1 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
3 PR rtl-optimization/83496
4 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
5 booleans to RTX. Call fix_reg_dead_note on every non-null element.
6 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
7 redundant insn, if any.
8 (relax_delay_slots): Likewise.
9 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
11 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
13 PR tree-optimization/83965
14 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
15 that grouped statements are part of a reduction chain. Return
16 true if the statement is not marked as a reduction itself but
18 (vect_recog_dot_prod_pattern): Don't check whether the statement
19 is part of a group here.
20 (vect_recog_sad_pattern): Likewise.
21 (vect_recog_widen_sum_pattern): Likewise.
23 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
26 * final.c (rest_of_clean_state): Also look for calls inside sequences.
28 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
31 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
33 (ix86_output_indirect_function_return): New prototype.
34 (ix86_split_simple_return_pop_internal): Likewise.
35 * config/i386/i386.c (indirect_return_via_cx): New.
36 (indirect_return_via_cx_bnd): Likewise.
37 (indirect_thunk_name): Handle return va CX_REG.
38 (output_indirect_thunk_function): Create alias for
39 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
40 (ix86_output_indirect_jmp): Remove the bool argument.
41 (ix86_output_indirect_function_return): New function.
42 (ix86_split_simple_return_pop_internal): Likewise.
43 * config/i386/i386.md (*indirect_jump): Don't pass false
44 to ix86_output_indirect_jmp.
45 (*tablejump_1): Likewise.
46 (simple_return_pop_internal): Change it to define_insn_and_split.
47 Call ix86_split_simple_return_pop_internal to split it for
49 (simple_return_indirect_internal): Call
50 ix86_output_indirect_function_return instead of
51 ix86_output_indirect_jmp.
53 2018-02-26 Jakub Jelinek <jakub@redhat.com>
56 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
57 memset and value initialization afterwards.
59 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
61 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
63 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
66 * common/config/aarch64/aarch64-common.c
67 (aarch_option_optimization_table[]): Switch
68 off fomit-frame-pointer
70 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
71 Chung-Ju Wu <jasonwucj@gmail.com>
73 * config/nds32/nds32-multiple.md (load_multiple): Disallow
75 (store_multiple): Ditto.
77 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
79 * config.gcc: Add --with-cpu support for nds32 target.
80 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
81 * config/nds32/nds32.opt: Add -mcpu= option.
83 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
85 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
86 isel=yes): Warn for these deprecated options.
88 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
90 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
91 ISA_2_5_MASKS_EMBEDDED.
93 2018-02-23 Jakub Jelinek <jakub@redhat.com>
95 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
96 p->max as pointers rather than using iterative_hash_expr.
98 2018-02-23 Carl Love <cel@us.ibm.com>
100 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
101 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
103 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
104 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
105 P8V_BUILTIN_VEC_VUNSIGNED2.
107 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
110 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
111 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
113 * lra-constraints.c (curr_insn_transform): Set up
114 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
116 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
117 * lra-eliminations.c (spill_pseudos): Ditto.
118 (process_insn_for_elimination): Ditto.
119 * lra-lives.c (reg_early_clobber_p): Use the new macros.
120 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
121 LRA_NON_CLOBBERED_ALT.
123 2018-02-22 Martin Sebor <msebor@redhat.com>
125 PR tree-optimization/84480
126 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
127 to maybe_diag_stxncpy_trunc. Call it.
128 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
129 from gimple_fold_builtin_strcpy. Print inlining stack.
130 (handle_builtin_stxncpy): Print inlining stack.
131 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
133 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
136 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
137 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
138 and -fcheck-pointer-bounds are used together.
139 (indirect_thunk_prefix): New enum.
140 (indirect_thunk_need_prefix): New function.
141 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
142 "_nt" instead of "_bnd" for NOTRACK prefix.
143 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
144 (output_indirect_thunk_function): Likewise.
146 (ix86_code_end): Update output_indirect_thunk_function calls.
147 (ix86_output_indirect_branch_via_reg): Replace
148 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
149 (ix86_output_indirect_branch_via_push): Likewise.
150 (ix86_output_function_return): Likewise.
151 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
152 incompatible with -fcf-protection=branch and
153 -fcheck-pointer-bounds.
155 2018-02-22 Steve Ellcey <sellcey@cavium.com>
158 * config/aarch64/aarch64.c (aarch64_print_address_internal):
159 Change gcc_assert call to output_operand_lossage.
161 2018-02-22 Steve Ellcey <sellcey@cavium.com>
163 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
165 2018-02-22 DJ Delorie <dj@redhat.com>
166 Sebastian Perta <sebastian.perta@renesas.com>
167 Oleg Endo <olegendo@gcc.gnu.org>
169 * config/rx/rx.c (rx_rtx_costs): New function.
