1 2018-01-23 Bin Cheng <bin.cheng@arm.com>
3 PR tree-optimization/82604
4 * tree-loop-distribution.c (enum partition_kind): New enum item
6 (partition_builtin_p): Support above new enum item.
7 (generate_code_for_partition): Ditto.
8 (compute_access_range): Differentiate cases that equality can be
9 proven at all loops, the innermost loops or no loops.
10 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
11 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
12 (finalize_partitions, distribute_loop): Don't fuse partition of
13 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
14 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
17 2018-01-23 Martin Liska <mliska@suse.cz>
19 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
20 order to ignore the predictor.
21 (PRED_POLYMORPHIC_CALL): Likewise.
22 (PRED_RECURSIVE_CALL): Likewise.
24 2018-01-23 Martin Liska <mliska@suse.cz>
26 * tree-profile.c (tree_profiling): Print function header to
27 aware reader which function we are working on.
28 * value-prof.c (gimple_find_values_to_profile): Do not print
29 not interesting value histograms.
31 2018-01-23 Martin Liska <mliska@suse.cz>
33 * profile-count.h (enum profile_quality): Add
34 profile_uninitialized as the first value. Do not number values
35 as they are zero based.
36 (profile_count::verify): Update sanity check.
37 (profile_probability::verify): Likewise.
39 2018-01-23 Nathan Sidwell <nathan@acm.org>
41 * doc/invoke.texi (ffor-scope): Deprecate.
43 2018-01-23 David Malcolm <dmalcolm@redhat.com>
45 PR tree-optimization/83510
46 * domwalk.c (set_all_edges_as_executable): New function.
47 (dom_walker::dom_walker): Convert bool param
48 "skip_unreachable_blocks" to enum reachability. Move setup of
49 edge flags to set_all_edges_as_executable and only do it when
50 reachability is REACHABLE_BLOCKS.
51 * domwalk.h (enum dom_walker::reachability): New enum.
52 (dom_walker::dom_walker): Convert bool param
53 "skip_unreachable_blocks" to enum reachability.
54 (set_all_edges_as_executable): New decl.
55 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
56 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
58 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
59 but converting true to REACHABLE_BLOCKS.
60 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
62 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
63 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
64 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
66 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
67 if check_all_array_refs will be called.
69 2018-01-23 David Malcolm <dmalcolm@redhat.com>
71 * tree.c (selftest::test_location_wrappers): Add more test
74 2018-01-23 David Malcolm <dmalcolm@redhat.com>
76 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
77 (selftest::test_bit_in_range): Likewise.
79 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
82 * doc/sourcebuild.texi (vect_float): Say that the selector
83 only describes the situation when -funsafe-math-optimizations is on.
84 (vect_float_strict): Document.
86 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
88 PR tree-optimization/83965
89 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
90 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
91 instead of checking only for a reduction.
92 (vect_recog_widen_sum_pattern): Likewise.
94 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
96 * predict.c (probably_never_executed): Only use precise profile info.
97 (compute_function_frequency): Skip after inlining hack since we now
98 have quality checking.
100 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
102 * profile-count.h (profile_probability::very_unlikely,
103 profile_probability::unlikely, profile_probability::even): Set
104 precision to guessed.
106 2018-01-23 Richard Biener <rguenther@suse.de>
108 PR tree-optimization/83963
109 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
110 Properly terminate dominator walk when crossing the exit edge not
111 when visiting its source block.
113 2018-01-23 Jakub Jelinek <jakub@redhat.com>
116 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
117 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
119 2018-01-22 Jakub Jelinek <jakub@redhat.com>
121 PR tree-optimization/83957
122 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
123 semicolon after for body surrounded by braces.
125 PR tree-optimization/83081
126 * profile-count.h (profile_probability::split): New method.
127 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
128 Use profile_probability::split.
129 (do_compare_rtx_and_jump): Fix adjustment of probabilities
130 when splitting a single conditional jump into 2.
132 2018-01-22 David Malcolm <dmalcolm@redhat.com>
134 PR tree-optimization/69452
135 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
138 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
140 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
141 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
142 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
144 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
146 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
147 * config/rl78/rl78.md: New define_expand "movdi"
148 * config/rl78/rl78.c: New function definition rl78_split_movdi
150 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
153 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
155 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
156 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
157 128-bit to produce an UNSPEC move to get the double word with the
158 signbit and then a shift directly to do signbit.
159 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
160 implementation with a new version that just does either a direct
161 move or a regular move. Move memory interface to separate insns.
162 Move insns so they are next to the expander.
163 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
164 with signbit move. Split big and little endian case.
165 (signbit<mode>2_dm_mem_le): Likewise.
166 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
167 (signbit<mode>2_dm2): Likewise.
169 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
171 * config/rl78/rl78.md: New define_expand "anddi3".
173 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
175 * config/rl78/rl78.md: New define_expand "umindi3".
177 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
179 * config/rl78/rl78.md: New define_expand "smindi3".
181 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
183 * config/rl78/rl78.md: New define_expand "smaxdi3".
185 2018-01-22 Carl Love <cel@us.ibm.com>
187 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
188 LVX_V1TI): Add macro expansion.
189 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
190 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
191 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
192 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
193 Change check to determine if the instruction is a byte reversing
194 entry. Fix typo in comment.
195 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
196 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
197 Add def_builtin calls for new builtins.
198 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
199 Add define_insn expansion.
201 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
203 * config/rl78/rl78.md: New define_expand "umaxdi3".
205 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
207 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
208 for non-QImode registers
210 2018-01-22 Richard Biener <rguenther@suse.de>
212 PR tree-optimization/83963
213 * graphite-scop-detection.c (scop_detection::get_sese): Delay
214 including the loop exit block.
215 (scop_detection::merge_sese): Likewise.
216 (scop_detection::add_scop): Do it here instead.
218 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
220 * doc/sourcebuild.texi (arm_softfloat): Document.
222 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
225 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
226 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
227 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
229 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
230 David Edelsohn <dje.gcc@gmail.com>
233 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
234 Change "crset eq" to "crset 2".
235 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
236 (*call_indirect_aix<mode>_nospec): Likewise.
237 (*call_value_indirect_aix<mode>_nospec): Likewise.
238 (*call_indirect_elfv2<mode>_nospec): Likewise.
239 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
240 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
241 change assembly output from . to $.
242 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
243 (indirect_jump<mode>_nospec): Change assembly output from . to $.
244 (*tablejump<mode>_internal1_nospec): Likewise.
246 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
249 * config/sh/sh_optimize_sett_clrt.cc:
250 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
252 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
254 PR tree-optimization/83940
255 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
256 offset_dt to vect_constant_def rather than vect_unknown_def_type.
257 (vect_check_load_store_mask): Add a mask_dt_out parameter and
258 use it to pass back the definition type.
259 (vect_check_store_rhs): Likewise rhs_dt_out.
260 (vect_build_gather_load_calls): Add a mask_dt argument and use
261 it instead of a call to vect_is_simple_use.
262 (vectorizable_store): Update calls to vect_check_load_store_mask
263 and vect_check_store_rhs. Use the dt returned by the latter instead
264 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
265 instead of calls to vect_is_simple_use. Pass the scalar rather
266 than the vector operand to vect_is_simple_use when handling
267 second and subsequent copies of an rhs value.
268 (vectorizable_load): Update calls to vect_check_load_store_mask
269 and vect_build_gather_load_calls. Use the cached mask_dt and
270 gs_info.offset_dt instead of calls to vect_is_simple_use.
272 2018-01-20 Jakub Jelinek <jakub@redhat.com>
275 * tree-emutls.c: Include gimplify.h.
276 (lower_emutls_2): New function.
277 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
278 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
279 it before further processing.
282 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
283 UINTVAL (trueop1) instead of INTVAL (op1).
285 2018-01-19 Jakub Jelinek <jakub@redhat.com>
289 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
290 INCOMING_FRAME_SP_OFFSET if not defined.
291 (scan_trace): Add ENTRY argument. If true and
292 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
293 emit a note to adjust the CFA offset.
294 (create_cfi_notes): Adjust scan_trace callers.
295 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
296 INCOMING_FRAME_SP_OFFSET in the CIE.
297 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
298 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
300 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
301 * doc/tm.texi: Regenerated.
303 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
305 PR rtl-optimization/83147
306 * lra-constraints.c (remove_inheritance_pseudos): Use
307 lra_substitute_pseudo_within_insn.
309 2018-01-19 Tom de Vries <tom@codesourcery.com>
310 Cesar Philippidis <cesar@codesourcery.com>
313 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
315 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
318 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
319 spaces for function labels.
321 2018-01-19 Martin Liska <mliska@suse.cz>
323 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
324 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
325 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
326 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
327 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
328 (PRED_CONST_RETURN): Change from 69 to 65.
329 (PRED_NULL_RETURN): Change from 91 to 71.
330 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
331 (PRED_LOOP_GUARD): Change from 66 to 73.
333 2018-01-19 Martin Liska <mliska@suse.cz>
335 * predict.c (predict_insn_def): Add new assert.
336 (struct branch_predictor): Change type to signed integer.
337 (test_prediction_value_range): Amend test to cover
339 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
340 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
341 (PRED_LOOP_ITERATIONS_MAX): Likewise.
342 (PRED_LOOP_IV_COMPARE): Likewise.
343 * predict.h (PROB_UNINITIALIZED): Define new constant.
345 2018-01-19 Martin Liska <mliska@suse.cz>
347 * predict.c (dump_prediction): Add new format for
348 analyze_brprob.py script which is enabled with -details
350 * profile-count.h (precise_p): New function.
352 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
354 PR tree-optimization/83922
355 * tree-vect-loop.c (vect_verify_full_masking): Return false if
356 there are no statements that need masking.
357 (vect_active_double_reduction_p): New function.
358 (vect_analyze_loop_operations): Use it when handling phis that
359 are not in the loop header.
361 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
363 PR tree-optimization/83914
364 * tree-vect-loop.c (vectorizable_induction): Don't convert
365 init_expr or apply the peeling adjustment for inductions
366 that are nested within the vectorized loop.
368 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
370 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
373 2018-01-18 Jakub Jelinek <jakub@redhat.com>
377 * function.h (gimplify_parameters): Add gimple_seq * argument.
378 * function.c: Include gimple.h and options.h.
379 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
380 for the added local temporaries if needed.
381 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
382 if there are any parameter cleanups, wrap whole body into a
383 try/finally with the cleanups.
385 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
388 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
389 Use GET_MODE_CLASS for scalar floating point.
391 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
395 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
396 Fix call of call_cgraph_insertion_hooks.
398 2018-01-18 Martin Sebor <msebor@redhat.com>
400 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
402 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
405 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
408 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
411 * common.opt: (-ffile-prefix-map): New option.
412 * opts.c (common_handle_option): Defer it.
413 * opts-global.c (handle_common_deferred_options): Handle it.
414 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
415 * file-prefix-map.h: New file.
416 (remap_debug_filename, add_debug_prefix_map): ...here.
417 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
418 * final.c (debug_prefix_map, add_debug_prefix_map
419 remap_debug_filename): Move to...
420 * file-prefix-map.c: New file.
421 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
422 generalize, get rid of alloca(), use strrchr() instead of strchr().
423 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
424 Implement in terms of add_prefix_map().
425 (remap_macro_filename, remap_debug_filename): Implement in term of
427 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
428 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
429 * dbxout.c: Include file-prefix-map.h.
430 * varasm.c: Likewise.
431 * vmsdbgout.c: Likewise.
432 * xcoffout.c: Likewise.
433 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
434 * doc/cppopts.texi (-fmacro-prefix-map): Document.
435 * doc/invoke.texi (-ffile-prefix-map): Document.
436 (-fdebug-prefix-map): Update description.
438 2018-01-18 Martin Liska <mliska@suse.cz>
440 * config/i386/i386.c (indirect_thunk_name): Document that also
442 (output_indirect_thunk): Document why both instructions
443 (pause and lfence) are generated.
445 2018-01-18 Richard Biener <rguenther@suse.de>
447 PR tree-optimization/83887
448 * graphite-scop-detection.c
449 (scop_detection::get_nearest_dom_with_single_entry): Remove.
450 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
451 (scop_detection::merge_sese): Re-implement with a flood-fill
452 algorithm that properly finds a SESE region if it exists.
454 2018-01-18 Jakub Jelinek <jakub@redhat.com>
457 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
458 pointer_diff optimizations use view_convert instead of convert.
460 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
462 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
463 Generate different code for -mno-speculate-indirect-jumps.
464 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
465 (*call_indirect_aix<mode>): Disable for
466 -mno-speculate-indirect-jumps.
467 (*call_indirect_aix<mode>_nospec): New define_insn.
468 (*call_value_indirect_aix<mode>): Disable for
469 -mno-speculate-indirect-jumps.
470 (*call_value_indirect_aix<mode>_nospec): New define_insn.
471 (*sibcall_nonlocal_sysv<mode>): Generate different code for
472 -mno-speculate-indirect-jumps.
473 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
475 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
477 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
478 long double type, set the flags for noting the default long double
479 type, even if we don't pass or return a long double type.
481 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
484 * ipa-inline.c (flatten_function): Do not overwrite final inlining
487 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
489 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
490 support for merge[hl].
491 (fold_mergehl_helper): New helper function.
492 (tree-vector-builder.h): New #include for tree_vector_builder usage.
493 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
494 (altivec_vmrglw_direct): Add xxmrglw insn.
496 2018-01-17 Andrew Waterman <andrew@sifive.com>
498 * config/riscv/riscv.c (riscv_conditional_register_usage): If
499 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
501 2018-01-17 David Malcolm <dmalcolm@redhat.com>
504 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
505 call the lto_location_cache before reading the
506 DECL_SOURCE_LOCATION of the types.
508 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
509 Richard Sandiford <richard.sandiford@linaro.org>
511 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
512 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
513 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
514 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
515 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
517 * config/aarch64/constraints.md (aarch64_movti_operand):
519 * config/aarch64/predicates.md (Uti): Add new constraint.
521 2018-01-17 Carl Love <cel@us.ibm.com>
522 * config/rs6000/vsx.md (define_expand xl_len_r,
523 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
524 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
526 (define_expand, define_insn): Move the shift left from the
527 define_insn to the define_expand for lxvl and stxvl instructions.
528 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
529 and XL_LEN_R definitions to PURE.
531 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
533 * config/i386/i386.c (indirect_thunk_name): Declare regno
534 as unsigned int. Compare regno with INVALID_REGNUM.
535 (output_indirect_thunk): Ditto.
536 (output_indirect_thunk_function): Ditto.
537 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
538 in the call to output_indirect_thunk_function.
540 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
543 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
544 rather than the size of inner_type to determine the stack slot size
545 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
547 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
550 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
553 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
555 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
556 endian Linux systems to optionally enable multilibs for selecting
557 the long double type if the user configured an explicit type.
558 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
559 have no long double multilibs if not defined.
560 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
561 warn if the user used -mabi={ieee,ibm}longdouble and we built
562 multilibs for long double.
563 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
564 appropriate multilib option.
565 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
567 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
568 for building long double multilibs.
569 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
571 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
573 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
576 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
578 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
581 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
584 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
587 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
589 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
590 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
592 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
594 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
595 different rtl trees depending on TARGET_64BIT.
596 (rs6000_gen_lvx): Likewise.
598 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
600 * config/visium/visium.md (nop): Tweak comment.
601 (hazard_nop): Likewise.
603 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
605 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
606 -mspeculate-indirect-jumps.
607 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
608 for -mno-speculate-indirect-jumps.
609 (*call_indirect_elfv2<mode>_nospec): New define_insn.
610 (*call_value_indirect_elfv2<mode>): Disable for
611 -mno-speculate-indirect-jumps.
612 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
613 (indirect_jump): Emit different RTL for
614 -mno-speculate-indirect-jumps.
615 (*indirect_jump<mode>): Disable for
616 -mno-speculate-indirect-jumps.
617 (*indirect_jump<mode>_nospec): New define_insn.
618 (tablejump): Emit different RTL for
619 -mno-speculate-indirect-jumps.
620 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
621 (tablejumpsi_nospec): New define_expand.
622 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
623 (tablejumpdi_nospec): New define_expand.
624 (*tablejump<mode>_internal1): Disable for
625 -mno-speculate-indirect-jumps.
626 (*tablejump<mode>_internal1_nospec): New define_insn.
627 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
630 2018-01-16 Artyom Skrobov tyomitch@gmail.com
632 * caller-save.c (insert_save): Drop unnecessary parameter. All
635 2018-01-16 Jakub Jelinek <jakub@redhat.com>
636 Richard Biener <rguenth@suse.de>
639 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
640 return early, inline manually is_gimple_sizepos. Make sure if we
641 call gimplify_expr we don't end up with a gimple constant.
642 * tree.c (variably_modified_type_p): Don't return true for
643 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
644 * gimplify.h (is_gimple_sizepos): Remove.
646 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
648 PR tree-optimization/83857
649 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
650 vectorizable_live_operation for pure SLP statements.
651 (vectorizable_live_operation): Handle PHIs.
653 2018-01-16 Richard Biener <rguenther@suse.de>
655 PR tree-optimization/83867
656 * tree-vect-stmts.c (vect_transform_stmt): Precompute
657 nested_in_vect_loop_p since the scalar stmt may get invalidated.
659 2018-01-16 Jakub Jelinek <jakub@redhat.com>
662 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
663 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
664 If off is not INTEGER_CST, issue a may not be aligned warning
665 rather than isn't aligned. Use isn%'t rather than isn't.
666 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
668 <case MULT_EXPR>: Improve the case when bottom and one of the
669 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
670 operand, in that case check if the other operand is multiple of
671 bottom divided by the INTEGER_CST operand.
673 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
676 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
677 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
678 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
679 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
680 * config/pa/pa.c (pa_function_arg_advance): Likewise.
681 (pa_function_arg, pa_arg_partial_bytes): Likewise.
682 (pa_function_arg_size): New function.
684 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
686 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
687 in a separate statement.
689 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
691 PR tree-optimization/83847
692 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
693 group gathers and scatters.
695 2018-01-16 Jakub Jelinek <jakub@redhat.com>
697 PR rtl-optimization/86620
698 * params.def (max-sched-ready-insns): Bump minimum value to 1.
700 PR rtl-optimization/83213
701 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
702 to last if both are JUMP_INSNs.
704 PR tree-optimization/83843
705 * gimple-ssa-store-merging.c
706 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
707 store_immediate_info for bswap/nop orig_stores.
709 2018-01-15 Andrew Waterman <andrew@sifive.com>
711 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
713 <UDIV>: Increase cost if !TARGET_DIV.
715 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
717 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
718 (define_attr "cr_logical_3op"): New.
719 (cceq_ior_compare): Adjust.
720 (cceq_ior_compare_complement): Adjust.
721 (*cceq_rev_compare): Adjust.
722 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
723 (is_cracked_insn): Adjust.
724 (insn_must_be_first_in_group): Adjust.
725 * config/rs6000/40x.md: Adjust.
726 * config/rs6000/440.md: Adjust.
727 * config/rs6000/476.md: Adjust.
728 * config/rs6000/601.md: Adjust.
729 * config/rs6000/603.md: Adjust.
730 * config/rs6000/6xx.md: Adjust.
731 * config/rs6000/7450.md: Adjust.
732 * config/rs6000/7xx.md: Adjust.
733 * config/rs6000/8540.md: Adjust.
734 * config/rs6000/cell.md: Adjust.
735 * config/rs6000/e300c2c3.md: Adjust.
736 * config/rs6000/e500mc.md: Adjust.
737 * config/rs6000/e500mc64.md: Adjust.
738 * config/rs6000/e5500.md: Adjust.
739 * config/rs6000/e6500.md: Adjust.
740 * config/rs6000/mpc.md: Adjust.
741 * config/rs6000/power4.md: Adjust.
742 * config/rs6000/power5.md: Adjust.
743 * config/rs6000/power6.md: Adjust.
