analyze_node_operations): Likewise.
4168 (vect_slp_analyze_bb_1): Likewise.
4169 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
4170 than an unsigned HOST_WIDE_INT.
4171 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
4172 (vectorizable_load): Handle polynomial vf.
4173 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
4175 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
4177 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4178 Alan Hayward <alan.hayward@arm.com>
4179 David Sherwood <david.sherwood@arm.com>
4181 * match.pd: Handle bit operations involving three constants
4182 and try to fold one pair.
4184 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4186 * tree-vect-loop-manip.c: Include gimple-fold.h.
4187 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
4188 niters_maybe_zero parameters. Handle other cases besides a step of 1.
4189 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
4190 Add a path that uses a step of VF instead of 1, but disable it
4192 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
4193 and niters_no_overflow parameters. Update calls to
4194 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
4195 Create a new SSA name if the latter choses to use a ste other
4196 than zero, and return it via niters_vector_mult_vf_var.
4197 * tree-vect-loop.c (vect_transform_loop): Update calls to
4198 vect_do_peeling, vect_gen_vector_loop_niters and
4199 slpeel_make_loop_iterate_ntimes.
4200 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
4201 (vect_gen_vector_loop_niters): Update declarations after above changes.
4203 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
4205 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
4206 128-bit round to integer instructions.
4207 (ceil<mode>2): Likewise.
4208 (btrunc<mode>2): Likewise.
4209 (round<mode>2): Likewise.
4211 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4213 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
4214 unaligned VSX load/store on P8/P9.
4215 (expand_block_clear): Allow the use of unaligned VSX
4216 load/store on P8/P9.
4218 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4220 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
4222 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
4223 swap associated with both a load and a store.
4225 2018-01-02 Andrew Waterman <andrew@sifive.com>
4227 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
4228 * config/riscv/riscv.md (clear_cache): Use it.
4230 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
4232 * web.c: Remove out-of-date comment.
4234 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4236 * expr.c (fixup_args_size_notes): Check that any existing
4237 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
4238 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
4239 (emit_single_push_insn): ...here.
4241 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4243 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
4244 (const_vector_encoded_nelts): New function.
4245 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
4246 (const_vector_int_elt, const_vector_elt): Declare.
4247 * emit-rtl.c (const_vector_int_elt_1): New function.
4248 (const_vector_elt): Likewise.
4249 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
4250 of CONST_VECTOR_ELT.
4252 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4254 * expr.c: Include rtx-vector-builder.h.
4255 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
4256 directly on the tree encoding.
4257 (const_vector_from_tree): Likewise.
4258 * optabs.c: Include rtx-vector-builder.h.
4259 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
4260 sequence of "u" values.
4261 * vec-perm-indices.c: Include rtx-vector-builder.h.
4262 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
4263 directly on the vec_perm_indices encoding.
4265 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4267 * doc/rtl.texi (const_vector): Describe new encoding scheme.
4268 * Makefile.in (OBJS): Add rtx-vector-builder.o.
4269 * rtx-vector-builder.h: New file.
4270 * rtx-vector-builder.c: Likewise.
4271 * rtl.h (rtx_def::u2): Add a const_vector field.
4272 (CONST_VECTOR_NPATTERNS): New macro.
4273 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
4274 (CONST_VECTOR_DUPLICATE_P): Likewise.
4275 (CONST_VECTOR_STEPPED_P): Likewise.
4276 (CONST_VECTOR_ENCODED_ELT): Likewise.
4277 (const_vec_duplicate_p): Check for a duplicated vector encoding.
4278 (unwrap_const_vec_duplicate): Likewise.
4279 (const_vec_series_p): Check for a non-duplicated vector encoding.
4280 Say that the function only returns true for integer vectors.
4281 * emit-rtl.c: Include rtx-vector-builder.h.
4282 (gen_const_vec_duplicate_1): Delete.
4283 (gen_const_vector): Call gen_const_vec_duplicate instead of
4284 gen_const_vec_duplicate_1.
4285 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
4286 (gen_const_vec_duplicate): Use rtx_vector_builder.
4287 (gen_const_vec_series): Likewise.
4288 (gen_rtx_CONST_VECTOR): Likewise.
4289 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
4290 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
4291 Build a new vector rather than modifying a CONST_VECTOR in-place.
4292 (handle_special_swappables): Update call accordingly.
