1 2018-02-27 Richard Biener <rguenther@suse.de>
3 PR tree-optimization/84512
4 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
5 Do not use the estimate returned from record_stmt_cost for
6 the scalar iteration cost but sum properly using add_stmt_cost.
8 2018-02-27 Richard Biener <rguenther@suse.de>
10 PR tree-optimization/84466
11 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
12 Adjust last change to less strictly validate use operands.
14 2018-02-27 Martin Liska <mliska@suse.cz>
17 * gcov.c (process_file): Allow partial overlap and consider it
18 also as group functions.
19 (output_lines): Properly calculate range of lines for a group.
21 2018-02-27 Martin Liska <mliska@suse.cz>
23 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
24 'ggc' suffixes. Change first column width.
25 (timer::print): Fix formatting of the column.
27 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
29 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
30 preserve inline entry blocks for the sake of debug inline
31 entry point markers alone.
32 (remove_unused_locals): Suggest in comments a better place to
33 force the preservation of inline entry blocks that are
34 otherwise unused, but do not preserve them.
36 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
38 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
40 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
43 * config/i386/constraints.md (Bs): Replace
44 ix86_indirect_branch_register with
45 TARGET_INDIRECT_BRANCH_REGISTER.
47 * config/i386/i386.md (indirect_jump): Likewise.
48 (tablejump): Likewise.
49 (*sibcall_memory): Likewise.
50 (*sibcall_value_memory): Likewise.
51 Peepholes of indirect call and jump via memory: Likewise.
52 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
53 (*sibcall_value_GOT_32): Likewise.
54 * config/i386/i386.opt: Likewise.
55 * config/i386/predicates.md (indirect_branch_operand): Likewise.
56 (GOT_memory_operand): Likewise.
57 (call_insn_operand): Likewise.
58 (sibcall_insn_operand): Likewise.
59 (GOT32_symbol_operand): Likewise.
60 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
62 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
64 PR rtl-optimization/83496
65 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
66 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
67 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
68 redundant insn, if any.
69 (relax_delay_slots): Likewise.
70 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
72 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
74 PR tree-optimization/83965
75 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
76 that grouped statements are part of a reduction chain. Return
77 true if the statement is not marked as a reduction itself but
79 (vect_recog_dot_prod_pattern): Don't check whether the statement
80 is part of a group here.
81 (vect_recog_sad_pattern): Likewise.
82 (vect_recog_widen_sum_pattern): Likewise.
84 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
87 * final.c (rest_of_clean_state): Also look for calls inside sequences.
89 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
92 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
94 (ix86_output_indirect_function_return): New prototype.
95 (ix86_split_simple_return_pop_internal): Likewise.
96 * config/i386/i386.c (indirect_return_via_cx): New.
97 (indirect_return_via_cx_bnd): Likewise.
98 (indirect_thunk_name): Handle return va CX_REG.
99 (output_indirect_thunk_function): Create alias for
100 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
101 (ix86_output_indirect_jmp): Remove the bool argument.
102 (ix86_output_indirect_function_return): New function.
103 (ix86_split_simple_return_pop_internal): Likewise.
104 * config/i386/i386.md (*indirect_jump): Don't pass false
105 to ix86_output_indirect_jmp.
106 (*tablejump_1): Likewise.
107 (simple_return_pop_internal): Change it to define_insn_and_split.
108 Call ix86_split_simple_return_pop_internal to split it for
110 (simple_return_indirect_internal): Call
111 ix86_output_indirect_function_return instead of
112 ix86_output_indirect_jmp.
114 2018-02-26 Jakub Jelinek <jakub@redhat.com>
117 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
118 memset and value initialization afterwards.
120 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
122 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
124 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
127 * common/config/aarch64/aarch64-common.c
128 (aarch_option_optimization_table[]): Switch
129 off fomit-frame-pointer
131 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
132 Chung-Ju Wu <jasonwucj@gmail.com>
134 * config/nds32/nds32-multiple.md (load_multiple): Disallow
136 (store_multiple): Ditto.
138 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
140 * config.gcc: Add --with-cpu support for nds32 target.
141 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
142 * config/nds32/nds32.opt: Add -mcpu= option.
144 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
146 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
147 isel=yes): Warn for these deprecated options.
149 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
151 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
152 ISA_2_5_MASKS_EMBEDDED.
154 2018-02-23 Jakub Jelinek <jakub@redhat.com>
156 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
157 p->max as pointers rather than using iterative_hash_expr.
159 2018-02-23 Carl Love <cel@us.ibm.com>
161 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
162 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
164 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
165 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
166 P8V_BUILTIN_VEC_VUNSIGNED2.
168 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
171 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
172 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
174 * lra-constraints.c (curr_insn_transform): Set up
175 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
177 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
178 * lra-eliminations.c (spill_pseudos): Ditto.
179 (process_insn_for_elimination): Ditto.
180 * lra-lives.c (reg_early_clobber_p): Use the new macros.
181 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
182 LRA_NON_CLOBBERED_ALT.
184 2018-02-22 Martin Sebor <msebor@redhat.com>
186 PR tree-optimization/84480
187 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
188 to maybe_diag_stxncpy_trunc. Call it.
189 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
190 from gimple_fold_builtin_strcpy. Print inlining stack.
191 (handle_builtin_stxncpy): Print inlining stack.
192 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
194 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
197 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
198 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
199 and -fcheck-pointer-bounds are used together.
200 (indirect_thunk_prefix): New enum.
201 (indirect_thunk_need_prefix): New function.
202 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
203 "_nt" instead of "_bnd" for NOTRACK prefix.
204 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
205 (output_indirect_thunk_function): Likewise.
207 (ix86_code_end): Update output_indirect_thunk_function calls.
208 (ix86_output_indirect_branch_via_reg): Replace
209 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
210 (ix86_output_indirect_branch_via_push): Likewise.
211 (ix86_output_function_return): Likewise.
212 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
213 incompatible with -fcf-protection=branch and
214 -fcheck-pointer-bounds.
216 2018-02-22 Steve Ellcey <sellcey@cavium.com>
219 * config/aarch64/aarch64.c (aarch64_print_address_internal):
220 Change gcc_assert call to output_operand_lossage.
222 2018-02-22 Steve Ellcey <sellcey@cavium.com>
224 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
226 2018-02-22 DJ Delorie <dj@redhat.com>
227 Sebastian Perta <sebastian.perta@renesas.com>
228 Oleg Endo <olegendo@gcc.gnu.org>
230 * config/rx/rx.c (rx_rtx_costs): New function.
231 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
233 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
235 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
237 2018-02-22 Martin Liska <mliska@suse.cz>
240 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
241 Add "native" as a possible value.
243 2018-02-22 Martin Liska <mliska@suse.cz>
246 * config/i386/i386.c (ix86_option_override_internal):
247 Add "native" as a possible value for -march and -mtune.
249 2018-02-22 Jakub Jelinek <jakub@redhat.com>
252 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
253 to all type variants.
255 PR tree-optimization/84503
256 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
257 width as info->bitpos + info->bitsize - start.
258 (merged_store_group::merge_overlapping): Simplify width computation.
259 (check_no_overlap): New function.
260 (imm_store_chain_info::try_coalesce_bswap): Compute expected
261 start + width and last_order of the group, fail if check_no_overlap
263 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
264 to group if check_no_overlap fails.
266 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
268 * config/rs6000/altivec.md: Delete contraint arguments to
269 define_expand, define_split, and define_peephole2, and in
270 define_insn_and_split if always unused.
271 * config/rs6000/darwin.md: Ditto.
272 * config/rs6000/dfp.md: Ditto.
273 * config/rs6000/rs6000.md: Ditto.
274 * config/rs6000/sync.md: Ditto.
275 * config/rs6000/vector.md: Ditto.
276 * config/rs6000/vsx.md: Ditto.
278 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
280 * config/rs6000/altivec.md: Write output control strings as braced
281 blocks instead of double-quoted strings.
282 * config/rs6000/darwin.md: Ditto.
283 * config/rs6000/rs6000.md: Ditto.
284 * config/rs6000/vector.md: Ditto.
285 * config/rs6000/vsx.md: Ditto.
287 2018-02-21 Jason Merrill <jason@redhat.com>
289 PR c++/84314 - ICE with templates and fastcall attribute.
290 * attribs.c (build_type_attribute_qual_variant): Remove assert.
292 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
294 * ipa-cp.c (determine_versionability): Fix comment typos.
296 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
299 * ipa-cp.c (determine_versionability): Do not version functions caling
302 2018-02-21 Martin Liska <mliska@suse.cz>
305 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
306 Add "native" as a possible value.
307 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
308 the macro when native cpu detection is available.
310 2018-02-21 Martin Liska <mliska@suse.cz>
313 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
314 Add "native" as a possible value.
315 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
316 when native cpu detection is available.
318 2018-02-21 Jakub Jelinek <jakub@redhat.com>
319 Martin Sebor <msebor@redhat.com>
321 PR tree-optimization/84478
322 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
324 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
325 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
326 support which is conservatively correct, for 2 only stay conservative
327 for maxlen. Formatting and comment capitalization fixes. Add STRICT
328 argument to the 2 argument get_range_strlen, adjust 6 arg
329 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
331 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
332 (gimple_fold_builtin_strlen): Pass true as last argument to
335 2018-02-20 Martin Sebor <msebor@redhat.com>
338 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
339 (builtin_memref::set_base_and_offset): Same. Handle inner references.
340 (builtin_memref::builtin_memref): Factor out parts into
341 set_base_and_offset and call it.
343 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
346 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
347 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
348 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
349 search at the associated MODE_INT.
351 2018-02-20 Jeff Law <law@redhat.com>
354 PR tree-optimization/81592
356 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
357 for range data rather than using global data.
358 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
359 range data rather than using global data.
360 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
361 pass it to children as needed.
362 (struct directive::fmtresult): Similarly.
363 (struct directive::set_width): Similarly.
364 (struct directive::set_precision): Similarly.
365 (format_integer, format_directive, parse_directive): Similarly.
366 (format_none): Accept unnamed vr_values parameter.
367 (format_percent, format_floating, format_character): Similarly.
368 (format_string, format_plain): Similarly.
369 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
370 the EVRP range analyzer for range data rather than using global data.
371 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
372 gimple-ssa-evrp-analyze.h
373 (class sprintf_dom_walker): Add after_dom_children member function.
374 Add evrp_range_analyzer member.
375 (sprintf_dom_walker::before_dom_children): Call into the EVRP
376 range analyzer as needed.
377 (sprintf_dom_walker::after_dom_children): New member function.
378 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
380 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
381 (evrp_range_analyzer::pop_to_marker): Likewise.
383 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
385 PR tree-optimization/84419
386 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
387 with the required type if its current type is compatible but
390 2018-02-20 Jakub Jelinek <jakub@redhat.com>
393 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
396 2018-02-20 Martin Liska <mliska@suse.cz>
399 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
400 possible values if we don't have a hint.
402 2018-02-20 Martin Liska <mliska@suse.cz>
406 * final.c (shorten_branches): Build align_tab array with one
408 * opts.c (finish_options): Add alignment option limit check.
409 (MAX_CODE_ALIGN): Likewise.
410 (MAX_CODE_ALIGN_VALUE): Likewise.
411 * doc/invoke.texi: Document maximum allowed option value for
412 all -falign-* options.
414 2018-02-19 Jakub Jelinek <jakub@redhat.com>
417 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
418 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
419 * var-tracking.c (emit_note_insn_var_location): Remove all references
420 to NOTE_INSN_CALL_ARG_LOCATION.
421 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
422 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
423 Use copy_rtx_if_shared.
424 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
425 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
426 (dwarf2out_var_location): Remove handling of
427 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
429 * final.c (final_scan_insn): Remove all references to
430 NOTE_INSN_CALL_ARG_LOCATION.
431 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
432 before dumping final insns.
433 * except.c (emit_note_eh_region_end): Remove all references to
434 NOTE_INSN_CALL_ARG_LOCATION.
435 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
436 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
437 * config/arc/arc.c (hwloop_optimize): Likewise.
438 * config/arm/arm.c (create_fix_barrier): Likewise.
439 * config/s390/s390.c (s390_chunkify_start): Likewise.
440 * config/sh/sh.c (find_barrier): Likewise.
441 * config/i386/i386.c (rest_of_insert_endbranch,
442 ix86_seh_fixup_eh_fallthru): Likewise.
443 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
444 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
445 * config/frv/frv.c (frv_function_prologue): Likewise.
446 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
448 (note_outside_basic_block_p): Remove all references to
449 NOTE_INSN_CALL_ARG_LOCATION.
450 * gengtype.c (adjust_field_rtx_def): Likewise.
451 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
453 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
454 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
457 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
460 PR tree-optimization/84452
461 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
462 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
465 2018-02-19 Martin Liska <mliska@suse.cz>
468 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
470 2018-02-19 Martin Liska <mliska@suse.cz>
471 Richard Sandiford <richard.sandiford@linaro.org>
473 PR tree-optimization/82491
474 * gimple-fold.c (get_base_constructor): Make earlier bail out
477 2018-02-19 Carl Love <cel@us.ibm.com>
479 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
480 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
482 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
485 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
487 * config/rl78/rl78.md (movdf): New define expand.
489 2018-02-19 Martin Liska <mliska@suse.cz>
492 * doc/invoke.texi: Fix typo.
493 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
495 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
497 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
498 handle rs6000_single_float and rs6000_double_float specially for
501 2018-02-16 Jeff Law <law@redhat.com>
503 * config/rx/rx.c (add_pop_cfi_notes): New function.;
506 2018-02-16 Jakub Jelinek <jakub@redhat.com>
509 * ipa-inline.c (inline_small_functions): Fix a typo.
511 2018-02-16 Nathan Sidwell <nathan@acm.org>
513 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
515 2018-02-16 Carl Love <cel@us.ibm.com>
517 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
518 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
519 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
520 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
521 expansion to P8V_BUILTIN_VEC_FLOAT2.
523 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
525 PR rtl-optimization/70023
526 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
527 src_regno into account.
529 2018-02-16 Carl Love <cel@us.ibm.com>
531 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
532 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
533 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
534 * config/rs6000/rs6000.c: Remove case statements for
535 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
536 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
537 and P9V_BUILTIN_VEC_VINSERT4B.
538 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
539 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
540 * config/rs6000/vsx.md:
541 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
544 2018-02-16 Carl Love <cel@us.ibm.com>
546 * config/rs6000/altivec.h: Add builtin names vec_extract4b
548 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
550 * config/rs6000/rs6000-c.c: Add the definitions for
551 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
552 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
553 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
554 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
555 definition for insert4b and define insn *insert3b_internal.
556 * doc/extend.texi: Add documentation for vec_extract4b.
558 2018-02-16 Nathan Sidwell <nathan@acm.org>
560 * doc/extend.texi (Backwards Compatibility): Mention friend
561 injection. Note for-scope is deprecated.
562 * doc/invoke.texi (-ffriend-injection): Deprecate.
564 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
566 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
567 that moved to I2, also allow destinations that are a paradoxical
568 subreg (instead of a normal reg).
570 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
573 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
576 2018-02-16 Richard Biener <rguenther@suse.de>
578 PR tree-optimization/84037
579 PR tree-optimization/84016
581 * config/i386/i386.c (ix86_builtin_vectorization_cost):
582 Adjust vec_construct for the fact we need additional higher latency
583 128bit inserts for AVX256 and AVX512 vector builds.
584 (ix86_add_stmt_cost): Scale vector construction cost for
587 2018-02-16 Richard Biener <rguenther@suse.de>
589 PR tree-optimization/84417
590 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
591 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
592 (non_rewritable_lvalue_p): Likewise, use poly-ints.
594 2018-02-16 Martin Liska <mliska@suse.cz>
597 * internal-fn.def (ASAN_CHECK): Set proper flags.
598 (ASAN_MARK): Likewise.
600 2018-02-16 Julia Koval <julia.koval@intel.com>
602 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
605 2018-02-16 Jakub Jelinek <jakub@redhat.com>
608 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
609 Use ++iter rather than iter++ for std::list iterators.
610 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
611 defer deleting them until all nodes in the forest are processed. Do
612 free even leaf nodes. Change to_process into auto_vec.
615 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
616 * vec.h (vec_default_construct): Use memset instead of placement new
617 if BROKEN_VALUE_INITIALIZATION is defined.
618 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
619 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
622 PR rtl-optimization/83723
623 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
624 * lra.c (lra_substitute_pseudo): Likewise. If true, use
625 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
627 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
629 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
631 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
633 PR rtl-optimization/81443
634 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
635 from inner REGs to paradoxical SUBREGs.
637 2018-02-16 Richard Biener <rguenther@suse.de>
639 PR tree-optimization/84399
640 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
641 For operands we can analyze at their definition make sure we can
642 analyze them at each use as well.
644 2018-02-16 Richard Biener <rguenther@suse.de>
646 PR tree-optimization/84190
647 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
648 volatile accesses if the decl isn't volatile.
650 2018-02-15 Jason Merrill <jason@redhat.com>
652 PR c++/84314 - ICE with templates and fastcall attribute.
653 * attribs.c (build_type_attribute_qual_variant): Don't clobber
654 TYPE_CANONICAL on an existing type.
656 2018-02-15 Jakub Jelinek <jakub@redhat.com>
658 PR tree-optimization/84383
659 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
660 dstoff nor call operand_equal_p if dstbase is NULL.
662 PR tree-optimization/84334
663 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
664 also a CONSTANT_CLASS_P, punt.
666 2018-02-14 Jim Wilson <jimw@sifive.com>
668 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
669 first SMALL_OPERAND check. New local min_second_step. Move assert
670 to where locals are set. Add TARGET_RVC support.
671 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
673 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
675 * doc/invoke.texi: Correct -Wformat-overflow code sample.
677 2018-02-14 Martin Sebor <msebor@redhat.com>
679 PR tree-optimization/83698
680 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
681 arrays constrain the offset range to their bounds.
682 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
683 (builtin_access::overlap): Avoid setting the size of overlap if it's
685 (maybe_diag_overlap): Also consider arrays when deciding what values
686 of offsets to include in diagnostics.
688 2018-02-14 Martin Sebor <msebor@redhat.com>
691 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
692 that correspond to the kind of a declaration.
694 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
697 * config/pa/pa.md: Load address of PIC label using the linkage table
698 if the label is nonlocal.
700 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
702 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
703 warning message if user requests -maltivec=be.
704 * doc/invoke.texi: Document deprecation of -maltivec=be.
706 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
709 * config/rs6000/rs6000-c.c: Update definitions for
710 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
711 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
713 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
716 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
717 add _get_ssp intrinsics. Remove argument from
718 __builtin_ia32_rdssp[d|q].
719 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
720 * config/i386/i386-builtin.def: Remove argument from
721 __builtin_ia32_rdssp[d|q].
722 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
723 ix86_expand_special_args_builtin for _rdssp[d|q].
724 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
725 Clear register before usage.
726 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
727 Add documentation for new _get_ssp and _inc_ssp intrinsics.
729 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
731 PR tree-optimization/84357
732 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
733 operand 1 of an ARRAY_REF too.
735 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
738 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
739 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
741 (set_of_reg): New struct.
742 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
743 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
744 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
746 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
747 Split into bitclr, bitset, bitinvert patterns if appropriate.
748 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
749 use rx_fuse_in_memory_bitop.
750 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
751 to named insn, correct maximum insn length.
753 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
756 * machmode.def: Define a complex mode for PARTIAL_INT.
757 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
759 * doc/rtl.texi: Document CSPImode.
760 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
762 (msp430_hard_regno_nregs_with_padding): Likewise.
764 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
767 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
769 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
771 PR rtl-optimization/84169
772 * combine.c (try_combine): New variable split_i2i3. Set it to true if
773 we generated a parallel as new i3 and we split that to new i2 and i3
774 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
775 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
776 those to i2, not i1. Partially rewrite this scan code.
778 2018-02-13 Jakub Jelinek <jakub@redhat.com>
781 * stor-layout.c (place_field): For variable length fields, adjust
782 offset_align afterwards not just based on the field's alignment,
783 but also on the size.
786 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
787 of exps and logs in the use_exp2 case.
789 2018-02-13 Jeff Law <law@redhat.com>
791 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
794 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
797 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
801 * common.opt (gas-loc-support, gas-locview-support): New.
802 (ginline-points, ginternal-reset-location-views): New.
803 * doc/invoke.texi: Document them. Use @itemx where intended.
804 (gvariable-location-views): Adjust.
805 * target.def (reset_location_view): New.
806 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
807 (TARGET_RESET_LOCATION_VIEW): New.
808 * doc/tm.texi: Rebuilt.
809 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
810 (dwarf2out_default_as_locview_support): New.
811 (output_asm_line_debug_info): Use option variables.
812 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
813 (output_loc_list): Likewise.
814 (add_high_low_attributes): Check option variables.
815 Don't output entry view attribute in strict mode.
816 (gen_inlined_subroutine_die): Check option variables.
817 (dwarf2out_inline_entry): Likewise.
818 (init_sections_and_labels): Likewise.
819 (dwarf2out_early_finish): Likewise.
820 (maybe_reset_location_view): New, from...
821 (dwarf2out_var_location): ... here. Call it.
822 * debug.h (dwarf2out_default_as_loc_support): Declare.
823 (dwarf2out_default_as_locview_support): Declare.
824 * hooks.c (hook_int_rtx_insn_0): New.
825 * hooks.h (hook_int_rtx_insn_0): Declare.
826 * toplev.c (process_options): Take -gas-loc-support and
827 -gas-locview-support from dwarf2out. Enable
828 -gvariable-location-views by default only with locview
829 assembler support. Enable -ginternal-reset-location-views by
830 default only if the target defines the corresponding hook.
831 Enable -ginline-points by default if location views are
832 enabled; force it disabled if statement frontiers are
834 * tree-inline.c (expand_call_inline): Check option variables.
835 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
837 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
839 PR tree-optimization/84321
840 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
841 handling. Also check whether the anti-range contains any values
842 that satisfy the mask; switch to a VR_RANGE if not.
844 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
847 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
849 2018-02-13 Martin Jambor <mjambor@suse.cz>
852 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
853 of call statements, also set location of a load to a temporary.
855 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
857 * config/rl78/rl78.c (add_vector_labels): New function.
858 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
859 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
860 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
861 which checks that no arguments are passed.
862 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
863 * doc/extend.texi: Documentation for the new attribute.
865 2018-02-13 Andreas Schwab <schwab@suse.de>
867 * config/riscv/linux.h (CPP_SPEC): Define.
869 2018-02-13 Jakub Jelinek <jakub@redhat.com>
872 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
873 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
874 OPTION_MASK_ISA_AES as first argument to def_builtin_const
875 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
876 instead of OPTION_MASK_ISA_PCLMUL as first argument to
877 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
878 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
879 temporarily for AES and PCLMUL builtins.
881 PR tree-optimization/84339
882 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
883 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
887 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
888 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
889 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
891 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
893 * omp-simd-clone.h: New file.
894 * omp-simd-clone.c: Include omp-simd-clone.h.
895 (expand_simd_clones): No longer static.
896 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
897 cgraph.h and omp-simd-clone.h.
898 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
899 (vect_recog_widen_shift_pattern): Formatting fix.
900 (vect_pattern_recog_1): Don't check optab for calls.
903 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
904 operands[2] into a REG before using gen_lowpart on it.
906 2018-02-12 Jeff Law <law@redhat.com>
909 * config/sh/sh.c (find_barrier): Consider a sibling call
912 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
913 successfully back substituting a reg.
915 2018-02-12 Richard Biener <rguenther@suse.de>
917 PR tree-optimization/84037
918 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
919 parameter, move visited init to caller.
920 (vect_slp_analyze_operations): Separate cost from validity
921 check, initialize visited once for all instances.
922 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
924 * tree-vect-stmts.c (vect_model_simple_cost): Make early
926 (vect_model_promotion_demotion_cost): Likewise.
927 (vectorizable_bswap): Guard cost modeling with !slp_node
928 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
930 (vectorizable_call): Likewise.
931 (vectorizable_conversion): Likewise.
932 (vectorizable_assignment): Likewise.
933 (vectorizable_shift): Likewise.
934 (vectorizable_operation): Likewise.
935 (vectorizable_store): Likewise.
936 (vectorizable_load): Likewise.
937 (vectorizable_condition): Likewise.
938 (vectorizable_comparison): Likewise.
940 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
943 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
944 (ASAN_MARK): Fix fnspec to account for return value, change pointer
945 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
947 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
950 * params.def (inline-min-speedup): Increase from 8 to 15.
951 (max-inline-insns-auto): Decrease from 40 to 30.
952 * ipa-split.c (consider_split): Add some buffer for function to
953 be considered inlining candidate.
954 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
957 2018-02-12 Richard Biener <rguenther@suse.de>
959 PR tree-optimization/84037
960 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
961 matched stmts if we cannot swap the non-matched ones.
963 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
965 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
966 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
967 _mm_maskz_scalef_round_ss): New intrinsics.
968 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
969 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
970 __builtin_ia32_scalefss_round): Remove.
971 (__builtin_ia32_scalefsd_mask_round,
972 __builtin_ia32_scalefss_mask_round): New intrinsics.
973 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
974 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
975 ((match_operand:VF_128 2 "<round_nimm_predicate>"
976 "<round_constraint>")): Changed to ...
977 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
978 "<round_scalar_constraint>")): ... this.
979 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
980 %0, %1, %2<round_op3>}"): Changed to ...
981 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
982 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
983 %2<round_scalar_mask_op3>}"): ... this.
984 * config/i386/subst.md (round_scalar_nimm_predicate): New.
986 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
988 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
989 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
990 (_mm_maskz_sqrt_round_ss): New intrinsics.
991 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
992 (__builtin_ia32_sqrtsd_mask_round)
993 (__builtin_ia32_sqrtss_mask_round): New builtins.
994 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
995 (__builtin_ia32_sqrtss_round): Remove.
996 (__builtin_ia32_sqrtsd_mask_round)
997 (__builtin_ia32_sqrtss_mask_round): New builtins.
998 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
999 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1000 ((match_operand:VF_128 1 "vector_operand"
1001 "xBm,<round_constraint>")): Changed to ...
1002 ((match_operand:VF_128 1 "vector_operand"
1003 "xBm,<round_scalar_constraint>")): ... this.
1004 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1005 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1006 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1007 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1008 %<iptr>1<round_scalar_mask_op3>}): ... this.
1009 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1010 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1012 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1015 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1016 Cast vec_cmpeq result to correct type.
1017 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1018 Cast vec_cmpgt result to correct type.
1020 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1022 * final.c (final_scan_insn_1): Renamed from...
1023 (final_scan_insn): ... this. New wrapper, to recover
1024 seen from the outermost call in recursive ones.
1025 * config/sparc/sparc.c (output_return): Drop seen from call.
1026 (output_sibcall): Likewise.
1027 * config/visium/visium.c (output_branch): Likewise.
1029 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1031 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1034 2018-02-10 Alan Modra <amodra@gmail.com>
1037 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1038 Specify LR as an input.