744 * config/rs6000/power7.md: Adjust.
745 * config/rs6000/power8.md: Adjust.
746 * config/rs6000/power9.md: Adjust.
747 * config/rs6000/rs64.md: Adjust.
748 * config/rs6000/titan.md: Adjust.
750 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
752 * config/i386/predicates.md (indirect_branch_operand): Rewrite
753 ix86_indirect_branch_register logic.
755 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
757 * config/i386/constraints.md (Bs): Update
758 ix86_indirect_branch_register check. Don't check
759 ix86_indirect_branch_register with GOT_memory_operand.
761 * config/i386/predicates.md (GOT_memory_operand): Don't check
762 ix86_indirect_branch_register here.
763 (GOT32_symbol_operand): Likewise.
765 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
767 * config/i386/predicates.md (constant_call_address_operand):
768 Rewrite ix86_indirect_branch_register logic.
769 (sibcall_insn_operand): Likewise.
771 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
773 * config/i386/constraints.md (Bs): Replace
774 ix86_indirect_branch_thunk_register with
775 ix86_indirect_branch_register.
777 * config/i386/i386.md (indirect_jump): Likewise.
778 (tablejump): Likewise.
779 (*sibcall_memory): Likewise.
780 (*sibcall_value_memory): Likewise.
781 Peepholes of indirect call and jump via memory: Likewise.
782 * config/i386/i386.opt: Likewise.
783 * config/i386/predicates.md (indirect_branch_operand): Likewise.
784 (GOT_memory_operand): Likewise.
785 (call_insn_operand): Likewise.
786 (sibcall_insn_operand): Likewise.
787 (GOT32_symbol_operand): Likewise.
789 2018-01-15 Jakub Jelinek <jakub@redhat.com>
792 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
793 type rather than type addr's type points to.
794 (expand_omp_atomic_mutex): Likewise.
795 (expand_omp_atomic): Likewise.
797 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
800 * config/i386/i386.c (output_indirect_thunk_function): Use
801 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
802 for __x86_return_thunk.
804 2018-01-15 Richard Biener <rguenther@suse.de>
807 * expmed.c (extract_bit_field_1): Fix typo.
809 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
812 * config/arm/iterators.md (VF): New mode iterator.
813 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
814 Remove integer-related logic from pattern.
815 (neon_vabd<mode>_3): Likewise.
817 2018-01-15 Jakub Jelinek <jakub@redhat.com>
820 * common.opt (fstrict-overflow): No longer an alias.
821 (fwrapv-pointer): New option.
822 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
823 also for pointer types based on flag_wrapv_pointer.
824 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
825 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
826 opts->x_flag_wrapv got set.
827 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
828 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
829 POINTER_TYPE_OVERFLOW_UNDEFINED.
830 * match.pd: Likewise in address comparison pattern.
831 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
833 2018-01-15 Richard Biener <rguenther@suse.de>
836 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
837 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
838 Reset type names to their identifier if their TYPE_DECL doesn't
839 have linkage (and thus is used for ODR and devirt).
840 (save_debug_info_for_decl): Remove.
841 (save_debug_info_for_type): Likewise.
842 (add_tree_to_fld_list): Adjust.
843 * tree-pretty-print.c (dump_generic_node): Make dumping of
844 type names more robust.
846 2018-01-15 Richard Biener <rguenther@suse.de>
848 * BASE-VER: Bump to 8.0.1.
850 2018-01-14 Martin Sebor <msebor@redhat.com>
853 * builtins.c (check_access): Avoid warning when the no-warning bit
856 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
858 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
859 * ira-color (allocno_hard_regs_compare): Likewise.
861 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
864 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
865 Use .pushsection/.popsection.
867 2018-01-14 Martin Sebor <msebor@redhat.com>
870 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
872 2018-01-14 Jakub Jelinek <jakub@redhat.com>
874 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
875 entry from extra_headers.
876 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
877 extra_headers, make the list bitwise identical to the i?86-*-* one.
879 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
881 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
882 -mcmodel=large with -mindirect-branch=thunk,
883 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
884 -mfunction-return=thunk-extern.
885 * doc/invoke.texi: Document -mcmodel=large is incompatible with
886 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
887 -mfunction-return=thunk and -mfunction-return=thunk-extern.
889 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
891 * config/i386/i386.c (print_reg): Print the name of the full
892 integer register without '%'.
893 (ix86_print_operand): Handle 'V'.
894 * doc/extend.texi: Document 'V' modifier.
896 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
898 * config/i386/constraints.md (Bs): Disallow memory operand for
899 -mindirect-branch-register.
901 * config/i386/predicates.md (indirect_branch_operand): Likewise.
902 (GOT_memory_operand): Likewise.
903 (call_insn_operand): Likewise.
904 (sibcall_insn_operand): Likewise.
905 (GOT32_symbol_operand): Likewise.
906 * config/i386/i386.md (indirect_jump): Call convert_memory_address
907 for -mindirect-branch-register.
908 (tablejump): Likewise.
909 (*sibcall_memory): Likewise.
910 (*sibcall_value_memory): Likewise.
911 Disallow peepholes of indirect call and jump via memory for
912 -mindirect-branch-register.
913 (*call_pop): Replace m with Bw.
914 (*call_value_pop): Likewise.
915 (*sibcall_pop_memory): Replace m with Bs.
916 * config/i386/i386.opt (mindirect-branch-register): New option.
917 * doc/invoke.texi: Document -mindirect-branch-register option.
919 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
921 * config/i386/i386-protos.h (ix86_output_function_return): New.
922 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
923 set function_return_type.
924 (indirect_thunk_name): Add ret_p to indicate thunk for function
926 (output_indirect_thunk_function): Pass false to
928 (ix86_output_indirect_branch_via_reg): Likewise.
929 (ix86_output_indirect_branch_via_push): Likewise.
930 (output_indirect_thunk_function): Create alias for function
931 return thunk if regno < 0.
932 (ix86_output_function_return): New function.
933 (ix86_handle_fndecl_attribute): Handle function_return.
934 (ix86_attribute_table): Add function_return.
935 * config/i386/i386.h (machine_function): Add
936 function_return_type.
937 * config/i386/i386.md (simple_return_internal): Use
938 ix86_output_function_return.
939 (simple_return_internal_long): Likewise.
940 * config/i386/i386.opt (mfunction-return=): New option.
941 (indirect_branch): Mention -mfunction-return=.
942 * doc/extend.texi: Document function_return function attribute.
943 * doc/invoke.texi: Document -mfunction-return= option.
945 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
947 * config/i386/i386-opts.h (indirect_branch): New.
948 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
949 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
950 with local indirect jump when converting indirect call and jump.
951 (ix86_set_indirect_branch_type): New.
952 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
953 (indirectlabelno): New.
954 (indirect_thunk_needed): Likewise.
955 (indirect_thunk_bnd_needed): Likewise.
956 (indirect_thunks_used): Likewise.
957 (indirect_thunks_bnd_used): Likewise.
958 (INDIRECT_LABEL): Likewise.
959 (indirect_thunk_name): Likewise.
960 (output_indirect_thunk): Likewise.
961 (output_indirect_thunk_function): Likewise.
962 (ix86_output_indirect_branch_via_reg): Likewise.
963 (ix86_output_indirect_branch_via_push): Likewise.
964 (ix86_output_indirect_branch): Likewise.
965 (ix86_output_indirect_jmp): Likewise.
966 (ix86_code_end): Call output_indirect_thunk_function if needed.
967 (ix86_output_call_insn): Call ix86_output_indirect_branch if
969 (ix86_handle_fndecl_attribute): Handle indirect_branch.
970 (ix86_attribute_table): Add indirect_branch.
971 * config/i386/i386.h (machine_function): Add indirect_branch_type
972 and has_local_indirect_jump.
973 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
975 (tablejump): Likewise.
976 (*indirect_jump): Use ix86_output_indirect_jmp.
977 (*tablejump_1): Likewise.
978 (simple_return_indirect_internal): Likewise.
979 * config/i386/i386.opt (mindirect-branch=): New option.
980 (indirect_branch): New.
983 (thunk-inline): Likewise.
984 (thunk-extern): Likewise.
985 * doc/extend.texi: Document indirect_branch function attribute.
986 * doc/invoke.texi: Document -mindirect-branch= option.
988 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
991 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
993 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
995 * ipa-inline.c (want_inline_small_function_p): Return false if
996 inlining has already failed with CIF_FINAL_ERROR.
997 (update_caller_keys): Call want_inline_small_function_p before
999 (update_callee_keys): Likewise.
1001 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1003 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1005 (rs6000_quadword_masked_address_p): Likewise.
1006 (quad_aligned_load_p): Likewise.
1007 (quad_aligned_store_p): Likewise.
1008 (const_load_sequence_p): Add comment to describe the outer-most loop.
1009 (mimic_memory_attributes_and_flags): New function.
1010 (rs6000_gen_stvx): Likewise.
1011 (replace_swapped_aligned_store): Likewise.
1012 (rs6000_gen_lvx): Likewise.
1013 (replace_swapped_aligned_load): Likewise.
1014 (replace_swapped_load_constant): Capitalize argument name in
1015 comment describing this function.
1016 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1017 and stores that access quad-word aligned addresses and replace
1018 with stvx or lvx instructions when appropriate.
1019 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1020 New function prototype.
1021 (rs6000_quadword_masked_address_p): Likewise.
1022 (rs6000_gen_lvx): Likewise.
1023 (rs6000_gen_stvx): Likewise.
1024 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1025 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1026 when memory address is aligned.
1027 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1028 this split to select lvx instruction when memory address is aligned.
1029 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1030 instruction when memory address is aligned.
1031 (*vsx_le_perm_load_v16qi): Likewise.
1032 (four unnamed splitters): Modify to select the stvx instruction
1033 when memory is aligned.
1035 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1037 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1038 which appears in the queue multiple times.
1040 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1041 Alan Hayward <alan.hayward@arm.com>
1042 David Sherwood <david.sherwood@arm.com>
1044 * tree-vectorizer.h (vec_lower_bound): New structure.
1045 (_loop_vec_info): Add check_nonzero and lower_bounds.
1046 (LOOP_VINFO_CHECK_NONZERO): New macro.
1047 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1048 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1049 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1050 fields. Make seg_len the distance travelled, not including the
1052 (dr_direction_indicator): Declare.
1053 (dr_zero_step_indicator): Likewise.
1054 (dr_known_forward_stride_p): Likewise.
1055 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1057 (runtime_alias_check_p): Allow runtime alias checks with
1059 (operator ==): Compare access_size and align.
1060 (prune_runtime_alias_test_list): Rework for new distinction between
1061 the access_size and seg_len.
1062 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1064 (get_segment_min_max): New function.
1065 (create_intersect_range_checks): Use it.
1066 (dr_step_indicator): New function.
1067 (dr_direction_indicator): Likewise.
1068 (dr_zero_step_indicator): Likewise.
1069 (dr_known_forward_stride_p): Likewise.
1070 * tree-loop-distribution.c (data_ref_segment_size): Return
1071 DR_STEP * (niters - 1).
1072 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1074 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1075 (vect_preserves_scalar_order_p): New function, split out from...
1076 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1077 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1078 (vect_vfa_access_size): New function.
1079 (vect_vfa_align): Likewise.
1080 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1081 (dump_lower_bound): New function.
1082 (vect_check_lower_bound): Likewise.
1083 (vect_small_gap_p): Likewise.
1084 (vectorizable_with_step_bound_p): Likewise.
1085 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1086 depencies if the vectorization factor is 1. Convert the checks
1087 for nonzero steps into checks on the bounds of DR_STEP. Try using
1088 a bunds check for variable steps if the minimum required step is
1089 relatively small. Update calls to the dr_with_seg_len
1090 constructor and to vect_compile_time_alias.
1091 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1093 (vect_loop_versioning): Call it.
1094 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1096 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1098 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1099 Alan Hayward <alan.hayward@arm.com>
1100 David Sherwood <david.sherwood@arm.com>
1102 * doc/sourcebuild.texi (vect_scatter_store): Document.
1103 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1105 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1107 * genopinit.c (main): Add supports_vec_scatter_store and
1108 supports_vec_scatter_store_cached to target_optabs.
1109 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1110 IFN_MASK_SCATTER_STORE.
1111 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1113 * internal-fn.h (internal_store_fn_p): Declare.
1114 (internal_fn_stored_value_index): Likewise.
1115 * internal-fn.c (scatter_store_direct): New macro.
1116 (expand_scatter_store_optab_fn): New function.
1117 (direct_scatter_store_optab_supported_p): New macro.
1118 (internal_store_fn_p): New function.
1119 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1120 IFN_MASK_SCATTER_STORE.
1121 (internal_fn_mask_index): Likewise.
1122 (internal_fn_stored_value_index): New function.
1123 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1125 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1126 * optabs-query.c (supports_vec_scatter_store_p): New function.
1127 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1128 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1129 true for scatter stores.
1130 (vect_gather_scatter_fn_p): Handle scatter stores too.
1131 (vect_check_gather_scatter): Consider using scatter stores if
1132 supports_vec_scatter_store_p.
1133 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1135 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1136 internal_fn_stored_value_index.
1137 (check_load_store_masking): Handle scatter stores too.
1138 (vect_get_store_rhs): Make public.
1139 (vectorizable_call): Use internal_store_fn_p.
1140 (vectorizable_store): Handle scatter store internal functions.
1141 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1142 when deciding whether the end of the group has been reached.
1143 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1144 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1145 (mask_scatter_store<mode>): New insns.
1147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1148 Alan Hayward <alan.hayward@arm.com>
1149 David Sherwood <david.sherwood@arm.com>
1151 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1152 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1153 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1155 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1156 Use vect_truncate_gather_scatter_offset if we can't treat the
1157 operation as a normal gather load or scatter store.
1158 (get_group_load_store_type): Take the gather_scatter_info
1159 as argument. Try using a gather load or scatter store for
1160 single-element groups.
1161 (get_load_store_type): Update calls to get_group_load_store_type
1162 and vect_use_strided_gather_scatters_p.
1164 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1165 Alan Hayward <alan.hayward@arm.com>
1166 David Sherwood <david.sherwood@arm.com>
1168 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1169 optional tree argument.
1170 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1172 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1173 but continue to use the current value as a fallback.
1174 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1175 to compare the updates.
1176 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1177 (get_load_store_type): Use it when handling a strided access.
1178 (vect_get_strided_load_store_ops): New function.
1179 (vect_get_data_ptr_increment): Likewise.
1180 (vectorizable_load): Handle strided gather loads. Always pass
1181 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1183 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1184 Alan Hayward <alan.hayward@arm.com>
1185 David Sherwood <david.sherwood@arm.com>
1187 * doc/md.texi (gather_load@var{m}): Document.
1188 (mask_gather_load@var{m}): Likewise.
1189 * genopinit.c (main): Add supports_vec_gather_load and
1190 supports_vec_gather_load_cached to target_optabs.
1191 * optabs-tree.c (init_tree_optimization_optabs): Use
1192 ggc_cleared_alloc to allocate target_optabs.
1193 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1194 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1196 * internal-fn.h (internal_load_fn_p): Declare.
1197 (internal_gather_scatter_fn_p): Likewise.
1198 (internal_fn_mask_index): Likewise.
1199 (internal_gather_scatter_fn_supported_p): Likewise.
1200 * internal-fn.c (gather_load_direct): New macro.
1201 (expand_gather_load_optab_fn): New function.
1202 (direct_gather_load_optab_supported_p): New macro.
1203 (direct_internal_fn_optab): New function.
1204 (internal_load_fn_p): Likewise.
1205 (internal_gather_scatter_fn_p): Likewise.
1206 (internal_fn_mask_index): Likewise.
1207 (internal_gather_scatter_fn_supported_p): Likewise.
1208 * optabs-query.c (supports_at_least_one_mode_p): New function.
1209 (supports_vec_gather_load_p): Likewise.
1210 * optabs-query.h (supports_vec_gather_load_p): Declare.
1211 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1212 and memory_type field.
1213 (NUM_PATTERNS): Bump to 15.
1214 * tree-vect-data-refs.c: Include internal-fn.h.
1215 (vect_gather_scatter_fn_p): New function.
1216 (vect_describe_gather_scatter_call): Likewise.
1217 (vect_check_gather_scatter): Try using internal functions for
1218 gather loads. Recognize existing calls to a gather load function.
1219 (vect_analyze_data_refs): Consider using gather loads if
1220 supports_vec_gather_load_p.
1221 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1222 (vect_get_gather_scatter_offset_type): Likewise.
1223 (vect_convert_mask_for_vectype): Likewise.
1224 (vect_add_conversion_to_patterm): Likewise.
1225 (vect_try_gather_scatter_pattern): Likewise.
1226 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1227 (vect_vect_recog_func_ptrs): Add it.
1228 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1229 internal_fn_mask_index and internal_gather_scatter_fn_p.
1230 (check_load_store_masking): Take the gather_scatter_info as an
1231 argument and handle gather loads.
1232 (vect_get_gather_scatter_ops): New function.
1233 (vectorizable_call): Check internal_load_fn_p.
1234 (vectorizable_load): Likewise. Handle gather load internal
1236 (vectorizable_store): Update call to check_load_store_masking.
1237 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1238 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1239 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1240 (aarch64_gather_scale_operand_d): New predicates.
1241 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1242 (mask_gather_load<mode>): New insns.
1244 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1245 Alan Hayward <alan.hayward@arm.com>
1246 David Sherwood <david.sherwood@arm.com>
1248 * optabs.def (fold_left_plus_optab): New optab.
1249 * doc/md.texi (fold_left_plus_@var{m}): Document.
1250 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1251 * internal-fn.c (fold_left_direct): Define.
1252 (expand_fold_left_optab_fn): Likewise.
1253 (direct_fold_left_optab_supported_p): Likewise.
1254 * fold-const-call.c (fold_const_fold_left): New function.
1255 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1256 * tree-parloops.c (valid_reduction_p): New function.
1257 (gather_scalar_reductions): Use it.
1258 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1259 (vect_finish_replace_stmt): Declare.
1260 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1261 (needs_fold_left_reduction_p): New function, split out from...
1262 (vect_is_simple_reduction): ...here. Accept reductions that
1263 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1264 (vect_force_simple_reduction): Also store the reduction type in
1265 the assignment's STMT_VINFO_REDUC_TYPE.
1266 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1267 (merge_with_identity): New function.
1268 (vect_expand_fold_left): Likewise.
1269 (vectorize_fold_left_reduction): Likewise.
1270 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1271 scalar phi in place for it. Check for target support and reject
1272 cases that would reassociate the operation. Defer the transform
1273 phase to vectorize_fold_left_reduction.
1274 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1275 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1276 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1278 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1280 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1281 call to ifc_temp_var.
1283 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1284 Alan Hayward <alan.hayward@arm.com>
1285 David Sherwood <david.sherwood@arm.com>
1287 * target.def (legitimize_address_displacement): Take the original
1288 offset as a poly_int.
1289 * targhooks.h (default_legitimize_address_displacement): Update
1291 * targhooks.c (default_legitimize_address_displacement): Likewise.
1292 * doc/tm.texi: Regenerate.
1293 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1294 as an argument, moving assert of ad->disp == ad->disp_term to...
1295 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1296 Try calling targetm.legitimize_address_displacement before expanding
1297 the address rather than afterwards, and adjust for the new interface.
1298 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1299 Match the new hook interface. Handle SVE addresses.
1300 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1303 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1305 * Makefile.in (OBJS): Add early-remat.o.
1306 * target.def (select_early_remat_modes): New hook.
1307 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1308 * doc/tm.texi: Regenerate.
1309 * targhooks.h (default_select_early_remat_modes): Declare.
1310 * targhooks.c (default_select_early_remat_modes): New function.
1311 * timevar.def (TV_EARLY_REMAT): New timevar.
1312 * passes.def (pass_early_remat): New pass.
1313 * tree-pass.h (make_pass_early_remat): Declare.