170 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
172 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
174 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
176 2018-02-22 Martin Liska <mliska@suse.cz>
179 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
180 Add "native" as a possible value.
182 2018-02-22 Martin Liska <mliska@suse.cz>
185 * config/i386/i386.c (ix86_option_override_internal):
186 Add "native" as a possible value for -march and -mtune.
188 2018-02-22 Jakub Jelinek <jakub@redhat.com>
191 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
192 to all type variants.
194 PR tree-optimization/84503
195 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
196 width as info->bitpos + info->bitsize - start.
197 (merged_store_group::merge_overlapping): Simplify width computation.
198 (check_no_overlap): New function.
199 (imm_store_chain_info::try_coalesce_bswap): Compute expected
200 start + width and last_order of the group, fail if check_no_overlap
202 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
203 to group if check_no_overlap fails.
205 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
207 * config/rs6000/altivec.md: Delete contraint arguments to
208 define_expand, define_split, and define_peephole2, and in
209 define_insn_and_split if always unused.
210 * config/rs6000/darwin.md: Ditto.
211 * config/rs6000/dfp.md: Ditto.
212 * config/rs6000/rs6000.md: Ditto.
213 * config/rs6000/sync.md: Ditto.
214 * config/rs6000/vector.md: Ditto.
215 * config/rs6000/vsx.md: Ditto.
217 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
219 * config/rs6000/altivec.md: Write output control strings as braced
220 blocks instead of double-quoted strings.
221 * config/rs6000/darwin.md: Ditto.
222 * config/rs6000/rs6000.md: Ditto.
223 * config/rs6000/vector.md: Ditto.
224 * config/rs6000/vsx.md: Ditto.
226 2018-02-21 Jason Merrill <jason@redhat.com>
228 PR c++/84314 - ICE with templates and fastcall attribute.
229 * attribs.c (build_type_attribute_qual_variant): Remove assert.
231 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
233 * ipa-cp.c (determine_versionability): Fix comment typos.
235 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
238 * ipa-cp.c (determine_versionability): Do not version functions caling
241 2018-02-21 Martin Liska <mliska@suse.cz>
244 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
245 Add "native" as a possible value.
246 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
247 the macro when native cpu detection is available.
249 2018-02-21 Martin Liska <mliska@suse.cz>
252 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
253 Add "native" as a possible value.
254 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
255 when native cpu detection is available.
257 2018-02-21 Jakub Jelinek <jakub@redhat.com>
258 Martin Sebor <msebor@redhat.com>
260 PR tree-optimization/84478
261 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
263 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
264 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
265 support which is conservatively correct, for 2 only stay conservative
266 for maxlen. Formatting and comment capitalization fixes. Add STRICT
267 argument to the 2 argument get_range_strlen, adjust 6 arg
268 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
270 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
271 (gimple_fold_builtin_strlen): Pass true as last argument to
274 2018-02-20 Martin Sebor <msebor@redhat.com>
277 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
278 (builtin_memref::set_base_and_offset): Same. Handle inner references.
279 (builtin_memref::builtin_memref): Factor out parts into
280 set_base_and_offset and call it.
282 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
285 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
286 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
287 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
288 search at the associated MODE_INT.
290 2018-02-20 Jeff Law <law@redhat.com>
293 PR tree-optimization/81592
295 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
296 for range data rather than using global data.
297 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
298 range data rather than using global data.
299 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
300 pass it to children as needed.
301 (struct directive::fmtresult): Similarly.
302 (struct directive::set_width): Similarly.
303 (struct directive::set_precision): Similarly.
304 (format_integer, format_directive, parse_directive): Similarly.
305 (format_none): Accept unnamed vr_values parameter.
306 (format_percent, format_floating, format_character): Similarly.
307 (format_string, format_plain): Similarly.
308 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
309 the EVRP range analyzer for range data rather than using global data.
310 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
311 gimple-ssa-evrp-analyze.h
312 (class sprintf_dom_walker): Add after_dom_children member function.
313 Add evrp_range_analyzer member.
314 (sprintf_dom_walker::before_dom_children): Call into the EVRP
315 range analyzer as needed.
316 (sprintf_dom_walker::after_dom_children): New member function.
317 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
319 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
320 (evrp_range_analyzer::pop_to_marker): Likewise.
322 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
324 PR tree-optimization/84419
325 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
326 with the required type if its current type is compatible but
329 2018-02-20 Jakub Jelinek <jakub@redhat.com>
332 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
335 2018-02-20 Martin Liska <mliska@suse.cz>
338 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
339 possible values if we don't have a hint.
341 2018-02-20 Martin Liska <mliska@suse.cz>
345 * final.c (shorten_branches): Build align_tab array with one
347 * opts.c (finish_options): Add alignment option limit check.
348 (MAX_CODE_ALIGN): Likewise.
349 (MAX_CODE_ALIGN_VALUE): Likewise.
350 * doc/invoke.texi: Document maximum allowed option value for
351 all -falign-* options.