1040 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1043 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1044 remove_member_access_dummy_vars): New functions.
1045 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1046 lower_omp_1, execute_lower_omp): Use them.
1048 PR rtl-optimization/84308
1049 * shrink-wrap.c (spread_components): Release todo vector.
1051 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1053 PR rtl-optimization/57193
1054 * ira-color.c (struct allocno_color_data): Add member
1055 conflict_allocno_hard_prefs.
1056 (update_conflict_allocno_hard_prefs): New.
1057 (bucket_allocno_compare_func): Add a preference based on
1058 conflict_allocno_hard_prefs.
1059 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1060 (color_allocnos): Remove a dead code. Initiate
1061 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1063 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1066 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1067 constraint from =wa to wa. Avoid a subreg on the output operand,
1068 instead use a pseudo and subreg it in a move.
1069 (p9_xxbrd_<mode>): Changed to ...
1070 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1071 (p9_xxbrd_v2df): New expander.
1072 (p9_xxbrw_<mode>): Changed to ...
1073 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1074 (p9_xxbrw_v4sf): New expander.
1076 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1078 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1080 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1083 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1084 multiply in 32-bit mode.
1085 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1086 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1089 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1091 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1092 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1093 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1094 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1096 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1099 * dwarf2out.c (is_trivial_indirect_ref): New function.
1100 (dwarf2out_late_global_decl): Do not generate a location
1101 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1102 and that are not defined in the current unit.
1104 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1106 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1107 instead of a libcall for UNORDERED.
1109 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1112 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1113 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1115 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1118 * config/s390/s390.c (s390_set_current_function): Invoke
1119 s390_indirect_branch_settings also if fndecl didn't change.
1121 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1123 * config/rs6000/rs6000.md (blockage): Set length to zero.
1125 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1127 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1129 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1132 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1133 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1137 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1138 PARALLEL incoming that failed vt_get_decl_and_offset check.
1141 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1142 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1144 (get_variable_section): For decls in named .bss* sections pass true as
1145 second argument to bss_initializer_p.
1147 2018-02-09 Marek Polacek <polacek@redhat.com>
1148 Jakub Jelinek <jakub@redhat.com>
1151 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1152 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1153 Sync some changes from cxx_fold_indirect_ref.
1155 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1157 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1159 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1160 (BLOCK_INLINE_ENTRY_LABEL): New.
1161 (dwarf2out_var_location): Disregard inline entry markers.
1162 (inline_entry_data): New struct.
1163 (inline_entry_data_hasher): New hashtable type.
1164 (inline_entry_data_hasher::hash): New.
1165 (inline_entry_data_hasher::equal): New.
1166 (inline_entry_data_table): New variable.
1167 (add_high_low_attributes): Add DW_AT_entry_pc and
1168 DW_AT_GNU_entry_view attributes if a pending entry is found
1169 in inline_entry_data_table. Add old entry_pc attribute only
1170 if debug nonbinding markers are disabled.
1171 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1172 markers are enabled.
1173 (block_within_block_p, dwarf2out_inline_entry): New.
1174 (dwarf2out_finish): Check that no entries remained in
1175 inline_entry_data_table.
1176 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1177 (final_scan_insn, notice_source_line): Likewise.
1178 (rest_of_clean_state): Skip inline entry markers.
1179 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1181 * gimple.c (gimple_build_debug_inline_entry): New.
1182 * gimple.h (enum gimple_debug_subcode): Add
1183 GIMPLE_DEBUG_INLINE_ENTRY.
1184 (gimple_build_debug_inline_entry): Declare.
1185 (gimple_debug_inline_entry_p): New.
1186 (gimple_debug_nonbind_marker_p): Adjust.
1187 * insn-notes.def (INLINE_ENTRY): New.
1188 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1189 inline entry marker notes.
1190 (print_insn): Likewise.
1191 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1192 (INSN_DEBUG_MARKER_KIND): Likewise.
1193 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1194 * tree-inline.c (expand_call_inline): Build and insert
1195 debug_inline_entry stmt.
1196 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1197 inline entry blocks early, if nonbind markers are enabled.
1198 (dump_scope_block): Dump fragment info.
1199 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1200 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1201 (gimple_build_debug_inline_entry): New.
1202 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1203 Enable/disable inline entry points too.
1204 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1205 (DEBUG_INSN): Describe inline entry markers.
1207 * common.opt (gvariable-location-views): New.
1208 (gvariable-location-views=incompat5): New.
1209 * config.in: Rebuilt.
1210 * configure: Rebuilt.
1211 * configure.ac: Test assembler for view support.
1212 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1213 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1214 * dwarf2out.c (var_loc_view): New typedef.
1215 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1216 (dwarf2out_locviews_in_attribute): New.
1217 (dwarf2out_locviews_in_loclist): New.
1218 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1219 (enum dw_line_info_opcode): Add LI_adv_address.
1220 (struct dw_line_info_table): Add view.
1221 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1222 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1223 (zero_view_p): New variable.
1224 (ZERO_VIEW_P): New macro.
1225 (output_asm_line_debug_info): New.
1226 (struct var_loc_node): Add view.
1227 (add_AT_view_list, AT_loc_list): New.
1228 (add_var_loc_to_decl): Add view param. Test it against last.
1229 (new_loc_list): Add view params. Record them.
1230 (AT_loc_list_ptr): Handle loc and view lists.
1231 (view_list_to_loc_list_val_node): New.
1232 (print_dw_val): Handle dw_val_class_view_list.
1233 (size_of_die): Likewise.
1234 (value_format): Likewise.
1235 (loc_list_has_views): New.
1236 (gen_llsym): Set vl_symbol too.
1237 (maybe_gen_llsym, skip_loc_list_entry): New.
1238 (dwarf2out_maybe_output_loclist_view_pair): New.
1239 (output_loc_list): Output view list or entries too.
1240 (output_view_list_offset): New.
1241 (output_die): Handle dw_val_class_view_list.
1242 (output_dwarf_version): New.
1243 (output_compilation_unit_header): Use it.
1244 (output_skeleton_debug_sections): Likewise.
1245 (output_rnglists, output_line_info): Likewise.
1246 (output_pubnames, output_aranges): Update version comments.
1247 (output_one_line_info_table): Output view numbers in asm comments.
1248 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1249 Call maybe_gen_llsym.
1250 (loc_list_from_tree_1): Adjust.
1251 (add_AT_location_description): Create view list attribute if
1252 needed, check it's absent otherwise.
1253 (convert_cfa_to_fb_loc_list): Adjust.
1254 (maybe_emit_file): Call output_asm_line_debug_info for test.
1255 (dwarf2out_var_location): Reset views as needed. Precompute
1256 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1257 attribute. Set view.
1258 (new_line_info_table): Reset next view.
1259 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1260 (dwarf2out_source_line): Likewise. Output view resets and labels to
1261 the assembler, or select appropriate line info opcodes.
1262 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1263 (optimize_string_length): Catch it. Adjust.
1264 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1265 dw_val_class_view_list, and remove it if no longer needed.
1266 (hash_loc_list): Hash view numbers.
1267 (loc_list_hasher::equal): Compare them.
1268 (optimize_location_lists): Check whether a view list symbol is
1269 needed, and whether the locview attribute is present, and
1270 whether they match. Remove the locview attribute if no longer
1272 (index_location_lists): Call skip_loc_list_entry for test.
1273 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1274 Use output_dwarf_version.
1275 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1276 (struct dw_val_node): Add val_view_list.
1277 * final.c (SEEN_NEXT_VIEW): New.
1278 (set_next_view_needed): New.
1279 (clear_next_view_needed): New.
1280 (maybe_output_next_view): New.
1281 (final_start_function): Rename to...
1282 (final_start_function_1): ... this. Take pointer to FIRST,
1283 add SEEN parameter. Emit param bindings in the initial view.
1284 (final_start_function): Reintroduce SEEN-less interface.
1285 (final): Rename to...
1286 (final_1): ... this. Take SEEN parameter. Output final pending
1287 next view at the end.
1288 (final): Reintroduce seen-less interface.
1289 (final_scan_insn): Output pending next view before switching
1290 sections or ending a block. Mark the next view as needed when
1291 outputting variable locations. Notify debug backend of section
1292 changes, and of location view changes.
1293 (rest_of_handle_final): Adjust.
1294 * toplev.c (process_options): Autodetect value for debug variable
1295 location views option. Warn on incompat5 without -gdwarf-5.
1296 * doc/invoke.texi (gvariable-location-views): New.
1297 (gvariable-location-views=incompat5): New.
1298 (gno-variable-location-views): New.
1300 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1302 PR tree-optimization/84136
1303 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1304 that the result of find_edge is non-NULL.
1306 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1309 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1310 storing integer register in SImode. Fix cost of 256 and 512
1311 byte aligned SSE register store.
1313 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1315 * config/i386/i386.c (ix86_multiplication_cost): Fix
1316 multiplication cost for TARGET_AVX512DQ.
1318 2018-02-08 Marek Polacek <polacek@redhat.com>
1320 PR tree-optimization/84238
1321 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1324 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1326 PR tree-optimization/84265
1327 * tree-vect-stmts.c (vectorizable_store): Don't treat
1328 VMAT_CONTIGUOUS accesses as grouped.
1329 (vectorizable_load): Likewise.
1331 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1333 PR tree-optimization/81635
1334 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1335 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1336 (test_round_for_mask): New functions.
1337 (wide_int_cc_tests): Call test_round_for_mask.
1338 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1339 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1340 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1341 range returned by get_range_info.
1343 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1346 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1347 * symtab.c: Include builtins.h
1348 (symtab_node::output_to_lto_symbol_table_p): Move here
1349 from lto-streamer-out.c:output_symbol_p.
1350 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1351 (output_symbol_p): Move all logic to symtab.c
1352 (produce_symtab): Update.
1354 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1356 * config/s390/s390-opts.h (enum indirect_branch): Define.
1357 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1358 (s390_indirect_branch_via_thunk)
1359 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1360 (enum s390_indirect_branch_type): Define.
1361 * config/s390/s390.c (struct s390_frame_layout, struct
1362 machine_function): Remove.
1363 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1364 (indirect_branch_table_label_no, indirect_branch_table_name):
1366 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1367 (enum s390_indirect_branch_option): Define.
1368 (s390_return_addr_from_memory): New function.
1369 (s390_handle_string_attribute): New function.
1370 (s390_attribute_table): Add new attribute handler.
1371 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1372 (s390_indirect_branch_via_thunk): New function.
1373 (s390_indirect_branch_via_inline_thunk): New function.
1374 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1375 sibling call optimization for non z10 compiles.
1376 (s390_emit_call): Force indirect branch target to be a single
1377 register. Add r1 clobber for non-z10 compiles.
1378 (s390_emit_epilogue): Emit return jump via return_use expander.
1379 (s390_reorg): Handle JUMP_INSNs as execute targets.
1380 (s390_option_override_internal): Perform validity checks for the
1381 new command line options.
1382 (s390_indirect_branch_attrvalue): New function.
1383 (s390_indirect_branch_settings): New function.
1384 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1385 (s390_output_indirect_thunk_function): New function.
1386 (s390_code_end): Implement target hook.
1387 (s390_case_values_threshold): Implement target hook.
1388 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1390 * config/s390/s390.h (struct s390_frame_layout)
1391 (struct machine_function): Move here from s390.c.
1392 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1393 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1394 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1395 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1396 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1397 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1398 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1399 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1400 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1401 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1402 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1403 (mnemonic attribute): Add values which aren't recognized
1405 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1406 pattern for branch conversion. Fix mnemonic attribute.
1407 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1408 indirect branch via thunk if requested.
1409 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1410 ("*indirect_jump"): Disable for branch conversion using out of
1412 ("indirect_jump_via_thunk<mode>_z10")
1413 ("indirect_jump_via_thunk<mode>")
1414 ("indirect_jump_via_inlinethunk<mode>_z10")
1415 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1416 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1417 ("casesi_jump_via_inlinethunk<mode>_z10")
1418 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1419 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1420 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1421 ("*indirect2_jump"): Disable for branch conversion.
1422 ("casesi_jump"): Turn into expander and expand patterns for branch
1424 ("return_use"): New expander.
1425 ("*return"): Emit return via thunk and rename it to ...
1426 ("*return<mode>"): ... this one.
1427 * config/s390/s390.opt: Add new options and and enum for the
1430 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1432 * lra-constraints.c (match_reload): Unconditionally use
1433 gen_lowpart_SUBREG, rather than selecting between that
1434 and equivalent gen_rtx_SUBREG code.
1436 2018-02-08 Richard Biener <rguenther@suse.de>
1438 PR tree-optimization/84233
1439 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1440 changed flag instead of boguously re-using phi_inserted.
1442 2018-02-08 Martin Jambor <mjambor@suse.cz>
1444 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1445 static local variables.
1447 2018-02-08 Richard Biener <rguenther@suse.de>
1449 PR tree-optimization/84278
1450 * tree-vect-stmts.c (vectorizable_store): When looking for
1451 smaller vector types to perform grouped strided loads/stores
1452 make sure the mode is supported by the target.
1453 (vectorizable_load): Likewise.
1455 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1457 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1458 Increase LDP/STP opportunities by adding adjacent callee-saves.
1460 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1462 PR rtl-optimization/84068
1463 PR rtl-optimization/83459
1464 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1466 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1468 PR tree-optimization/84224
1469 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1470 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1473 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1476 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1477 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1478 count, remove one USE.
1480 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1482 * doc/install.texi (Configuration): Document the
1483 --with-long-double-format={ibm,ieee} PowerPC configuration
1487 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1488 Convert from define_expand to be define_insn_and_split. Rework
1489 float/double/_Float128 conversions to QI/HI/SImode to work with
1490 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1491 conversions to QI/HImode types did a store and then a load to
1492 truncate the value. For conversions to VSX registers, don't split
1493 the insn, instead emit the code directly. Use the code iterator
1494 any_fix to combine signed and unsigned conversions.
1495 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1496 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1497 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1498 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1499 (fix_<mode>di2_hw): Likewise.
1500 (fixuns_<mode>di2_hw): Likewise.
1501 (fix_<mode>si2_hw): Likewise.
1502 (fixuns_<mode>si2_hw): Likewise.
1503 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1504 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1505 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1506 fix<uns>_trunc<SFDF:mode>si2_p8.
1507 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1509 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1510 (fix<uns>_<mode>_mem): Likewise.
1511 (fctiw<u>z_<mode>_mem): Likewise.
1512 (fix<uns>_<mode>_mem): Likewise.
1513 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1514 the register allocator from doing a direct move to the GPRs to do
1515 a store, and instead use the ISA 3.0 store byte/half-word from
1516 vector register instruction. For IEEE 128-bit floating point,
1517 also optimize stores of 32-bit ints.
1518 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1520 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1522 * genextract.c (push_pathstr_operand): New function to support
1524 (walk_rtx): Call push_pathstr_operand.
1525 (print_path): Support [a-zA-Z].
1527 2018-02-07 Richard Biener <rguenther@suse.de>
1529 PR tree-optimization/84037
1530 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1531 (cse_and_gimplify_to_preheader): Declare.
1532 (vect_get_place_in_interleaving_chain): Likewise.
1533 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1535 (_loop_vec_info::~_loop_vec_info): Delete it.
1536 (cse_and_gimplify_to_preheader): New function.
1537 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1538 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1539 (vectorizable_load): Likewise. For grouped stores always base
1540 the IV on the first element.
1541 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1542 condition before gimplifying.
1544 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1546 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1547 *DIV_EXPR and *MOD_EXPR.
1549 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1552 * config/i386/i386.c (ix86_option_override_internal): Mask out
1553 the CF_SET bit when checking -fcf-protection.
1555 2018-02-07 Tom de Vries <tom@codesourcery.com>
1558 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1561 2018-02-07 Richard Biener <rguenther@suse.de>
1563 PR tree-optimization/84204
1564 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1567 PR tree-optimization/84205
1568 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1569 special-case isl_ast_op_zdiv_r.
1571 PR tree-optimization/84223
1572 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1573 Only add conditions from within the region.
1574 (gather_bbs::after_dom_children): Adjust.
1576 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1579 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1580 * config/avr/avr.md: Only post-reload split REG-REG moves if
1581 either register is REGERAL_REG_P.
1583 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1585 PR tree-optimization/84235
1586 * tree-ssa-scopedtables.c
1587 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1588 if the subtraction is performed in floating point type where NaNs are
1589 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1590 build 1. Formatting fix.
1592 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1595 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1596 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1597 and skip it regardless of bb boundaries. Use CALL_P macro,
1598 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1599 unnecessarily, formatting fix.
1601 2018-02-06 Michael Collison <michael.collison@arm.com>
1603 * config/arm/thumb2.md:
1604 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1605 (*thumb_mov_notscc): Ditto.
1607 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1610 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1611 unsigned_fix, not "s".
1613 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1615 * configure.ac (gcc_fn_eh_frame_ro): New function.
1616 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1617 correct .eh_frame permissions.
1618 * configure: Regenerate.
1620 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1622 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1625 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1627 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1628 Display warning message for -mno-speculate-indirect-jumps.
1630 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1632 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1634 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1636 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1638 PR tree-optimization/84225
1639 * tree-eh.c (find_trapping_overflow): Only call
1640 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1642 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1645 * config/i386/i386.c: Reimplement the check of possible options
1646 -mibt/-mshstk conbination. Change error messages.
1647 * doc/invoke.texi: Fix a typo: remove extra '='.
1649 2018-02-06 Marek Polacek <polacek@redhat.com>
1651 PR tree-optimization/84228
1652 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1654 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1657 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1658 emitted arch directives.
1659 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1660 __ARM_FEATURE_COPROC before changing architectures.
1662 2018-02-06 Richard Biener <rguenther@suse.de>
1664 * config/i386/i386.c (print_reg): Fix typo.
1665 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1667 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1669 * configure: Regenerate.
1671 2018-02-05 Martin Sebor <msebor@redhat.com>
1673 PR tree-optimization/83369
1674 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1677 2018-02-05 Martin Liska <mliska@suse.cz>
1679 * doc/invoke.texi: Cherry-pick upstream r323995.
1681 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1683 * ira.c (ira_init_register_move_cost): Adjust comment.
1685 2018-02-05 Martin Liska <mliska@suse.cz>
1687 PR gcov-profile/84137
1688 * doc/gcov.texi: Fix typo in documentation.
1690 2018-02-05 Martin Liska <mliska@suse.cz>
1692 PR gcov-profile/83879
1693 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1694 using dlopen functionality.
1696 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1698 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1699 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1700 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1701 _mm_maskz_range_round_ss): New intrinsics.
1702 (__builtin_ia32_rangesd128_round)
1703 (__builtin_ia32_rangess128_round): Remove.
1704 (__builtin_ia32_rangesd128_mask_round,
1705 __builtin_ia32_rangess128_mask_round): New builtins.
1706 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1707 __builtin_ia32_rangess128_round): Remove.
1708 (__builtin_ia32_rangesd128_mask_round,
1709 __builtin_ia32_rangess128_mask_round): New builtins.
1710 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1711 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1712 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1713 "<round_saeonly_constraint>")): Changed to ...
1714 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1715 "<round_saeonly_scalar_constraint>")): ... this.
1716 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1717 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1718 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1719 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1720 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1722 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1724 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1726 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1727 Remove all values except native, 8540 and 8548.
1729 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1731 * config/i386/i386.c (ix86_output_function_return): Pass
1732 INVALID_REGNUM, instead of -1, as invalid register number to
1733 indirect_thunk_name and output_indirect_thunk.
1735 2018-02-02 Julia Koval <julia.koval@intel.com>
1737 * config.gcc: Add -march=icelake.
1738 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1739 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1740 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1741 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1742 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1743 (processor_target_table): Add icelake.
1744 (ix86_option_override_internal): Handle new PTAs.
1745 (get_builtin_code_for_version): Handle icelake.
1746 (M_INTEL_COREI7_ICELAKE): New.
1747 (fold_builtin_cpu): Handle icelake.
1748 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1749 * doc/invoke.texi: Add -march=icelake.
1751 2018-02-02 Julia Koval <julia.koval@intel.com>
1753 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1754 to wide_int_bitmask.
1755 * wide-int-bitmask.h: New.
1757 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1760 * config/i386/i386.md: Replace Pmode with word_mode in
1761 builtin_setjmp_setup and builtin_longjmp to support x32.
1763 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1767 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1769 (rs6000_supported_cpu_names): New static variable.
1770 (linux_cpu_translation_table): Likewise.
1771 (elf_platform) <cpu>: Define new static variable and use it.
1772 Translate kernel AT_PLATFORM name to canonical name if needed.
1773 Error if platform name is unknown.
1775 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1778 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1780 2018-02-01 Jeff Law <law@redhat.com>
1783 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1784 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1785 the scratch if RELEASE_VIA_POP is false.
1786 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1787 If we have to save a temporary register, decrement SIZE appropriately.
1788 Pass new arguments to release_scratch_register_on_entry.
1789 (ix86_adjust_stack_and_probe): Likewise.
1790 (ix86_emit_probe_stack_range): Pass new arguments to
1791 release_scratch_register_on_entry.
1793 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1795 PR rtl-optimization/84157
1796 * combine.c (change_zero_ext): Use REG_P predicate in
1797 front of HARD_REGISTER_P predicate.
1799 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1801 * config/avr/avr.c (avr_option_override): Move disabling of
1802 -fdelete-null-pointer-checks to...
1803 * common/config/avr/avr-common.c (avr_option_optimization_table):
1806 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1808 PR tree-optimization/81635
1809 * tree-data-ref.c (split_constant_offset_1): For types that
1810 wrap on overflow, try to use range info to prove that wrapping
1813 2018-02-01 Renlin Li <renlin.li@arm.com>
1816 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1818 (aarch64_register_move_cost): Likewise.
1819 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1821 (REG_CLASS_NAMES): Likewise.
1822 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1823 TAILCALL_ADDR_REGS. Remove IP registers.
1824 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1826 2018-02-01 Richard Biener <rguenther@suse.de>
1828 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1829 for specifying RPO order and allow NULL for that.
1830 * domwalk.c (dom_walker::dom_walker): Likewise.
1831 (dom_walker::walk): Handle NULL RPO order.
1832 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1834 (rewrite_update_dom_walker): Likewise.
1835 (mark_def_dom_walker): Likewise.
1837 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1839 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1840 (aarch64_maybe_expand_sve_subreg_move): Declare.
1841 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1842 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1844 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1845 that are semantically a reverse operation.
1846 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1847 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1848 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1850 (aarch64_can_change_mode_class): For big-endian, forbid changes
1851 between two SVE modes if they have different element sizes.
1853 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1855 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1856 the TImode handling for big-endian targets.
1858 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1860 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1861 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1863 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1864 Remove BSWAP handing for big-endian targets and use the form of
1865 LD1RQ appropariate for the mode.
1867 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1869 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1870 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1873 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1876 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1877 check for operands that need to go through aarch64_sve_reload_be.
1879 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1881 PR tree-optimization/81661
1882 PR tree-optimization/84117
1883 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1884 * tree-eh.c: Include gimplify.h.
1885 (find_trapping_overflow, replace_trapping_overflow,
1886 rewrite_to_non_trapping_overflow): New functions.
1887 * tree-vect-loop.c: Include tree-eh.h.
1888 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1889 * tree-data-ref.c: Include tree-eh.h.
1890 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1892 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1894 PR rtl-optimization/84123
1895 * combine.c (change_zero_ext): Check if hard register satisfies
1896 can_change_dest_mode before calling gen_lowpart_SUBREG.
1898 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1901 * ira.c (ira_init_register_move_cost): Remove assert.
1903 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1905 PR rtl-optimization/84071
1906 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1907 * doc/tm.texi: Regenerate.
1909 2018-01-31 Richard Biener <rguenther@suse.de>
1911 PR tree-optimization/84132
1912 * tree-data-ref.c (analyze_miv_subscript): Properly
1913 check whether evolution_function_is_affine_multivariate_p
1914 before calling gcd_of_steps_may_divide_p.
1916 2018-01-31 Julia Koval <julia.koval@intel.com>
1919 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1920 * config/i386/i386.md (rdpid_rex64) New.
1921 (rdpid): Make 32bit only.
1923 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1926 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1927 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1929 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1932 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1934 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1936 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1938 PR rtl-optimization/84071
1939 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1940 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1942 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1944 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1945 (arc_attribute_table): Add 'aux' attribute.
1946 (arc_in_small_data_p): Consider aux like variables.
1947 (arc_is_aux_reg_p): New function.
1948 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1949 (arc_get_aux_arg): New function.
1950 (prepare_move_operands): Handle aux-register access.
1951 (arc_handle_aux_attribute): New function.
1952 * doc/extend.texi (ARC Variable attributes): Add subsection.
1954 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1956 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1957 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1958 (arc_attribute_table): Add 'uncached' attribute.
1959 (arc_print_operand): Print '.di' flag for uncached memory
1961 (arc_in_small_data_p): Do not consider for small data the uncached
1963 (arc_is_uncached_mem_p): New function.
1964 * config/arc/predicates.md (compact_store_memory_operand): Check
1965 for uncached memory accesses.
1966 (nonvol_nonimm_operand): Likewise.
1967 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1969 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1972 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1973 falign-loops=): Add Optimization flag.
1975 2018-01-30 Jeff Law <law@redhat.com>
1978 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1979 INT_REGISTERS_SAVED. Check it prior to calling
1980 get_scratch_register_on_entry.
1981 (ix86_adjust_stack_and_probe): Similarly.
1982 (ix86_emit_probe_stack_range): Similarly.
1983 (ix86_expand_prologue): Corresponding changes.
1985 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1988 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1989 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1991 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1994 * lra-constraints.c (curr_insn_transform): Process AND in the
1997 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1999 PR rtl-optimization/83986
2000 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2001 dependence against last_pending_memory_flush in addition to
2004 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2006 PR tree-optimization/81611
2007 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2010 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2013 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2016 2018-01-30 Richard Biener <rguenther@suse.de>
2017 Jakub Jelinek <jakub@redhat.com>
2019 PR tree-optimization/84111
2020 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2021 inner loops added during recursion, as they don't have up-to-date
2024 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2027 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2028 (can_inline_edge_by_limits_p): ... here.
2029 (can_early_inline_edge_p, check_callers,
2030 update_caller_keys, update_callee_keys, recursive_inlining,
2031 add_new_edges_to_heap, speculation_useful_p,
2032 inline_small_functions,
2033 inline_small_functions, flatten_function,
2034 inline_to_all_callers_1): Update.
2036 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2038 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2041 2018-01-30 Richard Biener <rguenther@suse.de>
2043 PR tree-optimization/83008
2044 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2045 invariant and constant vector uses in stmts when they need
2048 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2051 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2052 * configure: Regenerate.
2054 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2056 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2058 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2059 Use gen_rtx_REG rather than gen_lowpart.
2061 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2063 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2064 rather than 0 when creating partial subregs.
2066 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2068 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2071 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2074 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2075 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2076 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2077 flags. This restores the settings used before the 2017-07-24.