1314 * early-remat.c: New file.
1315 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1317 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1319 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1320 Alan Hayward <alan.hayward@arm.com>
1321 David Sherwood <david.sherwood@arm.com>
1323 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1324 vfm1 with a bound_epilog parameter.
1325 (vect_do_peeling): Update calls accordingly, and move the prologue
1326 call earlier in the function. Treat the base bound_epilog as 0 for
1327 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1328 this base when peeling for gaps.
1329 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1330 with fully-masked loops.
1331 (vect_estimate_min_profitable_iters): Handle the single peeled
1332 iteration in that case.
1334 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1335 Alan Hayward <alan.hayward@arm.com>
1336 David Sherwood <david.sherwood@arm.com>
1338 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1339 single-element interleaving even if the size is not a power of 2.
1340 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1341 accesses for single-element interleaving if the group size is
1344 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1345 Alan Hayward <alan.hayward@arm.com>
1346 David Sherwood <david.sherwood@arm.com>
1348 * doc/md.texi (fold_extract_last_@var{m}): Document.
1349 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1350 * optabs.def (fold_extract_last_optab): New optab.
1351 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1352 * internal-fn.c (fold_extract_direct): New macro.
1353 (expand_fold_extract_optab_fn): Likewise.
1354 (direct_fold_extract_optab_supported_p): Likewise.
1355 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1356 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1357 EXTRACT_LAST_REDUCTION.
1358 (get_initial_def_for_reduction): Do not create an initial vector
1359 for EXTRACT_LAST_REDUCTION reductions.
1360 (vectorizable_reduction): Leave the scalar phi in place for
1361 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1362 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1363 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1364 transform phase to vectorizable_condition.
1365 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1367 (vect_finish_stmt_generation): ...here.
1368 (vect_finish_replace_stmt): New function.
1369 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1370 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1372 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1374 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1375 Alan Hayward <alan.hayward@arm.com>
1376 David Sherwood <david.sherwood@arm.com>
1378 * doc/md.texi (extract_last_@var{m}): Document.
1379 * optabs.def (extract_last_optab): New optab.
1380 * internal-fn.def (EXTRACT_LAST): New internal function.
1381 * internal-fn.c (cond_unary_direct): New macro.
1382 (expand_cond_unary_optab_fn): Likewise.
1383 (direct_cond_unary_optab_supported_p): Likewise.
1384 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1385 loops using EXTRACT_LAST.
1386 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1387 (extract_last_<mode>): ...this optab.
1388 (vec_extract<mode><Vel>): Update accordingly.
1390 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1391 Alan Hayward <alan.hayward@arm.com>
1392 David Sherwood <david.sherwood@arm.com>
1394 * target.def (empty_mask_is_expensive): New hook.
1395 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1396 * doc/tm.texi: Regenerate.
1397 * targhooks.h (default_empty_mask_is_expensive): Declare.
1398 * targhooks.c (default_empty_mask_is_expensive): New function.
1399 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1400 if the target says that empty masks are expensive.
1401 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1403 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1405 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1406 Alan Hayward <alan.hayward@arm.com>
1407 David Sherwood <david.sherwood@arm.com>
1409 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1410 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1411 (vect_use_loop_mask_for_alignment_p): New function.
1412 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1413 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1414 niters_skip argument. Make sure that the first niters_skip elements
1415 of the first iteration are inactive.
1416 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1417 Update call to vect_set_loop_masks_directly.
1418 (get_misalign_in_elems): New function, split out from...
1419 (vect_gen_prolog_loop_niters): ...here.
1420 (vect_update_init_of_dr): Take a code argument that specifies whether
1421 the adjustment should be added or subtracted.
1422 (vect_update_init_of_drs): Likewise.
1423 (vect_prepare_for_masked_peels): New function.
1424 (vect_do_peeling): Skip prologue peeling if we're using a mask
1425 instead. Update call to vect_update_inits_of_drs.
1426 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1428 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1429 alignment. Do not include the number of peeled iterations in
1430 the minimum threshold in that case.
1431 (vectorizable_induction): Adjust the start value down by
1432 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1433 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1434 Take the number of skipped iterations into account when calculating
1436 * tree-vect-stmts.c (vect_gen_while_not): New function.
1438 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1439 Alan Hayward <alan.hayward@arm.com>
1440 David Sherwood <david.sherwood@arm.com>
1442 * doc/sourcebuild.texi (vect_fully_masked): Document.
1443 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1445 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1447 (vect_analyze_loop_2): ...here. Don't check the vectorization
1448 factor against the number of loop iterations if the loop is
1451 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1452 Alan Hayward <alan.hayward@arm.com>
1453 David Sherwood <david.sherwood@arm.com>
1455 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1456 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1457 (dump_groups): Update accordingly.
1458 (iv_use::mem_type): New member variable.
1459 (address_p): New function.
1460 (record_use): Add a mem_type argument and initialize the new
1462 (record_group_use): Add a mem_type argument. Use address_p.
1463 Remove obsolete null checks of base_object. Update call to record_use.
1464 (find_interesting_uses_op): Update call to record_group_use.
1465 (find_interesting_uses_cond): Likewise.
1466 (find_interesting_uses_address): Likewise.
1467 (get_mem_type_for_internal_fn): New function.
1468 (find_address_like_use): Likewise.
1469 (find_interesting_uses_stmt): Try find_address_like_use before
1470 calling find_interesting_uses_op.
1471 (addr_offset_valid_p): Use the iv mem_type field as the type
1472 of the addressed memory.
1473 (add_autoinc_candidates): Likewise.
1474 (get_address_cost): Likewise.
1475 (split_small_address_groups_p): Use address_p.
1476 (split_address_groups): Likewise.
1477 (add_iv_candidate_for_use): Likewise.
1478 (autoinc_possible_for_pair): Likewise.
1479 (rewrite_groups): Likewise.
1480 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1481 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1482 (get_alias_ptr_type_for_ptr_address): New function.
1483 (rewrite_use_address): Rewrite address uses in calls that were
1484 identified by find_address_like_use.
1486 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1487 Alan Hayward <alan.hayward@arm.com>
1488 David Sherwood <david.sherwood@arm.com>
1490 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1492 * gimple-expr.h (is_gimple_addressable: Likewise.
1493 * gimple-expr.c (is_gimple_address): Likewise.
1494 * internal-fn.c (expand_call_mem_ref): New function.
1495 (expand_mask_load_optab_fn): Use it.
1496 (expand_mask_store_optab_fn): Likewise.
1498 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1499 Alan Hayward <alan.hayward@arm.com>
1500 David Sherwood <david.sherwood@arm.com>
1502 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1503 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1504 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1505 (cond_umax@var{mode}): Document.
1506 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1507 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1508 (cond_umin_optab, cond_umax_optab): New optabs.
1509 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1510 (COND_IOR, COND_XOR): New internal functions.
1511 * internal-fn.h (get_conditional_internal_fn): Declare.
1512 * internal-fn.c (cond_binary_direct): New macro.
1513 (expand_cond_binary_optab_fn): Likewise.
1514 (direct_cond_binary_optab_supported_p): Likewise.
1515 (get_conditional_internal_fn): New function.
1516 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1517 Cope with reduction statements that are vectorized as calls rather
1519 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1520 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1521 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1522 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1523 (UNSPEC_COND_EOR): New unspecs.
1524 (optab): Add mappings for them.
1525 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1526 (sve_int_op, sve_fp_op): New int attributes.
1528 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1529 Alan Hayward <alan.hayward@arm.com>
1530 David Sherwood <david.sherwood@arm.com>
1532 * optabs.def (while_ult_optab): New optab.
1533 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1534 * internal-fn.def (WHILE_ULT): New internal function.
1535 * internal-fn.h (direct_internal_fn_supported_p): New override
1536 that takes two types as argument.
1537 * internal-fn.c (while_direct): New macro.
1538 (expand_while_optab_fn): New function.
1539 (convert_optab_supported_p): Likewise.
1540 (direct_while_optab_supported_p): New macro.
1541 * wide-int.h (wi::udiv_ceil): New function.
1542 * tree-vectorizer.h (rgroup_masks): New structure.
1543 (vec_loop_masks): New typedef.
1544 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1546 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1547 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1548 (vect_max_vf): New function.
1549 (slpeel_make_loop_iterate_ntimes): Delete.
1550 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1551 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1552 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1553 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1554 internal-fn.h, stor-layout.h and optabs-query.h.
1555 (vect_set_loop_mask): New function.
1556 (add_preheader_seq): Likewise.
1557 (add_header_seq): Likewise.
1558 (interleave_supported_p): Likewise.
1559 (vect_maybe_permute_loop_masks): Likewise.
1560 (vect_set_loop_masks_directly): Likewise.
1561 (vect_set_loop_condition_masked): Likewise.
1562 (vect_set_loop_condition_unmasked): New function, split out from
1563 slpeel_make_loop_iterate_ntimes.
1564 (slpeel_make_loop_iterate_ntimes): Rename to..
1565 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1566 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1567 (vect_do_peeling): Update call accordingly.
1568 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1570 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1571 mask_compare_type, can_fully_mask_p and fully_masked_p.
1572 (release_vec_loop_masks): New function.
1573 (_loop_vec_info): Use it to free the loop masks.
1574 (can_produce_all_loop_masks_p): New function.
1575 (vect_get_max_nscalars_per_iter): Likewise.
1576 (vect_verify_full_masking): Likewise.
1577 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1578 retries, and free the mask rgroups before retrying. Check loop-wide
1579 reasons for disallowing fully-masked loops. Make the final decision
1580 about whether use a fully-masked loop or not.
1581 (vect_estimate_min_profitable_iters): Do not assume that peeling
1582 for the number of iterations will be needed for fully-masked loops.
1583 (vectorizable_reduction): Disable fully-masked loops.
1584 (vectorizable_live_operation): Likewise.
1585 (vect_halve_mask_nunits): New function.
1586 (vect_double_mask_nunits): Likewise.
1587 (vect_record_loop_mask): Likewise.
1588 (vect_get_loop_mask): Likewise.
1589 (vect_transform_loop): Handle the case in which the final loop
1590 iteration might handle a partial vector. Call vect_set_loop_condition
1591 instead of slpeel_make_loop_iterate_ntimes.
1592 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1593 (check_load_store_masking): New function.
1594 (prepare_load_store_mask): Likewise.
1595 (vectorizable_store): Handle fully-masked loops.
1596 (vectorizable_load): Likewise.
1597 (supportable_widening_operation): Use vect_halve_mask_nunits for
1599 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1600 (vect_gen_while): New function.
1601 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1602 (aarch64_uqdec<mode>): New insn.
1604 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1605 Alan Hayward <alan.hayward@arm.com>
1606 David Sherwood <david.sherwood@arm.com>
1608 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1609 (reduc_xor_scal_optab): New optabs.
1610 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1611 (reduc_xor_scal_@var{m}): Document.
1612 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1613 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1615 * fold-const-call.c (fold_const_call): Handle them.
1616 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
1617 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
1618 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
1619 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
1620 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
1621 (UNSPEC_XORV): New unspecs.
1622 (optab): Add entries for them.
1623 (BITWISEV): New int iterator.
1624 (bit_reduc_op): New int attributes.
1626 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1627 Alan Hayward <alan.hayward@arm.com>
1628 David Sherwood <david.sherwood@arm.com>
1630 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
1631 * internal-fn.def (VEC_SHL_INSERT): New internal function.
1632 * optabs.def (vec_shl_insert_optab): New optab.
1633 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
1634 (duplicate_and_interleave): Likewise.
1635 * tree-vect-loop.c: Include internal-fn.h.
1636 (neutral_op_for_slp_reduction): New function, split out from
1637 get_initial_defs_for_reduction.
1638 (get_initial_def_for_reduction): Handle option 2 for variable-length
1639 vectors by loading the neutral value into a vector and then shifting
1640 the initial value into element 0.
1641 (get_initial_defs_for_reduction): Replace the code argument with
1642 the neutral value calculated by neutral_op_for_slp_reduction.
1643 Use gimple_build_vector for constant-length vectors.
1644 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
1645 but the first group_size elements have a neutral value.
1646 Use duplicate_and_interleave otherwise.
1647 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
1648 Update call to get_initial_defs_for_reduction. Handle SLP
1649 reductions for variable-length vectors by creating one vector
1650 result for each scalar result, with the elements associated
1651 with other scalar results stubbed out with the neutral value.
1652 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
1653 Require IFN_VEC_SHL_INSERT for double reductions on
1654 variable-length vectors, or SLP reductions that have
1655 a neutral value. Require can_duplicate_and_interleave_p
1656 support for variable-length unchained SLP reductions if there
1657 is no neutral value, such as for MIN/MAX reductions. Also require
1658 the number of vector elements to be a multiple of the number of
1659 SLP statements when doing variable-length unchained SLP reductions.
1660 Update call to vect_create_epilog_for_reduction.
1661 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
1662 and remove initial values.
1663 (duplicate_and_interleave): Make public.
1664 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
1665 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
1667 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1668 Alan Hayward <alan.hayward@arm.com>
1669 David Sherwood <david.sherwood@arm.com>
1671 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
1672 (can_duplicate_and_interleave_p): New function.
1673 (vect_get_and_check_slp_defs): Take the vector of statements
1674 rather than just the current one. Remove excess parentheses.
1675 Restriction rejectinon of vect_constant_def and vect_external_def
1676 for variable-length vectors to boolean types, or types for which
1677 can_duplicate_and_interleave_p is false.
1678 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
1679 (duplicate_and_interleave): New function.
1680 (vect_get_constant_vectors): Use gimple_build_vector for
1681 constant-length vectors and suitable variable-length constant
1682 vectors. Use duplicate_and_interleave for other variable-length
1683 vectors. Don't defer the update when inserting new statements.
1685 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1686 Alan Hayward <alan.hayward@arm.com>
1687 David Sherwood <david.sherwood@arm.com>
1689 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
1690 min_profitable_iters doesn't go negative.
1692 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1693 Alan Hayward <alan.hayward@arm.com>
1694 David Sherwood <david.sherwood@arm.com>
1696 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
1697 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
1698 * optabs.def (vec_mask_load_lanes_optab): New optab.
1699 (vec_mask_store_lanes_optab): Likewise.
1700 * internal-fn.def (MASK_LOAD_LANES): New internal function.
1701 (MASK_STORE_LANES): Likewise.
1702 * internal-fn.c (mask_load_lanes_direct): New macro.
1703 (mask_store_lanes_direct): Likewise.
1704 (expand_mask_load_optab_fn): Handle masked operations.
1705 (expand_mask_load_lanes_optab_fn): New macro.
1706 (expand_mask_store_optab_fn): Handle masked operations.
1707 (expand_mask_store_lanes_optab_fn): New macro.
1708 (direct_mask_load_lanes_optab_supported_p): Likewise.
1709 (direct_mask_store_lanes_optab_supported_p): Likewise.
1710 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
1712 (vect_load_lanes_supported): Likewise.
1713 * tree-vect-data-refs.c (strip_conversion): New function.
1714 (can_group_stmts_p): Likewise.
1715 (vect_analyze_data_ref_accesses): Use it instead of checking
1716 for a pair of assignments.
1717 (vect_store_lanes_supported): Take a masked_p parameter.
1718 (vect_load_lanes_supported): Likewise.
1719 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
1720 vect_store_lanes_supported and vect_load_lanes_supported.
1721 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
1722 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
1723 parameter. Don't allow gaps for masked accesses.
1724 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
1725 and vect_load_lanes_supported.
1726 (get_load_store_type): Take a masked_p parameter and update
1727 call to get_group_load_store_type.
1728 (vectorizable_store): Update call to get_load_store_type.
1729 Handle IFN_MASK_STORE_LANES.
1730 (vectorizable_load): Update call to get_load_store_type.
1731 Handle IFN_MASK_LOAD_LANES.
1733 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1734 Alan Hayward <alan.hayward@arm.com>
1735 David Sherwood <david.sherwood@arm.com>
1737 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
1739 * config/aarch64/aarch64-protos.h
1740 (aarch64_sve_struct_memory_operand_p): Declare.
1741 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
1742 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
1743 (VPRED, vpred): Handle SVE structure modes.
1744 * config/aarch64/constraints.md (Utx): New constraint.
1745 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
1746 (aarch64_sve_struct_nonimmediate_operand): New predicates.
1747 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
1748 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
1749 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
1750 structure modes. Split into pieces after RA.
1751 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
1752 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
1754 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
1755 SVE structure modes.
1756 (aarch64_classify_address): Likewise.
1757 (sizetochar): Move earlier in file.
1758 (aarch64_print_operand): Handle SVE register lists.
1759 (aarch64_array_mode): New function.
1760 (aarch64_sve_struct_memory_operand_p): Likewise.
1761 (TARGET_ARRAY_MODE): Redefine.
1763 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1764 Alan Hayward <alan.hayward@arm.com>
1765 David Sherwood <david.sherwood@arm.com>
1767 * target.def (array_mode): New target hook.
1768 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
1769 * doc/tm.texi: Regenerate.
1770 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
1771 * hooks.c (hook_optmode_mode_uhwi_none): New function.
1772 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
1774 * stor-layout.c (mode_for_array): Likewise. Support polynomial
1777 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1778 Alan Hayward <alan.hayward@arm.com>
1779 David Sherwood <david.sherwood@arm.com>
1781 * fold-const.c (fold_binary_loc): Check the argument types
1782 rather than the result type when testing for a vector operation.
1784 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1786 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
1787 * doc/tm.texi: Regenerate.
1789 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1790 Alan Hayward <alan.hayward@arm.com>
1791 David Sherwood <david.sherwood@arm.com>
1793 * doc/invoke.texi (-msve-vector-bits=): Document new option.
1794 (sve): Document new AArch64 extension.
1795 * doc/md.texi (w): Extend the description of the AArch64
1796 constraint to include SVE vectors.
1797 (Upl, Upa): Document new AArch64 predicate constraints.
1798 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
1800 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
1801 (msve-vector-bits=): New option.
1802 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
1803 SVE when these are disabled.
1804 (sve): New extension.
1805 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
1806 modes. Adjust their number of units based on aarch64_sve_vg.
1807 (MAX_BITSIZE_MODE_ANY_MODE): Define.
1808 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
1809 aarch64_addr_query_type.
1810 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
1811 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
1812 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
1813 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
1814 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
1815 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
1816 (aarch64_simd_imm_zero_p): Delete.
1817 (aarch64_check_zero_based_sve_index_immediate): Declare.
1818 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
1819 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
1820 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
1821 (aarch64_sve_float_mul_immediate_p): Likewise.
1822 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
1824 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
1825 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
1826 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
1827 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
1828 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
1829 (aarch64_regmode_natural_size): Likewise.
1830 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
1831 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
1833 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
1834 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
1835 for VG and the SVE predicate registers.
1836 (V_ALIASES): Add a "z"-prefixed alias.
1837 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
1838 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
1839 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
1840 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
1841 (REG_CLASS_NAMES): Add entries for them.
1842 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
1843 and the predicate registers.
1844 (aarch64_sve_vg): Declare.
1845 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
1846 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
1847 (REGMODE_NATURAL_SIZE): Define.
1848 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
1850 * config/aarch64/aarch64.c: Include cfgrtl.h.
1851 (simd_immediate_info): Add a constructor for series vectors,
1852 and an associated step field.
1853 (aarch64_sve_vg): New variable.
1854 (aarch64_dbx_register_number): Handle VG and the predicate registers.
1855 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
1856 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
1857 (VEC_ANY_DATA, VEC_STRUCT): New constants.
1858 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
1859 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
1860 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
1861 (aarch64_get_mask_mode): New functions.
1862 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
1863 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
1864 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
1865 predicate modes and predicate registers. Explicitly restrict
1866 GPRs to modes of 16 bytes or smaller. Only allow FP registers
1867 to store a vector mode if it is recognized by
1868 aarch64_classify_vector_mode.
1869 (aarch64_regmode_natural_size): New function.