353 2018-02-19 Jakub Jelinek <jakub@redhat.com>
356 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
357 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
358 * var-tracking.c (emit_note_insn_var_location): Remove all references
359 to NOTE_INSN_CALL_ARG_LOCATION.
360 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
361 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
362 Use copy_rtx_if_shared.
363 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
364 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
365 (dwarf2out_var_location): Remove handling of
366 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
368 * final.c (final_scan_insn): Remove all references to
369 NOTE_INSN_CALL_ARG_LOCATION.
370 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
371 before dumping final insns.
372 * except.c (emit_note_eh_region_end): Remove all references to
373 NOTE_INSN_CALL_ARG_LOCATION.
374 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
375 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
376 * config/arc/arc.c (hwloop_optimize): Likewise.
377 * config/arm/arm.c (create_fix_barrier): Likewise.
378 * config/s390/s390.c (s390_chunkify_start): Likewise.
379 * config/sh/sh.c (find_barrier): Likewise.
380 * config/i386/i386.c (rest_of_insert_endbranch,
381 ix86_seh_fixup_eh_fallthru): Likewise.
382 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
383 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
384 * config/frv/frv.c (frv_function_prologue): Likewise.
385 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
387 (note_outside_basic_block_p): Remove all references to
388 NOTE_INSN_CALL_ARG_LOCATION.
389 * gengtype.c (adjust_field_rtx_def): Likewise.
390 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
392 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
393 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
396 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
399 PR tree-optimization/84452
400 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
401 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
404 2018-02-19 Martin Liska <mliska@suse.cz>
407 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
409 2018-02-19 Martin Liska <mliska@suse.cz>
410 Richard Sandiford <richard.sandiford@linaro.org>
412 PR tree-optimization/82491
413 * gimple-fold.c (get_base_constructor): Make earlier bail out
416 2018-02-19 Carl Love <cel@us.ibm.com>
418 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
419 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
421 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
424 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
426 * config/rl78/rl78.md (movdf): New define expand.
428 2018-02-19 Martin Liska <mliska@suse.cz>
431 * doc/invoke.texi: Fix typo.
432 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
434 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
436 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
437 handle rs6000_single_float and rs6000_double_float specially for
440 2018-02-16 Jeff Law <law@redhat.com>
442 * config/rx/rx.c (add_pop_cfi_notes): New function.;
445 2018-02-16 Jakub Jelinek <jakub@redhat.com>
448 * ipa-inline.c (inline_small_functions): Fix a typo.
450 2018-02-16 Nathan Sidwell <nathan@acm.org>
452 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
454 2018-02-16 Carl Love <cel@us.ibm.com>
456 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
457 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
458 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
459 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
460 expansion to P8V_BUILTIN_VEC_FLOAT2.
462 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
464 PR rtl-optimization/70023
465 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
466 src_regno into account.
468 2018-02-16 Carl Love <cel@us.ibm.com>
470 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
471 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
472 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
473 * config/rs6000/rs6000.c: Remove case statements for
474 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
475 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
476 and P9V_BUILTIN_VEC_VINSERT4B.
477 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
478 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
479 * config/rs6000/vsx.md:
480 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
483 2018-02-16 Carl Love <cel@us.ibm.com>
485 * config/rs6000/altivec.h: Add builtin names vec_extract4b
487 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
489 * config/rs6000/rs6000-c.c: Add the definitions for
490 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
491 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
492 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
493 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
494 definition for insert4b and define insn *insert3b_internal.
495 * doc/extend.texi: Add documentation for vec_extract4b.
497 2018-02-16 Nathan Sidwell <nathan@acm.org>
499 * doc/extend.texi (Backwards Compatibility): Mention friend
500 injection. Note for-scope is deprecated.
501 * doc/invoke.texi (-ffriend-injection): Deprecate.
503 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
505 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
506 that moved to I2, also allow destinations that are a paradoxical
507 subreg (instead of a normal reg).
509 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
512 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
515 2018-02-16 Richard Biener <rguenther@suse.de>
517 PR tree-optimization/84037
518 PR tree-optimization/84016
520 * config/i386/i386.c (ix86_builtin_vectorization_cost):
521 Adjust vec_construct for the fact we need additional higher latency
522 128bit inserts for AVX256 and AVX512 vector builds.
523 (ix86_add_stmt_cost): Scale vector construction cost for
526 2018-02-16 Richard Biener <rguenther@suse.de>
528 PR tree-optimization/84417
529 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
530 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
531 (non_rewritable_lvalue_p): Likewise, use poly-ints.
533 2018-02-16 Martin Liska <mliska@suse.cz>
536 * internal-fn.def (ASAN_CHECK): Set proper flags.
537 (ASAN_MARK): Likewise.
539 2018-02-16 Julia Koval <julia.koval@intel.com>
541 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
544 2018-02-16 Jakub Jelinek <jakub@redhat.com>
547 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
548 Use ++iter rather than iter++ for std::list iterators.
549 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
550 defer deleting them until all nodes in the forest are processed. Do
551 free even leaf nodes. Change to_process into auto_vec.