2078 Turning off pre increment/decrement/modify allows IVOPTS to
2079 optimize DF/SF loops where the index is an int.
2081 2018-01-29 Richard Biener <rguenther@suse.de>
2082 Kelvin Nilsen <kelvin@gcc.gnu.org>
2085 * tree-vect-stmts.c (vectorizable_call): Don't call
2086 targetm.vectorize_builtin_md_vectorized_function if callee is
2089 2018-01-22 Carl Love <cel@us.ibm.com>
2091 * doc/extend.tex: Fix typo in second arg in
2092 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2094 2018-01-29 Richard Biener <rguenther@suse.de>
2096 PR tree-optimization/84086
2097 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2098 (flush_ssaname_freelist): When SSA names were released reset
2099 the SCEV hash table.
2101 2018-01-29 Richard Biener <rguenther@suse.de>
2103 PR tree-optimization/84057
2104 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2105 removed paths when removing edges.
2107 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2109 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2110 -mfunction-return=@var{choice}.
2112 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2115 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2117 (layout::print_source_line): Likewise.
2118 (test_get_line_width_without_trailing_whitespace): Add test cases.
2120 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2123 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2126 2018-01-26 Jim Wilson <jimw@sifive.com>
2128 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2130 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2133 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2135 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2136 and CMP + SUB-immediate -> SUBS.
2138 2018-01-26 Martin Sebor <msebor@redhat.com>
2140 PR tree-optimization/83896
2141 * tree-ssa-strlen.c (get_string_len): Rename...
2142 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2143 Avoid assuming length is constant.
2144 (handle_char_store): Use HOST_WIDE_INT for string length.
2146 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2149 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2150 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2152 2018-01-26 Richard Biener <rguenther@suse.de>
2154 PR rtl-optimization/84003
2155 * dse.c (record_store): Only record redundant stores when
2156 the earlier store aliases at least all accesses the later one does.
2158 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2160 PR rtl-optimization/83985
2161 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2162 REG_CFA_RESTORE insns.
2163 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2164 REG_CFA_RESTORE insns here.
2167 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2168 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2170 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2172 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2173 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2174 (arc_init): Likewise.
2175 (arc_override_options): Likewise.
2176 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2178 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2180 * config/arc/arc.h (TARGET_DBNZ): Define.
2181 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2182 properly set the tune attribute.
2183 (dbnz): Use TARGET_DBNZ guard.
2184 * config/arc/arc.opt (mtune): Add core3 option.
2186 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2188 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2189 recognize new pic like addresses.
2190 (arc_delegitimize_address): Clean up.
2192 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2194 * config/arc/arc-arches.def: Option mrf16 valid for all
2196 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2197 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2198 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2199 * config/arc/arc-tables.opt: Regenerate.
2200 * config/arc/arc.c (arc_conditional_register_usage): Handle
2201 reduced register file case.
2202 (arc_file_start): Set must have build attributes.
2203 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2205 * config/arc/arc.opt (mrf16): Add new option.
2206 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2207 * config/arc/genmultilib.awk: Handle new mrf16 option.
2208 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2209 * config/arc/t-multilib: Regenerate.
2210 * doc/invoke.texi (ARC Options): Document mrf16 option.
2212 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2214 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2215 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2216 (arc_attribute_table): Add 'secure_call' attribute.
2217 (arc_print_operand): Print secure call operand.
2218 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2220 (arc_is_secure_call_p): New function. * config/arc/arc.md
2221 (call_i): Add support for sjli instruction.
2222 (call_value_i): Likewise.
2223 * config/arc/constraints.md (Csc): New constraint.
2225 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2226 John Eric Martin <John.Martin@emmicro-us.com>
2228 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2229 * config/arc/arc.c (_arc_jli_section): New struct.
2230 (arc_jli_section): New type.
2231 (rc_jli_sections): New static variable.
2232 (arc_handle_jli_attribute): New function.
2233 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2234 (arc_file_end): New function.
2235 (TARGET_ASM_FILE_END): Define.
2236 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2237 (arc_add_jli_section): New function.
2238 (jli_call_scan): Likewise.
2239 (arc_reorg): Call jli_call_scan.
2240 (arc_output_addsi): Remove 'S' from printing asm operand.
2241 (arc_is_jli_call_p): New function.
2242 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2244 (movhi_insn): Likewise.
2245 (movsi_insn): Likewise.
2246 (movsi_set_cc_insn): Likewise.
2247 (loadqi_update): Likewise.
2248 (load_zeroextendqisi_update): Likewise.
2249 (load_signextendqisi_update): Likewise.
2250 (loadhi_update): Likewise.
2251 (load_zeroextendhisi_update): Likewise.
2252 (load_signextendhisi_update): Likewise.
2253 (loadsi_update): Likewise.
2254 (loadsf_update): Likewise.
2255 (movsicc_insn): Likewise.
2256 (bset_insn): Likewise.
2257 (bxor_insn): Likewise.
2258 (bclr_insn): Likewise.
2259 (bmsk_insn): Likewise.
2260 (bicsi3_insn): Likewise.
2261 (cmpsi_cc_c_insn): Likewise.
2262 (movsi_ne): Likewise.
2263 (movsi_cond_exec): Likewise.
2264 (clrsbsi2): Likewise.
2275 (call_i): Remove 'S' asm letter, add jli instruction.
2276 (call_value_i): Likewise.
2277 * config/arc/arc.op (mjli-always): New option.
2278 * config/arc/constraints.md (Cji): New constraint.
2279 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2281 (subsf3_fpx): Likewise.
2282 (mulsf3_fpx): Likewise.
2283 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2285 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2287 * doc/invoke.texi (ARC): Document mjli-always option.
2289 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2291 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2292 avoid addition with 0 and use incw and decw where possible.
2294 2018-01-26 Richard Biener <rguenther@suse.de>
2296 PR tree-optimization/81082
2297 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2298 association if it requires casting to unsigned.
2299 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2300 from fold_plusminus_mult_expr to catch important cases late when
2301 range info is available.
2303 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2305 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2306 * configure.ac (hidden_linkonce): New test.
2307 * configure: Regenerate.
2308 * config.in: Regenerate.
2310 2018-01-26 Julia Koval <julia.koval@intel.com>
2312 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2313 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2314 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2315 _mm_mask_bitshuffle_epi64_mask): Fix type.
2316 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2317 USI_FTYPE_V4DI_V4DI_USI): Remove.
2318 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2319 __builtin_ia32_vpshufbitqmb256_mask,
2320 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2321 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2322 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2324 2018-01-26 Alan Modra <amodra@gmail.com>
2327 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2328 UNSPEC_VBPERMQ. Sort other unspecs.
2330 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2332 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2334 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2337 * predict.c (drop_profile): Do not push/pop cfun; update also
2339 (handle_missing_profiles): Fix logic looking for zero profiles.
2341 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2344 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2345 on functions with #pragma omp declare simd or functions with simd
2347 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2348 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2349 Remove trailing \n from warning_at calls.
2351 2018-01-25 Tom de Vries <tom@codesourcery.com>
2354 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2355 for neutered workers.
2357 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2360 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2361 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2363 2018-01-24 Jeff Law <law@redhat.com>
2366 * i386.c (get_probe_interval): Move to earlier point.
2367 (ix86_compute_frame_layout): If -fstack-clash-protection and
2368 the frame is larger than the probe interval, then use pushes
2369 to save registers rather than reg->mem moves.
2370 (ix86_expand_prologue): Remove conditional for int_registers_saved
2373 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2376 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2377 min/max for never referenced object.
2379 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2382 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2384 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2385 attributes from DECL_ATTRIBUTES (decl) without affecting
2386 DECL_ATTRIBUTES (current_function_decl).
2387 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2388 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2390 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2392 PR tree-optimization/83979
2393 * fold-const.c (fold_comparison): Use constant_boolean_node
2394 instead of boolean_{true,false}_node.
2396 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2398 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2401 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2403 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2404 Simplify the clause that sets the length attribute.
2405 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2406 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2407 clause that sets the length attribute.
2408 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2410 2018-01-24 Tom de Vries <tom@codesourcery.com>
2413 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2414 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2415 Add strict parameter.
2416 (prevent_branch_around_nothing): Insert dummy insn between branch to
2417 label and label with no ptx insn inbetween.
2418 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2420 2018-01-24 Tom de Vries <tom@codesourcery.com>
2423 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2424 for neutered threads in warp.
2425 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2427 2018-01-24 Richard Biener <rguenther@suse.de>
2429 PR tree-optimization/83176
2430 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2433 2018-01-24 Richard Biener <rguenther@suse.de>
2435 PR tree-optimization/82819
2436 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2437 code generating pluses that are no-ops in the target precision.
2439 2018-01-24 Richard Biener <rguenther@suse.de>
2442 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2444 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2446 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2447 to merge probabilities.
2448 * predict.c (probably_never_executed): Also mark as cold functions
2449 with global 0 profile and guessed local profile.
2450 * profile-count.c (profile_probability::combine_with_count): New
2452 * profile-count.h (profile_probability::operator*,
2453 profile_probability::operator*=, profile_probability::operator/,
2454 profile_probability::operator/=): Reduce precision to adjusted
2455 and set value to guessed on contradictory divisions.
2456 (profile_probability::combine_with_freq): Remove.
2457 (profile_probability::combine_wiht_count): Declare.
2458 (profile_count::force_nonzero):: Set to adjusted.
2459 (profile_count::probability_in):: Set quality to adjusted.
2460 * tree-ssa-tail-merge.c (replace_block_by): Use
2463 2018-01-23 Andrew Waterman <andrew@sifive.com>
2464 Jim Wilson <jimw@sifive.com>
2466 * config/riscv/riscv.c (riscv_stack_boundary): New.
2467 (riscv_option_override): Set riscv_stack_boundary. Handle
2468 riscv_preferred_stack_boundary_arg.
2469 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2470 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2471 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2472 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2473 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2474 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2476 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2479 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2480 of struct ix86_frame.
2481 (ix86_expand_epilogue): Likewise. Add a local variable for
2482 the reg_save_offset field in struct ix86_frame.
2484 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2486 PR tree-optimization/82604
2487 * tree-loop-distribution.c (enum partition_kind): New enum item
2488 PKIND_PARTIAL_MEMSET.
2489 (partition_builtin_p): Support above new enum item.
2490 (generate_code_for_partition): Ditto.
2491 (compute_access_range): Differentiate cases that equality can be
2492 proven at all loops, the innermost loops or no loops.
2493 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2494 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2495 (finalize_partitions, distribute_loop): Don't fuse partition of
2496 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2497 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2500 2018-01-23 Martin Liska <mliska@suse.cz>
2502 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2503 order to ignore the predictor.
2504 (PRED_POLYMORPHIC_CALL): Likewise.
2505 (PRED_RECURSIVE_CALL): Likewise.
2507 2018-01-23 Martin Liska <mliska@suse.cz>
2509 * tree-profile.c (tree_profiling): Print function header to
2510 aware reader which function we are working on.
2511 * value-prof.c (gimple_find_values_to_profile): Do not print
2512 not interesting value histograms.
2514 2018-01-23 Martin Liska <mliska@suse.cz>
2516 * profile-count.h (enum profile_quality): Add
2517 profile_uninitialized as the first value. Do not number values
2518 as they are zero based.
2519 (profile_count::verify): Update sanity check.
2520 (profile_probability::verify): Likewise.
2522 2018-01-23 Nathan Sidwell <nathan@acm.org>
2524 * doc/invoke.texi (ffor-scope): Deprecate.
2526 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2528 PR tree-optimization/83510
2529 * domwalk.c (set_all_edges_as_executable): New function.
2530 (dom_walker::dom_walker): Convert bool param
2531 "skip_unreachable_blocks" to enum reachability. Move setup of
2532 edge flags to set_all_edges_as_executable and only do it when
2533 reachability is REACHABLE_BLOCKS.
2534 * domwalk.h (enum dom_walker::reachability): New enum.
2535 (dom_walker::dom_walker): Convert bool param
2536 "skip_unreachable_blocks" to enum reachability.
2537 (set_all_edges_as_executable): New decl.
2538 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2539 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2541 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2542 but converting true to REACHABLE_BLOCKS.
2543 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2545 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2546 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2547 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2549 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2550 if check_all_array_refs will be called.
2552 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2554 * tree.c (selftest::test_location_wrappers): Add more test
2557 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2559 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2560 (selftest::test_bit_in_range): Likewise.
2562 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2565 * doc/sourcebuild.texi (vect_float): Say that the selector
2566 only describes the situation when -funsafe-math-optimizations is on.
2567 (vect_float_strict): Document.
2569 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2571 PR tree-optimization/83965
2572 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2573 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2574 instead of checking only for a reduction.
2575 (vect_recog_widen_sum_pattern): Likewise.
2577 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2579 * predict.c (probably_never_executed): Only use precise profile info.
2580 (compute_function_frequency): Skip after inlining hack since we now
2581 have quality checking.
2583 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2585 * profile-count.h (profile_probability::very_unlikely,
2586 profile_probability::unlikely, profile_probability::even): Set
2587 precision to guessed.
2589 2018-01-23 Richard Biener <rguenther@suse.de>
2591 PR tree-optimization/83963
2592 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2593 Properly terminate dominator walk when crossing the exit edge not
2594 when visiting its source block.
2596 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2599 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2600 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2602 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2604 PR tree-optimization/83957
2605 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2606 semicolon after for body surrounded by braces.
2608 PR tree-optimization/83081
2609 * profile-count.h (profile_probability::split): New method.
2610 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2611 Use profile_probability::split.
2612 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2613 when splitting a single conditional jump into 2.
2615 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2617 PR tree-optimization/69452
2618 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2621 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2623 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2624 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2625 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2627 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2629 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2630 * config/rl78/rl78.md (movdi): New define_expand.
2631 * config/rl78/rl78.c (rl78_split_movdi): New function.
2633 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2636 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2638 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2639 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2640 128-bit to produce an UNSPEC move to get the double word with the
2641 signbit and then a shift directly to do signbit.
2642 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2643 implementation with a new version that just does either a direct
2644 move or a regular move. Move memory interface to separate insns.
2645 Move insns so they are next to the expander.
2646 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2647 with signbit move. Split big and little endian case.
2648 (signbit<mode>2_dm_mem_le): Likewise.
2649 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2650 (signbit<mode>2_dm2): Likewise.
2652 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2654 * config/rl78/rl78.md (anddi3): New define_expand.
2656 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2658 * config/rl78/rl78.md (umindi3): New define_expand.
2660 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2662 * config/rl78/rl78.md (smindi3): New define_expand.
2664 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2666 * config/rl78/rl78.md (smaxdi3): New define_expand.
2668 2018-01-22 Carl Love <cel@us.ibm.com>
2670 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2671 LVX_V1TI): Add macro expansion.
2672 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2673 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2674 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2675 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2676 Change check to determine if the instruction is a byte reversing
2677 entry. Fix typo in comment.
2678 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2679 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2680 Add def_builtin calls for new builtins.
2681 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2682 Add define_insn expansion.
2684 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2686 * config/rl78/rl78.md (umaxdi3): New define_expand.
2688 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2690 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2691 for non-QImode registers.
2693 2018-01-22 Richard Biener <rguenther@suse.de>
2695 PR tree-optimization/83963
2696 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2697 including the loop exit block.
2698 (scop_detection::merge_sese): Likewise.
2699 (scop_detection::add_scop): Do it here instead.
2701 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2703 * doc/sourcebuild.texi (arm_softfloat): Document.
2705 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2708 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2709 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2710 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2712 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2713 David Edelsohn <dje.gcc@gmail.com>
2716 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2717 Change "crset eq" to "crset 2".
2718 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2719 (*call_indirect_aix<mode>_nospec): Likewise.
2720 (*call_value_indirect_aix<mode>_nospec): Likewise.
2721 (*call_indirect_elfv2<mode>_nospec): Likewise.
2722 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2723 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2724 change assembly output from . to $.
2725 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2726 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2727 (*tablejump<mode>_internal1_nospec): Likewise.
2729 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2732 * config/sh/sh_optimize_sett_clrt.cc:
2733 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2735 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2737 PR tree-optimization/83940
2738 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2739 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2740 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2741 use it to pass back the definition type.
2742 (vect_check_store_rhs): Likewise rhs_dt_out.
2743 (vect_build_gather_load_calls): Add a mask_dt argument and use
2744 it instead of a call to vect_is_simple_use.
2745 (vectorizable_store): Update calls to vect_check_load_store_mask
2746 and vect_check_store_rhs. Use the dt returned by the latter instead
2747 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2748 instead of calls to vect_is_simple_use. Pass the scalar rather
2749 than the vector operand to vect_is_simple_use when handling
2750 second and subsequent copies of an rhs value.
2751 (vectorizable_load): Update calls to vect_check_load_store_mask
2752 and vect_build_gather_load_calls. Use the cached mask_dt and
2753 gs_info.offset_dt instead of calls to vect_is_simple_use.
2755 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2758 * tree-emutls.c: Include gimplify.h.
2759 (lower_emutls_2): New function.
2760 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2761 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2762 it before further processing.
2765 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2766 UINTVAL (trueop1) instead of INTVAL (op1).
2768 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2772 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2773 INCOMING_FRAME_SP_OFFSET if not defined.
2774 (scan_trace): Add ENTRY argument. If true and
2775 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2776 emit a note to adjust the CFA offset.
2777 (create_cfi_notes): Adjust scan_trace callers.
2778 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2779 INCOMING_FRAME_SP_OFFSET in the CIE.
2780 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2781 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2783 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2784 * doc/tm.texi: Regenerated.
2786 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2788 PR rtl-optimization/83147
2789 * lra-constraints.c (remove_inheritance_pseudos): Use
2790 lra_substitute_pseudo_within_insn.
2792 2018-01-19 Tom de Vries <tom@codesourcery.com>
2793 Cesar Philippidis <cesar@codesourcery.com>
2796 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2798 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2801 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2802 spaces for function labels.
2804 2018-01-19 Martin Liska <mliska@suse.cz>
2806 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2807 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2808 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2809 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2810 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2811 (PRED_CONST_RETURN): Change from 69 to 65.
2812 (PRED_NULL_RETURN): Change from 91 to 71.
2813 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2814 (PRED_LOOP_GUARD): Change from 66 to 73.
2816 2018-01-19 Martin Liska <mliska@suse.cz>
2818 * predict.c (predict_insn_def): Add new assert.
2819 (struct branch_predictor): Change type to signed integer.
2820 (test_prediction_value_range): Amend test to cover
2822 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2823 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2824 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2825 (PRED_LOOP_IV_COMPARE): Likewise.
2826 * predict.h (PROB_UNINITIALIZED): Define new constant.
2828 2018-01-19 Martin Liska <mliska@suse.cz>
2830 * predict.c (dump_prediction): Add new format for
2831 analyze_brprob.py script which is enabled with -details
2833 * profile-count.h (precise_p): New function.
2835 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2837 PR tree-optimization/83922
2838 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2839 there are no statements that need masking.
2840 (vect_active_double_reduction_p): New function.
2841 (vect_analyze_loop_operations): Use it when handling phis that
2842 are not in the loop header.
2844 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2846 PR tree-optimization/83914
2847 * tree-vect-loop.c (vectorizable_induction): Don't convert
2848 init_expr or apply the peeling adjustment for inductions
2849 that are nested within the vectorized loop.
2851 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2853 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2856 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2860 * function.h (gimplify_parameters): Add gimple_seq * argument.
2861 * function.c: Include gimple.h and options.h.
2862 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2863 for the added local temporaries if needed.
2864 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2865 if there are any parameter cleanups, wrap whole body into a
2866 try/finally with the cleanups.
2868 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2871 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2872 Use GET_MODE_CLASS for scalar floating point.
2874 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2878 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2879 Fix call of call_cgraph_insertion_hooks.
2881 2018-01-18 Martin Sebor <msebor@redhat.com>
2883 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2885 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2888 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2891 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2894 * common.opt: (-ffile-prefix-map): New option.
2895 * opts.c (common_handle_option): Defer it.
2896 * opts-global.c (handle_common_deferred_options): Handle it.
2897 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2898 * file-prefix-map.h: New file.
2899 (remap_debug_filename, add_debug_prefix_map): ...here.
2900 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2901 * final.c (debug_prefix_map, add_debug_prefix_map
2902 remap_debug_filename): Move to...
2903 * file-prefix-map.c: New file.
2904 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2905 generalize, get rid of alloca(), use strrchr() instead of strchr().
2906 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2907 Implement in terms of add_prefix_map().
2908 (remap_macro_filename, remap_debug_filename): Implement in term of
2910 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2911 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2912 * dbxout.c: Include file-prefix-map.h.
2913 * varasm.c: Likewise.
2914 * vmsdbgout.c: Likewise.
2915 * xcoffout.c: Likewise.
2916 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2917 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2918 * doc/invoke.texi (-ffile-prefix-map): Document.
2919 (-fdebug-prefix-map): Update description.
2921 2018-01-18 Martin Liska <mliska@suse.cz>
2923 * config/i386/i386.c (indirect_thunk_name): Document that also
2925 (output_indirect_thunk): Document why both instructions
2926 (pause and lfence) are generated.
2928 2018-01-18 Richard Biener <rguenther@suse.de>
2930 PR tree-optimization/83887
2931 * graphite-scop-detection.c
2932 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2933 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2934 (scop_detection::merge_sese): Re-implement with a flood-fill
2935 algorithm that properly finds a SESE region if it exists.
2937 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2940 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2941 pointer_diff optimizations use view_convert instead of convert.
2943 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2945 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2946 Generate different code for -mno-speculate-indirect-jumps.
2947 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2948 (*call_indirect_aix<mode>): Disable for
2949 -mno-speculate-indirect-jumps.
2950 (*call_indirect_aix<mode>_nospec): New define_insn.
2951 (*call_value_indirect_aix<mode>): Disable for
2952 -mno-speculate-indirect-jumps.
2953 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2954 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2955 -mno-speculate-indirect-jumps.
2956 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2958 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2960 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2961 long double type, set the flags for noting the default long double
2962 type, even if we don't pass or return a long double type.
2964 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2967 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2970 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2972 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2973 support for merge[hl].
2974 (fold_mergehl_helper): New helper function.
2975 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2976 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2977 (altivec_vmrglw_direct): Add xxmrglw insn.
2979 2018-01-17 Andrew Waterman <andrew@sifive.com>
2981 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2982 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2984 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2987 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2988 call the lto_location_cache before reading the
2989 DECL_SOURCE_LOCATION of the types.
2991 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2992 Richard Sandiford <richard.sandiford@linaro.org>
2994 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2995 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2996 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2997 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2998 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3000 * config/aarch64/constraints.md (aarch64_movti_operand):
3002 * config/aarch64/predicates.md (Uti): Add new constraint.
3004 2018-01-17 Carl Love <cel@us.ibm.com>
3006 * config/rs6000/vsx.md (define_expand xl_len_r,
3007 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3008 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3010 (define_expand, define_insn): Move the shift left from the
3011 define_insn to the define_expand for lxvl and stxvl instructions.
3012 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3013 and XL_LEN_R definitions to PURE.
3015 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3017 * config/i386/i386.c (indirect_thunk_name): Declare regno
3018 as unsigned int. Compare regno with INVALID_REGNUM.
3019 (output_indirect_thunk): Ditto.
3020 (output_indirect_thunk_function): Ditto.
3021 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3022 in the call to output_indirect_thunk_function.
3024 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3027 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3028 rather than the size of inner_type to determine the stack slot size
3029 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3031 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3034 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3037 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3039 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3040 endian Linux systems to optionally enable multilibs for selecting
3041 the long double type if the user configured an explicit type.
3042 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3043 have no long double multilibs if not defined.
3044 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3045 warn if the user used -mabi={ieee,ibm}longdouble and we built
3046 multilibs for long double.
3047 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3048 appropriate multilib option.
3049 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3051 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3052 for building long double multilibs.
3053 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3055 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3057 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3060 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3062 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3065 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3068 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3071 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3073 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3074 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3076 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3078 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3079 different rtl trees depending on TARGET_64BIT.
3080 (rs6000_gen_lvx): Likewise.
3082 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3084 * config/visium/visium.md (nop): Tweak comment.
3085 (hazard_nop): Likewise.
3087 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3089 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3090 -mspeculate-indirect-jumps.
3091 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3092 for -mno-speculate-indirect-jumps.
3093 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3094 (*call_value_indirect_elfv2<mode>): Disable for
3095 -mno-speculate-indirect-jumps.
3096 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3097 (indirect_jump): Emit different RTL for
3098 -mno-speculate-indirect-jumps.
3099 (*indirect_jump<mode>): Disable for
3100 -mno-speculate-indirect-jumps.
3101 (*indirect_jump<mode>_nospec): New define_insn.
3102 (tablejump): Emit different RTL for
3103 -mno-speculate-indirect-jumps.
3104 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3105 (tablejumpsi_nospec): New define_expand.
3106 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3107 (tablejumpdi_nospec): New define_expand.
3108 (*tablejump<mode>_internal1): Disable for
3109 -mno-speculate-indirect-jumps.
3110 (*tablejump<mode>_internal1_nospec): New define_insn.
3111 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3114 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3116 * caller-save.c (insert_save): Drop unnecessary parameter. All
3119 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3120 Richard Biener <rguenth@suse.de>
3123 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3124 return early, inline manually is_gimple_sizepos. Make sure if we
3125 call gimplify_expr we don't end up with a gimple constant.
3126 * tree.c (variably_modified_type_p): Don't return true for
3127 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3128 * gimplify.h (is_gimple_sizepos): Remove.
3130 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3132 PR tree-optimization/83857
3133 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3134 vectorizable_live_operation for pure SLP statements.
3135 (vectorizable_live_operation): Handle PHIs.
3137 2018-01-16 Richard Biener <rguenther@suse.de>
3139 PR tree-optimization/83867
3140 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3141 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3143 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3146 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3147 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3148 If off is not INTEGER_CST, issue a may not be aligned warning
3149 rather than isn't aligned. Use isn%'t rather than isn't.
3150 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3152 <case MULT_EXPR>: Improve the case when bottom and one of the
3153 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3154 operand, in that case check if the other operand is multiple of
3155 bottom divided by the INTEGER_CST operand.
3157 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3160 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3161 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3162 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3163 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3164 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3165 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3166 (pa_function_arg_size): New function.
3168 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3170 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3171 in a separate statement.
3173 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3175 PR tree-optimization/83847
3176 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3177 group gathers and scatters.
3179 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3181 PR rtl-optimization/86620
3182 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3184 PR rtl-optimization/83213
3185 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3186 to last if both are JUMP_INSNs.
3188 PR tree-optimization/83843
3189 * gimple-ssa-store-merging.c
3190 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3191 store_immediate_info for bswap/nop orig_stores.
3193 2018-01-15 Andrew Waterman <andrew@sifive.com>
3195 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3197 <UDIV>: Increase cost if !TARGET_DIV.
3199 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3201 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3202 (define_attr "cr_logical_3op"): New.
3203 (cceq_ior_compare): Adjust.
3204 (cceq_ior_compare_complement): Adjust.
3205 (*cceq_rev_compare): Adjust.