4293 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
4294 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
4295 Build a new vector rather than modifying a CONST_VECTOR in-place.
4296 (handle_special_swappables): Update call accordingly.
4298 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4300 * simplify-rtx.c (simplify_const_binary_operation): Use
4301 CONST_VECTOR_ELT instead of XVECEXP.
4303 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4305 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
4306 the selector elements to be different from the data elements
4307 if the selector is a VECTOR_CST.
4308 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
4309 ssizetype for the selector.
4311 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4313 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
4314 before testing each element individually.
4315 * tree-vect-generic.c (lower_vec_perm): Likewise.
4317 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4319 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
4320 * selftest-run-tests.c (selftest::run_tests): Call it.
4321 * vector-builder.h (vector_builder::operator ==): New function.
4322 (vector_builder::operator !=): Likewise.
4323 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
4324 (vec_perm_indices::all_from_input_p): New function.
4325 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4326 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
4327 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
4328 instead of reading the VECTOR_CST directly. Detect whether both
4329 vector inputs are the same before constructing the vec_perm_indices,
4330 and update the number of inputs argument accordingly. Use the
4331 utility functions added above. Only construct sel2 if we need to.
4333 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4335 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
4336 the broadcast of the low byte.
4337 (expand_mult_highpart): Use an explicit encoding for the permutes.
4338 * optabs-query.c (can_mult_highpart_p): Likewise.
4339 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
4340 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4341 (vectorizable_bswap): Likewise.
4342 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
4343 explicit encoding for the power-of-2 permutes.
4344 (vect_permute_store_chain): Likewise.
4345 (vect_grouped_load_supported): Likewise.
4346 (vect_permute_load_chain): Likewise.
4348 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4350 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
4351 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
4352 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
4353 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4354 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
4355 (vect_gen_perm_mask_any): Likewise.
4357 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4359 * int-vector-builder.h: New file.
4360 * vec-perm-indices.h: Include int-vector-builder.h.
4361 (vec_perm_indices): Redefine as an int_vector_builder.
4362 (auto_vec_perm_indices): Delete.
4363 (vec_perm_builder): Redefine as a stand-alone class.
4364 (vec_perm_indices::vec_perm_indices): New function.
4365 (vec_perm_indices::clamp): Likewise.
4366 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
4367 (vec_perm_indices::new_vector): New function.
4368 (vec_perm_indices::new_expanded_vector): Update for new
4369 vec_perm_indices class.
4370 (vec_perm_indices::rotate_inputs): New function.
4371 (vec_perm_indices::all_in_range_p): Operate directly on the
4372 encoded form, without computing elided elements.
4373 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
4374 encoding. Update for new vec_perm_indices class.
4375 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
4376 the given vec_perm_builder.
4377 (expand_vec_perm_var): Update vec_perm_builder constructor.
4378 (expand_mult_highpart): Use vec_perm_builder instead of
4379 auto_vec_perm_indices.
4380 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
4381 vec_perm_indices instead of auto_vec_perm_indices. Use a single
4382 or double series encoding as appropriate.
4383 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
4384 vec_perm_indices instead of auto_vec_perm_indices.
4385 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4386 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4387 (vect_permute_store_chain): Likewise.
4388 (vect_grouped_load_supported): Likewise.
4389 (vect_permute_load_chain): Likewise.
4390 (vect_shift_permute_load_chain): Likewise.
4391 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
4392 (vect_transform_slp_perm_load): Likewise.
4393 (vect_schedule_slp_instance): Likewise.
4394 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4395 (vectorizable_mask_load_store): Likewise.
4396 (vectorizable_bswap): Likewise.
4397 (vectorizable_store): Likewise.
4398 (vectorizable_load): Likewise.
4399 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
4400 vec_perm_indices instead of auto_vec_perm_indices. Use
4401 tree_to_vec_perm_builder to read the vector from a tree.
4402 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
4403 vec_perm_builder instead of a vec_perm_indices.
4404 (have_whole_vector_shift): Use vec_perm_builder and
4405 vec_perm_indices instead of auto_vec_perm_indices. Leave the
4406 truncation to calc_vec_perm_mask_for_shift.
4407 (vect_create_epilog_for_reduction): Likewise.