554 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
555 * vec.h (vec_default_construct): Use memset instead of placement new
556 if BROKEN_VALUE_INITIALIZATION is defined.
557 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
558 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
561 PR rtl-optimization/83723
562 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
563 * lra.c (lra_substitute_pseudo): Likewise. If true, use
564 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
566 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
568 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
570 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
572 PR rtl-optimization/81443
573 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
574 from inner REGs to paradoxical SUBREGs.
576 2018-02-16 Richard Biener <rguenther@suse.de>
578 PR tree-optimization/84399
579 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
580 For operands we can analyze at their definition make sure we can
581 analyze them at each use as well.
583 2018-02-16 Richard Biener <rguenther@suse.de>
585 PR tree-optimization/84190
586 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
587 volatile accesses if the decl isn't volatile.
589 2018-02-15 Jason Merrill <jason@redhat.com>
591 PR c++/84314 - ICE with templates and fastcall attribute.
592 * attribs.c (build_type_attribute_qual_variant): Don't clobber
593 TYPE_CANONICAL on an existing type.
595 2018-02-15 Jakub Jelinek <jakub@redhat.com>
597 PR tree-optimization/84383
598 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
599 dstoff nor call operand_equal_p if dstbase is NULL.
601 PR tree-optimization/84334
602 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
603 also a CONSTANT_CLASS_P, punt.
605 2018-02-14 Jim Wilson <jimw@sifive.com>
607 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
608 first SMALL_OPERAND check. New local min_second_step. Move assert
609 to where locals are set. Add TARGET_RVC support.
610 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
612 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
614 * doc/invoke.texi: Correct -Wformat-overflow code sample.
616 2018-02-14 Martin Sebor <msebor@redhat.com>
618 PR tree-optimization/83698
619 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
620 arrays constrain the offset range to their bounds.
621 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
622 (builtin_access::overlap): Avoid setting the size of overlap if it's
624 (maybe_diag_overlap): Also consider arrays when deciding what values
625 of offsets to include in diagnostics.
627 2018-02-14 Martin Sebor <msebor@redhat.com>
630 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
631 that correspond to the kind of a declaration.
633 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
636 * config/pa/pa.md: Load address of PIC label using the linkage table
637 if the label is nonlocal.
639 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
641 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
642 warning message if user requests -maltivec=be.
643 * doc/invoke.texi: Document deprecation of -maltivec=be.
645 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
648 * config/rs6000/rs6000-c.c: Update definitions for
649 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
650 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
652 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
655 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
656 add _get_ssp intrinsics. Remove argument from
657 __builtin_ia32_rdssp[d|q].
658 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
659 * config/i386/i386-builtin.def: Remove argument from
660 __builtin_ia32_rdssp[d|q].
661 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
662 ix86_expand_special_args_builtin for _rdssp[d|q].
663 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
664 Clear register before usage.
665 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
666 Add documentation for new _get_ssp and _inc_ssp intrinsics.
668 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
670 PR tree-optimization/84357
671 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
672 operand 1 of an ARRAY_REF too.
674 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
677 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
678 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
680 (set_of_reg): New struct.
681 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
682 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
683 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
685 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
686 Split into bitclr, bitset, bitinvert patterns if appropriate.
687 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
688 use rx_fuse_in_memory_bitop.
689 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
690 to named insn, correct maximum insn length.
692 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
695 * machmode.def: Define a complex mode for PARTIAL_INT.
696 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
698 * doc/rtl.texi: Document CSPImode.
699 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
701 (msp430_hard_regno_nregs_with_padding): Likewise.
703 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
706 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
708 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
710 PR rtl-optimization/84169
711 * combine.c (try_combine): New variable split_i2i3. Set it to true if
712 we generated a parallel as new i3 and we split that to new i2 and i3
713 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
714 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
715 those to i2, not i1. Partially rewrite this scan code.
717 2018-02-13 Jakub Jelinek <jakub@redhat.com>
720 * stor-layout.c (place_field): For variable length fields, adjust
721 offset_align afterwards not just based on the field's alignment,
722 but also on the size.
725 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
726 of exps and logs in the use_exp2 case.
728 2018-02-13 Jeff Law <law@redhat.com>
730 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
733 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
736 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
740 * common.opt (gas-loc-support, gas-locview-support): New.
741 (ginline-points, ginternal-reset-location-views): New.
742 * doc/invoke.texi: Document them. Use @itemx where intended.
743 (gvariable-location-views): Adjust.
744 * target.def (reset_location_view): New.
745 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
746 (TARGET_RESET_LOCATION_VIEW): New.
747 * doc/tm.texi: Rebuilt.
748 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
749 (dwarf2out_default_as_locview_support): New.
750 (output_asm_line_debug_info): Use option variables.