3206 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3207 (is_cracked_insn): Adjust.
3208 (insn_must_be_first_in_group): Adjust.
3209 * config/rs6000/40x.md: Adjust.
3210 * config/rs6000/440.md: Adjust.
3211 * config/rs6000/476.md: Adjust.
3212 * config/rs6000/601.md: Adjust.
3213 * config/rs6000/603.md: Adjust.
3214 * config/rs6000/6xx.md: Adjust.
3215 * config/rs6000/7450.md: Adjust.
3216 * config/rs6000/7xx.md: Adjust.
3217 * config/rs6000/8540.md: Adjust.
3218 * config/rs6000/cell.md: Adjust.
3219 * config/rs6000/e300c2c3.md: Adjust.
3220 * config/rs6000/e500mc.md: Adjust.
3221 * config/rs6000/e500mc64.md: Adjust.
3222 * config/rs6000/e5500.md: Adjust.
3223 * config/rs6000/e6500.md: Adjust.
3224 * config/rs6000/mpc.md: Adjust.
3225 * config/rs6000/power4.md: Adjust.
3226 * config/rs6000/power5.md: Adjust.
3227 * config/rs6000/power6.md: Adjust.
3228 * config/rs6000/power7.md: Adjust.
3229 * config/rs6000/power8.md: Adjust.
3230 * config/rs6000/power9.md: Adjust.
3231 * config/rs6000/rs64.md: Adjust.
3232 * config/rs6000/titan.md: Adjust.
3234 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3236 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3237 ix86_indirect_branch_register logic.
3239 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3241 * config/i386/constraints.md (Bs): Update
3242 ix86_indirect_branch_register check. Don't check
3243 ix86_indirect_branch_register with GOT_memory_operand.
3245 * config/i386/predicates.md (GOT_memory_operand): Don't check
3246 ix86_indirect_branch_register here.
3247 (GOT32_symbol_operand): Likewise.
3249 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3251 * config/i386/predicates.md (constant_call_address_operand):
3252 Rewrite ix86_indirect_branch_register logic.
3253 (sibcall_insn_operand): Likewise.
3255 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3257 * config/i386/constraints.md (Bs): Replace
3258 ix86_indirect_branch_thunk_register with
3259 ix86_indirect_branch_register.
3261 * config/i386/i386.md (indirect_jump): Likewise.
3262 (tablejump): Likewise.
3263 (*sibcall_memory): Likewise.
3264 (*sibcall_value_memory): Likewise.
3265 Peepholes of indirect call and jump via memory: Likewise.
3266 * config/i386/i386.opt: Likewise.
3267 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3268 (GOT_memory_operand): Likewise.
3269 (call_insn_operand): Likewise.
3270 (sibcall_insn_operand): Likewise.
3271 (GOT32_symbol_operand): Likewise.
3273 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3276 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3277 type rather than type addr's type points to.
3278 (expand_omp_atomic_mutex): Likewise.
3279 (expand_omp_atomic): Likewise.
3281 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3284 * config/i386/i386.c (output_indirect_thunk_function): Use
3285 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3286 for __x86_return_thunk.
3288 2018-01-15 Richard Biener <rguenther@suse.de>
3291 * expmed.c (extract_bit_field_1): Fix typo.
3293 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3296 * config/arm/iterators.md (VF): New mode iterator.
3297 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3298 Remove integer-related logic from pattern.
3299 (neon_vabd<mode>_3): Likewise.
3301 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3304 * common.opt (fstrict-overflow): No longer an alias.
3305 (fwrapv-pointer): New option.
3306 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3307 also for pointer types based on flag_wrapv_pointer.
3308 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3309 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3310 opts->x_flag_wrapv got set.
3311 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3312 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3313 POINTER_TYPE_OVERFLOW_UNDEFINED.
3314 * match.pd: Likewise in address comparison pattern.
3315 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3317 2018-01-15 Richard Biener <rguenther@suse.de>
3320 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3321 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3322 Reset type names to their identifier if their TYPE_DECL doesn't
3323 have linkage (and thus is used for ODR and devirt).
3324 (save_debug_info_for_decl): Remove.
3325 (save_debug_info_for_type): Likewise.
3326 (add_tree_to_fld_list): Adjust.
3327 * tree-pretty-print.c (dump_generic_node): Make dumping of
3328 type names more robust.
3330 2018-01-15 Richard Biener <rguenther@suse.de>
3332 * BASE-VER: Bump to 8.0.1.
3334 2018-01-14 Martin Sebor <msebor@redhat.com>
3337 * builtins.c (check_access): Avoid warning when the no-warning bit
3340 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3342 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3343 * ira-color (allocno_hard_regs_compare): Likewise.
3345 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3348 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3349 Use .pushsection/.popsection.
3351 2018-01-14 Martin Sebor <msebor@redhat.com>
3354 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3356 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3358 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3359 entry from extra_headers.
3360 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3361 extra_headers, make the list bitwise identical to the i?86-*-* one.
3363 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3365 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3366 -mcmodel=large with -mindirect-branch=thunk,
3367 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3368 -mfunction-return=thunk-extern.
3369 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3370 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3371 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3373 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3375 * config/i386/i386.c (print_reg): Print the name of the full
3376 integer register without '%'.
3377 (ix86_print_operand): Handle 'V'.
3378 * doc/extend.texi: Document 'V' modifier.
3380 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3382 * config/i386/constraints.md (Bs): Disallow memory operand for
3383 -mindirect-branch-register.
3385 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3386 (GOT_memory_operand): Likewise.
3387 (call_insn_operand): Likewise.
3388 (sibcall_insn_operand): Likewise.
3389 (GOT32_symbol_operand): Likewise.
3390 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3391 for -mindirect-branch-register.
3392 (tablejump): Likewise.
3393 (*sibcall_memory): Likewise.
3394 (*sibcall_value_memory): Likewise.
3395 Disallow peepholes of indirect call and jump via memory for
3396 -mindirect-branch-register.
3397 (*call_pop): Replace m with Bw.
3398 (*call_value_pop): Likewise.
3399 (*sibcall_pop_memory): Replace m with Bs.
3400 * config/i386/i386.opt (mindirect-branch-register): New option.
3401 * doc/invoke.texi: Document -mindirect-branch-register option.
3403 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3405 * config/i386/i386-protos.h (ix86_output_function_return): New.
3406 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3407 set function_return_type.
3408 (indirect_thunk_name): Add ret_p to indicate thunk for function
3410 (output_indirect_thunk_function): Pass false to
3411 indirect_thunk_name.
3412 (ix86_output_indirect_branch_via_reg): Likewise.
3413 (ix86_output_indirect_branch_via_push): Likewise.
3414 (output_indirect_thunk_function): Create alias for function
3415 return thunk if regno < 0.
3416 (ix86_output_function_return): New function.
3417 (ix86_handle_fndecl_attribute): Handle function_return.
3418 (ix86_attribute_table): Add function_return.
3419 * config/i386/i386.h (machine_function): Add
3420 function_return_type.
3421 * config/i386/i386.md (simple_return_internal): Use
3422 ix86_output_function_return.
3423 (simple_return_internal_long): Likewise.
3424 * config/i386/i386.opt (mfunction-return=): New option.
3425 (indirect_branch): Mention -mfunction-return=.
3426 * doc/extend.texi: Document function_return function attribute.
3427 * doc/invoke.texi: Document -mfunction-return= option.
3429 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3431 * config/i386/i386-opts.h (indirect_branch): New.
3432 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3433 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3434 with local indirect jump when converting indirect call and jump.
3435 (ix86_set_indirect_branch_type): New.
3436 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3437 (indirectlabelno): New.
3438 (indirect_thunk_needed): Likewise.
3439 (indirect_thunk_bnd_needed): Likewise.
3440 (indirect_thunks_used): Likewise.
3441 (indirect_thunks_bnd_used): Likewise.
3442 (INDIRECT_LABEL): Likewise.
3443 (indirect_thunk_name): Likewise.
3444 (output_indirect_thunk): Likewise.
3445 (output_indirect_thunk_function): Likewise.
3446 (ix86_output_indirect_branch_via_reg): Likewise.
3447 (ix86_output_indirect_branch_via_push): Likewise.
3448 (ix86_output_indirect_branch): Likewise.
3449 (ix86_output_indirect_jmp): Likewise.
3450 (ix86_code_end): Call output_indirect_thunk_function if needed.
3451 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3453 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3454 (ix86_attribute_table): Add indirect_branch.
3455 * config/i386/i386.h (machine_function): Add indirect_branch_type
3456 and has_local_indirect_jump.
3457 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3459 (tablejump): Likewise.
3460 (*indirect_jump): Use ix86_output_indirect_jmp.
3461 (*tablejump_1): Likewise.
3462 (simple_return_indirect_internal): Likewise.
3463 * config/i386/i386.opt (mindirect-branch=): New option.
3464 (indirect_branch): New.
3467 (thunk-inline): Likewise.
3468 (thunk-extern): Likewise.
3469 * doc/extend.texi: Document indirect_branch function attribute.
3470 * doc/invoke.texi: Document -mindirect-branch= option.
3472 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3475 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3477 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3479 * ipa-inline.c (want_inline_small_function_p): Return false if
3480 inlining has already failed with CIF_FINAL_ERROR.
3481 (update_caller_keys): Call want_inline_small_function_p before
3483 (update_callee_keys): Likewise.
3485 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3487 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3489 (rs6000_quadword_masked_address_p): Likewise.
3490 (quad_aligned_load_p): Likewise.
3491 (quad_aligned_store_p): Likewise.
3492 (const_load_sequence_p): Add comment to describe the outer-most loop.
3493 (mimic_memory_attributes_and_flags): New function.
3494 (rs6000_gen_stvx): Likewise.
3495 (replace_swapped_aligned_store): Likewise.
3496 (rs6000_gen_lvx): Likewise.
3497 (replace_swapped_aligned_load): Likewise.
3498 (replace_swapped_load_constant): Capitalize argument name in
3499 comment describing this function.
3500 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3501 and stores that access quad-word aligned addresses and replace
3502 with stvx or lvx instructions when appropriate.
3503 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3504 New function prototype.
3505 (rs6000_quadword_masked_address_p): Likewise.
3506 (rs6000_gen_lvx): Likewise.
3507 (rs6000_gen_stvx): Likewise.
3508 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3509 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3510 when memory address is aligned.
3511 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3512 this split to select lvx instruction when memory address is aligned.
3513 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3514 instruction when memory address is aligned.
3515 (*vsx_le_perm_load_v16qi): Likewise.
3516 (four unnamed splitters): Modify to select the stvx instruction
3517 when memory is aligned.
3519 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3521 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3522 which appears in the queue multiple times.
3524 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3525 Alan Hayward <alan.hayward@arm.com>
3526 David Sherwood <david.sherwood@arm.com>
3528 * tree-vectorizer.h (vec_lower_bound): New structure.
3529 (_loop_vec_info): Add check_nonzero and lower_bounds.
3530 (LOOP_VINFO_CHECK_NONZERO): New macro.
3531 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3532 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3533 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3534 fields. Make seg_len the distance travelled, not including the
3536 (dr_direction_indicator): Declare.
3537 (dr_zero_step_indicator): Likewise.
3538 (dr_known_forward_stride_p): Likewise.
3539 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3541 (runtime_alias_check_p): Allow runtime alias checks with
3543 (operator ==): Compare access_size and align.
3544 (prune_runtime_alias_test_list): Rework for new distinction between
3545 the access_size and seg_len.
3546 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3548 (get_segment_min_max): New function.
3549 (create_intersect_range_checks): Use it.
3550 (dr_step_indicator): New function.
3551 (dr_direction_indicator): Likewise.
3552 (dr_zero_step_indicator): Likewise.
3553 (dr_known_forward_stride_p): Likewise.
3554 * tree-loop-distribution.c (data_ref_segment_size): Return
3555 DR_STEP * (niters - 1).
3556 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3558 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3559 (vect_preserves_scalar_order_p): New function, split out from...
3560 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3561 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3562 (vect_vfa_access_size): New function.
3563 (vect_vfa_align): Likewise.
3564 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3565 (dump_lower_bound): New function.
3566 (vect_check_lower_bound): Likewise.
3567 (vect_small_gap_p): Likewise.
3568 (vectorizable_with_step_bound_p): Likewise.
3569 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3570 depencies if the vectorization factor is 1. Convert the checks
3571 for nonzero steps into checks on the bounds of DR_STEP. Try using
3572 a bunds check for variable steps if the minimum required step is
3573 relatively small. Update calls to the dr_with_seg_len
3574 constructor and to vect_compile_time_alias.
3575 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3577 (vect_loop_versioning): Call it.
3578 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3580 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3582 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3583 Alan Hayward <alan.hayward@arm.com>
3584 David Sherwood <david.sherwood@arm.com>
3586 * doc/sourcebuild.texi (vect_scatter_store): Document.
3587 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3589 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3591 * genopinit.c (main): Add supports_vec_scatter_store and
3592 supports_vec_scatter_store_cached to target_optabs.
3593 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3594 IFN_MASK_SCATTER_STORE.
3595 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3597 * internal-fn.h (internal_store_fn_p): Declare.
3598 (internal_fn_stored_value_index): Likewise.
3599 * internal-fn.c (scatter_store_direct): New macro.
3600 (expand_scatter_store_optab_fn): New function.
3601 (direct_scatter_store_optab_supported_p): New macro.
3602 (internal_store_fn_p): New function.
3603 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3604 IFN_MASK_SCATTER_STORE.
3605 (internal_fn_mask_index): Likewise.
3606 (internal_fn_stored_value_index): New function.
3607 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3609 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3610 * optabs-query.c (supports_vec_scatter_store_p): New function.
3611 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3612 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3613 true for scatter stores.
3614 (vect_gather_scatter_fn_p): Handle scatter stores too.
3615 (vect_check_gather_scatter): Consider using scatter stores if
3616 supports_vec_scatter_store_p.
3617 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3619 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3620 internal_fn_stored_value_index.
3621 (check_load_store_masking): Handle scatter stores too.
3622 (vect_get_store_rhs): Make public.
3623 (vectorizable_call): Use internal_store_fn_p.
3624 (vectorizable_store): Handle scatter store internal functions.
3625 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3626 when deciding whether the end of the group has been reached.
3627 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3628 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3629 (mask_scatter_store<mode>): New insns.
3631 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3632 Alan Hayward <alan.hayward@arm.com>
3633 David Sherwood <david.sherwood@arm.com>
3635 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3636 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3637 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3639 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3640 Use vect_truncate_gather_scatter_offset if we can't treat the
3641 operation as a normal gather load or scatter store.
3642 (get_group_load_store_type): Take the gather_scatter_info
3643 as argument. Try using a gather load or scatter store for
3644 single-element groups.
3645 (get_load_store_type): Update calls to get_group_load_store_type
3646 and vect_use_strided_gather_scatters_p.
3648 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3649 Alan Hayward <alan.hayward@arm.com>
3650 David Sherwood <david.sherwood@arm.com>
3652 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3653 optional tree argument.
3654 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3656 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3657 but continue to use the current value as a fallback.
3658 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3659 to compare the updates.
3660 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3661 (get_load_store_type): Use it when handling a strided access.
3662 (vect_get_strided_load_store_ops): New function.
3663 (vect_get_data_ptr_increment): Likewise.
3664 (vectorizable_load): Handle strided gather loads. Always pass
3665 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3667 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3668 Alan Hayward <alan.hayward@arm.com>
3669 David Sherwood <david.sherwood@arm.com>
3671 * doc/md.texi (gather_load@var{m}): Document.
3672 (mask_gather_load@var{m}): Likewise.
3673 * genopinit.c (main): Add supports_vec_gather_load and
3674 supports_vec_gather_load_cached to target_optabs.
3675 * optabs-tree.c (init_tree_optimization_optabs): Use
3676 ggc_cleared_alloc to allocate target_optabs.
3677 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3678 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3680 * internal-fn.h (internal_load_fn_p): Declare.
3681 (internal_gather_scatter_fn_p): Likewise.
3682 (internal_fn_mask_index): Likewise.
3683 (internal_gather_scatter_fn_supported_p): Likewise.
3684 * internal-fn.c (gather_load_direct): New macro.
3685 (expand_gather_load_optab_fn): New function.
3686 (direct_gather_load_optab_supported_p): New macro.
3687 (direct_internal_fn_optab): New function.
3688 (internal_load_fn_p): Likewise.
3689 (internal_gather_scatter_fn_p): Likewise.
3690 (internal_fn_mask_index): Likewise.
3691 (internal_gather_scatter_fn_supported_p): Likewise.
3692 * optabs-query.c (supports_at_least_one_mode_p): New function.
3693 (supports_vec_gather_load_p): Likewise.
3694 * optabs-query.h (supports_vec_gather_load_p): Declare.
3695 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3696 and memory_type field.
3697 (NUM_PATTERNS): Bump to 15.
3698 * tree-vect-data-refs.c: Include internal-fn.h.
3699 (vect_gather_scatter_fn_p): New function.
3700 (vect_describe_gather_scatter_call): Likewise.
3701 (vect_check_gather_scatter): Try using internal functions for
3702 gather loads. Recognize existing calls to a gather load function.
3703 (vect_analyze_data_refs): Consider using gather loads if
3704 supports_vec_gather_load_p.
3705 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3706 (vect_get_gather_scatter_offset_type): Likewise.
3707 (vect_convert_mask_for_vectype): Likewise.
3708 (vect_add_conversion_to_patterm): Likewise.
3709 (vect_try_gather_scatter_pattern): Likewise.
3710 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3711 (vect_vect_recog_func_ptrs): Add it.
3712 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3713 internal_fn_mask_index and internal_gather_scatter_fn_p.
3714 (check_load_store_masking): Take the gather_scatter_info as an
3715 argument and handle gather loads.
3716 (vect_get_gather_scatter_ops): New function.
3717 (vectorizable_call): Check internal_load_fn_p.
3718 (vectorizable_load): Likewise. Handle gather load internal
3720 (vectorizable_store): Update call to check_load_store_masking.
3721 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3722 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3723 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3724 (aarch64_gather_scale_operand_d): New predicates.
3725 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3726 (mask_gather_load<mode>): New insns.
3728 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3729 Alan Hayward <alan.hayward@arm.com>
3730 David Sherwood <david.sherwood@arm.com>
3732 * optabs.def (fold_left_plus_optab): New optab.
3733 * doc/md.texi (fold_left_plus_@var{m}): Document.
3734 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3735 * internal-fn.c (fold_left_direct): Define.
3736 (expand_fold_left_optab_fn): Likewise.
3737 (direct_fold_left_optab_supported_p): Likewise.
3738 * fold-const-call.c (fold_const_fold_left): New function.
3739 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3740 * tree-parloops.c (valid_reduction_p): New function.
3741 (gather_scalar_reductions): Use it.
3742 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3743 (vect_finish_replace_stmt): Declare.
3744 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3745 (needs_fold_left_reduction_p): New function, split out from...
3746 (vect_is_simple_reduction): ...here. Accept reductions that
3747 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3748 (vect_force_simple_reduction): Also store the reduction type in
3749 the assignment's STMT_VINFO_REDUC_TYPE.
3750 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3751 (merge_with_identity): New function.
3752 (vect_expand_fold_left): Likewise.
3753 (vectorize_fold_left_reduction): Likewise.
3754 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3755 scalar phi in place for it. Check for target support and reject
3756 cases that would reassociate the operation. Defer the transform
3757 phase to vectorize_fold_left_reduction.
3758 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3759 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3760 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3762 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3764 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3765 call to ifc_temp_var.
3767 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3768 Alan Hayward <alan.hayward@arm.com>
3769 David Sherwood <david.sherwood@arm.com>
3771 * target.def (legitimize_address_displacement): Take the original
3772 offset as a poly_int.
3773 * targhooks.h (default_legitimize_address_displacement): Update
3775 * targhooks.c (default_legitimize_address_displacement): Likewise.
3776 * doc/tm.texi: Regenerate.
3777 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3778 as an argument, moving assert of ad->disp == ad->disp_term to...
3779 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3780 Try calling targetm.legitimize_address_displacement before expanding
3781 the address rather than afterwards, and adjust for the new interface.
3782 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3783 Match the new hook interface. Handle SVE addresses.
3784 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3787 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3789 * Makefile.in (OBJS): Add early-remat.o.
3790 * target.def (select_early_remat_modes): New hook.
3791 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3792 * doc/tm.texi: Regenerate.
3793 * targhooks.h (default_select_early_remat_modes): Declare.
3794 * targhooks.c (default_select_early_remat_modes): New function.
3795 * timevar.def (TV_EARLY_REMAT): New timevar.
3796 * passes.def (pass_early_remat): New pass.
3797 * tree-pass.h (make_pass_early_remat): Declare.
3798 * early-remat.c: New file.
3799 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3801 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3803 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3804 Alan Hayward <alan.hayward@arm.com>
3805 David Sherwood <david.sherwood@arm.com>
3807 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3808 vfm1 with a bound_epilog parameter.
3809 (vect_do_peeling): Update calls accordingly, and move the prologue
3810 call earlier in the function. Treat the base bound_epilog as 0 for
3811 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3812 this base when peeling for gaps.
3813 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3814 with fully-masked loops.
3815 (vect_estimate_min_profitable_iters): Handle the single peeled
3816 iteration in that case.
3818 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3819 Alan Hayward <alan.hayward@arm.com>
3820 David Sherwood <david.sherwood@arm.com>
3822 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3823 single-element interleaving even if the size is not a power of 2.
3824 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3825 accesses for single-element interleaving if the group size is
3828 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3829 Alan Hayward <alan.hayward@arm.com>
3830 David Sherwood <david.sherwood@arm.com>
3832 * doc/md.texi (fold_extract_last_@var{m}): Document.
3833 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3834 * optabs.def (fold_extract_last_optab): New optab.
3835 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3836 * internal-fn.c (fold_extract_direct): New macro.
3837 (expand_fold_extract_optab_fn): Likewise.
3838 (direct_fold_extract_optab_supported_p): Likewise.
3839 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3840 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3841 EXTRACT_LAST_REDUCTION.
3842 (get_initial_def_for_reduction): Do not create an initial vector
3843 for EXTRACT_LAST_REDUCTION reductions.
3844 (vectorizable_reduction): Leave the scalar phi in place for
3845 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3846 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3847 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3848 transform phase to vectorizable_condition.
3849 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3851 (vect_finish_stmt_generation): ...here.
3852 (vect_finish_replace_stmt): New function.
3853 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3854 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3856 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3858 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3859 Alan Hayward <alan.hayward@arm.com>
3860 David Sherwood <david.sherwood@arm.com>
3862 * doc/md.texi (extract_last_@var{m}): Document.
3863 * optabs.def (extract_last_optab): New optab.
3864 * internal-fn.def (EXTRACT_LAST): New internal function.
3865 * internal-fn.c (cond_unary_direct): New macro.
3866 (expand_cond_unary_optab_fn): Likewise.
3867 (direct_cond_unary_optab_supported_p): Likewise.
3868 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3869 loops using EXTRACT_LAST.
3870 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3871 (extract_last_<mode>): ...this optab.
3872 (vec_extract<mode><Vel>): Update accordingly.
3874 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3875 Alan Hayward <alan.hayward@arm.com>
3876 David Sherwood <david.sherwood@arm.com>
3878 * target.def (empty_mask_is_expensive): New hook.
3879 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3880 * doc/tm.texi: Regenerate.
3881 * targhooks.h (default_empty_mask_is_expensive): Declare.
3882 * targhooks.c (default_empty_mask_is_expensive): New function.
3883 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3884 if the target says that empty masks are expensive.
3885 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3887 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3889 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3890 Alan Hayward <alan.hayward@arm.com>
3891 David Sherwood <david.sherwood@arm.com>
3893 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3894 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3895 (vect_use_loop_mask_for_alignment_p): New function.
3896 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3897 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3898 niters_skip argument. Make sure that the first niters_skip elements
3899 of the first iteration are inactive.
3900 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3901 Update call to vect_set_loop_masks_directly.
3902 (get_misalign_in_elems): New function, split out from...
3903 (vect_gen_prolog_loop_niters): ...here.
3904 (vect_update_init_of_dr): Take a code argument that specifies whether
3905 the adjustment should be added or subtracted.
3906 (vect_update_init_of_drs): Likewise.
3907 (vect_prepare_for_masked_peels): New function.
3908 (vect_do_peeling): Skip prologue peeling if we're using a mask
3909 instead. Update call to vect_update_inits_of_drs.
3910 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3912 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3913 alignment. Do not include the number of peeled iterations in
3914 the minimum threshold in that case.
3915 (vectorizable_induction): Adjust the start value down by
3916 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3917 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3918 Take the number of skipped iterations into account when calculating
3920 * tree-vect-stmts.c (vect_gen_while_not): New function.
3922 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3923 Alan Hayward <alan.hayward@arm.com>
3924 David Sherwood <david.sherwood@arm.com>
3926 * doc/sourcebuild.texi (vect_fully_masked): Document.
3927 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3929 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3931 (vect_analyze_loop_2): ...here. Don't check the vectorization
3932 factor against the number of loop iterations if the loop is
3935 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3936 Alan Hayward <alan.hayward@arm.com>
3937 David Sherwood <david.sherwood@arm.com>
3939 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3940 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3941 (dump_groups): Update accordingly.
3942 (iv_use::mem_type): New member variable.
3943 (address_p): New function.
3944 (record_use): Add a mem_type argument and initialize the new
3946 (record_group_use): Add a mem_type argument. Use address_p.
3947 Remove obsolete null checks of base_object. Update call to record_use.
3948 (find_interesting_uses_op): Update call to record_group_use.
3949 (find_interesting_uses_cond): Likewise.
3950 (find_interesting_uses_address): Likewise.
3951 (get_mem_type_for_internal_fn): New function.
3952 (find_address_like_use): Likewise.
3953 (find_interesting_uses_stmt): Try find_address_like_use before
3954 calling find_interesting_uses_op.
3955 (addr_offset_valid_p): Use the iv mem_type field as the type
3956 of the addressed memory.
3957 (add_autoinc_candidates): Likewise.
3958 (get_address_cost): Likewise.
3959 (split_small_address_groups_p): Use address_p.
3960 (split_address_groups): Likewise.
3961 (add_iv_candidate_for_use): Likewise.
3962 (autoinc_possible_for_pair): Likewise.
3963 (rewrite_groups): Likewise.
3964 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3965 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3966 (get_alias_ptr_type_for_ptr_address): New function.
3967 (rewrite_use_address): Rewrite address uses in calls that were
3968 identified by find_address_like_use.
3970 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3971 Alan Hayward <alan.hayward@arm.com>
3972 David Sherwood <david.sherwood@arm.com>
3974 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3976 * gimple-expr.h (is_gimple_addressable: Likewise.
3977 * gimple-expr.c (is_gimple_address): Likewise.
3978 * internal-fn.c (expand_call_mem_ref): New function.
3979 (expand_mask_load_optab_fn): Use it.
3980 (expand_mask_store_optab_fn): Likewise.