1870 (aarch64_hard_regno_caller_save_mode): Return the original mode
1872 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
1873 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
1874 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
1875 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
1877 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
1878 does not overlap dest if the function is frame-related. Handle
1880 (aarch64_split_add_offset): New function.
1881 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
1882 them aarch64_add_offset.
1883 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
1884 and update call to aarch64_sub_sp.
1885 (aarch64_add_cfa_expression): New function.
1886 (aarch64_expand_prologue): Pass extra temporary registers to the
1887 functions above. Handle the case in which we need to emit new
1888 DW_CFA_expressions for registers that were originally saved
1889 relative to the stack pointer, but now have to be expressed
1890 relative to the frame pointer.
1891 (aarch64_output_mi_thunk): Pass extra temporary registers to the
1893 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
1894 IP0 and IP1 values for SVE frames.
1895 (aarch64_expand_vec_series): New function.
1896 (aarch64_expand_sve_widened_duplicate): Likewise.
1897 (aarch64_expand_sve_const_vector): Likewise.
1898 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
1899 Handle SVE constants. Use emit_move_insn to move a force_const_mem
1900 into the register, rather than emitting a SET directly.
1901 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
1902 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
1903 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
1904 (offset_9bit_signed_scaled_p): New functions.
1905 (aarch64_replicate_bitmask_imm): New function.
1906 (aarch64_bitmask_imm): Use it.
1907 (aarch64_cannot_force_const_mem): Reject expressions involving
1908 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
1909 (aarch64_classify_index): Handle SVE indices, by requiring
1910 a plain register index with a scale that matches the element size.
1911 (aarch64_classify_address): Handle SVE addresses. Assert that
1912 the mode of the address is VOIDmode or an integer mode.
1913 Update call to aarch64_classify_symbol.
1914 (aarch64_classify_symbolic_expression): Update call to
1915 aarch64_classify_symbol.
1916 (aarch64_const_vec_all_in_range_p): New function.
1917 (aarch64_print_vector_float_operand): Likewise.
1918 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
1919 "vN" for FP registers with SVE modes. Handle (const ...) vectors
1920 and the FP immediates 1.0 and 0.5.
1921 (aarch64_print_address_internal): Handle SVE addresses.
1922 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
1923 (aarch64_regno_regclass): Handle predicate registers.
1924 (aarch64_secondary_reload): Handle big-endian reloads of SVE
1926 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
1927 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
1928 (aarch64_convert_sve_vector_bits): New function.
1929 (aarch64_override_options): Use it to handle -msve-vector-bits=.
1930 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
1932 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
1933 Handle SVE vector and predicate modes. Accept VL-based constants
1934 that need only one temporary register, and VL offsets that require
1935 no temporary registers.
1936 (aarch64_conditional_register_usage): Mark the predicate registers
1937 as fixed if SVE isn't available.
1938 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
1939 Return true for SVE vector and predicate modes.
1940 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
1941 rather than an unsigned int. Handle SVE modes.
1942 (aarch64_preferred_simd_mode): Update call accordingly. Handle
1944 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
1946 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
1947 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
1948 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
1949 (aarch64_sve_float_mul_immediate_p): New functions.
1950 (aarch64_sve_valid_immediate): New function.
1951 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
1952 Explicitly reject structure modes. Check for INDEX constants.
1953 Handle PTRUE and PFALSE constants.
1954 (aarch64_check_zero_based_sve_index_immediate): New function.
1955 (aarch64_simd_imm_zero_p): Delete.
1956 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
1957 vector modes. Accept constants in the range of CNT[BHWD].
1958 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
1959 ask for an Advanced SIMD mode.
1960 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
1961 (aarch64_simd_vector_alignment): Handle SVE predicates.
1962 (aarch64_vectorize_preferred_vector_alignment): New function.
1963 (aarch64_simd_vector_alignment_reachable): Use it instead of
1965 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
1966 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
1968 (MAX_VECT_LEN): Delete.
1969 (expand_vec_perm_d): Add a vec_flags field.
1970 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
1971 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
1972 (aarch64_evpc_ext): Don't apply a big-endian lane correction
1974 (aarch64_evpc_rev): Rename to...
1975 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
1976 (aarch64_evpc_rev_global): New function.
1977 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
1978 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
1980 (aarch64_evpc_sve_tbl): New function.
1981 (aarch64_expand_vec_perm_const_1): Update after rename of
1982 aarch64_evpc_rev. Handle SVE permutes too, trying
1983 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
1984 than aarch64_evpc_tbl.
1985 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
1986 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
1987 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
1988 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
1989 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
1990 (aarch64_expand_sve_vcond): New functions.
1991 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
1992 of aarch64_vector_mode_p.
1993 (aarch64_dwarf_poly_indeterminate_value): New function.
1994 (aarch64_compute_pressure_classes): Likewise.
1995 (aarch64_can_change_mode_class): Likewise.
1996 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
1997 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
1998 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
1999 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2000 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2001 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2002 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2003 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2005 (Dn, Dl, Dr): Accept const as well as const_vector.
2006 (Dz): Likewise. Compare against CONST0_RTX.
2007 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2008 of "vector" where appropriate.
2009 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2010 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2011 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2012 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2013 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2014 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2015 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2016 (v_int_equiv): Extend to SVE modes.
2017 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2019 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2020 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2021 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2022 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2023 (SVE_COND_FP_CMP): New int iterators.
2024 (perm_hilo): Handle the new unpack unspecs.
2025 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2027 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2028 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2029 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2030 (aarch64_equality_operator, aarch64_constant_vector_operand)
2031 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2032 (aarch64_sve_nonimmediate_operand): Likewise.
2033 (aarch64_sve_general_operand): Likewise.
2034 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2035 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2036 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2037 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2038 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2039 (aarch64_sve_float_arith_immediate): Likewise.
2040 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2041 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2042 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2043 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2044 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2045 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2046 (aarch64_sve_float_arith_operand): Likewise.
2047 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2048 (aarch64_sve_float_mul_operand): Likewise.
2049 (aarch64_sve_vec_perm_operand): Likewise.
2050 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2051 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2052 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2053 as well as const_vector.
2054 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2055 in file. Use CONST0_RTX and CONSTM1_RTX.
2056 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2057 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2058 Use aarch64_simd_imm_zero.
2059 * config/aarch64/aarch64-sve.md: New file.
2060 * config/aarch64/aarch64.md: Include it.
2061 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2062 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2063 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2064 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2065 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2066 (sve): New attribute.
2067 (enabled): Disable instructions with the sve attribute unless
2069 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2070 aarch64_expand_mov_immediate.
2071 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2072 CNT[BHSD] immediates.
2073 (movti): Split CONST_POLY_INT moves into two halves.
2074 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2075 Split additions that need a temporary here if the destination
2076 is the stack pointer.
2077 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2078 (*add<mode>3_poly_1): New instruction.
2079 (set_clobber_cc): New expander.
2081 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2083 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2084 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2085 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2086 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2087 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2088 Change innermode from fixed_mode_size to machine_mode.
2089 (simplify_subreg): Update call accordingly. Handle a constant-sized
2090 subreg of a variable-length CONST_VECTOR.
2092 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2093 Alan Hayward <alan.hayward@arm.com>
2094 David Sherwood <david.sherwood@arm.com>
2096 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2097 (add_offset_to_base): New function, split out from...
2098 (create_mem_ref): ...here. When handling a scale other than 1,
2099 check first whether the address is valid without the offset.
2100 Add it into the base if so, leaving the index and scale as-is.
2102 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2105 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2106 fold_for_warn before checking if arg2 is INTEGER_CST.
2108 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2110 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2111 (store_multiple_operation): Delete.
2112 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2113 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2114 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2115 guarded by TARGET_STRING.
2116 (rs6000_output_load_multiple): Delete.
2117 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2118 OPTION_MASK_STRING / TARGET_STRING handling.
2119 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2120 (const rs6000_opt_masks) <"string">: Change mask to 0.
2121 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2122 (MASK_STRING): Delete.
2123 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2125 (load_multiple): Delete.
2132 (store_multiple): Delete.
2139 (movmemsi_8reg): Delete.
2140 (corresponding unnamed define_insn): Delete.
2141 (movmemsi_6reg): Delete.
2142 (corresponding unnamed define_insn): Delete.
2143 (movmemsi_4reg): Delete.
2144 (corresponding unnamed define_insn): Delete.
2145 (movmemsi_2reg): Delete.
2146 (corresponding unnamed define_insn): Delete.
2147 (movmemsi_1reg): Delete.
2148 (corresponding unnamed define_insn): Delete.
2149 * config/rs6000/rs6000.opt (mno-string): New.
2150 (mstring): Replace by deprecation warning stub.
2151 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2153 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2155 * regrename.c (regrename_do_replace): If replacing the same
2156 reg multiple times, try to reuse last created gen_raw_REG.
2159 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2160 main to workaround a bug in GDB.
2162 2018-01-12 Tom de Vries <tom@codesourcery.com>
2165 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2167 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2169 PR rtl-optimization/80481
2170 * ira-color.c (get_cap_member): New function.
2171 (allocnos_conflict_by_live_ranges_p): Use it.
2172 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2173 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2175 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2178 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2179 (*saddl_se_1): Ditto.
2181 (*ssubl_se_1): Ditto.
2183 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2185 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2186 rather than wi::to_widest for DR_INITs.
2187 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2188 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2189 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2191 (vect_analyze_group_access_1): Note that here.
2193 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2195 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2196 polynomial type sizes.
2198 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2200 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2201 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2202 (gimple_add_tmp_var): Likewise.
2204 2018-01-12 Martin Liska <mliska@suse.cz>
2206 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2207 (gimple_alloc_sizes): Likewise.
2208 (dump_gimple_statistics): Use PRIu64 in printf format.
2209 * gimple.h: Change uint64_t to int.
2211 2018-01-12 Martin Liska <mliska@suse.cz>
2213 * tree-core.h: Use uint64_t instead of int.
2214 * tree.c (tree_node_counts): Likewise.
2215 (tree_node_sizes): Likewise.
2216 (dump_tree_statistics): Use PRIu64 in printf format.
2218 2018-01-12 Martin Liska <mliska@suse.cz>
2220 * Makefile.in: As qsort_chk is implemented in vec.c, add
2221 vec.o to linkage of gencfn-macros.
2222 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2223 passing the info to record_node_allocation_statistics.
2224 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2226 * ggc-common.c (struct ggc_usage): Add operator== and use
2227 it in operator< and compare function.
2228 * mem-stats.h (struct mem_usage): Likewise.
2229 * vec.c (struct vec_usage): Remove operator< and compare
2230 function. Can be simply inherited.
2232 2018-01-12 Martin Jambor <mjambor@suse.cz>
2235 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2236 * tree-ssa-math-opts.c: Include domwalk.h.
2237 (convert_mult_to_fma_1): New function.
2238 (fma_transformation_info): New type.
2239 (fma_deferring_state): Likewise.
2240 (cancel_fma_deferring): New function.
2241 (result_of_phi): Likewise.
2242 (last_fma_candidate_feeds_initial_phi): Likewise.
2243 (convert_mult_to_fma): Added deferring logic, split actual
2244 transformation to convert_mult_to_fma_1.
2245 (math_opts_dom_walker): New type.
2246 (math_opts_dom_walker::after_dom_children): New method, body moved
2247 here from pass_optimize_widening_mul::execute, added deferring logic
2249 (pass_optimize_widening_mul::execute): Moved most of code to
2250 math_opts_dom_walker::after_dom_children.
2251 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2252 * config/i386/i386.c (ix86_option_override_internal): Added
2253 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2255 2018-01-12 Richard Biener <rguenther@suse.de>
2258 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2259 inline instance vars.
2261 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2264 * config/rx/rx.c (rx_is_restricted_memory_address):
2267 2018-01-12 Richard Biener <rguenther@suse.de>
2269 PR tree-optimization/80846
2270 * target.def (split_reduction): New target hook.
2271 * targhooks.c (default_split_reduction): New function.
2272 * targhooks.h (default_split_reduction): Declare.
2273 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2274 target requests first reduce vectors by combining low and high
2276 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2277 (get_vectype_for_scalar_type_and_size): Export.
2278 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2279 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2280 * doc/tm.texi: Regenerate.
2281 * config/i386/i386.c (ix86_split_reduction): Implement
2282 TARGET_VECTORIZE_SPLIT_REDUCTION.
2284 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2287 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2288 in PIC mode except for TARGET_VXWORKS_RTP.
2289 * config/sparc/sparc.c: Include cfgrtl.h.
2290 (TARGET_INIT_PIC_REG): Define.
2291 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2292 (sparc_pic_register_p): New predicate.
2293 (sparc_legitimate_address_p): Use it.
2294 (sparc_legitimize_pic_address): Likewise.
2295 (sparc_delegitimize_address): Likewise.
2296 (sparc_mode_dependent_address_p): Likewise.
2297 (gen_load_pcrel_sym): Remove 4th parameter.
2298 (load_got_register): Adjust call to above. Remove obsolete stuff.
2299 (sparc_expand_prologue): Do not call load_got_register here.
2300 (sparc_flat_expand_prologue): Likewise.
2301 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2302 (sparc_use_pseudo_pic_reg): New function.
2303 (sparc_init_pic_reg): Likewise.
2304 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2305 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2307 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2309 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2310 Add item for branch_cost.
2312 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2314 PR rtl-optimization/83565
2315 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2316 not extend the result to a larger mode for rotate operations.
2317 (num_sign_bit_copies1): Likewise.
2319 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2322 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2324 Use values-Xc.o for -pedantic.
2325 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2327 2018-01-12 Martin Liska <mliska@suse.cz>
2330 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2332 (possible_polymorphic_call_targets): Use it.
2333 (ipa_devirt): Likewise.
2335 2018-01-12 Martin Liska <mliska@suse.cz>
2337 * profile-count.h (enum profile_quality): Use 0 as invalid
2338 enum value of profile_quality.
2340 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2342 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2343 -mext-string options.
2345 2018-01-12 Richard Biener <rguenther@suse.de>
2347 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2348 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2349 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2351 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2353 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2355 * configure.ac (--with-long-double-format): Add support for the
2356 configuration option to change the default long double format on
2358 * config.gcc (powerpc*-linux*-*): Likewise.
2359 * configure: Regenerate.
2360 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2361 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2362 used without modification.
2364 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2366 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2367 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2368 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2369 MISC_BUILTIN_SPEC_BARRIER.
2370 (rs6000_init_builtins): Likewise.
2371 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2373 (speculation_barrier): New define_insn.
2374 * doc/extend.texi: Document __builtin_speculation_barrier.
2376 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2379 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2380 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2381 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2383 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2384 integral modes instead of "ss" and "sd".
2385 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2386 vectors with 32-bit and 64-bit elements.
2387 (vecdupssescalarmodesuffix): New mode attribute.
2388 (vec_dup<mode>): Use it.
2390 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2393 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2394 frame if argument is passed on stack.
2396 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2399 * ree.c (combine_reaching_defs): Optimize also
2400 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2401 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2403 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2406 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2408 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2411 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2412 after they are computed.
2414 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2416 PR tree-optimization/83695
2417 * gimple-loop-linterchange.cc
2418 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2419 reset cached scev information after interchange.
2420 (pass_linterchange::execute): Remove call to scev_reset_htab.
2422 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2424 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2425 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2426 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2427 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2428 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2429 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2430 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2431 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2432 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2433 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2434 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2435 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2436 (V_lane_reg): Likewise.
2437 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2439 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2440 (vfmal_lane_low<mode>_intrinsic,
2441 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2442 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2443 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2444 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2445 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2446 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2448 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2450 * config/arm/arm-cpus.in (fp16fml): New feature.
2451 (ALL_SIMD): Add fp16fml.
2452 (armv8.2-a): Add fp16fml as an option.
2453 (armv8.3-a): Likewise.
2454 (armv8.4-a): Add fp16fml as part of fp16.
2455 * config/arm/arm.h (TARGET_FP16FML): Define.
2456 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2458 * config/arm/arm-modes.def (V2HF): Define.
2459 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2460 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2461 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2462 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2463 vfmsl_low, vfmsl_high): New set of builtins.
2464 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2465 (vfml_op): New code attribute.
2466 (VFMLHALVES): New int iterator.
2467 (VFML, VFMLSEL): New mode attributes.
2468 (V_reg): Define mapping for V2HF.
2469 (V_hi, V_lo): New mode attributes.
2470 (VF_constraint): Likewise.
2471 (vfml_half, vfml_half_selector): New int attributes.
2472 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2474 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2475 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2477 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2478 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2479 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2480 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2482 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2483 Document new effective target and option set.
2485 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2487 * config/arm/arm-cpus.in (armv8_4): New feature.
2488 (ARMv8_4a): New fgroup.
2489 (armv8.4-a): New arch.
2490 * config/arm/arm-tables.opt: Regenerate.
2491 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2492 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2493 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2494 Add matching rules for -march=armv8.4-a and extensions.
2495 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2497 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2500 * config/rx/rx.md (BW): New mode attribute.
2501 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2503 2018-01-11 Richard Biener <rguenther@suse.de>
2505 PR tree-optimization/83435
2506 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2507 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2508 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2510 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2511 Alan Hayward <alan.hayward@arm.com>
2512 David Sherwood <david.sherwood@arm.com>
2514 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2516 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2517 (aarch64_print_address_internal): Use it to check for a zero offset.
2519 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2520 Alan Hayward <alan.hayward@arm.com>
2521 David Sherwood <david.sherwood@arm.com>
2523 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2524 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2525 Return a poly_int64 rather than a HOST_WIDE_INT.
2526 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2527 rather than a HOST_WIDE_INT.
2528 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2529 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2530 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2531 final_offset from HOST_WIDE_INT to poly_int64.
2532 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2533 to_constant when getting the number of units in an Advanced SIMD
2535 (aarch64_builtin_vectorized_function): Check for a constant number
2537 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2539 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2540 attribute instead of GET_MODE_NUNITS.
2541 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2542 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2543 GET_MODE_SIZE for fixed-size registers.
2544 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2545 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2546 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2547 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2548 (aarch64_print_operand, aarch64_print_address_internal)
2549 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2550 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2551 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2552 Handle polynomial GET_MODE_SIZE.
2553 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2554 wider than SImode without modification.
2555 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2556 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2557 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2558 passing and returning SVE modes.
2559 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2560 rather than GEN_INT.
2561 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2562 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2563 (aarch64_allocate_and_probe_stack_space): Likewise.
2564 (aarch64_layout_frame): Cope with polynomial offsets.
2565 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2566 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2568 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2569 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2570 poly_int64 rather than a HOST_WIDE_INT.
2571 (aarch64_get_separate_components, aarch64_process_components)
2572 (aarch64_expand_prologue, aarch64_expand_epilogue)
2573 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2574 (aarch64_anchor_offset): New function, split out from...
2575 (aarch64_legitimize_address): ...here.
2576 (aarch64_builtin_vectorization_cost): Handle polynomial
2577 TYPE_VECTOR_SUBPARTS.
2578 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2580 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2581 number of elements from the PARALLEL rather than the mode.
2582 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2583 rather than GET_MODE_BITSIZE.
2584 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2585 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2586 (aarch64_expand_vec_perm_const_1): Handle polynomial
2587 d->perm.length () and d->perm elements.
2588 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2589 Apply to_constant to d->perm elements.
2590 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2591 polynomial CONST_VECTOR_NUNITS.
2592 (aarch64_move_pointer): Take amount as a poly_int64 rather
2594 (aarch64_progress_pointer): Avoid temporary variable.
2595 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2596 the mode attribute instead of GET_MODE.
2598 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2599 Alan Hayward <alan.hayward@arm.com>
2600 David Sherwood <david.sherwood@arm.com>
2602 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2603 x exists before using it.
2604 (aarch64_add_constant_internal): Rename to...
2605 (aarch64_add_offset_1): ...this. Replace regnum with separate
2606 src and dest rtxes. Handle the case in which they're different,
2607 including when the offset is zero. Replace scratchreg with an rtx.