751 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
752 (output_loc_list): Likewise.
753 (add_high_low_attributes): Check option variables.
754 Don't output entry view attribute in strict mode.
755 (gen_inlined_subroutine_die): Check option variables.
756 (dwarf2out_inline_entry): Likewise.
757 (init_sections_and_labels): Likewise.
758 (dwarf2out_early_finish): Likewise.
759 (maybe_reset_location_view): New, from...
760 (dwarf2out_var_location): ... here. Call it.
761 * debug.h (dwarf2out_default_as_loc_support): Declare.
762 (dwarf2out_default_as_locview_support): Declare.
763 * hooks.c (hook_int_rtx_insn_0): New.
764 * hooks.h (hook_int_rtx_insn_0): Declare.
765 * toplev.c (process_options): Take -gas-loc-support and
766 -gas-locview-support from dwarf2out. Enable
767 -gvariable-location-views by default only with locview
768 assembler support. Enable -ginternal-reset-location-views by
769 default only if the target defines the corresponding hook.
770 Enable -ginline-points by default if location views are
771 enabled; force it disabled if statement frontiers are
773 * tree-inline.c (expand_call_inline): Check option variables.
774 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
776 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
778 PR tree-optimization/84321
779 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
780 handling. Also check whether the anti-range contains any values
781 that satisfy the mask; switch to a VR_RANGE if not.
783 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
786 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
788 2018-02-13 Martin Jambor <mjambor@suse.cz>
791 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
792 of call statements, also set location of a load to a temporary.
794 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
796 * config/rl78/rl78.c (add_vector_labels): New function.
797 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
798 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
799 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
800 which checks that no arguments are passed.
801 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
802 * doc/extend.texi: Documentation for the new attribute.
804 2018-02-13 Andreas Schwab <schwab@suse.de>
806 * config/riscv/linux.h (CPP_SPEC): Define.
808 2018-02-13 Jakub Jelinek <jakub@redhat.com>
811 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
812 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
813 OPTION_MASK_ISA_AES as first argument to def_builtin_const
814 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
815 instead of OPTION_MASK_ISA_PCLMUL as first argument to
816 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
817 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
818 temporarily for AES and PCLMUL builtins.
820 PR tree-optimization/84339
821 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
822 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
826 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
827 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
828 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
830 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
832 * omp-simd-clone.h: New file.
833 * omp-simd-clone.c: Include omp-simd-clone.h.
834 (expand_simd_clones): No longer static.
835 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
836 cgraph.h and omp-simd-clone.h.
837 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
838 (vect_recog_widen_shift_pattern): Formatting fix.
839 (vect_pattern_recog_1): Don't check optab for calls.
842 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
843 operands[2] into a REG before using gen_lowpart on it.
845 2018-02-12 Jeff Law <law@redhat.com>
848 * config/sh/sh.c (find_barrier): Consider a sibling call
851 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
852 successfully back substituting a reg.
854 2018-02-12 Richard Biener <rguenther@suse.de>
856 PR tree-optimization/84037
857 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
858 parameter, move visited init to caller.
859 (vect_slp_analyze_operations): Separate cost from validity
860 check, initialize visited once for all instances.
861 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
863 * tree-vect-stmts.c (vect_model_simple_cost): Make early
865 (vect_model_promotion_demotion_cost): Likewise.
866 (vectorizable_bswap): Guard cost modeling with !slp_node
867 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
869 (vectorizable_call): Likewise.
870 (vectorizable_conversion): Likewise.
871 (vectorizable_assignment): Likewise.
872 (vectorizable_shift): Likewise.
873 (vectorizable_operation): Likewise.
874 (vectorizable_store): Likewise.
875 (vectorizable_load): Likewise.
876 (vectorizable_condition): Likewise.
877 (vectorizable_comparison): Likewise.
879 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
882 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
883 (ASAN_MARK): Fix fnspec to account for return value, change pointer
884 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
886 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
889 * params.def (inline-min-speedup): Increase from 8 to 15.
890 (max-inline-insns-auto): Decrease from 40 to 30.
891 * ipa-split.c (consider_split): Add some buffer for function to
892 be considered inlining candidate.
893 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
896 2018-02-12 Richard Biener <rguenther@suse.de>
898 PR tree-optimization/84037
899 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
900 matched stmts if we cannot swap the non-matched ones.
902 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
904 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
905 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
906 _mm_maskz_scalef_round_ss): New intrinsics.
907 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
908 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
909 __builtin_ia32_scalefss_round): Remove.
910 (__builtin_ia32_scalefsd_mask_round,
911 __builtin_ia32_scalefss_mask_round): New intrinsics.
912 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
913 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
914 ((match_operand:VF_128 2 "<round_nimm_predicate>"
915 "<round_constraint>")): Changed to ...
916 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
917 "<round_scalar_constraint>")): ... this.