3982 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3983 Alan Hayward <alan.hayward@arm.com>
3984 David Sherwood <david.sherwood@arm.com>
3986 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3987 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3988 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3989 (cond_umax@var{mode}): Document.
3990 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3991 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3992 (cond_umin_optab, cond_umax_optab): New optabs.
3993 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3994 (COND_IOR, COND_XOR): New internal functions.
3995 * internal-fn.h (get_conditional_internal_fn): Declare.
3996 * internal-fn.c (cond_binary_direct): New macro.
3997 (expand_cond_binary_optab_fn): Likewise.
3998 (direct_cond_binary_optab_supported_p): Likewise.
3999 (get_conditional_internal_fn): New function.
4000 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4001 Cope with reduction statements that are vectorized as calls rather
4003 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4004 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4005 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4006 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4007 (UNSPEC_COND_EOR): New unspecs.
4008 (optab): Add mappings for them.
4009 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4010 (sve_int_op, sve_fp_op): New int attributes.
4012 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4013 Alan Hayward <alan.hayward@arm.com>
4014 David Sherwood <david.sherwood@arm.com>
4016 * optabs.def (while_ult_optab): New optab.
4017 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4018 * internal-fn.def (WHILE_ULT): New internal function.
4019 * internal-fn.h (direct_internal_fn_supported_p): New override
4020 that takes two types as argument.
4021 * internal-fn.c (while_direct): New macro.
4022 (expand_while_optab_fn): New function.
4023 (convert_optab_supported_p): Likewise.
4024 (direct_while_optab_supported_p): New macro.
4025 * wide-int.h (wi::udiv_ceil): New function.
4026 * tree-vectorizer.h (rgroup_masks): New structure.
4027 (vec_loop_masks): New typedef.
4028 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4030 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4031 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4032 (vect_max_vf): New function.
4033 (slpeel_make_loop_iterate_ntimes): Delete.
4034 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4035 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4036 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4037 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4038 internal-fn.h, stor-layout.h and optabs-query.h.
4039 (vect_set_loop_mask): New function.
4040 (add_preheader_seq): Likewise.
4041 (add_header_seq): Likewise.
4042 (interleave_supported_p): Likewise.
4043 (vect_maybe_permute_loop_masks): Likewise.
4044 (vect_set_loop_masks_directly): Likewise.
4045 (vect_set_loop_condition_masked): Likewise.
4046 (vect_set_loop_condition_unmasked): New function, split out from
4047 slpeel_make_loop_iterate_ntimes.
4048 (slpeel_make_loop_iterate_ntimes): Rename to..
4049 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4050 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4051 (vect_do_peeling): Update call accordingly.
4052 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4054 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4055 mask_compare_type, can_fully_mask_p and fully_masked_p.
4056 (release_vec_loop_masks): New function.
4057 (_loop_vec_info): Use it to free the loop masks.
4058 (can_produce_all_loop_masks_p): New function.
4059 (vect_get_max_nscalars_per_iter): Likewise.
4060 (vect_verify_full_masking): Likewise.
4061 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4062 retries, and free the mask rgroups before retrying. Check loop-wide
4063 reasons for disallowing fully-masked loops. Make the final decision
4064 about whether use a fully-masked loop or not.
4065 (vect_estimate_min_profitable_iters): Do not assume that peeling
4066 for the number of iterations will be needed for fully-masked loops.
4067 (vectorizable_reduction): Disable fully-masked loops.
4068 (vectorizable_live_operation): Likewise.
4069 (vect_halve_mask_nunits): New function.
4070 (vect_double_mask_nunits): Likewise.
4071 (vect_record_loop_mask): Likewise.
4072 (vect_get_loop_mask): Likewise.
4073 (vect_transform_loop): Handle the case in which the final loop
4074 iteration might handle a partial vector. Call vect_set_loop_condition
4075 instead of slpeel_make_loop_iterate_ntimes.
4076 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4077 (check_load_store_masking): New function.
4078 (prepare_load_store_mask): Likewise.
4079 (vectorizable_store): Handle fully-masked loops.
4080 (vectorizable_load): Likewise.
4081 (supportable_widening_operation): Use vect_halve_mask_nunits for
4083 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4084 (vect_gen_while): New function.
4085 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4086 (aarch64_uqdec<mode>): New insn.
4088 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4089 Alan Hayward <alan.hayward@arm.com>
4090 David Sherwood <david.sherwood@arm.com>
4092 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4093 (reduc_xor_scal_optab): New optabs.
4094 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4095 (reduc_xor_scal_@var{m}): Document.
4096 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4097 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4099 * fold-const-call.c (fold_const_call): Handle them.
4100 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4101 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4102 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4103 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4104 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4105 (UNSPEC_XORV): New unspecs.
4106 (optab): Add entries for them.
4107 (BITWISEV): New int iterator.
4108 (bit_reduc_op): New int attributes.
4110 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4111 Alan Hayward <alan.hayward@arm.com>
4112 David Sherwood <david.sherwood@arm.com>
4114 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4115 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4116 * optabs.def (vec_shl_insert_optab): New optab.
4117 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4118 (duplicate_and_interleave): Likewise.
4119 * tree-vect-loop.c: Include internal-fn.h.
4120 (neutral_op_for_slp_reduction): New function, split out from
4121 get_initial_defs_for_reduction.
4122 (get_initial_def_for_reduction): Handle option 2 for variable-length
4123 vectors by loading the neutral value into a vector and then shifting
4124 the initial value into element 0.
4125 (get_initial_defs_for_reduction): Replace the code argument with
4126 the neutral value calculated by neutral_op_for_slp_reduction.
4127 Use gimple_build_vector for constant-length vectors.
4128 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4129 but the first group_size elements have a neutral value.
4130 Use duplicate_and_interleave otherwise.
4131 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4132 Update call to get_initial_defs_for_reduction. Handle SLP
4133 reductions for variable-length vectors by creating one vector
4134 result for each scalar result, with the elements associated
4135 with other scalar results stubbed out with the neutral value.
4136 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4137 Require IFN_VEC_SHL_INSERT for double reductions on
4138 variable-length vectors, or SLP reductions that have
4139 a neutral value. Require can_duplicate_and_interleave_p
4140 support for variable-length unchained SLP reductions if there
4141 is no neutral value, such as for MIN/MAX reductions. Also require
4142 the number of vector elements to be a multiple of the number of
4143 SLP statements when doing variable-length unchained SLP reductions.
4144 Update call to vect_create_epilog_for_reduction.
4145 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4146 and remove initial values.
4147 (duplicate_and_interleave): Make public.
4148 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4149 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4151 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4152 Alan Hayward <alan.hayward@arm.com>
4153 David Sherwood <david.sherwood@arm.com>
4155 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4156 (can_duplicate_and_interleave_p): New function.
4157 (vect_get_and_check_slp_defs): Take the vector of statements
4158 rather than just the current one. Remove excess parentheses.
4159 Restriction rejectinon of vect_constant_def and vect_external_def
4160 for variable-length vectors to boolean types, or types for which
4161 can_duplicate_and_interleave_p is false.
4162 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4163 (duplicate_and_interleave): New function.
4164 (vect_get_constant_vectors): Use gimple_build_vector for
4165 constant-length vectors and suitable variable-length constant
4166 vectors. Use duplicate_and_interleave for other variable-length
4167 vectors. Don't defer the update when inserting new statements.
4169 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4170 Alan Hayward <alan.hayward@arm.com>
4171 David Sherwood <david.sherwood@arm.com>
4173 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4174 min_profitable_iters doesn't go negative.
4176 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4177 Alan Hayward <alan.hayward@arm.com>
4178 David Sherwood <david.sherwood@arm.com>
4180 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4181 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4182 * optabs.def (vec_mask_load_lanes_optab): New optab.
4183 (vec_mask_store_lanes_optab): Likewise.
4184 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4185 (MASK_STORE_LANES): Likewise.
4186 * internal-fn.c (mask_load_lanes_direct): New macro.
4187 (mask_store_lanes_direct): Likewise.
4188 (expand_mask_load_optab_fn): Handle masked operations.
4189 (expand_mask_load_lanes_optab_fn): New macro.
4190 (expand_mask_store_optab_fn): Handle masked operations.
4191 (expand_mask_store_lanes_optab_fn): New macro.
4192 (direct_mask_load_lanes_optab_supported_p): Likewise.
4193 (direct_mask_store_lanes_optab_supported_p): Likewise.
4194 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4196 (vect_load_lanes_supported): Likewise.
4197 * tree-vect-data-refs.c (strip_conversion): New function.
4198 (can_group_stmts_p): Likewise.
4199 (vect_analyze_data_ref_accesses): Use it instead of checking
4200 for a pair of assignments.
4201 (vect_store_lanes_supported): Take a masked_p parameter.
4202 (vect_load_lanes_supported): Likewise.
4203 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4204 vect_store_lanes_supported and vect_load_lanes_supported.
4205 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4206 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4207 parameter. Don't allow gaps for masked accesses.
4208 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4209 and vect_load_lanes_supported.
4210 (get_load_store_type): Take a masked_p parameter and update
4211 call to get_group_load_store_type.
4212 (vectorizable_store): Update call to get_load_store_type.
4213 Handle IFN_MASK_STORE_LANES.
4214 (vectorizable_load): Update call to get_load_store_type.
4215 Handle IFN_MASK_LOAD_LANES.
4217 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4218 Alan Hayward <alan.hayward@arm.com>
4219 David Sherwood <david.sherwood@arm.com>
4221 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4223 * config/aarch64/aarch64-protos.h
4224 (aarch64_sve_struct_memory_operand_p): Declare.
4225 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4226 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4227 (VPRED, vpred): Handle SVE structure modes.
4228 * config/aarch64/constraints.md (Utx): New constraint.
4229 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4230 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4231 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4232 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4233 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4234 structure modes. Split into pieces after RA.
4235 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4236 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4238 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4239 SVE structure modes.
4240 (aarch64_classify_address): Likewise.
4241 (sizetochar): Move earlier in file.
4242 (aarch64_print_operand): Handle SVE register lists.
4243 (aarch64_array_mode): New function.
4244 (aarch64_sve_struct_memory_operand_p): Likewise.
4245 (TARGET_ARRAY_MODE): Redefine.
4247 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4248 Alan Hayward <alan.hayward@arm.com>
4249 David Sherwood <david.sherwood@arm.com>
4251 * target.def (array_mode): New target hook.
4252 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4253 * doc/tm.texi: Regenerate.
4254 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4255 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4256 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4258 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4261 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4262 Alan Hayward <alan.hayward@arm.com>
4263 David Sherwood <david.sherwood@arm.com>
4265 * fold-const.c (fold_binary_loc): Check the argument types
4266 rather than the result type when testing for a vector operation.
4268 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4270 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4271 * doc/tm.texi: Regenerate.
4273 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4274 Alan Hayward <alan.hayward@arm.com>
4275 David Sherwood <david.sherwood@arm.com>
4277 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4278 (sve): Document new AArch64 extension.
4279 * doc/md.texi (w): Extend the description of the AArch64
4280 constraint to include SVE vectors.
4281 (Upl, Upa): Document new AArch64 predicate constraints.
4282 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4284 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4285 (msve-vector-bits=): New option.
4286 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4287 SVE when these are disabled.
4288 (sve): New extension.
4289 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4290 modes. Adjust their number of units based on aarch64_sve_vg.
4291 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4292 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4293 aarch64_addr_query_type.
4294 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4295 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4296 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4297 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4298 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4299 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4300 (aarch64_simd_imm_zero_p): Delete.
4301 (aarch64_check_zero_based_sve_index_immediate): Declare.
4302 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4303 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4304 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4305 (aarch64_sve_float_mul_immediate_p): Likewise.
4306 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4308 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4309 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4310 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4311 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4312 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4313 (aarch64_regmode_natural_size): Likewise.
4314 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4315 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4317 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4318 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4319 for VG and the SVE predicate registers.
4320 (V_ALIASES): Add a "z"-prefixed alias.
4321 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4322 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4323 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4324 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4325 (REG_CLASS_NAMES): Add entries for them.
4326 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4327 and the predicate registers.
4328 (aarch64_sve_vg): Declare.
4329 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4330 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4331 (REGMODE_NATURAL_SIZE): Define.
4332 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4334 * config/aarch64/aarch64.c: Include cfgrtl.h.
4335 (simd_immediate_info): Add a constructor for series vectors,
4336 and an associated step field.
4337 (aarch64_sve_vg): New variable.
4338 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4339 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4340 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4341 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4342 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4343 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4344 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4345 (aarch64_get_mask_mode): New functions.
4346 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4347 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4348 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4349 predicate modes and predicate registers. Explicitly restrict
4350 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4351 to store a vector mode if it is recognized by
4352 aarch64_classify_vector_mode.
4353 (aarch64_regmode_natural_size): New function.
4354 (aarch64_hard_regno_caller_save_mode): Return the original mode
4356 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4357 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4358 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4359 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4361 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4362 does not overlap dest if the function is frame-related. Handle
4364 (aarch64_split_add_offset): New function.
4365 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4366 them aarch64_add_offset.
4367 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4368 and update call to aarch64_sub_sp.
4369 (aarch64_add_cfa_expression): New function.
4370 (aarch64_expand_prologue): Pass extra temporary registers to the
4371 functions above. Handle the case in which we need to emit new
4372 DW_CFA_expressions for registers that were originally saved
4373 relative to the stack pointer, but now have to be expressed
4374 relative to the frame pointer.
4375 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4377 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4378 IP0 and IP1 values for SVE frames.
4379 (aarch64_expand_vec_series): New function.
4380 (aarch64_expand_sve_widened_duplicate): Likewise.
4381 (aarch64_expand_sve_const_vector): Likewise.
4382 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4383 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4384 into the register, rather than emitting a SET directly.
4385 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4386 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4387 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4388 (offset_9bit_signed_scaled_p): New functions.
4389 (aarch64_replicate_bitmask_imm): New function.
4390 (aarch64_bitmask_imm): Use it.
4391 (aarch64_cannot_force_const_mem): Reject expressions involving
4392 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4393 (aarch64_classify_index): Handle SVE indices, by requiring
4394 a plain register index with a scale that matches the element size.
4395 (aarch64_classify_address): Handle SVE addresses. Assert that
4396 the mode of the address is VOIDmode or an integer mode.
4397 Update call to aarch64_classify_symbol.
4398 (aarch64_classify_symbolic_expression): Update call to
4399 aarch64_classify_symbol.
4400 (aarch64_const_vec_all_in_range_p): New function.
4401 (aarch64_print_vector_float_operand): Likewise.
4402 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4403 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4404 and the FP immediates 1.0 and 0.5.
4405 (aarch64_print_address_internal): Handle SVE addresses.
4406 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4407 (aarch64_regno_regclass): Handle predicate registers.
4408 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4410 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4411 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4412 (aarch64_convert_sve_vector_bits): New function.
4413 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4414 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4416 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4417 Handle SVE vector and predicate modes. Accept VL-based constants
4418 that need only one temporary register, and VL offsets that require
4419 no temporary registers.
4420 (aarch64_conditional_register_usage): Mark the predicate registers
4421 as fixed if SVE isn't available.
4422 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4423 Return true for SVE vector and predicate modes.
4424 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4425 rather than an unsigned int. Handle SVE modes.
4426 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4428 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4430 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4431 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4432 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4433 (aarch64_sve_float_mul_immediate_p): New functions.
4434 (aarch64_sve_valid_immediate): New function.
4435 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4436 Explicitly reject structure modes. Check for INDEX constants.
4437 Handle PTRUE and PFALSE constants.
4438 (aarch64_check_zero_based_sve_index_immediate): New function.
4439 (aarch64_simd_imm_zero_p): Delete.
4440 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4441 vector modes. Accept constants in the range of CNT[BHWD].
4442 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4443 ask for an Advanced SIMD mode.
4444 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4445 (aarch64_simd_vector_alignment): Handle SVE predicates.
4446 (aarch64_vectorize_preferred_vector_alignment): New function.
4447 (aarch64_simd_vector_alignment_reachable): Use it instead of
4449 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4450 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4452 (MAX_VECT_LEN): Delete.
4453 (expand_vec_perm_d): Add a vec_flags field.
4454 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4455 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4456 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4458 (aarch64_evpc_rev): Rename to...
4459 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4460 (aarch64_evpc_rev_global): New function.
4461 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4462 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4464 (aarch64_evpc_sve_tbl): New function.
4465 (aarch64_expand_vec_perm_const_1): Update after rename of
4466 aarch64_evpc_rev. Handle SVE permutes too, trying
4467 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4468 than aarch64_evpc_tbl.
4469 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4470 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4471 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4472 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4473 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4474 (aarch64_expand_sve_vcond): New functions.
4475 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4476 of aarch64_vector_mode_p.
4477 (aarch64_dwarf_poly_indeterminate_value): New function.
4478 (aarch64_compute_pressure_classes): Likewise.
4479 (aarch64_can_change_mode_class): Likewise.
4480 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4481 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4482 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4483 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4484 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4485 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4486 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4487 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4489 (Dn, Dl, Dr): Accept const as well as const_vector.
4490 (Dz): Likewise. Compare against CONST0_RTX.
4491 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4492 of "vector" where appropriate.
4493 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4494 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4495 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4496 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4497 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4498 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4499 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4500 (v_int_equiv): Extend to SVE modes.
4501 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4503 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4504 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4505 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4506 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4507 (SVE_COND_FP_CMP): New int iterators.
4508 (perm_hilo): Handle the new unpack unspecs.
4509 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4511 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4512 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4513 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4514 (aarch64_equality_operator, aarch64_constant_vector_operand)
4515 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4516 (aarch64_sve_nonimmediate_operand): Likewise.
4517 (aarch64_sve_general_operand): Likewise.
4518 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4519 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4520 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4521 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4522 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4523 (aarch64_sve_float_arith_immediate): Likewise.
4524 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4525 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4526 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4527 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4528 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4529 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4530 (aarch64_sve_float_arith_operand): Likewise.
4531 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4532 (aarch64_sve_float_mul_operand): Likewise.
4533 (aarch64_sve_vec_perm_operand): Likewise.
4534 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4535 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4536 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4537 as well as const_vector.
4538 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4539 in file. Use CONST0_RTX and CONSTM1_RTX.
4540 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4541 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4542 Use aarch64_simd_imm_zero.
4543 * config/aarch64/aarch64-sve.md: New file.
4544 * config/aarch64/aarch64.md: Include it.
4545 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4546 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4547 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4548 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4549 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4550 (sve): New attribute.
4551 (enabled): Disable instructions with the sve attribute unless
4553 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4554 aarch64_expand_mov_immediate.
4555 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4556 CNT[BHSD] immediates.
4557 (movti): Split CONST_POLY_INT moves into two halves.
4558 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4559 Split additions that need a temporary here if the destination
4560 is the stack pointer.
4561 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4562 (*add<mode>3_poly_1): New instruction.
4563 (set_clobber_cc): New expander.
4565 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4567 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4568 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4569 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4570 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4571 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4572 Change innermode from fixed_mode_size to machine_mode.
4573 (simplify_subreg): Update call accordingly. Handle a constant-sized
4574 subreg of a variable-length CONST_VECTOR.
4576 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4577 Alan Hayward <alan.hayward@arm.com>
4578 David Sherwood <david.sherwood@arm.com>
4580 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4581 (add_offset_to_base): New function, split out from...
4582 (create_mem_ref): ...here. When handling a scale other than 1,
4583 check first whether the address is valid without the offset.
4584 Add it into the base if so, leaving the index and scale as-is.
4586 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4589 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4590 fold_for_warn before checking if arg2 is INTEGER_CST.
4592 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4594 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4595 (store_multiple_operation): Delete.
4596 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4597 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4598 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4599 guarded by TARGET_STRING.
4600 (rs6000_output_load_multiple): Delete.
4601 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4602 OPTION_MASK_STRING / TARGET_STRING handling.
4603 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4604 (const rs6000_opt_masks) <"string">: Change mask to 0.
4605 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4606 (MASK_STRING): Delete.
4607 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4609 (load_multiple): Delete.
4616 (store_multiple): Delete.
4623 (movmemsi_8reg): Delete.
4624 (corresponding unnamed define_insn): Delete.
4625 (movmemsi_6reg): Delete.
4626 (corresponding unnamed define_insn): Delete.
4627 (movmemsi_4reg): Delete.
4628 (corresponding unnamed define_insn): Delete.
4629 (movmemsi_2reg): Delete.
4630 (corresponding unnamed define_insn): Delete.
4631 (movmemsi_1reg): Delete.
4632 (corresponding unnamed define_insn): Delete.
4633 * config/rs6000/rs6000.opt (mno-string): New.
4634 (mstring): Replace by deprecation warning stub.
4635 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4637 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4639 * regrename.c (regrename_do_replace): If replacing the same
4640 reg multiple times, try to reuse last created gen_raw_REG.
4643 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4644 main to workaround a bug in GDB.
4646 2018-01-12 Tom de Vries <tom@codesourcery.com>
4649 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4651 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4653 PR rtl-optimization/80481
4654 * ira-color.c (get_cap_member): New function.
4655 (allocnos_conflict_by_live_ranges_p): Use it.
4656 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4657 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4659 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4662 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4663 (*saddl_se_1): Ditto.
4665 (*ssubl_se_1): Ditto.
4667 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4669 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4670 rather than wi::to_widest for DR_INITs.
4671 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4672 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4673 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4675 (vect_analyze_group_access_1): Note that here.
4677 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4679 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4680 polynomial type sizes.
4682 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4684 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4685 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4686 (gimple_add_tmp_var): Likewise.
4688 2018-01-12 Martin Liska <mliska@suse.cz>
4690 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4691 (gimple_alloc_sizes): Likewise.
4692 (dump_gimple_statistics): Use PRIu64 in printf format.
4693 * gimple.h: Change uint64_t to int.
4695 2018-01-12 Martin Liska <mliska@suse.cz>
4697 * tree-core.h: Use uint64_t instead of int.
4698 * tree.c (tree_node_counts): Likewise.
4699 (tree_node_sizes): Likewise.
4700 (dump_tree_statistics): Use PRIu64 in printf format.
4702 2018-01-12 Martin Liska <mliska@suse.cz>
4704 * Makefile.in: As qsort_chk is implemented in vec.c, add
4705 vec.o to linkage of gencfn-macros.
4706 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4707 passing the info to record_node_allocation_statistics.
4708 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4710 * ggc-common.c (struct ggc_usage): Add operator== and use
4711 it in operator< and compare function.
4712 * mem-stats.h (struct mem_usage): Likewise.
4713 * vec.c (struct vec_usage): Remove operator< and compare
4714 function. Can be simply inherited.
4716 2018-01-12 Martin Jambor <mjambor@suse.cz>
4719 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4720 * tree-ssa-math-opts.c: Include domwalk.h.
4721 (convert_mult_to_fma_1): New function.
4722 (fma_transformation_info): New type.
4723 (fma_deferring_state): Likewise.
4724 (cancel_fma_deferring): New function.
4725 (result_of_phi): Likewise.
4726 (last_fma_candidate_feeds_initial_phi): Likewise.
4727 (convert_mult_to_fma): Added deferring logic, split actual
4728 transformation to convert_mult_to_fma_1.
4729 (math_opts_dom_walker): New type.
4730 (math_opts_dom_walker::after_dom_children): New method, body moved
4731 here from pass_optimize_widening_mul::execute, added deferring logic
4733 (pass_optimize_widening_mul::execute): Moved most of code to
4734 math_opts_dom_walker::after_dom_children.
4735 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4736 * config/i386/i386.c (ix86_option_override_internal): Added
4737 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4739 2018-01-12 Richard Biener <rguenther@suse.de>
4742 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4743 inline instance vars.
4745 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4748 * config/rx/rx.c (rx_is_restricted_memory_address):
4751 2018-01-12 Richard Biener <rguenther@suse.de>
4753 PR tree-optimization/80846
4754 * target.def (split_reduction): New target hook.
4755 * targhooks.c (default_split_reduction): New function.
4756 * targhooks.h (default_split_reduction): Declare.
4757 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4758 target requests first reduce vectors by combining low and high
4760 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4761 (get_vectype_for_scalar_type_and_size): Export.
4762 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4763 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4764 * doc/tm.texi: Regenerate.
4765 * config/i386/i386.c (ix86_split_reduction): Implement
4766 TARGET_VECTORIZE_SPLIT_REDUCTION.
4768 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4771 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4772 in PIC mode except for TARGET_VXWORKS_RTP.
4773 * config/sparc/sparc.c: Include cfgrtl.h.
4774 (TARGET_INIT_PIC_REG): Define.
4775 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4776 (sparc_pic_register_p): New predicate.
4777 (sparc_legitimate_address_p): Use it.
4778 (sparc_legitimize_pic_address): Likewise.
4779 (sparc_delegitimize_address): Likewise.
4780 (sparc_mode_dependent_address_p): Likewise.
4781 (gen_load_pcrel_sym): Remove 4th parameter.
4782 (load_got_register): Adjust call to above. Remove obsolete stuff.
4783 (sparc_expand_prologue): Do not call load_got_register here.
4784 (sparc_flat_expand_prologue): Likewise.
4785 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4786 (sparc_use_pseudo_pic_reg): New function.
4787 (sparc_init_pic_reg): Likewise.
4788 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4789 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4791 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4793 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4794 Add item for branch_cost.
4796 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4798 PR rtl-optimization/83565
4799 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4800 not extend the result to a larger mode for rotate operations.
4801 (num_sign_bit_copies1): Likewise.
4803 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4806 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4808 Use values-Xc.o for -pedantic.
4809 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4811 2018-01-12 Martin Liska <mliska@suse.cz>
4814 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4816 (possible_polymorphic_call_targets): Use it.
4817 (ipa_devirt): Likewise.
4819 2018-01-12 Martin Liska <mliska@suse.cz>
4821 * profile-count.h (enum profile_quality): Use 0 as invalid
4822 enum value of profile_quality.
4824 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4826 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4827 -mext-string options.
4829 2018-01-12 Richard Biener <rguenther@suse.de>
4831 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4832 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4833 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4835 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4837 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4839 * configure.ac (--with-long-double-format): Add support for the
4840 configuration option to change the default long double format on
4842 * config.gcc (powerpc*-linux*-*): Likewise.
4843 * configure: Regenerate.
4844 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4845 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4846 used without modification.
4848 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4850 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4851 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4852 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4853 MISC_BUILTIN_SPEC_BARRIER.
4854 (rs6000_init_builtins): Likewise.
4855 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4857 (speculation_barrier): New define_insn.
4858 * doc/extend.texi: Document __builtin_speculation_barrier.
4860 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4863 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4864 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4865 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4867 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4868 integral modes instead of "ss" and "sd".
4869 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4870 vectors with 32-bit and 64-bit elements.
4871 (vecdupssescalarmodesuffix): New mode attribute.
4872 (vec_dup<mode>): Use it.
4874 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4877 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4878 frame if argument is passed on stack.
4880 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4883 * ree.c (combine_reaching_defs): Optimize also
4884 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4885 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4887 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4890 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4892 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4895 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4896 after they are computed.