4408 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
4409 from auto_vec_perm_indices to vec_perm_indices.
4410 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
4411 instead of changing individual elements.
4412 (aarch64_vectorize_vec_perm_const): Use new_vector to install
4413 the vector in d.perm.
4414 * config/arm/arm.c (expand_vec_perm_d::perm): Change
4415 from auto_vec_perm_indices to vec_perm_indices.
4416 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
4417 instead of changing individual elements.
4418 (arm_vectorize_vec_perm_const): Use new_vector to install
4419 the vector in d.perm.
4420 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
4421 Update vec_perm_builder constructor.
4422 (rs6000_expand_interleave): Likewise.
4423 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
4424 (rs6000_expand_interleave): Likewise.
4426 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4428 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
4429 to qimode could truncate the indices.
4430 * optabs.c (expand_vec_perm_var): Likewise.
4432 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4434 * Makefile.in (OBJS): Add vec-perm-indices.o.
4435 * vec-perm-indices.h: New file.
4436 * vec-perm-indices.c: Likewise.
4437 * target.h (vec_perm_indices): Replace with a forward class
4439 (auto_vec_perm_indices): Move to vec-perm-indices.h.
4440 * optabs.h: Include vec-perm-indices.h.
4441 (expand_vec_perm): Delete.
4442 (selector_fits_mode_p, expand_vec_perm_var): Declare.
4443 (expand_vec_perm_const): Declare.
4444 * target.def (vec_perm_const_ok): Replace with...
4445 (vec_perm_const): ...this new hook.
4446 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
4447 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
4448 * doc/tm.texi: Regenerate.
4449 * optabs.def (vec_perm_const): Delete.
4450 * doc/md.texi (vec_perm_const): Likewise.
4451 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
4452 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
4453 expand_vec_perm for constant permutation vectors. Assert that
4454 the mode of variable permutation vectors is the integer equivalent
4455 of the mode that is being permuted.
4456 * optabs-query.h (selector_fits_mode_p): Declare.
4457 * optabs-query.c: Include vec-perm-indices.h.
4458 (selector_fits_mode_p): New function.
4459 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
4460 is defined, instead of checking whether the vec_perm_const_optab
4461 exists. Use targetm.vectorize.vec_perm_const instead of
4462 targetm.vectorize.vec_perm_const_ok. Check whether the indices
4463 fit in the vector mode before using a variable permute.
4464 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
4465 vec_perm_indices instead of an rtx.
4466 (expand_vec_perm): Replace with...
4467 (expand_vec_perm_const): ...this new function. Take the selector
4468 as a vec_perm_indices rather than an rtx. Also take the mode of
4469 the selector. Update call to shift_amt_for_vec_perm_mask.
4470 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
4471 Use vec_perm_indices::new_expanded_vector to expand the original
4472 selector into bytes. Check whether the indices fit in the vector
4473 mode before using a variable permute.
4474 (expand_vec_perm_var): Make global.
4475 (expand_mult_highpart): Use expand_vec_perm_const.
4476 * fold-const.c: Includes vec-perm-indices.h.
4477 * tree-ssa-forwprop.c: Likewise.
4478 * tree-vect-data-refs.c: Likewise.
4479 * tree-vect-generic.c: Likewise.
4480 * tree-vect-loop.c: Likewise.
4481 * tree-vect-slp.c: Likewise.
4482 * tree-vect-stmts.c: Likewise.
4483 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
4485 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
4486 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
4487 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
4488 (aarch64_vectorize_vec_perm_const): ...this new function.
4489 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
4490 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
4491 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
4492 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
4493 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
4494 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
4495 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
4497 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
4498 check for NEON modes.
4499 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
4500 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
4501 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
4502 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
4504 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
4505 the old VEC_PERM_CONST conditions.
4506 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
4507 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
4508 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
4509 (ia64_vectorize_vec_perm_const_ok): Merge into...
4510 (ia64_vectorize_vec_perm_const): ...this new function.
4511 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
4512 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
4513 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
4514 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
4515 * config/mips/mips.c (mips_expand_vec_perm_const)
4516 (mips_vectorize_vec_perm_const_ok): Merge into...
4517 (mips_vectorize_vec_perm_const): ...this new function.