2608 Use 2 additions if there is no spare register into which we can
2609 move a 16-bit constant.
2610 (aarch64_add_constant): Delete.
2611 (aarch64_add_offset): Replace reg with separate src and dest
2612 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2613 Use aarch64_add_offset_1.
2614 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2615 an rtx rather than an int. Take the delta as a poly_int64
2616 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
2617 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
2618 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
2619 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
2620 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
2622 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
2623 aarch64_add_constant.
2625 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2627 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
2628 Use scalar_float_mode.
2630 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2632 * config/aarch64/aarch64-simd.md
2633 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
2634 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
2635 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
2636 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
2637 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
2638 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
2639 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
2640 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
2641 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
2642 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
2644 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2647 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
2648 targ_options->x_arm_arch_string is non NULL.
2650 2018-01-11 Tamar Christina <tamar.christina@arm.com>
2652 * config/aarch64/aarch64.h
2653 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
2655 2018-01-11 Sudakshina Das <sudi.das@arm.com>
2658 * expmed.c (emit_store_flag_force): Swap if const op0
2659 and change VOIDmode to mode of op0.
2661 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2663 PR rtl-optimization/83761
2664 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
2665 than bytes to mode_for_size.
2667 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
2670 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
2671 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
2674 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
2677 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
2678 when in layout mode.
2679 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
2680 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
2683 2018-01-10 Michael Collison <michael.collison@arm.com>
2685 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
2686 * config/aarch64/aarch64-option-extension.def: Add
2687 AARCH64_OPT_EXTENSION of 'fp16fml'.
2688 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2689 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
2690 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
2691 * config/aarch64/constraints.md (Ui7): New constraint.
2692 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
2693 (VFMLA_SEL_W): Ditto.
2696 (VFMLA16_LOW): New int iterator.
2697 (VFMLA16_HIGH): Ditto.
2698 (UNSPEC_FMLAL): New unspec.
2699 (UNSPEC_FMLSL): Ditto.
2700 (UNSPEC_FMLAL2): Ditto.
2701 (UNSPEC_FMLSL2): Ditto.
2702 (f16mac): New code attribute.
2703 * config/aarch64/aarch64-simd-builtins.def
2704 (aarch64_fmlal_lowv2sf): Ditto.
2705 (aarch64_fmlsl_lowv2sf): Ditto.
2706 (aarch64_fmlalq_lowv4sf): Ditto.
2707 (aarch64_fmlslq_lowv4sf): Ditto.
2708 (aarch64_fmlal_highv2sf): Ditto.
2709 (aarch64_fmlsl_highv2sf): Ditto.
2710 (aarch64_fmlalq_highv4sf): Ditto.
2711 (aarch64_fmlslq_highv4sf): Ditto.
2712 (aarch64_fmlal_lane_lowv2sf): Ditto.
2713 (aarch64_fmlsl_lane_lowv2sf): Ditto.
2714 (aarch64_fmlal_laneq_lowv2sf): Ditto.
2715 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
2716 (aarch64_fmlalq_lane_lowv4sf): Ditto.
2717 (aarch64_fmlsl_lane_lowv4sf): Ditto.
2718 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
2719 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
2720 (aarch64_fmlal_lane_highv2sf): Ditto.
2721 (aarch64_fmlsl_lane_highv2sf): Ditto.
2722 (aarch64_fmlal_laneq_highv2sf): Ditto.
2723 (aarch64_fmlsl_laneq_highv2sf): Ditto.
2724 (aarch64_fmlalq_lane_highv4sf): Ditto.
2725 (aarch64_fmlsl_lane_highv4sf): Ditto.
2726 (aarch64_fmlalq_laneq_highv4sf): Ditto.
2727 (aarch64_fmlsl_laneq_highv4sf): Ditto.
2728 * config/aarch64/aarch64-simd.md:
2729 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
2730 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
2731 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
2732 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
2733 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
2734 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
2735 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
2736 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
2737 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
2738 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
2739 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
2740 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
2741 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
2742 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
2743 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
2744 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
2745 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
2746 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
2747 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
2748 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
2749 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
2750 (vfmlsl_low_u32): Ditto.
2751 (vfmlalq_low_u32): Ditto.
2752 (vfmlslq_low_u32): Ditto.
2753 (vfmlal_high_u32): Ditto.
2754 (vfmlsl_high_u32): Ditto.
2755 (vfmlalq_high_u32): Ditto.
2756 (vfmlslq_high_u32): Ditto.
2757 (vfmlal_lane_low_u32): Ditto.
2758 (vfmlsl_lane_low_u32): Ditto.
2759 (vfmlal_laneq_low_u32): Ditto.
2760 (vfmlsl_laneq_low_u32): Ditto.
2761 (vfmlalq_lane_low_u32): Ditto.
2762 (vfmlslq_lane_low_u32): Ditto.
2763 (vfmlalq_laneq_low_u32): Ditto.
2764 (vfmlslq_laneq_low_u32): Ditto.
2765 (vfmlal_lane_high_u32): Ditto.
2766 (vfmlsl_lane_high_u32): Ditto.
2767 (vfmlal_laneq_high_u32): Ditto.
2768 (vfmlsl_laneq_high_u32): Ditto.
2769 (vfmlalq_lane_high_u32): Ditto.
2770 (vfmlslq_lane_high_u32): Ditto.
2771 (vfmlalq_laneq_high_u32): Ditto.
2772 (vfmlslq_laneq_high_u32): Ditto.
2773 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
2774 (AARCH64_FL_FOR_ARCH8_4): New.
2775 (AARCH64_ISA_F16FML): New ISA flag.
2776 (TARGET_F16FML): New feature flag for fp16fml.
2777 (doc/invoke.texi): Document new fp16fml option.
2779 2018-01-10 Michael Collison <michael.collison@arm.com>
2781 * config/aarch64/aarch64-builtins.c:
2782 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
2783 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2784 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
2785 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
2786 (AARCH64_ISA_SHA3): New ISA flag.
2787 (TARGET_SHA3): New feature flag for sha3.
2788 * config/aarch64/iterators.md (sha512_op): New int attribute.
2789 (CRYPTO_SHA512): New int iterator.
2790 (UNSPEC_SHA512H): New unspec.
2791 (UNSPEC_SHA512H2): Ditto.
2792 (UNSPEC_SHA512SU0): Ditto.
2793 (UNSPEC_SHA512SU1): Ditto.
2794 * config/aarch64/aarch64-simd-builtins.def
2795 (aarch64_crypto_sha512hqv2di): New builtin.
2796 (aarch64_crypto_sha512h2qv2di): Ditto.
2797 (aarch64_crypto_sha512su0qv2di): Ditto.
2798 (aarch64_crypto_sha512su1qv2di): Ditto.
2799 (aarch64_eor3qv8hi): Ditto.
2800 (aarch64_rax1qv2di): Ditto.
2801 (aarch64_xarqv2di): Ditto.
2802 (aarch64_bcaxqv8hi): Ditto.
2803 * config/aarch64/aarch64-simd.md:
2804 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
2805 (aarch64_crypto_sha512su0qv2di): Ditto.
2806 (aarch64_crypto_sha512su1qv2di): Ditto.
2807 (aarch64_eor3qv8hi): Ditto.
2808 (aarch64_rax1qv2di): Ditto.
2809 (aarch64_xarqv2di): Ditto.
2810 (aarch64_bcaxqv8hi): Ditto.
2811 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
2812 (vsha512h2q_u64): Ditto.
2813 (vsha512su0q_u64): Ditto.
2814 (vsha512su1q_u64): Ditto.
2815 (veor3q_u16): Ditto.
2816 (vrax1q_u64): Ditto.
2818 (vbcaxq_u16): Ditto.
2819 * config/arm/types.md (crypto_sha512): New type attribute.
2820 (crypto_sha3): Ditto.
2821 (doc/invoke.texi): Document new sha3 option.
2823 2018-01-10 Michael Collison <michael.collison@arm.com>
2825 * config/aarch64/aarch64-builtins.c:
2826 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
2827 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2828 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
2829 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
2830 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
2831 (AARCH64_ISA_SM4): New ISA flag.
2832 (TARGET_SM4): New feature flag for sm4.
2833 * config/aarch64/aarch64-simd-builtins.def
2834 (aarch64_sm3ss1qv4si): Ditto.
2835 (aarch64_sm3tt1aq4si): Ditto.
2836 (aarch64_sm3tt1bq4si): Ditto.
2837 (aarch64_sm3tt2aq4si): Ditto.
2838 (aarch64_sm3tt2bq4si): Ditto.
2839 (aarch64_sm3partw1qv4si): Ditto.
2840 (aarch64_sm3partw2qv4si): Ditto.
2841 (aarch64_sm4eqv4si): Ditto.
2842 (aarch64_sm4ekeyqv4si): Ditto.
2843 * config/aarch64/aarch64-simd.md:
2844 (aarch64_sm3ss1qv4si): Ditto.
2845 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
2846 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
2847 (aarch64_sm4eqv4si): Ditto.
2848 (aarch64_sm4ekeyqv4si): Ditto.
2849 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
2850 (sm3part_op): Ditto.
2851 (CRYPTO_SM3TT): Ditto.
2852 (CRYPTO_SM3PART): Ditto.
2853 (UNSPEC_SM3SS1): New unspec.
2854 (UNSPEC_SM3TT1A): Ditto.
2855 (UNSPEC_SM3TT1B): Ditto.
2856 (UNSPEC_SM3TT2A): Ditto.
2857 (UNSPEC_SM3TT2B): Ditto.
2858 (UNSPEC_SM3PARTW1): Ditto.
2859 (UNSPEC_SM3PARTW2): Ditto.
2860 (UNSPEC_SM4E): Ditto.
2861 (UNSPEC_SM4EKEY): Ditto.
2862 * config/aarch64/constraints.md (Ui2): New constraint.
2863 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
2864 * config/arm/types.md (crypto_sm3): New type attribute.
2865 (crypto_sm4): Ditto.
2866 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
2867 (vsm3tt1aq_u32): Ditto.
2868 (vsm3tt1bq_u32): Ditto.
2869 (vsm3tt2aq_u32): Ditto.
2870 (vsm3tt2bq_u32): Ditto.
2871 (vsm3partw1q_u32): Ditto.
2872 (vsm3partw2q_u32): Ditto.
2873 (vsm4eq_u32): Ditto.
2874 (vsm4ekeyq_u32): Ditto.
2875 (doc/invoke.texi): Document new sm4 option.
2877 2018-01-10 Michael Collison <michael.collison@arm.com>
2879 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
2880 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
2881 (AARCH64_FL_FOR_ARCH8_4): New.
2882 (AARCH64_FL_V8_4): New flag.
2883 (doc/invoke.texi): Document new armv8.4-a option.
2885 2018-01-10 Michael Collison <michael.collison@arm.com>
2887 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
2888 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
2889 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
2890 * config/aarch64/aarch64-option-extension.def: Add
2891 AARCH64_OPT_EXTENSION of 'sha2'.
2892 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
2893 (crypto): Disable sha2 and aes if crypto disabled.
2894 (crypto): Enable aes and sha2 if enabled.
2895 (simd): Disable sha2 and aes if simd disabled.
2896 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
2898 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
2899 (TARGET_SHA2): New feature flag for sha2.
2900 (TARGET_AES): New feature flag for aes.
2901 * config/aarch64/aarch64-simd.md:
2902 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
2903 conditional on TARGET_AES.
2904 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
2905 (aarch64_crypto_sha1hsi): Make pattern conditional
2907 (aarch64_crypto_sha1hv4si): Ditto.
2908 (aarch64_be_crypto_sha1hv4si): Ditto.
2909 (aarch64_crypto_sha1su1v4si): Ditto.
2910 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
2911 (aarch64_crypto_sha1su0v4si): Ditto.
2912 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
2913 (aarch64_crypto_sha256su0v4si): Ditto.
2914 (aarch64_crypto_sha256su1v4si): Ditto.
2915 (doc/invoke.texi): Document new aes and sha2 options.
2917 2018-01-10 Martin Sebor <msebor@redhat.com>
2919 PR tree-optimization/83781
2920 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
2923 2018-01-11 Martin Sebor <msebor@gmail.com>
2924 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2926 PR tree-optimization/83501
2927 PR tree-optimization/81703
2929 * tree-ssa-strlen.c (get_string_cst): Rename...
2930 (get_string_len): ...to this. Handle global constants.
2931 (handle_char_store): Adjust.
2933 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
2934 Jim Wilson <jimw@sifive.com>
2936 * config/riscv/riscv-protos.h (riscv_output_return): New.
2937 * config/riscv/riscv.c (struct machine_function): New naked_p field.
2938 (riscv_attribute_table, riscv_output_return),
2939 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
2940 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
2941 (riscv_compute_frame_info): Only compute frame->mask if not a naked
2943 (riscv_expand_prologue): Add early return for naked function.
2944 (riscv_expand_epilogue): Likewise.
2945 (riscv_function_ok_for_sibcall): Return false for naked function.
2946 (riscv_set_current_function): New.
2947 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
2948 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
2949 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
2950 * doc/extend.texi (RISC-V Function Attributes): New.
2952 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
2954 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
2955 check for 128-bit long double before checking TCmode.
2956 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
2957 128-bit long doubles before checking TFmode or TCmode.
2958 (FLOAT128_IBM_P): Likewise.
2960 2018-01-10 Martin Sebor <msebor@redhat.com>
2962 PR tree-optimization/83671
2963 * builtins.c (c_strlen): Unconditionally return zero for the empty
2965 Use -Warray-bounds for warnings.
2966 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
2967 for non-constant array indices with COMPONENT_REF, arrays of
2968 arrays, and pointers to arrays.
2969 (gimple_fold_builtin_strlen): Determine and set length range for
2970 non-constant character arrays.
2972 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
2975 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
2978 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
2980 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
2982 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
2985 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
2986 VECTOR_MEM_ALTIVEC_OR_VSX_P.
2987 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
2988 indexed_or_indirect_operand predicate.
2989 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
2990 (*vsx_le_perm_load_v8hi): Likewise.
2991 (*vsx_le_perm_load_v16qi): Likewise.
2992 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
2993 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
2994 (*vsx_le_perm_store_v8hi): Likewise.
2995 (*vsx_le_perm_store_v16qi): Likewise.
2996 (eight unnamed splitters): Likewise.
2998 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3000 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3001 * config/rs6000/emmintrin.h: Likewise.
3002 * config/rs6000/mmintrin.h: Likewise.
3003 * config/rs6000/xmmintrin.h: Likewise.
3005 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3008 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3010 * tree.c (tree_nop_conversion): Return true for location wrapper
3012 (maybe_wrap_with_location): New function.
3013 (selftest::check_strip_nops): New function.
3014 (selftest::test_location_wrappers): New function.
3015 (selftest::tree_c_tests): Call it.
3016 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3017 (maybe_wrap_with_location): New decl.
3018 (EXPR_LOCATION_WRAPPER_P): New macro.
3019 (location_wrapper_p): New inline function.
3020 (tree_strip_any_location_wrapper): New inline function.
3022 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3025 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3026 stack_realign_offset for the largest alignment of stack slot
3028 (ix86_find_max_used_stack_alignment): New function.
3029 (ix86_finalize_stack_frame_flags): Use it. Set
3030 max_used_stack_alignment if we don't realign stack.
3031 * config/i386/i386.h (machine_function): Add
3032 max_used_stack_alignment.
3034 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3036 * config/arm/arm.opt (-mbranch-cost): New option.
3037 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3040 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3043 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3044 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3046 2018-01-10 Richard Biener <rguenther@suse.de>
3049 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3050 early out so it also covers the case where we have a non-NULL
3053 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3055 PR tree-optimization/83753
3056 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3057 for non-strided grouped accesses if the number of elements is 1.
3059 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3062 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3063 * i386.h (TARGET_USE_GATHER): Define.
3064 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3066 2018-01-10 Martin Liska <mliska@suse.cz>
3069 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3070 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3072 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3073 CLEANUP_NO_PARTITIONING is not set.
3075 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3077 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3078 for vectors, as a partial revert of r254296.
3079 * rtl.h (const_vec_p): Delete.
3080 (const_vec_duplicate_p): Don't test for vector CONSTs.
3081 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3082 * expmed.c (make_tree): Likewise.
3085 * common.md (E, F): Use CONSTANT_P instead of checking for
3087 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3088 checking for CONST_VECTOR.
3090 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3093 * predict.c (force_edge_cold): Handle in more sane way edges
3096 2018-01-09 Carl Love <cel@us.ibm.com>
3098 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3100 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3101 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3102 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3103 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3104 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3105 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3106 * config/rs6000/rs6000-protos.h: Add extern defition for
3107 rs6000_generate_float2_double_code.
3108 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3110 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3111 (float2_v2df): Add define_expand.
3113 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3116 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3117 op_mode in the force_to_mode call.
3119 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3121 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3122 instead of checking each element individually.
3123 (aarch64_evpc_uzp): Likewise.
3124 (aarch64_evpc_zip): Likewise.
3125 (aarch64_evpc_ext): Likewise.
3126 (aarch64_evpc_rev): Likewise.
3127 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3128 instead of checking each element individually. Return true without
3130 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3131 whether all selected elements come from the same input, instead of
3132 checking each element individually. Remove calls to gen_rtx_REG,
3133 start_sequence and end_sequence and instead assert that no rtl is
3136 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3138 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3139 order of HIGH and CONST checks.
3141 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3143 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3144 if the destination isn't an SSA_NAME.
3146 2018-01-09 Richard Biener <rguenther@suse.de>
3148 PR tree-optimization/83668
3149 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3151 (canonicalize_loop_form): ... here, renamed from ...
3152 (canonicalize_loop_closed_ssa_form): ... this and amended to
3153 swap successor edges for loop exit blocks to make us use
3154 the RPO order we need for initial schedule generation.
3156 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3158 PR tree-optimization/64811
3159 * match.pd: When optimizing comparisons with Inf, avoid
3160 introducing or losing exceptions from comparisons with NaN.
3162 2018-01-09 Martin Liska <mliska@suse.cz>
3165 * asan.c (shadow_mem_size): Add gcc_assert.
3167 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3169 Don't save registers in main().
3172 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3173 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3174 * config/avr/avr.c (avr_set_current_function): Don't error if
3175 naked, OS_task or OS_main are specified at the same time.
3176 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3178 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3180 * common/config/avr/avr-common.c (avr_option_optimization_table):
3181 Switch on -mmain-is-OS_task for optimizing compilations.
3183 2018-01-09 Richard Biener <rguenther@suse.de>
3185 PR tree-optimization/83572
3186 * graphite.c: Include cfganal.h.
3187 (graphite_transform_loops): Connect infinite loops to exit
3188 and remove fake edges at the end.
3190 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3192 * ipa-inline.c (edge_badness): Revert accidental checkin.
3194 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3197 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3198 symbols; not inline clones.
3200 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3203 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3204 hard registers. Formatting fixes.
3206 PR preprocessor/83722
3207 * gcc.c (try_generate_repro): Pass
3208 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3209 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3212 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3213 Kito Cheng <kito.cheng@gmail.com>
3215 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3216 (riscv_leaf_function_p): Delete.
3217 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3219 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3221 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3223 (do_ifelse): New function.
3224 (do_isel): New function.
3225 (do_sub3): New function.
3226 (do_add3): New function.
3227 (do_load_mask_compare): New function.
3228 (do_overlap_load_compare): New function.
3229 (expand_compare_loop): New function.
3230 (expand_block_compare): Call expand_compare_loop() when appropriate.
3231 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3233 (-mblock-compare-inline-loop-limit): New option.
3235 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3238 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3239 Reverse order of second and third operands in first alternative.
3240 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3241 of first and second elements in UNSPEC_VPERMR vector.
3242 (altivec_expand_vec_perm_le): Likewise.