4898 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4900 PR tree-optimization/83695
4901 * gimple-loop-linterchange.cc
4902 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4903 reset cached scev information after interchange.
4904 (pass_linterchange::execute): Remove call to scev_reset_htab.
4906 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4908 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4909 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4910 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4911 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4912 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4913 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4914 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4915 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4916 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4917 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4918 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4919 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4920 (V_lane_reg): Likewise.
4921 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4923 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4924 (vfmal_lane_low<mode>_intrinsic,
4925 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4926 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4927 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4928 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4929 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4930 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4932 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4934 * config/arm/arm-cpus.in (fp16fml): New feature.
4935 (ALL_SIMD): Add fp16fml.
4936 (armv8.2-a): Add fp16fml as an option.
4937 (armv8.3-a): Likewise.
4938 (armv8.4-a): Add fp16fml as part of fp16.
4939 * config/arm/arm.h (TARGET_FP16FML): Define.
4940 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4942 * config/arm/arm-modes.def (V2HF): Define.
4943 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4944 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4945 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4946 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4947 vfmsl_low, vfmsl_high): New set of builtins.
4948 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4949 (vfml_op): New code attribute.
4950 (VFMLHALVES): New int iterator.
4951 (VFML, VFMLSEL): New mode attributes.
4952 (V_reg): Define mapping for V2HF.
4953 (V_hi, V_lo): New mode attributes.
4954 (VF_constraint): Likewise.
4955 (vfml_half, vfml_half_selector): New int attributes.
4956 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4958 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4959 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4961 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4962 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4963 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4964 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4966 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4967 Document new effective target and option set.
4969 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4971 * config/arm/arm-cpus.in (armv8_4): New feature.
4972 (ARMv8_4a): New fgroup.
4973 (armv8.4-a): New arch.
4974 * config/arm/arm-tables.opt: Regenerate.
4975 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4976 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4977 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4978 Add matching rules for -march=armv8.4-a and extensions.
4979 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4981 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4984 * config/rx/rx.md (BW): New mode attribute.
4985 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4987 2018-01-11 Richard Biener <rguenther@suse.de>
4989 PR tree-optimization/83435
4990 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4991 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4992 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4994 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4995 Alan Hayward <alan.hayward@arm.com>
4996 David Sherwood <david.sherwood@arm.com>
4998 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5000 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5001 (aarch64_print_address_internal): Use it to check for a zero offset.
5003 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5004 Alan Hayward <alan.hayward@arm.com>
5005 David Sherwood <david.sherwood@arm.com>
5007 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5008 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5009 Return a poly_int64 rather than a HOST_WIDE_INT.
5010 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5011 rather than a HOST_WIDE_INT.
5012 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5013 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5014 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5015 final_offset from HOST_WIDE_INT to poly_int64.
5016 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5017 to_constant when getting the number of units in an Advanced SIMD
5019 (aarch64_builtin_vectorized_function): Check for a constant number
5021 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5023 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5024 attribute instead of GET_MODE_NUNITS.
5025 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5026 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5027 GET_MODE_SIZE for fixed-size registers.
5028 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5029 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5030 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5031 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5032 (aarch64_print_operand, aarch64_print_address_internal)
5033 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5034 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5035 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5036 Handle polynomial GET_MODE_SIZE.
5037 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5038 wider than SImode without modification.
5039 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5040 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5041 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5042 passing and returning SVE modes.
5043 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5044 rather than GEN_INT.
5045 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5046 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5047 (aarch64_allocate_and_probe_stack_space): Likewise.
5048 (aarch64_layout_frame): Cope with polynomial offsets.
5049 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5050 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5052 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5053 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5054 poly_int64 rather than a HOST_WIDE_INT.
5055 (aarch64_get_separate_components, aarch64_process_components)
5056 (aarch64_expand_prologue, aarch64_expand_epilogue)
5057 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5058 (aarch64_anchor_offset): New function, split out from...
5059 (aarch64_legitimize_address): ...here.
5060 (aarch64_builtin_vectorization_cost): Handle polynomial
5061 TYPE_VECTOR_SUBPARTS.
5062 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5064 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5065 number of elements from the PARALLEL rather than the mode.
5066 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5067 rather than GET_MODE_BITSIZE.
5068 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5069 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5070 (aarch64_expand_vec_perm_const_1): Handle polynomial
5071 d->perm.length () and d->perm elements.
5072 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5073 Apply to_constant to d->perm elements.
5074 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5075 polynomial CONST_VECTOR_NUNITS.
5076 (aarch64_move_pointer): Take amount as a poly_int64 rather
5078 (aarch64_progress_pointer): Avoid temporary variable.
5079 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5080 the mode attribute instead of GET_MODE.
5082 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5083 Alan Hayward <alan.hayward@arm.com>
5084 David Sherwood <david.sherwood@arm.com>
5086 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5087 x exists before using it.
5088 (aarch64_add_constant_internal): Rename to...
5089 (aarch64_add_offset_1): ...this. Replace regnum with separate
5090 src and dest rtxes. Handle the case in which they're different,
5091 including when the offset is zero. Replace scratchreg with an rtx.
5092 Use 2 additions if there is no spare register into which we can
5093 move a 16-bit constant.
5094 (aarch64_add_constant): Delete.
5095 (aarch64_add_offset): Replace reg with separate src and dest
5096 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5097 Use aarch64_add_offset_1.
5098 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5099 an rtx rather than an int. Take the delta as a poly_int64
5100 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5101 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5102 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5103 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5104 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5106 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5107 aarch64_add_constant.
5109 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5111 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5112 Use scalar_float_mode.
5114 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5116 * config/aarch64/aarch64-simd.md
5117 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5118 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5119 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5120 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5121 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5122 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5123 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5124 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5125 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5126 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5128 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5131 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5132 targ_options->x_arm_arch_string is non NULL.
5134 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5136 * config/aarch64/aarch64.h
5137 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5139 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5142 * expmed.c (emit_store_flag_force): Swap if const op0
5143 and change VOIDmode to mode of op0.
5145 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5147 PR rtl-optimization/83761
5148 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5149 than bytes to mode_for_size.
5151 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5154 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5155 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5158 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5161 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5162 when in layout mode.
5163 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5164 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5167 2018-01-10 Michael Collison <michael.collison@arm.com>
5169 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5170 * config/aarch64/aarch64-option-extension.def: Add
5171 AARCH64_OPT_EXTENSION of 'fp16fml'.
5172 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5173 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5174 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5175 * config/aarch64/constraints.md (Ui7): New constraint.
5176 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5177 (VFMLA_SEL_W): Ditto.
5180 (VFMLA16_LOW): New int iterator.
5181 (VFMLA16_HIGH): Ditto.
5182 (UNSPEC_FMLAL): New unspec.
5183 (UNSPEC_FMLSL): Ditto.
5184 (UNSPEC_FMLAL2): Ditto.
5185 (UNSPEC_FMLSL2): Ditto.
5186 (f16mac): New code attribute.
5187 * config/aarch64/aarch64-simd-builtins.def
5188 (aarch64_fmlal_lowv2sf): Ditto.
5189 (aarch64_fmlsl_lowv2sf): Ditto.
5190 (aarch64_fmlalq_lowv4sf): Ditto.
5191 (aarch64_fmlslq_lowv4sf): Ditto.
5192 (aarch64_fmlal_highv2sf): Ditto.
5193 (aarch64_fmlsl_highv2sf): Ditto.
5194 (aarch64_fmlalq_highv4sf): Ditto.
5195 (aarch64_fmlslq_highv4sf): Ditto.
5196 (aarch64_fmlal_lane_lowv2sf): Ditto.
5197 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5198 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5199 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5200 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5201 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5202 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5203 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5204 (aarch64_fmlal_lane_highv2sf): Ditto.
5205 (aarch64_fmlsl_lane_highv2sf): Ditto.
5206 (aarch64_fmlal_laneq_highv2sf): Ditto.
5207 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5208 (aarch64_fmlalq_lane_highv4sf): Ditto.
5209 (aarch64_fmlsl_lane_highv4sf): Ditto.
5210 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5211 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5212 * config/aarch64/aarch64-simd.md:
5213 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5214 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5215 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5216 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5217 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5218 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5219 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5220 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5221 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5222 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5223 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5224 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5225 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5226 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5227 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5228 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5229 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5230 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5231 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5232 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5233 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5234 (vfmlsl_low_u32): Ditto.
5235 (vfmlalq_low_u32): Ditto.
5236 (vfmlslq_low_u32): Ditto.
5237 (vfmlal_high_u32): Ditto.
5238 (vfmlsl_high_u32): Ditto.
5239 (vfmlalq_high_u32): Ditto.
5240 (vfmlslq_high_u32): Ditto.
5241 (vfmlal_lane_low_u32): Ditto.
5242 (vfmlsl_lane_low_u32): Ditto.
5243 (vfmlal_laneq_low_u32): Ditto.
5244 (vfmlsl_laneq_low_u32): Ditto.
5245 (vfmlalq_lane_low_u32): Ditto.
5246 (vfmlslq_lane_low_u32): Ditto.
5247 (vfmlalq_laneq_low_u32): Ditto.
5248 (vfmlslq_laneq_low_u32): Ditto.
5249 (vfmlal_lane_high_u32): Ditto.
5250 (vfmlsl_lane_high_u32): Ditto.
5251 (vfmlal_laneq_high_u32): Ditto.
5252 (vfmlsl_laneq_high_u32): Ditto.
5253 (vfmlalq_lane_high_u32): Ditto.
5254 (vfmlslq_lane_high_u32): Ditto.
5255 (vfmlalq_laneq_high_u32): Ditto.
5256 (vfmlslq_laneq_high_u32): Ditto.
5257 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5258 (AARCH64_FL_FOR_ARCH8_4): New.
5259 (AARCH64_ISA_F16FML): New ISA flag.
5260 (TARGET_F16FML): New feature flag for fp16fml.
5261 (doc/invoke.texi): Document new fp16fml option.
5263 2018-01-10 Michael Collison <michael.collison@arm.com>
5265 * config/aarch64/aarch64-builtins.c:
5266 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5267 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5268 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5269 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5270 (AARCH64_ISA_SHA3): New ISA flag.
5271 (TARGET_SHA3): New feature flag for sha3.
5272 * config/aarch64/iterators.md (sha512_op): New int attribute.
5273 (CRYPTO_SHA512): New int iterator.
5274 (UNSPEC_SHA512H): New unspec.
5275 (UNSPEC_SHA512H2): Ditto.
5276 (UNSPEC_SHA512SU0): Ditto.
5277 (UNSPEC_SHA512SU1): Ditto.
5278 * config/aarch64/aarch64-simd-builtins.def
5279 (aarch64_crypto_sha512hqv2di): New builtin.
5280 (aarch64_crypto_sha512h2qv2di): Ditto.
5281 (aarch64_crypto_sha512su0qv2di): Ditto.
5282 (aarch64_crypto_sha512su1qv2di): Ditto.
5283 (aarch64_eor3qv8hi): Ditto.
5284 (aarch64_rax1qv2di): Ditto.
5285 (aarch64_xarqv2di): Ditto.
5286 (aarch64_bcaxqv8hi): Ditto.
5287 * config/aarch64/aarch64-simd.md:
5288 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5289 (aarch64_crypto_sha512su0qv2di): Ditto.
5290 (aarch64_crypto_sha512su1qv2di): Ditto.
5291 (aarch64_eor3qv8hi): Ditto.
5292 (aarch64_rax1qv2di): Ditto.
5293 (aarch64_xarqv2di): Ditto.
5294 (aarch64_bcaxqv8hi): Ditto.
5295 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5296 (vsha512h2q_u64): Ditto.
5297 (vsha512su0q_u64): Ditto.
5298 (vsha512su1q_u64): Ditto.
5299 (veor3q_u16): Ditto.
5300 (vrax1q_u64): Ditto.
5302 (vbcaxq_u16): Ditto.
5303 * config/arm/types.md (crypto_sha512): New type attribute.
5304 (crypto_sha3): Ditto.
5305 (doc/invoke.texi): Document new sha3 option.
5307 2018-01-10 Michael Collison <michael.collison@arm.com>
5309 * config/aarch64/aarch64-builtins.c:
5310 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5311 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5312 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5313 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5314 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5315 (AARCH64_ISA_SM4): New ISA flag.
5316 (TARGET_SM4): New feature flag for sm4.
5317 * config/aarch64/aarch64-simd-builtins.def
5318 (aarch64_sm3ss1qv4si): Ditto.
5319 (aarch64_sm3tt1aq4si): Ditto.
5320 (aarch64_sm3tt1bq4si): Ditto.
5321 (aarch64_sm3tt2aq4si): Ditto.
5322 (aarch64_sm3tt2bq4si): Ditto.
5323 (aarch64_sm3partw1qv4si): Ditto.
5324 (aarch64_sm3partw2qv4si): Ditto.
5325 (aarch64_sm4eqv4si): Ditto.
5326 (aarch64_sm4ekeyqv4si): Ditto.
5327 * config/aarch64/aarch64-simd.md:
5328 (aarch64_sm3ss1qv4si): Ditto.
5329 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5330 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5331 (aarch64_sm4eqv4si): Ditto.
5332 (aarch64_sm4ekeyqv4si): Ditto.
5333 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5334 (sm3part_op): Ditto.
5335 (CRYPTO_SM3TT): Ditto.
5336 (CRYPTO_SM3PART): Ditto.
5337 (UNSPEC_SM3SS1): New unspec.
5338 (UNSPEC_SM3TT1A): Ditto.
5339 (UNSPEC_SM3TT1B): Ditto.
5340 (UNSPEC_SM3TT2A): Ditto.
5341 (UNSPEC_SM3TT2B): Ditto.
5342 (UNSPEC_SM3PARTW1): Ditto.
5343 (UNSPEC_SM3PARTW2): Ditto.
5344 (UNSPEC_SM4E): Ditto.
5345 (UNSPEC_SM4EKEY): Ditto.
5346 * config/aarch64/constraints.md (Ui2): New constraint.
5347 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5348 * config/arm/types.md (crypto_sm3): New type attribute.
5349 (crypto_sm4): Ditto.
5350 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5351 (vsm3tt1aq_u32): Ditto.
5352 (vsm3tt1bq_u32): Ditto.
5353 (vsm3tt2aq_u32): Ditto.
5354 (vsm3tt2bq_u32): Ditto.
5355 (vsm3partw1q_u32): Ditto.
5356 (vsm3partw2q_u32): Ditto.
5357 (vsm4eq_u32): Ditto.
5358 (vsm4ekeyq_u32): Ditto.
5359 (doc/invoke.texi): Document new sm4 option.
5361 2018-01-10 Michael Collison <michael.collison@arm.com>
5363 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5364 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5365 (AARCH64_FL_FOR_ARCH8_4): New.
5366 (AARCH64_FL_V8_4): New flag.
5367 (doc/invoke.texi): Document new armv8.4-a option.
5369 2018-01-10 Michael Collison <michael.collison@arm.com>
5371 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5372 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5373 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5374 * config/aarch64/aarch64-option-extension.def: Add
5375 AARCH64_OPT_EXTENSION of 'sha2'.
5376 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5377 (crypto): Disable sha2 and aes if crypto disabled.
5378 (crypto): Enable aes and sha2 if enabled.
5379 (simd): Disable sha2 and aes if simd disabled.
5380 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5382 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5383 (TARGET_SHA2): New feature flag for sha2.
5384 (TARGET_AES): New feature flag for aes.
5385 * config/aarch64/aarch64-simd.md:
5386 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5387 conditional on TARGET_AES.
5388 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5389 (aarch64_crypto_sha1hsi): Make pattern conditional
5391 (aarch64_crypto_sha1hv4si): Ditto.
5392 (aarch64_be_crypto_sha1hv4si): Ditto.
5393 (aarch64_crypto_sha1su1v4si): Ditto.
5394 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5395 (aarch64_crypto_sha1su0v4si): Ditto.
5396 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5397 (aarch64_crypto_sha256su0v4si): Ditto.
5398 (aarch64_crypto_sha256su1v4si): Ditto.
5399 (doc/invoke.texi): Document new aes and sha2 options.
5401 2018-01-10 Martin Sebor <msebor@redhat.com>
5403 PR tree-optimization/83781
5404 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5407 2018-01-11 Martin Sebor <msebor@gmail.com>
5408 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5410 PR tree-optimization/83501
5411 PR tree-optimization/81703
5413 * tree-ssa-strlen.c (get_string_cst): Rename...
5414 (get_string_len): ...to this. Handle global constants.
5415 (handle_char_store): Adjust.
5417 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5418 Jim Wilson <jimw@sifive.com>
5420 * config/riscv/riscv-protos.h (riscv_output_return): New.
5421 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5422 (riscv_attribute_table, riscv_output_return),
5423 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5424 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5425 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5427 (riscv_expand_prologue): Add early return for naked function.
5428 (riscv_expand_epilogue): Likewise.
5429 (riscv_function_ok_for_sibcall): Return false for naked function.
5430 (riscv_set_current_function): New.
5431 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5432 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5433 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5434 * doc/extend.texi (RISC-V Function Attributes): New.
5436 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5438 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5439 check for 128-bit long double before checking TCmode.
5440 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5441 128-bit long doubles before checking TFmode or TCmode.
5442 (FLOAT128_IBM_P): Likewise.
5444 2018-01-10 Martin Sebor <msebor@redhat.com>
5446 PR tree-optimization/83671
5447 * builtins.c (c_strlen): Unconditionally return zero for the empty
5449 Use -Warray-bounds for warnings.
5450 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5451 for non-constant array indices with COMPONENT_REF, arrays of
5452 arrays, and pointers to arrays.
5453 (gimple_fold_builtin_strlen): Determine and set length range for
5454 non-constant character arrays.
5456 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5459 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5462 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5464 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5466 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5469 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5470 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5471 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5472 indexed_or_indirect_operand predicate.
5473 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5474 (*vsx_le_perm_load_v8hi): Likewise.
5475 (*vsx_le_perm_load_v16qi): Likewise.
5476 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5477 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5478 (*vsx_le_perm_store_v8hi): Likewise.
5479 (*vsx_le_perm_store_v16qi): Likewise.
5480 (eight unnamed splitters): Likewise.
5482 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5484 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5485 * config/rs6000/emmintrin.h: Likewise.
5486 * config/rs6000/mmintrin.h: Likewise.
5487 * config/rs6000/xmmintrin.h: Likewise.
5489 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5492 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5494 * tree.c (tree_nop_conversion): Return true for location wrapper
5496 (maybe_wrap_with_location): New function.
5497 (selftest::check_strip_nops): New function.
5498 (selftest::test_location_wrappers): New function.
5499 (selftest::tree_c_tests): Call it.
5500 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5501 (maybe_wrap_with_location): New decl.
5502 (EXPR_LOCATION_WRAPPER_P): New macro.
5503 (location_wrapper_p): New inline function.
5504 (tree_strip_any_location_wrapper): New inline function.
5506 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5509 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5510 stack_realign_offset for the largest alignment of stack slot
5512 (ix86_find_max_used_stack_alignment): New function.
5513 (ix86_finalize_stack_frame_flags): Use it. Set
5514 max_used_stack_alignment if we don't realign stack.
5515 * config/i386/i386.h (machine_function): Add
5516 max_used_stack_alignment.
5518 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5520 * config/arm/arm.opt (-mbranch-cost): New option.
5521 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5524 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5527 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5528 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5530 2018-01-10 Richard Biener <rguenther@suse.de>
5533 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5534 early out so it also covers the case where we have a non-NULL
5537 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5539 PR tree-optimization/83753
5540 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5541 for non-strided grouped accesses if the number of elements is 1.
5543 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5546 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5547 * i386.h (TARGET_USE_GATHER): Define.
5548 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5550 2018-01-10 Martin Liska <mliska@suse.cz>
5553 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5554 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5556 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5557 CLEANUP_NO_PARTITIONING is not set.
5559 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5561 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5562 for vectors, as a partial revert of r254296.
5563 * rtl.h (const_vec_p): Delete.
5564 (const_vec_duplicate_p): Don't test for vector CONSTs.
5565 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5566 * expmed.c (make_tree): Likewise.
5569 * common.md (E, F): Use CONSTANT_P instead of checking for
5571 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5572 checking for CONST_VECTOR.
5574 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5577 * predict.c (force_edge_cold): Handle in more sane way edges
5580 2018-01-09 Carl Love <cel@us.ibm.com>
5582 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5584 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5585 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5586 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5587 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5588 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5589 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5590 * config/rs6000/rs6000-protos.h: Add extern defition for
5591 rs6000_generate_float2_double_code.
5592 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5594 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5595 (float2_v2df): Add define_expand.
5597 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5600 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5601 op_mode in the force_to_mode call.
5603 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5605 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5606 instead of checking each element individually.
5607 (aarch64_evpc_uzp): Likewise.
5608 (aarch64_evpc_zip): Likewise.
5609 (aarch64_evpc_ext): Likewise.
5610 (aarch64_evpc_rev): Likewise.
5611 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5612 instead of checking each element individually. Return true without
5614 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5615 whether all selected elements come from the same input, instead of
5616 checking each element individually. Remove calls to gen_rtx_REG,
5617 start_sequence and end_sequence and instead assert that no rtl is
5620 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5622 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5623 order of HIGH and CONST checks.
5625 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5627 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5628 if the destination isn't an SSA_NAME.
5630 2018-01-09 Richard Biener <rguenther@suse.de>
5632 PR tree-optimization/83668
5633 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5635 (canonicalize_loop_form): ... here, renamed from ...
5636 (canonicalize_loop_closed_ssa_form): ... this and amended to
5637 swap successor edges for loop exit blocks to make us use
5638 the RPO order we need for initial schedule generation.
5640 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5642 PR tree-optimization/64811
5643 * match.pd: When optimizing comparisons with Inf, avoid
5644 introducing or losing exceptions from comparisons with NaN.
5646 2018-01-09 Martin Liska <mliska@suse.cz>
5649 * asan.c (shadow_mem_size): Add gcc_assert.
5651 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5653 Don't save registers in main().
5656 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5657 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5658 * config/avr/avr.c (avr_set_current_function): Don't error if
5659 naked, OS_task or OS_main are specified at the same time.
5660 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5662 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5664 * common/config/avr/avr-common.c (avr_option_optimization_table):
5665 Switch on -mmain-is-OS_task for optimizing compilations.
5667 2018-01-09 Richard Biener <rguenther@suse.de>
5669 PR tree-optimization/83572
5670 * graphite.c: Include cfganal.h.
5671 (graphite_transform_loops): Connect infinite loops to exit
5672 and remove fake edges at the end.
5674 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5676 * ipa-inline.c (edge_badness): Revert accidental checkin.
5678 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5681 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5682 symbols; not inline clones.
5684 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5687 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5688 hard registers. Formatting fixes.
5690 PR preprocessor/83722
5691 * gcc.c (try_generate_repro): Pass
5692 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5693 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5696 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5697 Kito Cheng <kito.cheng@gmail.com>
5699 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5700 (riscv_leaf_function_p): Delete.
5701 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5703 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5705 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5707 (do_ifelse): New function.
5708 (do_isel): New function.
5709 (do_sub3): New function.
5710 (do_add3): New function.
5711 (do_load_mask_compare): New function.
5712 (do_overlap_load_compare): New function.
5713 (expand_compare_loop): New function.
5714 (expand_block_compare): Call expand_compare_loop() when appropriate.
5715 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5717 (-mblock-compare-inline-loop-limit): New option.
5719 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5722 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5723 Reverse order of second and third operands in first alternative.
5724 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5725 of first and second elements in UNSPEC_VPERMR vector.
5726 (altivec_expand_vec_perm_le): Likewise.
5728 2018-01-08 Jeff Law <law@redhat.com>
5730 PR rtl-optimizatin/81308
5731 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5732 (process_switch): If group_case_labels makes a change, then set
5734 (pass_convert_switch::execute): If a switch is converted, then
5735 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5737 PR rtl-optimization/81308
5738 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5741 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5743 PR target/83663 - Revert r255946
5744 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5745 generation for cases where splatting a value is not useful.
5746 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5747 across a vec_duplicate and a paradoxical subreg forming a vector
5748 mode to a vec_concat.
5750 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5752 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5753 -march=armv8.3-a variants.
5754 * config/arm/t-multilib: Likewise.
5755 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5757 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5759 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5761 (cceq_ior_compare_complement): Give it a name so I can use it, and
5762 change boolean_or_operator predicate to boolean_operator so it can
5763 be used to generate a crand.
5764 (eqne): New code iterator.
5765 (bd/bd_neg): New code_attrs.
5766 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5767 a single define_insn.
5768 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5769 decrement (bdnzt/bdnzf/bdzt/bdzf).
5770 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5771 with the new names of the branch decrement patterns, and added the
5772 names of the branch decrement conditional patterns.
5774 2018-01-08 Richard Biener <rguenther@suse.de>
5776 PR tree-optimization/83563
5777 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5780 2018-01-08 Richard Biener <rguenther@suse.de>
5783 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5785 2018-01-08 Richard Biener <rguenther@suse.de>
5787 PR tree-optimization/83685
5788 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5789 references to abnormals.
5791 2018-01-08 Richard Biener <rguenther@suse.de>
5794 * dwarf2out.c (output_indirect_strings): Handle empty
5795 skeleton_debug_str_hash.
5796 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5798 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5800 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5801 (emit_store_direct): Likewise.
5802 (arc_trampoline_adjust_address): Likewise.
5803 (arc_asm_trampoline_template): New function.
5804 (arc_initialize_trampoline): Use asm_trampoline_template.
5805 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5806 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5807 * config/arc/arc.md (flush_icache): Delete pattern.
5809 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5811 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5812 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5815 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5818 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5819 by not USED_FOR_TARGET.
5820 (make_pass_resolve_sw_modes): Likewise.
5822 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5824 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5827 2018-01-08 Richard Biener <rguenther@suse.de>
5830 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5832 2018-01-08 Richard Biener <rguenther@suse.de>
5835 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5837 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5840 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5841 basic blocks with a small number of successors.
5842 (convert_control_dep_chain_into_preds): Improve handling of
5844 (dump_predicates): Split apart into...
5845 (dump_pred_chain): ...here...
5846 (dump_pred_info): ...and here.
5847 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5848 (can_chain_union_be_invalidated_p): Improve check for invalidation
5850 (uninit_uses_cannot_happen): Avoid unnecessary if
5851 convert_control_dep_chain_into_preds yielded nothing.
5853 2018-01-06 Martin Sebor <msebor@redhat.com>
5855 PR tree-optimization/83640
5856 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5857 subtracting negative offset from size.
5858 (builtin_access::overlap): Adjust offset bounds of the access to fall
5859 within the size of the object if possible.
5861 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5863 PR rtl-optimization/83699
5864 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5865 extract_bit_field_as_subreg to cases in which the extracted
5866 value is also a vector.
5868 * lra-constraints.c (process_alt_operands): Test for the equivalence
5869 substitutions when detecting a possible reload cycle.
5871 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5874 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5875 by default if flag_selective_schedling{,2}. Formatting fixes.