918 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
919 %0, %1, %2<round_op3>}"): Changed to ...
920 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
921 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
922 %2<round_scalar_mask_op3>}"): ... this.
923 * config/i386/subst.md (round_scalar_nimm_predicate): New.
925 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
927 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
928 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
929 (_mm_maskz_sqrt_round_ss): New intrinsics.
930 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
931 (__builtin_ia32_sqrtsd_mask_round)
932 (__builtin_ia32_sqrtss_mask_round): New builtins.
933 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
934 (__builtin_ia32_sqrtss_round): Remove.
935 (__builtin_ia32_sqrtsd_mask_round)
936 (__builtin_ia32_sqrtss_mask_round): New builtins.
937 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
938 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
939 ((match_operand:VF_128 1 "vector_operand"
940 "xBm,<round_constraint>")): Changed to ...
941 ((match_operand:VF_128 1 "vector_operand"
942 "xBm,<round_scalar_constraint>")): ... this.
943 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
944 %0, %2, %<iptr>1<round_op3>}): Changed to ...
945 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
946 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
947 %<iptr>1<round_scalar_mask_op3>}): ... this.
948 ((set_attr "prefix" "<round_prefix>")): Changed to ...
949 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
951 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
954 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
955 Cast vec_cmpeq result to correct type.
956 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
957 Cast vec_cmpgt result to correct type.
959 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
961 * final.c (final_scan_insn_1): Renamed from...
962 (final_scan_insn): ... this. New wrapper, to recover
963 seen from the outermost call in recursive ones.
964 * config/sparc/sparc.c (output_return): Drop seen from call.
965 (output_sibcall): Likewise.
966 * config/visium/visium.c (output_branch): Likewise.
968 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
970 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
973 2018-02-10 Alan Modra <amodra@gmail.com>
976 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
977 Specify LR as an input.
979 2018-02-10 Jakub Jelinek <jakub@redhat.com>
982 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
983 remove_member_access_dummy_vars): New functions.
984 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
985 lower_omp_1, execute_lower_omp): Use them.
987 PR rtl-optimization/84308
988 * shrink-wrap.c (spread_components): Release todo vector.
990 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
992 PR rtl-optimization/57193
993 * ira-color.c (struct allocno_color_data): Add member
994 conflict_allocno_hard_prefs.
995 (update_conflict_allocno_hard_prefs): New.
996 (bucket_allocno_compare_func): Add a preference based on
997 conflict_allocno_hard_prefs.
998 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
999 (color_allocnos): Remove a dead code. Initiate
1000 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1002 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1005 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1006 constraint from =wa to wa. Avoid a subreg on the output operand,
1007 instead use a pseudo and subreg it in a move.
1008 (p9_xxbrd_<mode>): Changed to ...
1009 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1010 (p9_xxbrd_v2df): New expander.
1011 (p9_xxbrw_<mode>): Changed to ...
1012 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1013 (p9_xxbrw_v4sf): New expander.
1015 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1017 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1019 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1022 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1023 multiply in 32-bit mode.
1024 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1025 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1028 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1030 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1031 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1032 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1033 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1035 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1038 * dwarf2out.c (is_trivial_indirect_ref): New function.
1039 (dwarf2out_late_global_decl): Do not generate a location
1040 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1041 and that are not defined in the current unit.
1043 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1045 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1046 instead of a libcall for UNORDERED.
1048 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1051 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1052 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1054 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1057 * config/s390/s390.c (s390_set_current_function): Invoke
1058 s390_indirect_branch_settings also if fndecl didn't change.
1060 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1062 * config/rs6000/rs6000.md (blockage): Set length to zero.
1064 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1066 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1068 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1071 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1072 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1076 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1077 PARALLEL incoming that failed vt_get_decl_and_offset check.
1080 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1081 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1083 (get_variable_section): For decls in named .bss* sections pass true as
1084 second argument to bss_initializer_p.
1086 2018-02-09 Marek Polacek <polacek@redhat.com>
1087 Jakub Jelinek <jakub@redhat.com>
1090 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1091 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1092 Sync some changes from cxx_fold_indirect_ref.
1094 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1096 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1098 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1099 (BLOCK_INLINE_ENTRY_LABEL): New.
1100 (dwarf2out_var_location): Disregard inline entry markers.
1101 (inline_entry_data): New struct.
1102 (inline_entry_data_hasher): New hashtable type.
1103 (inline_entry_data_hasher::hash): New.
1104 (inline_entry_data_hasher::equal): New.
1105 (inline_entry_data_table): New variable.
1106 (add_high_low_attributes): Add DW_AT_entry_pc and
1107 DW_AT_GNU_entry_view attributes if a pending entry is found
1108 in inline_entry_data_table. Add old entry_pc attribute only
1109 if debug nonbinding markers are disabled.
1110 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1111 markers are enabled.
1112 (block_within_block_p, dwarf2out_inline_entry): New.