3244 2017-01-08 Jeff Law <law@redhat.com>
3246 PR rtl-optimizatin/81308
3247 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3248 (process_switch): If group_case_labels makes a change, then set
3250 (pass_convert_switch::execute): If a switch is converted, then
3251 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3253 PR rtl-optimization/81308
3254 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3257 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3259 PR target/83663 - Revert r255946
3260 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3261 generation for cases where splatting a value is not useful.
3262 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3263 across a vec_duplicate and a paradoxical subreg forming a vector
3264 mode to a vec_concat.
3266 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3268 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3269 -march=armv8.3-a variants.
3270 * config/arm/t-multilib: Likewise.
3271 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3273 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3275 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3277 (cceq_ior_compare_complement): Give it a name so I can use it, and
3278 change boolean_or_operator predicate to boolean_operator so it can
3279 be used to generate a crand.
3280 (eqne): New code iterator.
3281 (bd/bd_neg): New code_attrs.
3282 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3283 a single define_insn.
3284 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3285 decrement (bdnzt/bdnzf/bdzt/bdzf).
3286 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3287 with the new names of the branch decrement patterns, and added the
3288 names of the branch decrement conditional patterns.
3290 2018-01-08 Richard Biener <rguenther@suse.de>
3292 PR tree-optimization/83563
3293 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3296 2018-01-08 Richard Biener <rguenther@suse.de>
3299 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3301 2018-01-08 Richard Biener <rguenther@suse.de>
3303 PR tree-optimization/83685
3304 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3305 references to abnormals.
3307 2018-01-08 Richard Biener <rguenther@suse.de>
3310 * dwarf2out.c (output_indirect_strings): Handle empty
3311 skeleton_debug_str_hash.
3312 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3314 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3316 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3317 (emit_store_direct): Likewise.
3318 (arc_trampoline_adjust_address): Likewise.
3319 (arc_asm_trampoline_template): New function.
3320 (arc_initialize_trampoline): Use asm_trampoline_template.
3321 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3322 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3323 * config/arc/arc.md (flush_icache): Delete pattern.
3325 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3327 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3328 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3331 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3334 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3335 by not USED_FOR_TARGET.
3336 (make_pass_resolve_sw_modes): Likewise.
3338 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3340 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3343 2018-01-08 Richard Biener <rguenther@suse.de>
3346 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3348 2018-01-08 Richard Biener <rguenther@suse.de>
3351 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3353 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3356 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3357 basic blocks with a small number of successors.
3358 (convert_control_dep_chain_into_preds): Improve handling of
3360 (dump_predicates): Split apart into...
3361 (dump_pred_chain): ...here...
3362 (dump_pred_info): ...and here.
3363 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3364 (can_chain_union_be_invalidated_p): Improve check for invalidation
3366 (uninit_uses_cannot_happen): Avoid unnecessary if
3367 convert_control_dep_chain_into_preds yielded nothing.
3369 2018-01-06 Martin Sebor <msebor@redhat.com>
3371 PR tree-optimization/83640
3372 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3373 subtracting negative offset from size.
3374 (builtin_access::overlap): Adjust offset bounds of the access to fall
3375 within the size of the object if possible.
3377 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3379 PR rtl-optimization/83699
3380 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3381 extract_bit_field_as_subreg to cases in which the extracted
3382 value is also a vector.
3384 * lra-constraints.c (process_alt_operands): Test for the equivalence
3385 substitutions when detecting a possible reload cycle.
3387 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3390 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3391 by default if flag_selective_schedling{,2}. Formatting fixes.
3393 PR rtl-optimization/83682
3394 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3395 if it has non-VECTOR_MODE element mode.
3396 (vec_duplicate_p): Likewise.
3399 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3400 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3402 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3405 * config/i386/i386-builtin.def
3406 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3407 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3408 Require also OPTION_MASK_ISA_AVX512F in addition to
3409 OPTION_MASK_ISA_GFNI.
3410 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3411 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3412 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3413 to OPTION_MASK_ISA_GFNI.
3414 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3415 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3416 OPTION_MASK_ISA_AVX512BW.
3417 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3418 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3419 addition to OPTION_MASK_ISA_GFNI.
3420 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3421 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3422 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3423 to OPTION_MASK_ISA_GFNI.
3424 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3425 a requirement for all ISAs rather than any of them with a few
3427 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3429 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3430 bitmasks to be enabled with 3 exceptions, instead of requiring any
3431 enabled ISA with lots of exceptions.
3432 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3433 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3434 Change avx512bw in isa attribute to avx512f.
3435 * config/i386/sgxintrin.h: Add license boilerplate.
3436 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3437 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3438 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3439 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3441 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3442 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3443 temporarily sse2 rather than sse if not enabled already.
3446 * config/i386/sse.md (VI248_VLBW): Rename to ...
3447 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3448 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3449 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3450 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3451 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3452 mode iterator instead of VI248_VLBW.
3454 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3456 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3457 (record_modified): Skip clobbers; add debug output.
3458 (param_change_prob): Use sreal frequencies.
3460 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3462 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3463 punt for user-aligned variables.
3465 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3467 * tree-chrec.c (chrec_contains_symbols): Return true for
3470 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3473 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3474 of (x|y) == x for BICS pattern.
3476 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3478 PR tree-optimization/83605
3479 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3480 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3483 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3485 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3486 * config/epiphany/rtems.h: New file.
3488 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3489 Uros Bizjak <ubizjak@gmail.com>
3492 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3493 QIreg_operand instead of register_operand predicate.
3494 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3495 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3496 comments instead of -fmitigate[-_]rop.
3498 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3501 * cgraphunit.c (symbol_table::compile): Switch to text_section
3502 before calling assembly_start debug hook.
3503 * run-rtl-passes.c (run_rtl_passes): Likewise.
3506 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3508 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3509 range_int_cst_p rather than !symbolic_range_p before calling
3510 extract_range_from_multiplicative_op_1.
3512 2017-01-04 Jeff Law <law@redhat.com>
3514 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3515 redundant test in assertion.
3517 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3519 * doc/rtl.texi: Document machine_mode wrapper classes.
3521 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3523 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3526 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3528 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3529 the VEC_PERM_EXPR fold to fail.
3531 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3534 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3535 to switched_sections.
3537 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3540 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3543 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3546 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3547 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3549 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3552 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3553 is BLKmode and bitpos not zero or mode change is needed.
3555 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3558 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3561 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3564 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3565 instead of MULT rtx. Update all corresponding splitters.
3567 (*ssub<modesuffix>): Ditto.
3569 (*cmp_sadd_di): Update split patterns.
3570 (*cmp_sadd_si): Ditto.
3571 (*cmp_sadd_sidi): Ditto.
3572 (*cmp_ssub_di): Ditto.
3573 (*cmp_ssub_si): Ditto.
3574 (*cmp_ssub_sidi): Ditto.
3575 * config/alpha/predicates.md (const23_operand): New predicate.
3576 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3577 Look for ASHIFT, not MULT inner operand.
3578 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3580 2018-01-04 Martin Liska <mliska@suse.cz>
3582 PR gcov-profile/83669
3583 * gcov.c (output_intermediate_file): Add version to intermediate
3585 * doc/gcov.texi: Document new field 'version' in intermediate
3586 file format. Fix location of '-k' option of gcov command.
3588 2018-01-04 Martin Liska <mliska@suse.cz>
3591 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3593 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3595 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3597 2018-01-03 Martin Sebor <msebor@redhat.com>
3599 PR tree-optimization/83655
3600 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3601 checking calls with invalid arguments.
3603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3605 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3606 (vectorizable_mask_load_store): Delete.
3607 (vectorizable_call): Return false for masked loads and stores.
3608 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3609 instead of gimple_assign_rhs1.
3610 (vectorizable_load): Handle IFN_MASK_LOAD.
3611 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3613 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3615 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
3617 (vectorizable_mask_load_store): ...here.
3618 (vectorizable_load): ...and here.
3620 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3622 * tree-vect-stmts.c (vect_build_all_ones_mask)
3623 (vect_build_zero_merge_argument): New functions, split out from...
3624 (vectorizable_load): ...here.
3626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3628 * tree-vect-stmts.c (vect_check_store_rhs): New function,
3630 (vectorizable_mask_load_store): ...here.
3631 (vectorizable_store): ...and here.
3633 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3635 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
3637 (vectorizable_mask_load_store): ...here.
3639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3641 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
3642 (vect_model_store_cost): Take a vec_load_store_type instead of a
3644 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
3645 (vect_model_store_cost): Take a vec_load_store_type instead of a
3647 (vectorizable_mask_load_store): Update accordingly.
3648 (vectorizable_store): Likewise.
3649 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
3651 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3653 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
3654 IFN_MASK_LOAD calls here rather than...
3655 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
3657 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3658 Alan Hayward <alan.hayward@arm.com>
3659 David Sherwood <david.sherwood@arm.com>
3661 * expmed.c (extract_bit_field_1): For vector extracts,
3662 fall back to extract_bit_field_as_subreg if vec_extract
3665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3666 Alan Hayward <alan.hayward@arm.com>
3667 David Sherwood <david.sherwood@arm.com>
3669 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
3670 they are variable or constant sized.
3671 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
3672 slots for constant-sized data.
3674 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3675 Alan Hayward <alan.hayward@arm.com>
3676 David Sherwood <david.sherwood@arm.com>
3678 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
3679 handling COND_EXPRs with boolean comparisons, try to find a better
3680 basis for the mask type than the boolean itself.
3682 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3684 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
3685 is calculated and how it can be overridden.
3686 * genmodes.c (max_bitsize_mode_any_mode): New variable.
3687 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
3689 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
3692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3693 Alan Hayward <alan.hayward@arm.com>
3694 David Sherwood <david.sherwood@arm.com>
3696 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
3697 Remove the mode argument.
3698 (aarch64_simd_valid_immediate): Remove the mode and inverse
3700 * config/aarch64/iterators.md (bitsize): New iterator.
3701 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
3702 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
3703 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
3704 aarch64_simd_valid_immediate.
3705 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
3706 (aarch64_reg_or_bic_imm): Likewise.
3707 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
3708 with an insn_type enum and msl with a modifier_type enum.
3709 Replace element_width with a scalar_mode. Change the shift
3710 to unsigned int. Add constructors for scalar_float_mode and
3711 scalar_int_mode elements.
3712 (aarch64_vect_float_const_representable_p): Delete.
3713 (aarch64_can_const_movi_rtx_p)
3714 (aarch64_simd_scalar_immediate_valid_for_move)
3715 (aarch64_simd_make_constant): Update call to
3716 aarch64_simd_valid_immediate.
3717 (aarch64_advsimd_valid_immediate_hs): New function.
3718 (aarch64_advsimd_valid_immediate): Likewise.
3719 (aarch64_simd_valid_immediate): Remove mode and inverse
3720 arguments. Rewrite to use the above. Use const_vec_duplicate_p
3721 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
3722 and aarch64_float_const_representable_p on the result.
3723 (aarch64_output_simd_mov_immediate): Remove mode argument.
3724 Update call to aarch64_simd_valid_immediate and use of
3725 simd_immediate_info.
3726 (aarch64_output_scalar_simd_mov_immediate): Update call
3729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3730 Alan Hayward <alan.hayward@arm.com>
3731 David Sherwood <david.sherwood@arm.com>
3733 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
3734 (mode_nunits): Likewise CONST_MODE_NUNITS.
3735 * machmode.def (ADJUST_NUNITS): Document.
3736 * genmodes.c (mode_data::need_nunits_adj): New field.
3737 (blank_mode): Update accordingly.
3738 (adj_nunits): New variable.
3739 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
3741 (emit_mode_size_inline): Set need_bytesize_adj for all modes
3742 listed in adj_nunits.
3743 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
3744 listed in adj_nunits. Don't emit case statements for such modes.
3745 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
3746 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
3747 nothing if adj_nunits is nonnull.
3748 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
3749 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
3750 (emit_mode_fbit): Update use of print_maybe_const_decl.
3751 (emit_move_size): Likewise. Treat the array as non-const
3753 (emit_mode_adjustments): Handle adj_nunits.
3755 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3757 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
3758 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
3759 (VECTOR_MODES): Use it.
3760 (make_vector_modes): Take the prefix as an argument.
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 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
3767 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
3768 for MODE_VECTOR_BOOL.
3769 * machmode.def (VECTOR_BOOL_MODE): Document.
3770 * genmodes.c (VECTOR_BOOL_MODE): New macro.
3771 (make_vector_bool_mode): New function.
3772 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
3774 * lto-streamer-in.c (lto_input_mode_table): Likewise.
3775 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
3777 * stor-layout.c (int_mode_for_mode): Likewise.
3778 * tree.c (build_vector_type_for_mode): Likewise.
3779 * varasm.c (output_constant_pool_2): Likewise.
3780 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
3781 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
3782 for MODE_VECTOR_BOOL.
3783 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
3784 of mode class checks.
3785 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
3786 instead of a list of mode class checks.
3787 (expand_vector_scalar_condition): Likewise.
3788 (type_for_widest_vector_mode): Handle BImode as an inner mode.
3790 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3791 Alan Hayward <alan.hayward@arm.com>
3792 David Sherwood <david.sherwood@arm.com>
3794 * machmode.h (mode_size): Change from unsigned short to
3796 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
3797 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
3798 or if measurement_type is not polynomial.
3799 (fixed_size_mode::includes_p): Check for constant-sized modes.
3800 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
3801 return a poly_uint16 rather than an unsigned short.
3802 (emit_mode_size): Change the type of mode_size from unsigned short
3803 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
3804 (emit_mode_adjustments): Cope with polynomial vector sizes.
3805 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
3807 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
3809 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
3810 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
3811 * caller-save.c (setup_save_areas): Likewise.
3812 (replace_reg_with_saved_mem): Likewise.
3813 * calls.c (emit_library_call_value_1): Likewise.
3814 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
3815 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
3816 (gen_lowpart_for_combine): Likewise.
3817 * convert.c (convert_to_integer_1): Likewise.
3818 * cse.c (equiv_constant, cse_insn): Likewise.
3819 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
3820 (cselib_subst_to_values): Likewise.
3821 * dce.c (word_dce_process_block): Likewise.
3822 * df-problems.c (df_word_lr_mark_ref): Likewise.
3823 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
3824 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
3825 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
3826 (rtl_for_decl_location): Likewise.
3827 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
3828 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
3829 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
3830 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
3831 (expand_expr_real_1): Likewise.
3832 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
3833 (pad_below): Likewise.
3834 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
3835 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
3836 * ira.c (get_subreg_tracking_sizes): Likewise.
3837 * ira-build.c (ira_create_allocno_objects): Likewise.
3838 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
3839 (ira_sort_regnos_for_alter_reg): Likewise.
3840 * ira-costs.c (record_operand_costs): Likewise.
3841 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
3842 (resolve_simple_move): Likewise.
3843 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
3844 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
3845 (lra_constraints): Likewise.
3846 (CONST_POOL_OK_P): Reject variable-sized modes.
3847 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
3848 (add_pseudo_to_slot, lra_spill): Likewise.
3849 * omp-low.c (omp_clause_aligned_alignment): Likewise.
3850 * optabs-query.c (get_best_extraction_insn): Likewise.
3851 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
3852 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
3853 (expand_mult_highpart, valid_multiword_target_p): Likewise.
3854 * recog.c (offsettable_address_addr_space_p): Likewise.
3855 * regcprop.c (maybe_mode_change): Likewise.
3856 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
3857 * regrename.c (build_def_use): Likewise.
3858 * regstat.c (dump_reg_info): Likewise.
3859 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
3860 (find_reloads, find_reloads_subreg_address): Likewise.
3861 * reload1.c (eliminate_regs_1): Likewise.
3862 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
3863 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
3864 (simplify_binary_operation_1, simplify_subreg): Likewise.
3865 * targhooks.c (default_function_arg_padding): Likewise.
3866 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
3867 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
3868 (verify_gimple_assign_ternary): Likewise.
3869 * tree-inline.c (estimate_move_cost): Likewise.
3870 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3871 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
3872 (get_address_cost_ainc): Likewise.
3873 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
3874 (vect_supportable_dr_alignment): Likewise.
3875 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
3876 (vectorizable_reduction): Likewise.
3877 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
3878 (vectorizable_operation, vectorizable_load): Likewise.
3879 * tree.c (build_same_sized_truth_vector_type): Likewise.
3880 * valtrack.c (cleanup_auto_inc_dec): Likewise.
3881 * var-tracking.c (emit_note_insn_var_location): Likewise.
3882 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
3883 (ADDR_VEC_ALIGN): Likewise.
3885 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3886 Alan Hayward <alan.hayward@arm.com>
3887 David Sherwood <david.sherwood@arm.com>
3889 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
3891 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
3892 or if measurement_type is polynomial.
3893 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
3894 * combine.c (make_extraction): Likewise.
3895 * dse.c (find_shift_sequence): Likewise.
3896 * dwarf2out.c (mem_loc_descriptor): Likewise.
3897 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
3898 (extract_bit_field, extract_low_bits): Likewise.
3899 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
3900 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
3901 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
3902 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
3903 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
3904 * reload.c (find_reloads): Likewise.
3905 * reload1.c (alter_reg): Likewise.
3906 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
3907 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
3908 * tree-if-conv.c (predicate_mem_writes): Likewise.
3909 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
3910 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
3911 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
3912 * valtrack.c (dead_debug_insert_temp): Likewise.
3913 * varasm.c (mergeable_constant_section): Likewise.
3914 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
3916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3917 Alan Hayward <alan.hayward@arm.com>
3918 David Sherwood <david.sherwood@arm.com>
3920 * expr.c (expand_assignment): Cope with polynomial mode sizes
3921 when assigning to a CONCAT.
3923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3924 Alan Hayward <alan.hayward@arm.com>
3925 David Sherwood <david.sherwood@arm.com>
3927 * machmode.h (mode_precision): Change from unsigned short to
3929 (mode_to_precision): Return a poly_uint16 rather than an unsigned
3931 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
3932 or if measurement_type is not polynomial.
3933 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
3934 in which the mode is already known to be a scalar_int_mode.
3935 * genmodes.c (emit_mode_precision): Change the type of mode_precision
3936 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
3938 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
3939 for GET_MODE_PRECISION.
3940 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
3941 for GET_MODE_PRECISION.
3942 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
3944 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
3945 (expand_field_assignment, make_extraction): Likewise.
3946 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
3947 (get_last_value): Likewise.
3948 * convert.c (convert_to_integer_1): Likewise.
3949 * cse.c (cse_insn): Likewise.
3950 * expr.c (expand_expr_real_1): Likewise.
3951 * lra-constraints.c (simplify_operand_subreg): Likewise.
3952 * optabs-query.c (can_atomic_load_p): Likewise.
3953 * optabs.c (expand_atomic_load): Likewise.
3954 (expand_atomic_store): Likewise.
3955 * ree.c (combine_reaching_defs): Likewise.
3956 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
3957 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
3958 * tree.h (type_has_mode_precision_p): Likewise.
3959 * ubsan.c (instrument_si_overflow): Likewise.
3961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3962 Alan Hayward <alan.hayward@arm.com>
3963 David Sherwood <david.sherwood@arm.com>
3965 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
3966 polynomial numbers of units.
3967 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
3968 (valid_vector_subparts_p): New function.
3969 (build_vector_type): Remove temporary shim and take the number
3970 of units as a poly_uint64 rather than an int.
3971 (build_opaque_vector_type): Take the number of units as a
3972 poly_uint64 rather than an int.
3973 * tree.c (build_vector_from_ctor): Handle polynomial
3974 TYPE_VECTOR_SUBPARTS.
3975 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
3976 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
3977 (build_vector_from_val): If the number of units is variable,
3978 use build_vec_duplicate_cst for constant operands and
3979 VEC_DUPLICATE_EXPR otherwise.
3980 (make_vector_type): Remove temporary is_constant ().
3981 (build_vector_type, build_opaque_vector_type): Take the number of
3982 units as a poly_uint64 rather than an int.
3983 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
3985 * cfgexpand.c (expand_debug_expr): Likewise.