5877 PR rtl-optimization/83682
5878 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5879 if it has non-VECTOR_MODE element mode.
5880 (vec_duplicate_p): Likewise.
5883 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5884 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5886 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5889 * config/i386/i386-builtin.def
5890 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5891 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5892 Require also OPTION_MASK_ISA_AVX512F in addition to
5893 OPTION_MASK_ISA_GFNI.
5894 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5895 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5896 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5897 to OPTION_MASK_ISA_GFNI.
5898 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5899 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5900 OPTION_MASK_ISA_AVX512BW.
5901 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5902 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5903 addition to OPTION_MASK_ISA_GFNI.
5904 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5905 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5906 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5907 to OPTION_MASK_ISA_GFNI.
5908 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5909 a requirement for all ISAs rather than any of them with a few
5911 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5913 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5914 bitmasks to be enabled with 3 exceptions, instead of requiring any
5915 enabled ISA with lots of exceptions.
5916 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5917 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5918 Change avx512bw in isa attribute to avx512f.
5919 * config/i386/sgxintrin.h: Add license boilerplate.
5920 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5921 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5922 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5923 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5925 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5926 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5927 temporarily sse2 rather than sse if not enabled already.
5930 * config/i386/sse.md (VI248_VLBW): Rename to ...
5931 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5932 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5933 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5934 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5935 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5936 mode iterator instead of VI248_VLBW.
5938 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5940 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5941 (record_modified): Skip clobbers; add debug output.
5942 (param_change_prob): Use sreal frequencies.
5944 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5946 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5947 punt for user-aligned variables.
5949 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5951 * tree-chrec.c (chrec_contains_symbols): Return true for
5954 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5957 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5958 of (x|y) == x for BICS pattern.
5960 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5962 PR tree-optimization/83605
5963 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5964 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5967 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5969 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5970 * config/epiphany/rtems.h: New file.
5972 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5973 Uros Bizjak <ubizjak@gmail.com>
5976 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5977 QIreg_operand instead of register_operand predicate.
5978 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5979 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5980 comments instead of -fmitigate[-_]rop.
5982 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5985 * cgraphunit.c (symbol_table::compile): Switch to text_section
5986 before calling assembly_start debug hook.
5987 * run-rtl-passes.c (run_rtl_passes): Likewise.
5990 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5992 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5993 range_int_cst_p rather than !symbolic_range_p before calling
5994 extract_range_from_multiplicative_op_1.
5996 2018-01-04 Jeff Law <law@redhat.com>
5998 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5999 redundant test in assertion.
6001 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6003 * doc/rtl.texi: Document machine_mode wrapper classes.
6005 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6007 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6010 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6012 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6013 the VEC_PERM_EXPR fold to fail.
6015 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6018 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6019 to switched_sections.
6021 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6024 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6027 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6030 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6031 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6033 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6036 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6037 is BLKmode and bitpos not zero or mode change is needed.
6039 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6042 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6045 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6048 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6049 instead of MULT rtx. Update all corresponding splitters.
6051 (*ssub<modesuffix>): Ditto.
6053 (*cmp_sadd_di): Update split patterns.
6054 (*cmp_sadd_si): Ditto.
6055 (*cmp_sadd_sidi): Ditto.
6056 (*cmp_ssub_di): Ditto.
6057 (*cmp_ssub_si): Ditto.
6058 (*cmp_ssub_sidi): Ditto.
6059 * config/alpha/predicates.md (const23_operand): New predicate.
6060 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6061 Look for ASHIFT, not MULT inner operand.
6062 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6064 2018-01-04 Martin Liska <mliska@suse.cz>
6066 PR gcov-profile/83669
6067 * gcov.c (output_intermediate_file): Add version to intermediate
6069 * doc/gcov.texi: Document new field 'version' in intermediate
6070 file format. Fix location of '-k' option of gcov command.
6072 2018-01-04 Martin Liska <mliska@suse.cz>
6075 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6077 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6079 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6081 2018-01-03 Martin Sebor <msebor@redhat.com>
6083 PR tree-optimization/83655
6084 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6085 checking calls with invalid arguments.
6087 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6089 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6090 (vectorizable_mask_load_store): Delete.
6091 (vectorizable_call): Return false for masked loads and stores.
6092 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6093 instead of gimple_assign_rhs1.
6094 (vectorizable_load): Handle IFN_MASK_LOAD.
6095 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6097 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6099 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6101 (vectorizable_mask_load_store): ...here.
6102 (vectorizable_load): ...and here.
6104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6106 * tree-vect-stmts.c (vect_build_all_ones_mask)
6107 (vect_build_zero_merge_argument): New functions, split out from...
6108 (vectorizable_load): ...here.
6110 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6112 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6114 (vectorizable_mask_load_store): ...here.
6115 (vectorizable_store): ...and here.
6117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6119 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6121 (vectorizable_mask_load_store): ...here.
6123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6125 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6126 (vect_model_store_cost): Take a vec_load_store_type instead of a
6128 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6129 (vect_model_store_cost): Take a vec_load_store_type instead of a
6131 (vectorizable_mask_load_store): Update accordingly.
6132 (vectorizable_store): Likewise.
6133 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6137 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6138 IFN_MASK_LOAD calls here rather than...
6139 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6142 Alan Hayward <alan.hayward@arm.com>
6143 David Sherwood <david.sherwood@arm.com>
6145 * expmed.c (extract_bit_field_1): For vector extracts,
6146 fall back to extract_bit_field_as_subreg if vec_extract
6149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6150 Alan Hayward <alan.hayward@arm.com>
6151 David Sherwood <david.sherwood@arm.com>
6153 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6154 they are variable or constant sized.
6155 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6156 slots for constant-sized data.
6158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6159 Alan Hayward <alan.hayward@arm.com>
6160 David Sherwood <david.sherwood@arm.com>
6162 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6163 handling COND_EXPRs with boolean comparisons, try to find a better
6164 basis for the mask type than the boolean itself.
6166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6168 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6169 is calculated and how it can be overridden.
6170 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6171 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6173 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6177 Alan Hayward <alan.hayward@arm.com>
6178 David Sherwood <david.sherwood@arm.com>
6180 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6181 Remove the mode argument.
6182 (aarch64_simd_valid_immediate): Remove the mode and inverse
6184 * config/aarch64/iterators.md (bitsize): New iterator.
6185 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6186 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6187 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6188 aarch64_simd_valid_immediate.
6189 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6190 (aarch64_reg_or_bic_imm): Likewise.
6191 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6192 with an insn_type enum and msl with a modifier_type enum.
6193 Replace element_width with a scalar_mode. Change the shift
6194 to unsigned int. Add constructors for scalar_float_mode and
6195 scalar_int_mode elements.
6196 (aarch64_vect_float_const_representable_p): Delete.
6197 (aarch64_can_const_movi_rtx_p)
6198 (aarch64_simd_scalar_immediate_valid_for_move)
6199 (aarch64_simd_make_constant): Update call to
6200 aarch64_simd_valid_immediate.
6201 (aarch64_advsimd_valid_immediate_hs): New function.
6202 (aarch64_advsimd_valid_immediate): Likewise.
6203 (aarch64_simd_valid_immediate): Remove mode and inverse
6204 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6205 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6206 and aarch64_float_const_representable_p on the result.
6207 (aarch64_output_simd_mov_immediate): Remove mode argument.
6208 Update call to aarch64_simd_valid_immediate and use of
6209 simd_immediate_info.
6210 (aarch64_output_scalar_simd_mov_immediate): Update call
6213 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6214 Alan Hayward <alan.hayward@arm.com>
6215 David Sherwood <david.sherwood@arm.com>
6217 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6218 (mode_nunits): Likewise CONST_MODE_NUNITS.
6219 * machmode.def (ADJUST_NUNITS): Document.
6220 * genmodes.c (mode_data::need_nunits_adj): New field.
6221 (blank_mode): Update accordingly.
6222 (adj_nunits): New variable.
6223 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6225 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6226 listed in adj_nunits.
6227 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6228 listed in adj_nunits. Don't emit case statements for such modes.
6229 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6230 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6231 nothing if adj_nunits is nonnull.
6232 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6233 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6234 (emit_mode_fbit): Update use of print_maybe_const_decl.
6235 (emit_move_size): Likewise. Treat the array as non-const
6237 (emit_mode_adjustments): Handle adj_nunits.
6239 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6241 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6242 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6243 (VECTOR_MODES): Use it.
6244 (make_vector_modes): Take the prefix as an argument.
6246 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6247 Alan Hayward <alan.hayward@arm.com>
6248 David Sherwood <david.sherwood@arm.com>
6250 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6251 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6252 for MODE_VECTOR_BOOL.
6253 * machmode.def (VECTOR_BOOL_MODE): Document.
6254 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6255 (make_vector_bool_mode): New function.
6256 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6258 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6259 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6261 * stor-layout.c (int_mode_for_mode): Likewise.
6262 * tree.c (build_vector_type_for_mode): Likewise.
6263 * varasm.c (output_constant_pool_2): Likewise.
6264 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6265 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6266 for MODE_VECTOR_BOOL.
6267 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6268 of mode class checks.
6269 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6270 instead of a list of mode class checks.
6271 (expand_vector_scalar_condition): Likewise.
6272 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6275 Alan Hayward <alan.hayward@arm.com>
6276 David Sherwood <david.sherwood@arm.com>
6278 * machmode.h (mode_size): Change from unsigned short to
6280 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6281 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6282 or if measurement_type is not polynomial.
6283 (fixed_size_mode::includes_p): Check for constant-sized modes.
6284 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6285 return a poly_uint16 rather than an unsigned short.
6286 (emit_mode_size): Change the type of mode_size from unsigned short
6287 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6288 (emit_mode_adjustments): Cope with polynomial vector sizes.
6289 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6291 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6293 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6294 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6295 * caller-save.c (setup_save_areas): Likewise.
6296 (replace_reg_with_saved_mem): Likewise.
6297 * calls.c (emit_library_call_value_1): Likewise.
6298 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6299 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6300 (gen_lowpart_for_combine): Likewise.
6301 * convert.c (convert_to_integer_1): Likewise.
6302 * cse.c (equiv_constant, cse_insn): Likewise.
6303 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6304 (cselib_subst_to_values): Likewise.
6305 * dce.c (word_dce_process_block): Likewise.
6306 * df-problems.c (df_word_lr_mark_ref): Likewise.
6307 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6308 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6309 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6310 (rtl_for_decl_location): Likewise.
6311 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6312 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6313 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6314 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6315 (expand_expr_real_1): Likewise.
6316 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6317 (pad_below): Likewise.
6318 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6319 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6320 * ira.c (get_subreg_tracking_sizes): Likewise.
6321 * ira-build.c (ira_create_allocno_objects): Likewise.
6322 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6323 (ira_sort_regnos_for_alter_reg): Likewise.
6324 * ira-costs.c (record_operand_costs): Likewise.
6325 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6326 (resolve_simple_move): Likewise.
6327 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6328 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6329 (lra_constraints): Likewise.
6330 (CONST_POOL_OK_P): Reject variable-sized modes.
6331 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6332 (add_pseudo_to_slot, lra_spill): Likewise.
6333 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6334 * optabs-query.c (get_best_extraction_insn): Likewise.
6335 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6336 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6337 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6338 * recog.c (offsettable_address_addr_space_p): Likewise.
6339 * regcprop.c (maybe_mode_change): Likewise.
6340 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6341 * regrename.c (build_def_use): Likewise.
6342 * regstat.c (dump_reg_info): Likewise.
6343 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6344 (find_reloads, find_reloads_subreg_address): Likewise.
6345 * reload1.c (eliminate_regs_1): Likewise.
6346 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6347 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6348 (simplify_binary_operation_1, simplify_subreg): Likewise.
6349 * targhooks.c (default_function_arg_padding): Likewise.
6350 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6351 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6352 (verify_gimple_assign_ternary): Likewise.
6353 * tree-inline.c (estimate_move_cost): Likewise.
6354 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6355 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6356 (get_address_cost_ainc): Likewise.
6357 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6358 (vect_supportable_dr_alignment): Likewise.
6359 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6360 (vectorizable_reduction): Likewise.
6361 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6362 (vectorizable_operation, vectorizable_load): Likewise.
6363 * tree.c (build_same_sized_truth_vector_type): Likewise.
6364 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6365 * var-tracking.c (emit_note_insn_var_location): Likewise.
6366 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6367 (ADDR_VEC_ALIGN): Likewise.
6369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6370 Alan Hayward <alan.hayward@arm.com>
6371 David Sherwood <david.sherwood@arm.com>
6373 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6375 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6376 or if measurement_type is polynomial.
6377 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6378 * combine.c (make_extraction): Likewise.
6379 * dse.c (find_shift_sequence): Likewise.
6380 * dwarf2out.c (mem_loc_descriptor): Likewise.
6381 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6382 (extract_bit_field, extract_low_bits): Likewise.
6383 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6384 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6385 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6386 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6387 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6388 * reload.c (find_reloads): Likewise.
6389 * reload1.c (alter_reg): Likewise.
6390 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6391 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6392 * tree-if-conv.c (predicate_mem_writes): Likewise.
6393 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6394 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6395 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6396 * valtrack.c (dead_debug_insert_temp): Likewise.
6397 * varasm.c (mergeable_constant_section): Likewise.
6398 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6401 Alan Hayward <alan.hayward@arm.com>
6402 David Sherwood <david.sherwood@arm.com>
6404 * expr.c (expand_assignment): Cope with polynomial mode sizes
6405 when assigning to a CONCAT.
6407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6408 Alan Hayward <alan.hayward@arm.com>
6409 David Sherwood <david.sherwood@arm.com>
6411 * machmode.h (mode_precision): Change from unsigned short to
6413 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6415 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6416 or if measurement_type is not polynomial.
6417 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6418 in which the mode is already known to be a scalar_int_mode.
6419 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6420 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6422 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6423 for GET_MODE_PRECISION.
6424 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6425 for GET_MODE_PRECISION.
6426 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6428 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6429 (expand_field_assignment, make_extraction): Likewise.
6430 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6431 (get_last_value): Likewise.
6432 * convert.c (convert_to_integer_1): Likewise.
6433 * cse.c (cse_insn): Likewise.
6434 * expr.c (expand_expr_real_1): Likewise.
6435 * lra-constraints.c (simplify_operand_subreg): Likewise.
6436 * optabs-query.c (can_atomic_load_p): Likewise.
6437 * optabs.c (expand_atomic_load): Likewise.
6438 (expand_atomic_store): Likewise.
6439 * ree.c (combine_reaching_defs): Likewise.
6440 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6441 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6442 * tree.h (type_has_mode_precision_p): Likewise.
6443 * ubsan.c (instrument_si_overflow): Likewise.
6445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6446 Alan Hayward <alan.hayward@arm.com>
6447 David Sherwood <david.sherwood@arm.com>
6449 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6450 polynomial numbers of units.
6451 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6452 (valid_vector_subparts_p): New function.
6453 (build_vector_type): Remove temporary shim and take the number
6454 of units as a poly_uint64 rather than an int.
6455 (build_opaque_vector_type): Take the number of units as a
6456 poly_uint64 rather than an int.
6457 * tree.c (build_vector_from_ctor): Handle polynomial
6458 TYPE_VECTOR_SUBPARTS.
6459 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6460 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6461 (build_vector_from_val): If the number of units is variable,
6462 use build_vec_duplicate_cst for constant operands and
6463 VEC_DUPLICATE_EXPR otherwise.
6464 (make_vector_type): Remove temporary is_constant ().
6465 (build_vector_type, build_opaque_vector_type): Take the number of
6466 units as a poly_uint64 rather than an int.
6467 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6469 * cfgexpand.c (expand_debug_expr): Likewise.
6470 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6471 (store_constructor, expand_expr_real_1): Likewise.
6472 (const_scalar_mask_from_tree): Likewise.
6473 * fold-const-call.c (fold_const_reduction): Likewise.
6474 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6475 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6476 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6477 (fold_relational_const): Likewise.
6478 (native_interpret_vector): Likewise. Change the size from an
6479 int to an unsigned int.
6480 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6481 TYPE_VECTOR_SUBPARTS.
6482 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6483 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6484 duplicating a non-constant operand into a variable-length vector.
6485 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6486 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6487 * ipa-icf.c (sem_variable::equals): Likewise.
6488 * match.pd: Likewise.
6489 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6490 * print-tree.c (print_node): Likewise.
6491 * stor-layout.c (layout_type): Likewise.
6492 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6493 * tree-cfg.c (verify_gimple_comparison): Likewise.
6494 (verify_gimple_assign_binary): Likewise.
6495 (verify_gimple_assign_ternary): Likewise.
6496 (verify_gimple_assign_single): Likewise.
6497 * tree-pretty-print.c (dump_generic_node): Likewise.
6498 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6499 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6500 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6501 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6502 (vect_shift_permute_load_chain): Likewise.
6503 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6504 (expand_vector_condition, optimize_vector_constructor): Likewise.
6505 (lower_vec_perm, get_compute_type): Likewise.
6506 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6507 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6508 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6509 (vect_recog_mask_conversion_pattern): Likewise.
6510 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6511 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6512 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6513 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6514 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6515 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6516 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6517 (supportable_widening_operation): Likewise.
6518 (supportable_narrowing_operation): Likewise.
6519 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6521 * varasm.c (output_constant): Likewise.
6523 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6524 Alan Hayward <alan.hayward@arm.com>
6525 David Sherwood <david.sherwood@arm.com>
6527 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6528 so that both the length == 3 and length != 3 cases set up their
6529 own permute vectors. Add comments explaining why we know the
6530 number of elements is constant.
6531 (vect_permute_load_chain): Likewise.
6533 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6534 Alan Hayward <alan.hayward@arm.com>
6535 David Sherwood <david.sherwood@arm.com>
6537 * machmode.h (mode_nunits): Change from unsigned char to
6539 (ONLY_FIXED_SIZE_MODES): New macro.
6540 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6541 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6542 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6544 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6545 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6546 or if measurement_type is not polynomial.
6547 * genmodes.c (ZERO_COEFFS): New macro.
6548 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6550 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6551 Use ZERO_COEFFS when emitting initializers.
6552 * data-streamer.h (bp_pack_poly_value): New function.
6553 (bp_unpack_poly_value): Likewise.
6554 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6555 for GET_MODE_NUNITS.
6556 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6557 for GET_MODE_NUNITS.
6558 * tree.c (make_vector_type): Remove temporary shim and make
6559 the real function take the number of units as a poly_uint64
6561 (build_vector_type_for_mode): Handle polynomial nunits.
6562 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6563 * emit-rtl.c (const_vec_series_p_1): Likewise.
6564 (gen_rtx_CONST_VECTOR): Likewise.
6565 * fold-const.c (test_vec_duplicate_folding): Likewise.
6566 * genrecog.c (validate_pattern): Likewise.
6567 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6568 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6569 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6570 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6571 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6572 * rtlanal.c (subreg_get_info): Likewise.
6573 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6574 (vect_grouped_load_supported): Likewise.
6575 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6576 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6577 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6578 (simplify_const_unary_operation, simplify_binary_operation_1)
6579 (simplify_const_binary_operation, simplify_ternary_operation)
6580 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6581 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6582 instead of CONST_VECTOR_NUNITS.
6583 * varasm.c (output_constant_pool_2): Likewise.
6584 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6585 explicit-encoded elements in the XVEC for variable-length vectors.
6587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6589 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6591 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6592 Alan Hayward <alan.hayward@arm.com>
6593 David Sherwood <david.sherwood@arm.com>
6595 * coretypes.h (fixed_size_mode): Declare.
6596 (fixed_size_mode_pod): New typedef.
6597 * builtins.h (target_builtins::x_apply_args_mode)
6598 (target_builtins::x_apply_result_mode): Change type to
6599 fixed_size_mode_pod.
6600 * builtins.c (apply_args_size, apply_result_size, result_vector)
6601 (expand_builtin_apply_args_1, expand_builtin_apply)
6602 (expand_builtin_return): Update accordingly.
6604 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6606 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6607 * cselib.c (cselib_hash_rtx): Likewise.
6608 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6609 CONST_VECTOR encoding.
6611 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6612 Jeff Law <law@redhat.com>
6615 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6616 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6617 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6618 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6621 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6622 explicitly probe *sp in a noreturn function if there were any callee
6623 register saves or frame pointer is needed.
6625 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6628 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6629 BLKmode for ternary, binary or unary expressions.
6632 * var-tracking.c (delete_vta_debug_insn): New inline function.
6633 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6634 insns from get_insns () to NULL instead of each bb separately.
6635 Use delete_vta_debug_insn. No longer static.
6636 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6637 delete_vta_debug_insns callers.
6638 * rtl.h (delete_vta_debug_insns): Declare.
6639 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6640 instead of variable_tracking_main.
6642 2018-01-03 Martin Sebor <msebor@redhat.com>
6644 PR tree-optimization/83603
6645 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6646 arguments past the endof the argument list in functions declared
6647 without a prototype.
6648 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6649 Avoid checking when arguments are null.
6651 2018-01-03 Martin Sebor <msebor@redhat.com>
6654 * doc/extend.texi (attribute const): Fix a typo.
6655 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6656 issuing -Wsuggest-attribute for void functions.
6658 2018-01-03 Martin Sebor <msebor@redhat.com>
6660 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6661 offset_int::from instead of wide_int::to_shwi.
6662 (maybe_diag_overlap): Remove assertion.
6663 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6664 * gimple-ssa-sprintf.c (format_directive): Same.
6665 (parse_directive): Same.
6666 (sprintf_dom_walker::compute_format_length): Same.
6667 (try_substitute_return_value): Same.
6669 2018-01-03 Jeff Law <law@redhat.com>
6672 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6673 non-constant residual for zero at runtime and avoid probing in
6674 that case. Reorganize code for trailing problem to mirror handling
6677 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6679 PR tree-optimization/83501
6680 * tree-ssa-strlen.c (get_string_cst): New.
6681 (handle_char_store): Call get_string_cst.
6683 2018-01-03 Martin Liska <mliska@suse.cz>
6685 PR tree-optimization/83593
6686 * tree-ssa-strlen.c: Include tree-cfg.h.
6687 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6688 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6689 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6691 (strlen_dom_walker::before_dom_children): Call
6692 gimple_purge_dead_eh_edges. Dump tranformation with details
6694 (strlen_dom_walker::before_dom_children): Update call by adding
6695 new argument cleanup_eh.
6696 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6698 2018-01-03 Martin Liska <mliska@suse.cz>
6701 * cif-code.def (VARIADIC_THUNK): New enum value.
6702 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6705 2018-01-03 Jan Beulich <jbeulich@suse.com>
6707 * sse.md (mov<mode>_internal): Tighten condition for when to use
6708 vmovdqu<ssescalarsize> for TI and OI modes.
6710 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6712 Update copyright years.
6714 2018-01-03 Martin Liska <mliska@suse.cz>
6717 * ipa-visibility.c (function_and_variable_visibility): Skip
6718 functions with noipa attribure.
6720 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6722 * gcc.c (process_command): Update copyright notice dates.
6723 * gcov-dump.c (print_version): Ditto.
6724 * gcov.c (print_version): Ditto.
6725 * gcov-tool.c (print_version): Ditto.
6726 * gengtype.c (create_file): Ditto.
6727 * doc/cpp.texi: Bump @copying's copyright year.
6728 * doc/cppinternals.texi: Ditto.
6729 * doc/gcc.texi: Ditto.
6730 * doc/gccint.texi: Ditto.
6731 * doc/gcov.texi: Ditto.
6732 * doc/install.texi: Ditto.
6733 * doc/invoke.texi: Ditto.
6735 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6737 * vector-builder.h (vector_builder::m_full_nelts): Change from
6738 unsigned int to poly_uint64.
6739 (vector_builder::full_nelts): Update prototype accordingly.
6740 (vector_builder::new_vector): Likewise.
6741 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6742 (vector_builder::operator ==): Likewise.
6743 (vector_builder::finalize): Likewise.
6744 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6745 Take the number of elements as a poly_uint64 rather than an
6747 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6748 from unsigned int to poly_uint64.
6749 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6750 (vec_perm_indices::new_vector): Likewise.
6751 (vec_perm_indices::length): Likewise.
6752 (vec_perm_indices::nelts_per_input): Likewise.
6753 (vec_perm_indices::input_nelts): Likewise.
6754 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6755 number of elements per input as a poly_uint64 rather than an
6756 unsigned int. Use the original encoding for variable-length
6757 vectors, rather than clamping each individual element.
6758 For the second and subsequent elements in each pattern,
6759 clamp the step and base before clamping their sum.
6760 (vec_perm_indices::series_p): Handle polynomial element counts.
6761 (vec_perm_indices::all_in_range_p): Likewise.
6762 (vec_perm_indices_to_tree): Likewise.
6763 (vec_perm_indices_to_rtx): Likewise.
6764 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6765 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6766 (tree_vector_builder::new_binary_operation): Handle polynomial
6767 element counts. Return false if we need to know the number
6768 of elements at compile time.
6769 * fold-const.c (fold_vec_perm): Punt if the number of elements
6770 isn't known at compile time.
6772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6774 * vec-perm-indices.h (vec_perm_builder): Change element type
6775 from HOST_WIDE_INT to poly_int64.
6776 (vec_perm_indices::element_type): Update accordingly.
6777 (vec_perm_indices::clamp): Handle polynomial element_types.
6778 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6779 (vec_perm_indices::all_in_range_p): Likewise.
6780 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6782 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6783 polynomial vec_perm_indices element types.
6784 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6785 * fold-const.c (fold_vec_perm): Likewise.
6786 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6787 * tree-vect-generic.c (lower_vec_perm): Likewise.
6788 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6789 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6790 element type to HOST_WIDE_INT.
6792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6793 Alan Hayward <alan.hayward@arm.com>
6794 David Sherwood <david.sherwood@arm.com>
6796 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6797 rather than an int. Use plus_constant.
6798 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6799 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6801 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6802 Alan Hayward <alan.hayward@arm.com>
6803 David Sherwood <david.sherwood@arm.com>
6805 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6806 a HOST_WIDE_INT to a poly_int64.
6808 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6809 Alan Hayward <alan.hayward@arm.com>
6810 David Sherwood <david.sherwood@arm.com>
6812 * calls.c (load_register_parameters): Cope with polynomial
6813 mode sizes. Require a constant size for BLKmode parameters
6814 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6815 forces a parameter to be padded at the lsb end in order to
6816 fill a complete number of words, require the parameter size
6817 to be ordered wrt UNITS_PER_WORD.
6819 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6820 Alan Hayward <alan.hayward@arm.com>
6821 David Sherwood <david.sherwood@arm.com>
6823 * reload1.c (spill_stack_slot_width): Change element type
6824 from unsigned int to poly_uint64_pod.
6825 (alter_reg): Treat mode sizes as polynomial.
6827 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6828 Alan Hayward <alan.hayward@arm.com>
6829 David Sherwood <david.sherwood@arm.com>
6831 * reload.c (complex_word_subreg_p): New function.
6832 (reload_inner_reg_of_subreg, push_reload): Use it.