1113 (dwarf2out_finish): Check that no entries remained in
1114 inline_entry_data_table.
1115 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1116 (final_scan_insn, notice_source_line): Likewise.
1117 (rest_of_clean_state): Skip inline entry markers.
1118 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1120 * gimple.c (gimple_build_debug_inline_entry): New.
1121 * gimple.h (enum gimple_debug_subcode): Add
1122 GIMPLE_DEBUG_INLINE_ENTRY.
1123 (gimple_build_debug_inline_entry): Declare.
1124 (gimple_debug_inline_entry_p): New.
1125 (gimple_debug_nonbind_marker_p): Adjust.
1126 * insn-notes.def (INLINE_ENTRY): New.
1127 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1128 inline entry marker notes.
1129 (print_insn): Likewise.
1130 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1131 (INSN_DEBUG_MARKER_KIND): Likewise.
1132 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1133 * tree-inline.c (expand_call_inline): Build and insert
1134 debug_inline_entry stmt.
1135 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1136 inline entry blocks early, if nonbind markers are enabled.
1137 (dump_scope_block): Dump fragment info.
1138 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1139 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1140 (gimple_build_debug_inline_entry): New.
1141 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1142 Enable/disable inline entry points too.
1143 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1144 (DEBUG_INSN): Describe inline entry markers.
1146 * common.opt (gvariable-location-views): New.
1147 (gvariable-location-views=incompat5): New.
1148 * config.in: Rebuilt.
1149 * configure: Rebuilt.
1150 * configure.ac: Test assembler for view support.
1151 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1152 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1153 * dwarf2out.c (var_loc_view): New typedef.
1154 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1155 (dwarf2out_locviews_in_attribute): New.
1156 (dwarf2out_locviews_in_loclist): New.
1157 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1158 (enum dw_line_info_opcode): Add LI_adv_address.
1159 (struct dw_line_info_table): Add view.
1160 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1161 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1162 (zero_view_p): New variable.
1163 (ZERO_VIEW_P): New macro.
1164 (output_asm_line_debug_info): New.
1165 (struct var_loc_node): Add view.
1166 (add_AT_view_list, AT_loc_list): New.
1167 (add_var_loc_to_decl): Add view param. Test it against last.
1168 (new_loc_list): Add view params. Record them.
1169 (AT_loc_list_ptr): Handle loc and view lists.
1170 (view_list_to_loc_list_val_node): New.
1171 (print_dw_val): Handle dw_val_class_view_list.
1172 (size_of_die): Likewise.
1173 (value_format): Likewise.
1174 (loc_list_has_views): New.
1175 (gen_llsym): Set vl_symbol too.
1176 (maybe_gen_llsym, skip_loc_list_entry): New.
1177 (dwarf2out_maybe_output_loclist_view_pair): New.
1178 (output_loc_list): Output view list or entries too.
1179 (output_view_list_offset): New.
1180 (output_die): Handle dw_val_class_view_list.
1181 (output_dwarf_version): New.
1182 (output_compilation_unit_header): Use it.
1183 (output_skeleton_debug_sections): Likewise.
1184 (output_rnglists, output_line_info): Likewise.
1185 (output_pubnames, output_aranges): Update version comments.
1186 (output_one_line_info_table): Output view numbers in asm comments.
1187 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1188 Call maybe_gen_llsym.
1189 (loc_list_from_tree_1): Adjust.
1190 (add_AT_location_description): Create view list attribute if
1191 needed, check it's absent otherwise.
1192 (convert_cfa_to_fb_loc_list): Adjust.
1193 (maybe_emit_file): Call output_asm_line_debug_info for test.
1194 (dwarf2out_var_location): Reset views as needed. Precompute
1195 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1196 attribute. Set view.
1197 (new_line_info_table): Reset next view.
1198 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1199 (dwarf2out_source_line): Likewise. Output view resets and labels to
1200 the assembler, or select appropriate line info opcodes.
1201 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1202 (optimize_string_length): Catch it. Adjust.
1203 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1204 dw_val_class_view_list, and remove it if no longer needed.
1205 (hash_loc_list): Hash view numbers.
1206 (loc_list_hasher::equal): Compare them.
1207 (optimize_location_lists): Check whether a view list symbol is
1208 needed, and whether the locview attribute is present, and
1209 whether they match. Remove the locview attribute if no longer
1211 (index_location_lists): Call skip_loc_list_entry for test.
1212 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1213 Use output_dwarf_version.
1214 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1215 (struct dw_val_node): Add val_view_list.
1216 * final.c (SEEN_NEXT_VIEW): New.
1217 (set_next_view_needed): New.
1218 (clear_next_view_needed): New.
1219 (maybe_output_next_view): New.
1220 (final_start_function): Rename to...
1221 (final_start_function_1): ... this. Take pointer to FIRST,
1222 add SEEN parameter. Emit param bindings in the initial view.