3986 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
3987 (store_constructor, expand_expr_real_1): Likewise.
3988 (const_scalar_mask_from_tree): Likewise.
3989 * fold-const-call.c (fold_const_reduction): Likewise.
3990 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
3991 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
3992 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
3993 (fold_relational_const): Likewise.
3994 (native_interpret_vector): Likewise. Change the size from an
3995 int to an unsigned int.
3996 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
3997 TYPE_VECTOR_SUBPARTS.
3998 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
3999 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4000 duplicating a non-constant operand into a variable-length vector.
4001 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4002 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4003 * ipa-icf.c (sem_variable::equals): Likewise.
4004 * match.pd: Likewise.
4005 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4006 * print-tree.c (print_node): Likewise.
4007 * stor-layout.c (layout_type): Likewise.
4008 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4009 * tree-cfg.c (verify_gimple_comparison): Likewise.
4010 (verify_gimple_assign_binary): Likewise.
4011 (verify_gimple_assign_ternary): Likewise.
4012 (verify_gimple_assign_single): Likewise.
4013 * tree-pretty-print.c (dump_generic_node): Likewise.
4014 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4015 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4016 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4017 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4018 (vect_shift_permute_load_chain): Likewise.
4019 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4020 (expand_vector_condition, optimize_vector_constructor): Likewise.
4021 (lower_vec_perm, get_compute_type): Likewise.
4022 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4023 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4024 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4025 (vect_recog_mask_conversion_pattern): Likewise.
4026 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4027 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4028 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4029 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4030 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4031 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4032 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4033 (supportable_widening_operation): Likewise.
4034 (supportable_narrowing_operation): Likewise.
4035 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4037 * varasm.c (output_constant): Likewise.
4039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4040 Alan Hayward <alan.hayward@arm.com>
4041 David Sherwood <david.sherwood@arm.com>
4043 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4044 so that both the length == 3 and length != 3 cases set up their
4045 own permute vectors. Add comments explaining why we know the
4046 number of elements is constant.
4047 (vect_permute_load_chain): Likewise.
4049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4050 Alan Hayward <alan.hayward@arm.com>
4051 David Sherwood <david.sherwood@arm.com>
4053 * machmode.h (mode_nunits): Change from unsigned char to
4055 (ONLY_FIXED_SIZE_MODES): New macro.
4056 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4057 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4058 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4060 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4061 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4062 or if measurement_type is not polynomial.
4063 * genmodes.c (ZERO_COEFFS): New macro.
4064 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4066 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4067 Use ZERO_COEFFS when emitting initializers.
4068 * data-streamer.h (bp_pack_poly_value): New function.
4069 (bp_unpack_poly_value): Likewise.
4070 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4071 for GET_MODE_NUNITS.
4072 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4073 for GET_MODE_NUNITS.
4074 * tree.c (make_vector_type): Remove temporary shim and make
4075 the real function take the number of units as a poly_uint64
4077 (build_vector_type_for_mode): Handle polynomial nunits.
4078 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4079 * emit-rtl.c (const_vec_series_p_1): Likewise.
4080 (gen_rtx_CONST_VECTOR): Likewise.
4081 * fold-const.c (test_vec_duplicate_folding): Likewise.
4082 * genrecog.c (validate_pattern): Likewise.
4083 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4084 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4085 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4086 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4087 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4088 * rtlanal.c (subreg_get_info): Likewise.
4089 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4090 (vect_grouped_load_supported): Likewise.
4091 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4092 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4093 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4094 (simplify_const_unary_operation, simplify_binary_operation_1)
4095 (simplify_const_binary_operation, simplify_ternary_operation)
4096 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4097 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4098 instead of CONST_VECTOR_NUNITS.
4099 * varasm.c (output_constant_pool_2): Likewise.
4100 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4101 explicit-encoded elements in the XVEC for variable-length vectors.
4103 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4105 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4108 Alan Hayward <alan.hayward@arm.com>
4109 David Sherwood <david.sherwood@arm.com>
4111 * coretypes.h (fixed_size_mode): Declare.
4112 (fixed_size_mode_pod): New typedef.
4113 * builtins.h (target_builtins::x_apply_args_mode)
4114 (target_builtins::x_apply_result_mode): Change type to
4115 fixed_size_mode_pod.
4116 * builtins.c (apply_args_size, apply_result_size, result_vector)
4117 (expand_builtin_apply_args_1, expand_builtin_apply)
4118 (expand_builtin_return): Update accordingly.
4120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4122 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4123 * cselib.c (cselib_hash_rtx): Likewise.
4124 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4125 CONST_VECTOR encoding.
4127 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4128 Jeff Law <law@redhat.com>
4131 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4132 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4133 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4134 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4137 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4138 explicitly probe *sp in a noreturn function if there were any callee
4139 register saves or frame pointer is needed.
4141 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4144 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4145 BLKmode for ternary, binary or unary expressions.
4148 * var-tracking.c (delete_vta_debug_insn): New inline function.
4149 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4150 insns from get_insns () to NULL instead of each bb separately.
4151 Use delete_vta_debug_insn. No longer static.
4152 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4153 delete_vta_debug_insns callers.
4154 * rtl.h (delete_vta_debug_insns): Declare.
4155 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4156 instead of variable_tracking_main.
4158 2018-01-03 Martin Sebor <msebor@redhat.com>
4160 PR tree-optimization/83603
4161 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4162 arguments past the endof the argument list in functions declared
4163 without a prototype.
4164 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4165 Avoid checking when arguments are null.
4167 2018-01-03 Martin Sebor <msebor@redhat.com>
4170 * doc/extend.texi (attribute const): Fix a typo.
4171 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4172 issuing -Wsuggest-attribute for void functions.
4174 2018-01-03 Martin Sebor <msebor@redhat.com>
4176 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4177 offset_int::from instead of wide_int::to_shwi.
4178 (maybe_diag_overlap): Remove assertion.
4179 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4180 * gimple-ssa-sprintf.c (format_directive): Same.
4181 (parse_directive): Same.
4182 (sprintf_dom_walker::compute_format_length): Same.
4183 (try_substitute_return_value): Same.
4185 2017-01-03 Jeff Law <law@redhat.com>
4188 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4189 non-constant residual for zero at runtime and avoid probing in
4190 that case. Reorganize code for trailing problem to mirror handling
4193 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4195 PR tree-optimization/83501
4196 * tree-ssa-strlen.c (get_string_cst): New.
4197 (handle_char_store): Call get_string_cst.
4199 2018-01-03 Martin Liska <mliska@suse.cz>
4201 PR tree-optimization/83593
4202 * tree-ssa-strlen.c: Include tree-cfg.h.
4203 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4204 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4205 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4207 (strlen_dom_walker::before_dom_children): Call
4208 gimple_purge_dead_eh_edges. Dump tranformation with details
4210 (strlen_dom_walker::before_dom_children): Update call by adding
4211 new argument cleanup_eh.
4212 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4214 2018-01-03 Martin Liska <mliska@suse.cz>
4217 * cif-code.def (VARIADIC_THUNK): New enum value.
4218 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4221 2018-01-03 Jan Beulich <jbeulich@suse.com>
4223 * sse.md (mov<mode>_internal): Tighten condition for when to use
4224 vmovdqu<ssescalarsize> for TI and OI modes.
4226 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4228 Update copyright years.
4230 2018-01-03 Martin Liska <mliska@suse.cz>
4233 * ipa-visibility.c (function_and_variable_visibility): Skip
4234 functions with noipa attribure.
4236 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4238 * gcc.c (process_command): Update copyright notice dates.
4239 * gcov-dump.c (print_version): Ditto.
4240 * gcov.c (print_version): Ditto.
4241 * gcov-tool.c (print_version): Ditto.
4242 * gengtype.c (create_file): Ditto.
4243 * doc/cpp.texi: Bump @copying's copyright year.
4244 * doc/cppinternals.texi: Ditto.
4245 * doc/gcc.texi: Ditto.
4246 * doc/gccint.texi: Ditto.
4247 * doc/gcov.texi: Ditto.
4248 * doc/install.texi: Ditto.
4249 * doc/invoke.texi: Ditto.
4251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4253 * vector-builder.h (vector_builder::m_full_nelts): Change from
4254 unsigned int to poly_uint64.
4255 (vector_builder::full_nelts): Update prototype accordingly.
4256 (vector_builder::new_vector): Likewise.
4257 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4258 (vector_builder::operator ==): Likewise.
4259 (vector_builder::finalize): Likewise.
4260 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4261 Take the number of elements as a poly_uint64 rather than an
4263 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4264 from unsigned int to poly_uint64.
4265 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4266 (vec_perm_indices::new_vector): Likewise.
4267 (vec_perm_indices::length): Likewise.
4268 (vec_perm_indices::nelts_per_input): Likewise.
4269 (vec_perm_indices::input_nelts): Likewise.
4270 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4271 number of elements per input as a poly_uint64 rather than an
4272 unsigned int. Use the original encoding for variable-length
4273 vectors, rather than clamping each individual element.
4274 For the second and subsequent elements in each pattern,
4275 clamp the step and base before clamping their sum.
4276 (vec_perm_indices::series_p): Handle polynomial element counts.
4277 (vec_perm_indices::all_in_range_p): Likewise.
4278 (vec_perm_indices_to_tree): Likewise.
4279 (vec_perm_indices_to_rtx): Likewise.
4280 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4281 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4282 (tree_vector_builder::new_binary_operation): Handle polynomial
4283 element counts. Return false if we need to know the number
4284 of elements at compile time.
4285 * fold-const.c (fold_vec_perm): Punt if the number of elements
4286 isn't known at compile time.
4288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4290 * vec-perm-indices.h (vec_perm_builder): Change element type
4291 from HOST_WIDE_INT to poly_int64.
4292 (vec_perm_indices::element_type): Update accordingly.
4293 (vec_perm_indices::clamp): Handle polynomial element_types.
4294 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4295 (vec_perm_indices::all_in_range_p): Likewise.
4296 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4298 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4299 polynomial vec_perm_indices element types.
4300 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4301 * fold-const.c (fold_vec_perm): Likewise.
4302 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4303 * tree-vect-generic.c (lower_vec_perm): Likewise.
4304 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4305 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4306 element type to HOST_WIDE_INT.
4308 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4309 Alan Hayward <alan.hayward@arm.com>
4310 David Sherwood <david.sherwood@arm.com>
4312 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4313 rather than an int. Use plus_constant.
4314 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4315 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4317 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4318 Alan Hayward <alan.hayward@arm.com>
4319 David Sherwood <david.sherwood@arm.com>
4321 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4322 a HOST_WIDE_INT to a poly_int64.
4324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4325 Alan Hayward <alan.hayward@arm.com>
4326 David Sherwood <david.sherwood@arm.com>
4328 * calls.c (load_register_parameters): Cope with polynomial
4329 mode sizes. Require a constant size for BLKmode parameters
4330 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4331 forces a parameter to be padded at the lsb end in order to
4332 fill a complete number of words, require the parameter size
4333 to be ordered wrt UNITS_PER_WORD.
4335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4336 Alan Hayward <alan.hayward@arm.com>
4337 David Sherwood <david.sherwood@arm.com>
4339 * reload1.c (spill_stack_slot_width): Change element type
4340 from unsigned int to poly_uint64_pod.
4341 (alter_reg): Treat mode sizes as polynomial.
4343 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4344 Alan Hayward <alan.hayward@arm.com>
4345 David Sherwood <david.sherwood@arm.com>
4347 * reload.c (complex_word_subreg_p): New function.
4348 (reload_inner_reg_of_subreg, push_reload): Use it.
4350 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4351 Alan Hayward <alan.hayward@arm.com>
4352 David Sherwood <david.sherwood@arm.com>
4354 * lra-constraints.c (process_alt_operands): Reject matched
4355 operands whose sizes aren't ordered.
4356 (match_reload): Refer to this check here.
4358 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4359 Alan Hayward <alan.hayward@arm.com>
4360 David Sherwood <david.sherwood@arm.com>
4362 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4363 that the mode size is in the set {1, 2, 4, 8, 16}.
4365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4366 Alan Hayward <alan.hayward@arm.com>
4367 David Sherwood <david.sherwood@arm.com>
4369 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4370 Use plus_constant instead of gen_rtx_PLUS.
4372 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4373 Alan Hayward <alan.hayward@arm.com>
4374 David Sherwood <david.sherwood@arm.com>
4376 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4377 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4378 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4379 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4380 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4381 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4382 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4383 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4384 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4385 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4387 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4388 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4389 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4390 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4391 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4392 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4393 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4394 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4395 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4396 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4398 * expr.c (emit_move_resolve_push): Treat the input and result
4399 of PUSH_ROUNDING as a poly_int64.
4400 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4401 (emit_push_insn): Likewise.
4402 * lra-eliminations.c (mark_not_eliminable): Likewise.
4403 * recog.c (push_operand): Likewise.
4404 * reload1.c (elimination_effects): Likewise.
4405 * rtlanal.c (nonzero_bits1): Likewise.
4406 * calls.c (store_one_arg): Likewise. Require the padding to be
4407 known at compile time.
4409 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4410 Alan Hayward <alan.hayward@arm.com>
4411 David Sherwood <david.sherwood@arm.com>
4413 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4414 Use plus_constant instead of gen_rtx_PLUS.
4416 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4417 Alan Hayward <alan.hayward@arm.com>
4418 David Sherwood <david.sherwood@arm.com>
4420 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4424 Alan Hayward <alan.hayward@arm.com>
4425 David Sherwood <david.sherwood@arm.com>
4427 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4428 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4429 via stack temporaries. Treat the mode size as polynomial too.
4431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4432 Alan Hayward <alan.hayward@arm.com>
4433 David Sherwood <david.sherwood@arm.com>
4435 * expr.c (expand_expr_real_2): When handling conversions involving
4436 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4437 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4438 as a poly_uint64 too.
4440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4441 Alan Hayward <alan.hayward@arm.com>
4442 David Sherwood <david.sherwood@arm.com>
4444 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4447 Alan Hayward <alan.hayward@arm.com>
4448 David Sherwood <david.sherwood@arm.com>
4450 * combine.c (can_change_dest_mode): Handle polynomial
4451 REGMODE_NATURAL_SIZE.
4452 * expmed.c (store_bit_field_1): Likewise.
4453 * expr.c (store_constructor): Likewise.
4454 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4455 and polynomial REGMODE_NATURAL_SIZE.
4456 (gen_lowpart_common): Likewise.
4457 * reginfo.c (record_subregs_of_mode): Likewise.
4458 * rtlanal.c (read_modify_subreg_p): Likewise.
4460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4461 Alan Hayward <alan.hayward@arm.com>
4462 David Sherwood <david.sherwood@arm.com>
4464 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4465 numbers of elements.
4467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4468 Alan Hayward <alan.hayward@arm.com>
4469 David Sherwood <david.sherwood@arm.com>
4471 * match.pd: Cope with polynomial numbers of vector elements.
4473 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4474 Alan Hayward <alan.hayward@arm.com>
4475 David Sherwood <david.sherwood@arm.com>
4477 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4478 in a POINTER_PLUS_EXPR.
4480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4481 Alan Hayward <alan.hayward@arm.com>
4482 David Sherwood <david.sherwood@arm.com>
4484 * omp-simd-clone.c (simd_clone_subparts): New function.
4485 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4486 (ipa_simd_modify_function_body): Likewise.
4488 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4489 Alan Hayward <alan.hayward@arm.com>
4490 David Sherwood <david.sherwood@arm.com>
4492 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4493 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4494 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4495 (expand_vector_condition, vector_element): Likewise.
4496 (subparts_gt): New function.
4497 (get_compute_type): Use subparts_gt.
4498 (count_type_subparts): Delete.
4499 (expand_vector_operations_1): Use subparts_gt instead of
4500 count_type_subparts.
4502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4503 Alan Hayward <alan.hayward@arm.com>
4504 David Sherwood <david.sherwood@arm.com>
4506 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4507 (vect_compile_time_alias): ...this new function. Do the calculation
4508 on poly_ints rather than trees.
4509 (vect_prune_runtime_alias_test_list): Update call accordingly.
4511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4512 Alan Hayward <alan.hayward@arm.com>
4513 David Sherwood <david.sherwood@arm.com>
4515 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4517 (vect_schedule_slp_instance): Likewise.
4519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4520 Alan Hayward <alan.hayward@arm.com>
4521 David Sherwood <david.sherwood@arm.com>
4523 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4524 constant and extern definitions for variable-length vectors.
4525 (vect_get_constant_vectors): Note that the number of units
4526 is known to be constant.
4528 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4529 Alan Hayward <alan.hayward@arm.com>
4530 David Sherwood <david.sherwood@arm.com>
4532 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4533 of units as polynomial. Choose between WIDE and NARROW based
4536 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4537 Alan Hayward <alan.hayward@arm.com>
4538 David Sherwood <david.sherwood@arm.com>
4540 * tree-vect-stmts.c (simd_clone_subparts): New function.
4541 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4544 Alan Hayward <alan.hayward@arm.com>
4545 David Sherwood <david.sherwood@arm.com>
4547 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4548 vectors as polynomial. Use build_index_vector for
4551 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4552 Alan Hayward <alan.hayward@arm.com>
4553 David Sherwood <david.sherwood@arm.com>
4555 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4556 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4557 for variable-length vectors.
4558 (vectorizable_mask_load_store): Treat the number of units as
4559 polynomial, asserting that it is constant if the condition has
4560 already been enforced.
4561 (vectorizable_store, vectorizable_load): Likewise.
4563 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4564 Alan Hayward <alan.hayward@arm.com>
4565 David Sherwood <david.sherwood@arm.com>
4567 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4568 of units as polynomial. Punt if we can't tell at compile time
4569 which vector contains the final result.
4571 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4572 Alan Hayward <alan.hayward@arm.com>
4573 David Sherwood <david.sherwood@arm.com>
4575 * tree-vect-loop.c (vectorizable_induction): Treat the number
4576 of units as polynomial. Punt on SLP inductions. Use an integer
4577 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4578 cast of such a series for variable-length floating-point
4581 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4582 Alan Hayward <alan.hayward@arm.com>
4583 David Sherwood <david.sherwood@arm.com>
4585 * tree.h (build_index_vector): Declare.
4586 * tree.c (build_index_vector): New function.
4587 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4588 of units as polynomial, forcibly converting it to a constant if
4589 vectorizable_reduction has already enforced the condition.
4590 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4591 to create a {1,2,3,...} vector.
4592 (vectorizable_reduction): Treat the number of units as polynomial.
4593 Choose vectype_in based on the largest scalar element size rather
4594 than the smallest number of units. Enforce the restrictions
4597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4598 Alan Hayward <alan.hayward@arm.com>
4599 David Sherwood <david.sherwood@arm.com>
4601 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4602 number of units as polynomial.
4604 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4605 Alan Hayward <alan.hayward@arm.com>
4606 David Sherwood <david.sherwood@arm.com>
4608 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4609 * target.def (autovectorize_vector_sizes): Return the vector sizes
4610 by pointer, using vector_sizes rather than a bitmask.
4611 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4612 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4613 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4615 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
4616 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
4617 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
4618 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
4619 * omp-general.c (omp_max_vf): Likewise.
4620 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4621 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
4622 * tree-vect-loop.c (vect_analyze_loop): Likewise.
4623 * tree-vect-slp.c (vect_slp_bb): Likewise.
4624 * doc/tm.texi: Regenerate.
4625 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
4627 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
4628 the vector size as a poly_uint64 rather than an unsigned int.
4629 (current_vector_size): Change from an unsigned int to a poly_uint64.
4630 (get_vectype_for_scalar_type): Update accordingly.
4631 * tree.h (build_truth_vector_type): Take the size and number of
4632 units as a poly_uint64 rather than an unsigned int.
4633 (build_vector_type): Add a temporary overload that takes
4634 the number of units as a poly_uint64 rather than an unsigned int.