6834 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6835 Alan Hayward <alan.hayward@arm.com>
6836 David Sherwood <david.sherwood@arm.com>
6838 * lra-constraints.c (process_alt_operands): Reject matched
6839 operands whose sizes aren't ordered.
6840 (match_reload): Refer to this check here.
6842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6843 Alan Hayward <alan.hayward@arm.com>
6844 David Sherwood <david.sherwood@arm.com>
6846 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6847 that the mode size is in the set {1, 2, 4, 8, 16}.
6849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6850 Alan Hayward <alan.hayward@arm.com>
6851 David Sherwood <david.sherwood@arm.com>
6853 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6854 Use plus_constant instead of gen_rtx_PLUS.
6856 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6857 Alan Hayward <alan.hayward@arm.com>
6858 David Sherwood <david.sherwood@arm.com>
6860 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6861 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6862 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6863 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6864 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6865 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6866 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6867 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6868 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6869 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6871 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6872 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6873 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6874 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6875 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6876 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6877 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6878 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6879 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6880 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6882 * expr.c (emit_move_resolve_push): Treat the input and result
6883 of PUSH_ROUNDING as a poly_int64.
6884 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6885 (emit_push_insn): Likewise.
6886 * lra-eliminations.c (mark_not_eliminable): Likewise.
6887 * recog.c (push_operand): Likewise.
6888 * reload1.c (elimination_effects): Likewise.
6889 * rtlanal.c (nonzero_bits1): Likewise.
6890 * calls.c (store_one_arg): Likewise. Require the padding to be
6891 known at compile time.
6893 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6894 Alan Hayward <alan.hayward@arm.com>
6895 David Sherwood <david.sherwood@arm.com>
6897 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6898 Use plus_constant instead of gen_rtx_PLUS.
6900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6901 Alan Hayward <alan.hayward@arm.com>
6902 David Sherwood <david.sherwood@arm.com>
6904 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6908 Alan Hayward <alan.hayward@arm.com>
6909 David Sherwood <david.sherwood@arm.com>
6911 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6912 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6913 via stack temporaries. Treat the mode size as polynomial too.
6915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6916 Alan Hayward <alan.hayward@arm.com>
6917 David Sherwood <david.sherwood@arm.com>
6919 * expr.c (expand_expr_real_2): When handling conversions involving
6920 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6921 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6922 as a poly_uint64 too.
6924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6925 Alan Hayward <alan.hayward@arm.com>
6926 David Sherwood <david.sherwood@arm.com>
6928 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6930 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6931 Alan Hayward <alan.hayward@arm.com>
6932 David Sherwood <david.sherwood@arm.com>
6934 * combine.c (can_change_dest_mode): Handle polynomial
6935 REGMODE_NATURAL_SIZE.
6936 * expmed.c (store_bit_field_1): Likewise.
6937 * expr.c (store_constructor): Likewise.
6938 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6939 and polynomial REGMODE_NATURAL_SIZE.
6940 (gen_lowpart_common): Likewise.
6941 * reginfo.c (record_subregs_of_mode): Likewise.
6942 * rtlanal.c (read_modify_subreg_p): Likewise.
6944 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6945 Alan Hayward <alan.hayward@arm.com>
6946 David Sherwood <david.sherwood@arm.com>
6948 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6949 numbers of elements.
6951 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6952 Alan Hayward <alan.hayward@arm.com>
6953 David Sherwood <david.sherwood@arm.com>
6955 * match.pd: Cope with polynomial numbers of vector elements.
6957 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6958 Alan Hayward <alan.hayward@arm.com>
6959 David Sherwood <david.sherwood@arm.com>
6961 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6962 in a POINTER_PLUS_EXPR.
6964 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6965 Alan Hayward <alan.hayward@arm.com>
6966 David Sherwood <david.sherwood@arm.com>
6968 * omp-simd-clone.c (simd_clone_subparts): New function.
6969 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6970 (ipa_simd_modify_function_body): Likewise.
6972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6973 Alan Hayward <alan.hayward@arm.com>
6974 David Sherwood <david.sherwood@arm.com>
6976 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6977 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6978 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6979 (expand_vector_condition, vector_element): Likewise.
6980 (subparts_gt): New function.
6981 (get_compute_type): Use subparts_gt.
6982 (count_type_subparts): Delete.
6983 (expand_vector_operations_1): Use subparts_gt instead of
6984 count_type_subparts.
6986 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6987 Alan Hayward <alan.hayward@arm.com>
6988 David Sherwood <david.sherwood@arm.com>
6990 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6991 (vect_compile_time_alias): ...this new function. Do the calculation
6992 on poly_ints rather than trees.
6993 (vect_prune_runtime_alias_test_list): Update call accordingly.
6995 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6996 Alan Hayward <alan.hayward@arm.com>
6997 David Sherwood <david.sherwood@arm.com>
6999 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7001 (vect_schedule_slp_instance): Likewise.
7003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7004 Alan Hayward <alan.hayward@arm.com>
7005 David Sherwood <david.sherwood@arm.com>
7007 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7008 constant and extern definitions for variable-length vectors.
7009 (vect_get_constant_vectors): Note that the number of units
7010 is known to be constant.
7012 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7013 Alan Hayward <alan.hayward@arm.com>
7014 David Sherwood <david.sherwood@arm.com>
7016 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7017 of units as polynomial. Choose between WIDE and NARROW based
7020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7021 Alan Hayward <alan.hayward@arm.com>
7022 David Sherwood <david.sherwood@arm.com>
7024 * tree-vect-stmts.c (simd_clone_subparts): New function.
7025 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7027 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7028 Alan Hayward <alan.hayward@arm.com>
7029 David Sherwood <david.sherwood@arm.com>
7031 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7032 vectors as polynomial. Use build_index_vector for
7035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7036 Alan Hayward <alan.hayward@arm.com>
7037 David Sherwood <david.sherwood@arm.com>
7039 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7040 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7041 for variable-length vectors.
7042 (vectorizable_mask_load_store): Treat the number of units as
7043 polynomial, asserting that it is constant if the condition has
7044 already been enforced.
7045 (vectorizable_store, vectorizable_load): Likewise.
7047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7048 Alan Hayward <alan.hayward@arm.com>
7049 David Sherwood <david.sherwood@arm.com>
7051 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7052 of units as polynomial. Punt if we can't tell at compile time
7053 which vector contains the final result.
7055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7056 Alan Hayward <alan.hayward@arm.com>
7057 David Sherwood <david.sherwood@arm.com>
7059 * tree-vect-loop.c (vectorizable_induction): Treat the number
7060 of units as polynomial. Punt on SLP inductions. Use an integer
7061 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7062 cast of such a series for variable-length floating-point
7065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7066 Alan Hayward <alan.hayward@arm.com>
7067 David Sherwood <david.sherwood@arm.com>
7069 * tree.h (build_index_vector): Declare.
7070 * tree.c (build_index_vector): New function.
7071 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7072 of units as polynomial, forcibly converting it to a constant if
7073 vectorizable_reduction has already enforced the condition.
7074 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7075 to create a {1,2,3,...} vector.
7076 (vectorizable_reduction): Treat the number of units as polynomial.
7077 Choose vectype_in based on the largest scalar element size rather
7078 than the smallest number of units. Enforce the restrictions
7081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7082 Alan Hayward <alan.hayward@arm.com>
7083 David Sherwood <david.sherwood@arm.com>
7085 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7086 number of units as polynomial.
7088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7089 Alan Hayward <alan.hayward@arm.com>
7090 David Sherwood <david.sherwood@arm.com>
7092 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7093 * target.def (autovectorize_vector_sizes): Return the vector sizes
7094 by pointer, using vector_sizes rather than a bitmask.
7095 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7096 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7097 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7099 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7100 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7101 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7102 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7103 * omp-general.c (omp_max_vf): Likewise.
7104 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7105 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7106 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7107 * tree-vect-slp.c (vect_slp_bb): Likewise.
7108 * doc/tm.texi: Regenerate.
7109 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7111 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7112 the vector size as a poly_uint64 rather than an unsigned int.
7113 (current_vector_size): Change from an unsigned int to a poly_uint64.
7114 (get_vectype_for_scalar_type): Update accordingly.
7115 * tree.h (build_truth_vector_type): Take the size and number of
7116 units as a poly_uint64 rather than an unsigned int.
7117 (build_vector_type): Add a temporary overload that takes
7118 the number of units as a poly_uint64 rather than an unsigned int.
7119 * tree.c (make_vector_type): Likewise.
7120 (build_truth_vector_type): Take the number of units as a poly_uint64
7121 rather than an unsigned int.
7123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7124 Alan Hayward <alan.hayward@arm.com>
7125 David Sherwood <david.sherwood@arm.com>
7127 * target.def (get_mask_mode): Take the number of units and length
7128 as poly_uint64s rather than unsigned ints.
7129 * targhooks.h (default_get_mask_mode): Update accordingly.
7130 * targhooks.c (default_get_mask_mode): Likewise.
7131 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7132 * doc/tm.texi: Regenerate.
7134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7135 Alan Hayward <alan.hayward@arm.com>
7136 David Sherwood <david.sherwood@arm.com>
7138 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7139 * omp-general.c (omp_max_vf): Likewise.
7140 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7141 (expand_omp_simd): Handle polynomial safelen.
7142 * omp-low.c (omplow_simd_context): Add a default constructor.
7143 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7144 (lower_rec_simd_input_clauses): Update accordingly.
7145 (lower_rec_input_clauses): Likewise.
7147 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7148 Alan Hayward <alan.hayward@arm.com>
7149 David Sherwood <david.sherwood@arm.com>
7151 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7152 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7153 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7154 (vect_analyze_slp_cost): Likewise.
7155 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7156 (vect_model_load_cost): Likewise.
7158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7159 Alan Hayward <alan.hayward@arm.com>
7160 David Sherwood <david.sherwood@arm.com>
7162 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7163 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7164 from an unsigned int * to a poly_uint64_pod *.
7165 (calculate_unrolling_factor): New function.
7166 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7168 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7169 Alan Hayward <alan.hayward@arm.com>
7170 David Sherwood <david.sherwood@arm.com>
7172 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7173 from an unsigned int to a poly_uint64.
7174 (_loop_vec_info::slp_unrolling_factor): Likewise.
7175 (_loop_vec_info::vectorization_factor): Change from an int
7177 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7178 (vect_get_num_vectors): New function.
7179 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7180 (vect_get_num_copies): Use vect_get_num_vectors.
7181 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7182 to an unsigned int *.
7183 (vect_analyze_data_refs): Change min_vf from an int * to a
7185 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7186 than an unsigned HOST_WIDE_INT.
7187 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7188 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7189 to an unsigned int *.
7190 (vect_analyze_data_ref_dependences): Likewise.
7191 (vect_compute_data_ref_alignment): Handle polynomial vf.
7192 (vect_enhance_data_refs_alignment): Likewise.
7193 (vect_prune_runtime_alias_test_list): Likewise.
7194 (vect_shift_permute_load_chain): Likewise.
7195 (vect_supportable_dr_alignment): Likewise.
7196 (dependence_distance_ge_vf): Take the vectorization factor as a
7197 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7198 (vect_analyze_data_refs): Change min_vf from an int * to a
7200 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7201 vfm1 as a poly_uint64 rather than an int. Make the same change
7202 for the returned bound_scalar.
7203 (vect_gen_vector_loop_niters): Handle polynomial vf.
7204 (vect_do_peeling): Likewise. Update call to
7205 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7206 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7208 * tree-vect-loop.c (vect_determine_vectorization_factor)
7209 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7210 (vect_get_known_peeling_cost): Likewise.
7211 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7212 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7213 (vect_transform_loop): Likewise. Use the lowest possible VF when
7214 updating the upper bounds of the loop.
7215 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7217 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7218 polynomial unroll factors.
7219 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7220 (vect_make_slp_decision): Likewise.
7221 (vect_supported_load_permutation_p): Likewise, and polynomial
7223 (vect_analyze_slp_cost): Handle polynomial vf.
7224 (vect_slp_analyze_node_operations): Likewise.
7225 (vect_slp_analyze_bb_1): Likewise.
7226 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7227 than an unsigned HOST_WIDE_INT.
7228 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7229 (vectorizable_load): Handle polynomial vf.
7230 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7232 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7235 Alan Hayward <alan.hayward@arm.com>
7236 David Sherwood <david.sherwood@arm.com>
7238 * match.pd: Handle bit operations involving three constants
7239 and try to fold one pair.
7241 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7243 * tree-vect-loop-manip.c: Include gimple-fold.h.
7244 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7245 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7246 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7247 Add a path that uses a step of VF instead of 1, but disable it
7249 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7250 and niters_no_overflow parameters. Update calls to
7251 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7252 Create a new SSA name if the latter choses to use a ste other
7253 than zero, and return it via niters_vector_mult_vf_var.
7254 * tree-vect-loop.c (vect_transform_loop): Update calls to
7255 vect_do_peeling, vect_gen_vector_loop_niters and
7256 slpeel_make_loop_iterate_ntimes.
7257 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7258 (vect_gen_vector_loop_niters): Update declarations after above changes.
7260 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7262 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7263 128-bit round to integer instructions.
7264 (ceil<mode>2): Likewise.
7265 (btrunc<mode>2): Likewise.
7266 (round<mode>2): Likewise.
7268 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7270 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7271 unaligned VSX load/store on P8/P9.
7272 (expand_block_clear): Allow the use of unaligned VSX
7273 load/store on P8/P9.
7275 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7277 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7279 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7280 swap associated with both a load and a store.
7282 2018-01-02 Andrew Waterman <andrew@sifive.com>
7284 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7285 * config/riscv/riscv.md (clear_cache): Use it.
7287 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7289 * web.c: Remove out-of-date comment.
7291 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7293 * expr.c (fixup_args_size_notes): Check that any existing
7294 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7295 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7296 (emit_single_push_insn): ...here.
7298 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7300 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7301 (const_vector_encoded_nelts): New function.
7302 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7303 (const_vector_int_elt, const_vector_elt): Declare.
7304 * emit-rtl.c (const_vector_int_elt_1): New function.
7305 (const_vector_elt): Likewise.
7306 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7307 of CONST_VECTOR_ELT.
7309 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7311 * expr.c: Include rtx-vector-builder.h.
7312 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7313 directly on the tree encoding.
7314 (const_vector_from_tree): Likewise.
7315 * optabs.c: Include rtx-vector-builder.h.
7316 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7317 sequence of "u" values.
7318 * vec-perm-indices.c: Include rtx-vector-builder.h.
7319 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7320 directly on the vec_perm_indices encoding.
7322 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7324 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7325 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7326 * rtx-vector-builder.h: New file.
7327 * rtx-vector-builder.c: Likewise.
7328 * rtl.h (rtx_def::u2): Add a const_vector field.
7329 (CONST_VECTOR_NPATTERNS): New macro.
7330 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7331 (CONST_VECTOR_DUPLICATE_P): Likewise.
7332 (CONST_VECTOR_STEPPED_P): Likewise.
7333 (CONST_VECTOR_ENCODED_ELT): Likewise.
7334 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7335 (unwrap_const_vec_duplicate): Likewise.
7336 (const_vec_series_p): Check for a non-duplicated vector encoding.
7337 Say that the function only returns true for integer vectors.
7338 * emit-rtl.c: Include rtx-vector-builder.h.
7339 (gen_const_vec_duplicate_1): Delete.
7340 (gen_const_vector): Call gen_const_vec_duplicate instead of
7341 gen_const_vec_duplicate_1.
7342 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7343 (gen_const_vec_duplicate): Use rtx_vector_builder.
7344 (gen_const_vec_series): Likewise.
7345 (gen_rtx_CONST_VECTOR): Likewise.
7346 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7347 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7348 Build a new vector rather than modifying a CONST_VECTOR in-place.
7349 (handle_special_swappables): Update call accordingly.
7350 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7351 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7352 Build a new vector rather than modifying a CONST_VECTOR in-place.
7353 (handle_special_swappables): Update call accordingly.
7355 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7357 * simplify-rtx.c (simplify_const_binary_operation): Use
7358 CONST_VECTOR_ELT instead of XVECEXP.
7360 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7362 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7363 the selector elements to be different from the data elements
7364 if the selector is a VECTOR_CST.
7365 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7366 ssizetype for the selector.
7368 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7370 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7371 before testing each element individually.
7372 * tree-vect-generic.c (lower_vec_perm): Likewise.
7374 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7376 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7377 * selftest-run-tests.c (selftest::run_tests): Call it.
7378 * vector-builder.h (vector_builder::operator ==): New function.
7379 (vector_builder::operator !=): Likewise.
7380 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7381 (vec_perm_indices::all_from_input_p): New function.
7382 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7383 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7384 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7385 instead of reading the VECTOR_CST directly. Detect whether both
7386 vector inputs are the same before constructing the vec_perm_indices,
7387 and update the number of inputs argument accordingly. Use the
7388 utility functions added above. Only construct sel2 if we need to.
7390 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7392 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7393 the broadcast of the low byte.
7394 (expand_mult_highpart): Use an explicit encoding for the permutes.
7395 * optabs-query.c (can_mult_highpart_p): Likewise.
7396 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7397 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7398 (vectorizable_bswap): Likewise.
7399 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7400 explicit encoding for the power-of-2 permutes.
7401 (vect_permute_store_chain): Likewise.
7402 (vect_grouped_load_supported): Likewise.
7403 (vect_permute_load_chain): Likewise.
7405 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7407 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7408 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7409 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7410 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7411 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7412 (vect_gen_perm_mask_any): Likewise.
7414 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7416 * int-vector-builder.h: New file.
7417 * vec-perm-indices.h: Include int-vector-builder.h.
7418 (vec_perm_indices): Redefine as an int_vector_builder.
7419 (auto_vec_perm_indices): Delete.
7420 (vec_perm_builder): Redefine as a stand-alone class.
7421 (vec_perm_indices::vec_perm_indices): New function.
7422 (vec_perm_indices::clamp): Likewise.
7423 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7424 (vec_perm_indices::new_vector): New function.
7425 (vec_perm_indices::new_expanded_vector): Update for new
7426 vec_perm_indices class.
7427 (vec_perm_indices::rotate_inputs): New function.
7428 (vec_perm_indices::all_in_range_p): Operate directly on the
7429 encoded form, without computing elided elements.
7430 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7431 encoding. Update for new vec_perm_indices class.
7432 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7433 the given vec_perm_builder.
7434 (expand_vec_perm_var): Update vec_perm_builder constructor.
7435 (expand_mult_highpart): Use vec_perm_builder instead of
7436 auto_vec_perm_indices.
7437 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7438 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7439 or double series encoding as appropriate.
7440 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7441 vec_perm_indices instead of auto_vec_perm_indices.
7442 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7443 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7444 (vect_permute_store_chain): Likewise.
7445 (vect_grouped_load_supported): Likewise.
7446 (vect_permute_load_chain): Likewise.
7447 (vect_shift_permute_load_chain): Likewise.
7448 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7449 (vect_transform_slp_perm_load): Likewise.
7450 (vect_schedule_slp_instance): Likewise.
7451 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7452 (vectorizable_mask_load_store): Likewise.
7453 (vectorizable_bswap): Likewise.
7454 (vectorizable_store): Likewise.
7455 (vectorizable_load): Likewise.
7456 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7457 vec_perm_indices instead of auto_vec_perm_indices. Use
7458 tree_to_vec_perm_builder to read the vector from a tree.
7459 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7460 vec_perm_builder instead of a vec_perm_indices.
7461 (have_whole_vector_shift): Use vec_perm_builder and
7462 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7463 truncation to calc_vec_perm_mask_for_shift.
7464 (vect_create_epilog_for_reduction): Likewise.
7465 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7466 from auto_vec_perm_indices to vec_perm_indices.
7467 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7468 instead of changing individual elements.
7469 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7470 the vector in d.perm.
7471 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7472 from auto_vec_perm_indices to vec_perm_indices.
7473 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7474 instead of changing individual elements.
7475 (arm_vectorize_vec_perm_const): Use new_vector to install
7476 the vector in d.perm.
7477 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7478 Update vec_perm_builder constructor.
7479 (rs6000_expand_interleave): Likewise.
7480 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7481 (rs6000_expand_interleave): Likewise.
7483 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7485 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7486 to qimode could truncate the indices.
7487 * optabs.c (expand_vec_perm_var): Likewise.
7489 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7491 * Makefile.in (OBJS): Add vec-perm-indices.o.
7492 * vec-perm-indices.h: New file.
7493 * vec-perm-indices.c: Likewise.
7494 * target.h (vec_perm_indices): Replace with a forward class
7496 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7497 * optabs.h: Include vec-perm-indices.h.
7498 (expand_vec_perm): Delete.
7499 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7500 (expand_vec_perm_const): Declare.
7501 * target.def (vec_perm_const_ok): Replace with...
7502 (vec_perm_const): ...this new hook.
7503 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7504 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7505 * doc/tm.texi: Regenerate.
7506 * optabs.def (vec_perm_const): Delete.
7507 * doc/md.texi (vec_perm_const): Likewise.
7508 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7509 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7510 expand_vec_perm for constant permutation vectors. Assert that
7511 the mode of variable permutation vectors is the integer equivalent
7512 of the mode that is being permuted.
7513 * optabs-query.h (selector_fits_mode_p): Declare.
7514 * optabs-query.c: Include vec-perm-indices.h.
7515 (selector_fits_mode_p): New function.
7516 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7517 is defined, instead of checking whether the vec_perm_const_optab
7518 exists. Use targetm.vectorize.vec_perm_const instead of
7519 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7520 fit in the vector mode before using a variable permute.
7521 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7522 vec_perm_indices instead of an rtx.
7523 (expand_vec_perm): Replace with...
7524 (expand_vec_perm_const): ...this new function. Take the selector
7525 as a vec_perm_indices rather than an rtx. Also take the mode of
7526 the selector. Update call to shift_amt_for_vec_perm_mask.
7527 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7528 Use vec_perm_indices::new_expanded_vector to expand the original
7529 selector into bytes. Check whether the indices fit in the vector
7530 mode before using a variable permute.
7531 (expand_vec_perm_var): Make global.
7532 (expand_mult_highpart): Use expand_vec_perm_const.
7533 * fold-const.c: Includes vec-perm-indices.h.
7534 * tree-ssa-forwprop.c: Likewise.
7535 * tree-vect-data-refs.c: Likewise.
7536 * tree-vect-generic.c: Likewise.
7537 * tree-vect-loop.c: Likewise.
7538 * tree-vect-slp.c: Likewise.
7539 * tree-vect-stmts.c: Likewise.
7540 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7542 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7543 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7544 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7545 (aarch64_vectorize_vec_perm_const): ...this new function.
7546 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7547 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7548 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7549 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7550 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7551 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7552 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7554 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7555 check for NEON modes.
7556 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7557 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7558 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7559 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7561 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7562 the old VEC_PERM_CONST conditions.
7563 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7564 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7565 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7566 (ia64_vectorize_vec_perm_const_ok): Merge into...
7567 (ia64_vectorize_vec_perm_const): ...this new function.
7568 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7569 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7570 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7571 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7572 * config/mips/mips.c (mips_expand_vec_perm_const)
7573 (mips_vectorize_vec_perm_const_ok): Merge into...
7574 (mips_vectorize_vec_perm_const): ...this new function.
7575 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7576 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7577 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7578 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7579 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7580 (rs6000_expand_vec_perm_const): Delete.
7581 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7583 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7584 (altivec_expand_vec_perm_const_le): Take each operand individually.
7585 Operate on constant selectors rather than rtxes.
7586 (altivec_expand_vec_perm_const): Likewise. Update call to
7587 altivec_expand_vec_perm_const_le.
7588 (rs6000_expand_vec_perm_const): Delete.
7589 (rs6000_vectorize_vec_perm_const_ok): Delete.
7590 (rs6000_vectorize_vec_perm_const): New function.
7591 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7592 an element count and rtx array.
7593 (rs6000_expand_extract_even): Update call accordingly.
7594 (rs6000_expand_interleave): Likewise.
7595 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7596 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7597 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7598 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7599 (rs6000_expand_vec_perm_const): Delete.
7600 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7601 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7602 (altivec_expand_vec_perm_const_le): Take each operand individually.
7603 Operate on constant selectors rather than rtxes.
7604 (altivec_expand_vec_perm_const): Likewise. Update call to
7605 altivec_expand_vec_perm_const_le.
7606 (rs6000_expand_vec_perm_const): Delete.
7607 (rs6000_vectorize_vec_perm_const_ok): Delete.
7608 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7609 reference to the SPE evmerge intructions.
7610 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7611 an element count and rtx array.
7612 (rs6000_expand_extract_even): Update call accordingly.
7613 (rs6000_expand_interleave): Likewise.
7614 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7615 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7617 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7619 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7621 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7622 vector mode and that that mode matches the mode of the data
7624 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7625 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7626 directly using expand_vec_perm_1 when forcing selectors into
7628 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7630 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7632 * optabs-query.h (can_vec_perm_p): Delete.
7633 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7634 * optabs-query.c (can_vec_perm_p): Split into...
7635 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7636 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7637 particular selector is valid.
7638 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7639 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7640 (vect_grouped_load_supported): Likewise.
7641 (vect_shift_permute_load_chain): Likewise.
7642 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7643 (vect_transform_slp_perm_load): Likewise.
7644 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7645 (vectorizable_bswap): Likewise.
7646 (vect_gen_perm_mask_checked): Likewise.
7647 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7648 implementations of variable permutation vectors into account
7649 when deciding which selector to use.
7650 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7651 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7652 with a false third argument.
7653 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7654 to test whether the constant selector is valid and can_vec_perm_var_p
7655 to test whether a variable selector is valid.
7657 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7659 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7660 * optabs-query.c (can_vec_perm_p): Likewise.
7661 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7662 instead of vec_perm_indices.
7663 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7664 (vect_gen_perm_mask_checked): Likewise,
7665 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7666 (vect_gen_perm_mask_checked): Likewise,
7668 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7670 * optabs-query.h (qimode_for_vec_perm): Declare.
7671 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7672 (qimode_for_vec_perm): ...this new function.
7673 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7675 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7677 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7678 does not have a conditional at the top.
7680 2018-01-02 Richard Biener <rguenther@suse.de>
7682 * ipa-inline.c (big_speedup_p): Fix expression.
7684 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7687 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7690 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7694 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7695 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7696 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7697 cond_taken_branch_cost 3->4.
7699 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7701 PR tree-optimization/83581
7702 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7703 TODO_cleanup_cfg if any changes have been made.
7706 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7707 convert_modes if target mode has the right side, but different mode
7711 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7712 last argument when extracting from CONCAT. If either from_real or
7713 from_imag is NULL, use expansion through memory. If result is not
7714 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7715 the parts directly to inner mode, if even that fails, use expansion
7719 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7720 check for bswap in mode rather than HImode and use that in expand_unop
7723 Copyright (C) 2018 Free Software Foundation, Inc.
7725 Copying and distribution of this file, with or without modification,
7726 are permitted in any medium without royalty provided the copyright
7727 notice and this notice are preserved.