4518 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
4519 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
4520 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
4521 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
4522 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
4523 (rs6000_expand_vec_perm_const): Delete.
4524 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
4526 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
4527 (altivec_expand_vec_perm_const_le): Take each operand individually.
4528 Operate on constant selectors rather than rtxes.
4529 (altivec_expand_vec_perm_const): Likewise. Update call to
4530 altivec_expand_vec_perm_const_le.
4531 (rs6000_expand_vec_perm_const): Delete.
4532 (rs6000_vectorize_vec_perm_const_ok): Delete.
4533 (rs6000_vectorize_vec_perm_const): New function.
4534 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
4535 an element count and rtx array.
4536 (rs6000_expand_extract_even): Update call accordingly.
4537 (rs6000_expand_interleave): Likewise.
4538 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
4539 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
4540 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
4541 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
4542 (rs6000_expand_vec_perm_const): Delete.
4543 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
4544 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
4545 (altivec_expand_vec_perm_const_le): Take each operand individually.
4546 Operate on constant selectors rather than rtxes.
4547 (altivec_expand_vec_perm_const): Likewise. Update call to
4548 altivec_expand_vec_perm_const_le.
4549 (rs6000_expand_vec_perm_const): Delete.
4550 (rs6000_vectorize_vec_perm_const_ok): Delete.
4551 (rs6000_vectorize_vec_perm_const): New function. Remove stray
4552 reference to the SPE evmerge intructions.
4553 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
4554 an element count and rtx array.
4555 (rs6000_expand_extract_even): Update call accordingly.
4556 (rs6000_expand_interleave): Likewise.
4557 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
4558 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
4560 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
4562 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4564 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
4565 vector mode and that that mode matches the mode of the data
4567 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
4568 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
4569 directly using expand_vec_perm_1 when forcing selectors into
4571 (expand_vec_perm_var): New function, split out from expand_vec_perm.
4573 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4575 * optabs-query.h (can_vec_perm_p): Delete.
4576 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
4577 * optabs-query.c (can_vec_perm_p): Split into...
4578 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
4579 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
4580 particular selector is valid.
4581 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4582 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4583 (vect_grouped_load_supported): Likewise.
4584 (vect_shift_permute_load_chain): Likewise.
4585 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
4586 (vect_transform_slp_perm_load): Likewise.
4587 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4588 (vectorizable_bswap): Likewise.
4589 (vect_gen_perm_mask_checked): Likewise.
4590 * fold-const.c (fold_ternary_loc): Likewise. Don't take
4591 implementations of variable permutation vectors into account
4592 when deciding which selector to use.
4593 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
4594 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
4595 with a false third argument.
4596 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
4597 to test whether the constant selector is valid and can_vec_perm_var_p
4598 to test whether a variable selector is valid.
4600 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4602 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
4603 * optabs-query.c (can_vec_perm_p): Likewise.
4604 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
4605 instead of vec_perm_indices.
4606 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
4607 (vect_gen_perm_mask_checked): Likewise,
4608 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
4609 (vect_gen_perm_mask_checked): Likewise,
4611 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4613 * optabs-query.h (qimode_for_vec_perm): Declare.
4614 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
4615 (qimode_for_vec_perm): ...this new function.
4616 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
4618 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4620 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
4621 does not have a conditional at the top.
4623 2018-01-02 Richard Biener <rguenther@suse.de>
4625 * ipa-inline.c (big_speedup_p): Fix expression.
4627 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
4630 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
4633 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
4637 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
4638 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
4639 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
4640 cond_taken_branch_cost 3->4.
4642 2018-01-01 Jakub Jelinek <jakub@redhat.com>
4644 PR tree-optimization/83581
4645 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
4646 TODO_cleanup_cfg if any changes have been made.
4649 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
4650 convert_modes if target mode has the right side, but different mode
4654 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
4655 last argument when extracting from CONCAT. If either from_real or
4656 from_imag is NULL, use expansion through memory. If result is not
4657 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
4658 the parts directly to inner mode, if even that fails, use expansion
4662 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
4663 check for bswap in mode rather than HImode and use that in expand_unop
4666 Copyright (C) 2018 Free Software Foundation, Inc.
4668 Copying and distribution of this file, with or without modification,
4669 are permitted in any medium without royalty provided the copyright
4670 notice and this notice are preserved.