4635 * tree.c (make_vector_type): Likewise.
4636 (build_truth_vector_type): Take the number of units as a poly_uint64
4637 rather than an unsigned int.
4639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4640 Alan Hayward <alan.hayward@arm.com>
4641 David Sherwood <david.sherwood@arm.com>
4643 * target.def (get_mask_mode): Take the number of units and length
4644 as poly_uint64s rather than unsigned ints.
4645 * targhooks.h (default_get_mask_mode): Update accordingly.
4646 * targhooks.c (default_get_mask_mode): Likewise.
4647 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
4648 * doc/tm.texi: Regenerate.
4650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4651 Alan Hayward <alan.hayward@arm.com>
4652 David Sherwood <david.sherwood@arm.com>
4654 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
4655 * omp-general.c (omp_max_vf): Likewise.
4656 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
4657 (expand_omp_simd): Handle polynomial safelen.
4658 * omp-low.c (omplow_simd_context): Add a default constructor.
4659 (omplow_simd_context::max_vf): Change from int to poly_uint64.
4660 (lower_rec_simd_input_clauses): Update accordingly.
4661 (lower_rec_input_clauses): Likewise.
4663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4664 Alan Hayward <alan.hayward@arm.com>
4665 David Sherwood <david.sherwood@arm.com>
4667 * tree-vectorizer.h (vect_nunits_for_cost): New function.
4668 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
4669 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
4670 (vect_analyze_slp_cost): Likewise.
4671 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
4672 (vect_model_load_cost): Likewise.
4674 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4675 Alan Hayward <alan.hayward@arm.com>
4676 David Sherwood <david.sherwood@arm.com>
4678 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
4679 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
4680 from an unsigned int * to a poly_uint64_pod *.
4681 (calculate_unrolling_factor): New function.
4682 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
4684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4685 Alan Hayward <alan.hayward@arm.com>
4686 David Sherwood <david.sherwood@arm.com>
4688 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
4689 from an unsigned int to a poly_uint64.
4690 (_loop_vec_info::slp_unrolling_factor): Likewise.
4691 (_loop_vec_info::vectorization_factor): Change from an int
4693 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
4694 (vect_get_num_vectors): New function.
4695 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
4696 (vect_get_num_copies): Use vect_get_num_vectors.
4697 (vect_analyze_data_ref_dependences): Change max_vf from an int *
4698 to an unsigned int *.
4699 (vect_analyze_data_refs): Change min_vf from an int * to a
4701 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
4702 than an unsigned HOST_WIDE_INT.
4703 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
4704 (vect_analyze_data_ref_dependence): Change max_vf from an int *
4705 to an unsigned int *.
4706 (vect_analyze_data_ref_dependences): Likewise.
4707 (vect_compute_data_ref_alignment): Handle polynomial vf.
4708 (vect_enhance_data_refs_alignment): Likewise.
4709 (vect_prune_runtime_alias_test_list): Likewise.
4710 (vect_shift_permute_load_chain): Likewise.
4711 (vect_supportable_dr_alignment): Likewise.
4712 (dependence_distance_ge_vf): Take the vectorization factor as a
4713 poly_uint64 rather than an unsigned HOST_WIDE_INT.
4714 (vect_analyze_data_refs): Change min_vf from an int * to a
4716 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
4717 vfm1 as a poly_uint64 rather than an int. Make the same change
4718 for the returned bound_scalar.
4719 (vect_gen_vector_loop_niters): Handle polynomial vf.
4720 (vect_do_peeling): Likewise. Update call to
4721 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
4722 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
4724 * tree-vect-loop.c (vect_determine_vectorization_factor)
4725 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
4726 (vect_get_known_peeling_cost): Likewise.
4727 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
4728 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
4729 (vect_transform_loop): Likewise. Use the lowest possible VF when
4730 updating the upper bounds of the loop.
4731 (vect_min_worthwhile_factor): Make static. Return an unsigned int
4733 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
4734 polynomial unroll factors.
4735 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
4736 (vect_make_slp_decision): Likewise.
4737 (vect_supported_load_permutation_p): Likewise, and polynomial
4739 (vect_analyze_slp_cost): Handle polynomial vf.
4740 (vect_slp_analyze_node_operations): Likewise.
4741 (vect_slp_analyze_bb_1): Likewise.
4742 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
4743 than an unsigned HOST_WIDE_INT.
4744 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
4745 (vectorizable_load): Handle polynomial vf.
4746 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
4748 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
4750 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4751 Alan Hayward <alan.hayward@arm.com>
4752 David Sherwood <david.sherwood@arm.com>
4754 * match.pd: Handle bit operations involving three constants
4755 and try to fold one pair.
4757 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4759 * tree-vect-loop-manip.c: Include gimple-fold.h.
4760 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
4761 niters_maybe_zero parameters. Handle other cases besides a step of 1.
4762 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
4763 Add a path that uses a step of VF instead of 1, but disable it
4765 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
4766 and niters_no_overflow parameters. Update calls to
4767 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
4768 Create a new SSA name if the latter choses to use a ste other
4769 than zero, and return it via niters_vector_mult_vf_var.
4770 * tree-vect-loop.c (vect_transform_loop): Update calls to
4771 vect_do_peeling, vect_gen_vector_loop_niters and
4772 slpeel_make_loop_iterate_ntimes.
4773 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
4774 (vect_gen_vector_loop_niters): Update declarations after above changes.
4776 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
4778 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
4779 128-bit round to integer instructions.
4780 (ceil<mode>2): Likewise.
4781 (btrunc<mode>2): Likewise.
4782 (round<mode>2): Likewise.
4784 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4786 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
4787 unaligned VSX load/store on P8/P9.
4788 (expand_block_clear): Allow the use of unaligned VSX
4789 load/store on P8/P9.
4791 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4793 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
4795 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
4796 swap associated with both a load and a store.
4798 2018-01-02 Andrew Waterman <andrew@sifive.com>
4800 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
4801 * config/riscv/riscv.md (clear_cache): Use it.
4803 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
4805 * web.c: Remove out-of-date comment.
4807 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4809 * expr.c (fixup_args_size_notes): Check that any existing
4810 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
4811 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
4812 (emit_single_push_insn): ...here.
4814 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4816 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
4817 (const_vector_encoded_nelts): New function.
4818 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
4819 (const_vector_int_elt, const_vector_elt): Declare.
4820 * emit-rtl.c (const_vector_int_elt_1): New function.
4821 (const_vector_elt): Likewise.
4822 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
4823 of CONST_VECTOR_ELT.
4825 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4827 * expr.c: Include rtx-vector-builder.h.
4828 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
4829 directly on the tree encoding.
4830 (const_vector_from_tree): Likewise.
4831 * optabs.c: Include rtx-vector-builder.h.
4832 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
4833 sequence of "u" values.
4834 * vec-perm-indices.c: Include rtx-vector-builder.h.
4835 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
4836 directly on the vec_perm_indices encoding.
4838 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4840 * doc/rtl.texi (const_vector): Describe new encoding scheme.
4841 * Makefile.in (OBJS): Add rtx-vector-builder.o.
4842 * rtx-vector-builder.h: New file.
4843 * rtx-vector-builder.c: Likewise.
4844 * rtl.h (rtx_def::u2): Add a const_vector field.
4845 (CONST_VECTOR_NPATTERNS): New macro.
4846 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
4847 (CONST_VECTOR_DUPLICATE_P): Likewise.
4848 (CONST_VECTOR_STEPPED_P): Likewise.
4849 (CONST_VECTOR_ENCODED_ELT): Likewise.
4850 (const_vec_duplicate_p): Check for a duplicated vector encoding.
4851 (unwrap_const_vec_duplicate): Likewise.
4852 (const_vec_series_p): Check for a non-duplicated vector encoding.
4853 Say that the function only returns true for integer vectors.
4854 * emit-rtl.c: Include rtx-vector-builder.h.
4855 (gen_const_vec_duplicate_1): Delete.
4856 (gen_const_vector): Call gen_const_vec_duplicate instead of
4857 gen_const_vec_duplicate_1.
4858 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
4859 (gen_const_vec_duplicate): Use rtx_vector_builder.
4860 (gen_const_vec_series): Likewise.
4861 (gen_rtx_CONST_VECTOR): Likewise.
4862 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
4863 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
4864 Build a new vector rather than modifying a CONST_VECTOR in-place.
4865 (handle_special_swappables): Update call accordingly.
4866 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
4867 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
4868 Build a new vector rather than modifying a CONST_VECTOR in-place.
4869 (handle_special_swappables): Update call accordingly.
4871 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4873 * simplify-rtx.c (simplify_const_binary_operation): Use
4874 CONST_VECTOR_ELT instead of XVECEXP.
4876 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4878 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
4879 the selector elements to be different from the data elements
4880 if the selector is a VECTOR_CST.
4881 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
4882 ssizetype for the selector.
4884 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4886 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
4887 before testing each element individually.
4888 * tree-vect-generic.c (lower_vec_perm): Likewise.
4890 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4892 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
4893 * selftest-run-tests.c (selftest::run_tests): Call it.
4894 * vector-builder.h (vector_builder::operator ==): New function.
4895 (vector_builder::operator !=): Likewise.
4896 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
4897 (vec_perm_indices::all_from_input_p): New function.
4898 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4899 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
4900 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
4901 instead of reading the VECTOR_CST directly. Detect whether both
4902 vector inputs are the same before constructing the vec_perm_indices,
4903 and update the number of inputs argument accordingly. Use the
4904 utility functions added above. Only construct sel2 if we need to.
4906 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4908 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
4909 the broadcast of the low byte.
4910 (expand_mult_highpart): Use an explicit encoding for the permutes.
4911 * optabs-query.c (can_mult_highpart_p): Likewise.
4912 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
4913 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4914 (vectorizable_bswap): Likewise.
4915 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
4916 explicit encoding for the power-of-2 permutes.
4917 (vect_permute_store_chain): Likewise.
4918 (vect_grouped_load_supported): Likewise.
4919 (vect_permute_load_chain): Likewise.
4921 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4923 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
4924 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
4925 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
4926 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4927 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
4928 (vect_gen_perm_mask_any): Likewise.
4930 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
4932 * int-vector-builder.h: New file.
4933 * vec-perm-indices.h: Include int-vector-builder.h.
4934 (vec_perm_indices): Redefine as an int_vector_builder.
4935 (auto_vec_perm_indices): Delete.
4936 (vec_perm_builder): Redefine as a stand-alone class.
4937 (vec_perm_indices::vec_perm_indices): New function.
4938 (vec_perm_indices::clamp): Likewise.
4939 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
4940 (vec_perm_indices::new_vector): New function.
4941 (vec_perm_indices::new_expanded_vector): Update for new
4942 vec_perm_indices class.
4943 (vec_perm_indices::rotate_inputs): New function.
4944 (vec_perm_indices::all_in_range_p): Operate directly on the
4945 encoded form, without computing elided elements.
4946 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
4947 encoding. Update for new vec_perm_indices class.
4948 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
4949 the given vec_perm_builder.
4950 (expand_vec_perm_var): Update vec_perm_builder constructor.
4951 (expand_mult_highpart): Use vec_perm_builder instead of
4952 auto_vec_perm_indices.
4953 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
4954 vec_perm_indices instead of auto_vec_perm_indices. Use a single
4955 or double series encoding as appropriate.
4956 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
4957 vec_perm_indices instead of auto_vec_perm_indices.
4958 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4959 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4960 (vect_permute_store_chain): Likewise.
4961 (vect_grouped_load_supported): Likewise.
4962 (vect_permute_load_chain): Likewise.
4963 (vect_shift_permute_load_chain): Likewise.
4964 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
4965 (vect_transform_slp_perm_load): Likewise.
4966 (vect_schedule_slp_instance): Likewise.
4967 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4968 (vectorizable_mask_load_store): Likewise.
4969 (vectorizable_bswap): Likewise.
4970 (vectorizable_store): Likewise.
4971 (vectorizable_load): Likewise.
4972 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
4973 vec_perm_indices instead of auto_vec_perm_indices. Use
4974 tree_to_vec_perm_builder to read the vector from a tree.
4975 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
4976 vec_perm_builder instead of a vec_perm_indices.
4977 (have_whole_vector_shift): Use vec_perm_builder and
4978 vec_perm_indices instead of auto_vec_perm_indices. Leave the
4979 truncation to calc_vec_perm_mask_for_shift.
4980 (vect_create_epilog_for_reduction): Likewise.
4981 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
4982 from auto_vec_perm_indices to vec_perm_indices.
4983 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
4984 instead of changing individual elements.
4985 (aarch64_vectorize_vec_perm_const): Use new_vector to install
4986 the vector in d.perm.
4987 * config/arm/arm.c (expand_vec_perm_d::perm): Change
4988 from auto_vec_perm_indices to vec_perm_indices.
4989 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
4990 instead of changing individual elements.
4991 (arm_vectorize_vec_perm_const): Use new_vector to install
4992 the vector in d.perm.
4993 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
4994 Update vec_perm_builder constructor.
4995 (rs6000_expand_interleave): Likewise.
4996 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
4997 (rs6000_expand_interleave): Likewise.
4999 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5001 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5002 to qimode could truncate the indices.
5003 * optabs.c (expand_vec_perm_var): Likewise.
5005 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5007 * Makefile.in (OBJS): Add vec-perm-indices.o.
5008 * vec-perm-indices.h: New file.
5009 * vec-perm-indices.c: Likewise.
5010 * target.h (vec_perm_indices): Replace with a forward class
5012 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5013 * optabs.h: Include vec-perm-indices.h.
5014 (expand_vec_perm): Delete.
5015 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5016 (expand_vec_perm_const): Declare.
5017 * target.def (vec_perm_const_ok): Replace with...
5018 (vec_perm_const): ...this new hook.
5019 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5020 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5021 * doc/tm.texi: Regenerate.
5022 * optabs.def (vec_perm_const): Delete.
5023 * doc/md.texi (vec_perm_const): Likewise.
5024 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5025 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5026 expand_vec_perm for constant permutation vectors. Assert that
5027 the mode of variable permutation vectors is the integer equivalent
5028 of the mode that is being permuted.
5029 * optabs-query.h (selector_fits_mode_p): Declare.
5030 * optabs-query.c: Include vec-perm-indices.h.
5031 (selector_fits_mode_p): New function.
5032 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5033 is defined, instead of checking whether the vec_perm_const_optab
5034 exists. Use targetm.vectorize.vec_perm_const instead of
5035 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5036 fit in the vector mode before using a variable permute.
5037 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5038 vec_perm_indices instead of an rtx.
5039 (expand_vec_perm): Replace with...
5040 (expand_vec_perm_const): ...this new function. Take the selector
5041 as a vec_perm_indices rather than an rtx. Also take the mode of
5042 the selector. Update call to shift_amt_for_vec_perm_mask.
5043 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5044 Use vec_perm_indices::new_expanded_vector to expand the original
5045 selector into bytes. Check whether the indices fit in the vector
5046 mode before using a variable permute.
5047 (expand_vec_perm_var): Make global.
5048 (expand_mult_highpart): Use expand_vec_perm_const.
5049 * fold-const.c: Includes vec-perm-indices.h.
5050 * tree-ssa-forwprop.c: Likewise.
5051 * tree-vect-data-refs.c: Likewise.
5052 * tree-vect-generic.c: Likewise.
5053 * tree-vect-loop.c: Likewise.
5054 * tree-vect-slp.c: Likewise.
5055 * tree-vect-stmts.c: Likewise.
5056 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5058 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5059 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5060 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5061 (aarch64_vectorize_vec_perm_const): ...this new function.
5062 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5063 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5064 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5065 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5066 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5067 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5068 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5070 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5071 check for NEON modes.
5072 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5073 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5074 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5075 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5077 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5078 the old VEC_PERM_CONST conditions.
5079 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5080 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5081 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5082 (ia64_vectorize_vec_perm_const_ok): Merge into...
5083 (ia64_vectorize_vec_perm_const): ...this new function.
5084 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5085 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5086 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5087 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5088 * config/mips/mips.c (mips_expand_vec_perm_const)
5089 (mips_vectorize_vec_perm_const_ok): Merge into...
5090 (mips_vectorize_vec_perm_const): ...this new function.
5091 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5092 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5093 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5094 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5095 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5096 (rs6000_expand_vec_perm_const): Delete.
5097 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5099 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5100 (altivec_expand_vec_perm_const_le): Take each operand individually.
5101 Operate on constant selectors rather than rtxes.
5102 (altivec_expand_vec_perm_const): Likewise. Update call to
5103 altivec_expand_vec_perm_const_le.
5104 (rs6000_expand_vec_perm_const): Delete.
5105 (rs6000_vectorize_vec_perm_const_ok): Delete.
5106 (rs6000_vectorize_vec_perm_const): New function.
5107 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5108 an element count and rtx array.
5109 (rs6000_expand_extract_even): Update call accordingly.
5110 (rs6000_expand_interleave): Likewise.
5111 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5112 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5113 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5114 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5115 (rs6000_expand_vec_perm_const): Delete.
5116 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5117 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5118 (altivec_expand_vec_perm_const_le): Take each operand individually.
5119 Operate on constant selectors rather than rtxes.
5120 (altivec_expand_vec_perm_const): Likewise. Update call to
5121 altivec_expand_vec_perm_const_le.
5122 (rs6000_expand_vec_perm_const): Delete.
5123 (rs6000_vectorize_vec_perm_const_ok): Delete.
5124 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5125 reference to the SPE evmerge intructions.
5126 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5127 an element count and rtx array.
5128 (rs6000_expand_extract_even): Update call accordingly.
5129 (rs6000_expand_interleave): Likewise.
5130 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5131 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5133 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5135 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5137 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5138 vector mode and that that mode matches the mode of the data
5140 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5141 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5142 directly using expand_vec_perm_1 when forcing selectors into
5144 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5146 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5148 * optabs-query.h (can_vec_perm_p): Delete.
5149 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5150 * optabs-query.c (can_vec_perm_p): Split into...
5151 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5152 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5153 particular selector is valid.
5154 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5155 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5156 (vect_grouped_load_supported): Likewise.
5157 (vect_shift_permute_load_chain): Likewise.
5158 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5159 (vect_transform_slp_perm_load): Likewise.
5160 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5161 (vectorizable_bswap): Likewise.
5162 (vect_gen_perm_mask_checked): Likewise.
5163 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5164 implementations of variable permutation vectors into account
5165 when deciding which selector to use.
5166 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5167 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5168 with a false third argument.
5169 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5170 to test whether the constant selector is valid and can_vec_perm_var_p
5171 to test whether a variable selector is valid.
5173 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5175 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5176 * optabs-query.c (can_vec_perm_p): Likewise.
5177 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5178 instead of vec_perm_indices.
5179 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5180 (vect_gen_perm_mask_checked): Likewise,
5181 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5182 (vect_gen_perm_mask_checked): Likewise,
5184 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5186 * optabs-query.h (qimode_for_vec_perm): Declare.
5187 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5188 (qimode_for_vec_perm): ...this new function.
5189 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5191 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5193 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5194 does not have a conditional at the top.
5196 2018-01-02 Richard Biener <rguenther@suse.de>
5198 * ipa-inline.c (big_speedup_p): Fix expression.
5200 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5203 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5206 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5210 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5211 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5212 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5213 cond_taken_branch_cost 3->4.
5215 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5217 PR tree-optimization/83581
5218 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5219 TODO_cleanup_cfg if any changes have been made.
5222 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5223 convert_modes if target mode has the right side, but different mode
5227 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5228 last argument when extracting from CONCAT. If either from_real or
5229 from_imag is NULL, use expansion through memory. If result is not
5230 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5231 the parts directly to inner mode, if even that fails, use expansion
5235 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5236 check for bswap in mode rather than HImode and use that in expand_unop
5239 Copyright (C) 2018 Free Software Foundation, Inc.
5241 Copying and distribution of this file, with or without modification,
5242 are permitted in any medium without royalty provided the copyright
5243 notice and this notice are preserved.