1223 (final_start_function): Reintroduce SEEN-less interface.
1224 (final): Rename to...
1225 (final_1): ... this. Take SEEN parameter. Output final pending
1226 next view at the end.
1227 (final): Reintroduce seen-less interface.
1228 (final_scan_insn): Output pending next view before switching
1229 sections or ending a block. Mark the next view as needed when
1230 outputting variable locations. Notify debug backend of section
1231 changes, and of location view changes.
1232 (rest_of_handle_final): Adjust.
1233 * toplev.c (process_options): Autodetect value for debug variable
1234 location views option. Warn on incompat5 without -gdwarf-5.
1235 * doc/invoke.texi (gvariable-location-views): New.
1236 (gvariable-location-views=incompat5): New.
1237 (gno-variable-location-views): New.
1239 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1241 PR tree-optimization/84136
1242 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1243 that the result of find_edge is non-NULL.
1245 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1248 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1249 storing integer register in SImode. Fix cost of 256 and 512
1250 byte aligned SSE register store.
1252 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1254 * config/i386/i386.c (ix86_multiplication_cost): Fix
1255 multiplication cost for TARGET_AVX512DQ.
1257 2018-02-08 Marek Polacek <polacek@redhat.com>
1259 PR tree-optimization/84238
1260 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1263 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1265 PR tree-optimization/84265
1266 * tree-vect-stmts.c (vectorizable_store): Don't treat
1267 VMAT_CONTIGUOUS accesses as grouped.
1268 (vectorizable_load): Likewise.
1270 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1272 PR tree-optimization/81635
1273 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1274 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1275 (test_round_for_mask): New functions.
1276 (wide_int_cc_tests): Call test_round_for_mask.
1277 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1278 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1279 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1280 range returned by get_range_info.
1282 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1285 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1286 * symtab.c: Include builtins.h
1287 (symtab_node::output_to_lto_symbol_table_p): Move here
1288 from lto-streamer-out.c:output_symbol_p.
1289 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1290 (output_symbol_p): Move all logic to symtab.c
1291 (produce_symtab): Update.
1293 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1295 * config/s390/s390-opts.h (enum indirect_branch): Define.
1296 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1297 (s390_indirect_branch_via_thunk)
1298 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1299 (enum s390_indirect_branch_type): Define.
1300 * config/s390/s390.c (struct s390_frame_layout, struct
1301 machine_function): Remove.
1302 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1303 (indirect_branch_table_label_no, indirect_branch_table_name):
1305 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1306 (enum s390_indirect_branch_option): Define.
1307 (s390_return_addr_from_memory): New function.
1308 (s390_handle_string_attribute): New function.
1309 (s390_attribute_table): Add new attribute handler.
1310 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1311 (s390_indirect_branch_via_thunk): New function.
1312 (s390_indirect_branch_via_inline_thunk): New function.
1313 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1314 sibling call optimization for non z10 compiles.
1315 (s390_emit_call): Force indirect branch target to be a single
1316 register. Add r1 clobber for non-z10 compiles.
1317 (s390_emit_epilogue): Emit return jump via return_use expander.
1318 (s390_reorg): Handle JUMP_INSNs as execute targets.
1319 (s390_option_override_internal): Perform validity checks for the
1320 new command line options.
1321 (s390_indirect_branch_attrvalue): New function.
1322 (s390_indirect_branch_settings): New function.
1323 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1324 (s390_output_indirect_thunk_function): New function.
1325 (s390_code_end): Implement target hook.
1326 (s390_case_values_threshold): Implement target hook.
1327 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1329 * config/s390/s390.h (struct s390_frame_layout)
1330 (struct machine_function): Move here from s390.c.
1331 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1332 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1333 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1334 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1335 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1336 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1337 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1338 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1339 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1340 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1341 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1342 (mnemonic attribute): Add values which aren't recognized
1344 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1345 pattern for branch conversion. Fix mnemonic attribute.
1346 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1347 indirect branch via thunk if requested.
1348 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1349 ("*indirect_jump"): Disable for branch conversion using out of
1351 ("indirect_jump_via_thunk<mode>_z10")
1352 ("indirect_jump_via_thunk<mode>")
1353 ("indirect_jump_via_inlinethunk<mode>_z10")
1354 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1355 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1356 ("casesi_jump_via_inlinethunk<mode>_z10")
1357 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1358 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1359 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1360 ("*indirect2_jump"): Disable for branch conversion.
1361 ("casesi_jump"): Turn into expander and expand patterns for branch
1363 ("return_use"): New expander.
1364 ("*return"): Emit return via thunk and rename it to ...
1365 ("*return<mode>"): ... this one.
1366 * config/s390/s390.opt: Add new options and and enum for the
1369 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1371 * lra-constraints.c (match_reload): Unconditionally use
1372 gen_lowpart_SUBREG, rather than selecting between that
1373 and equivalent gen_rtx_SUBREG code.
1375 2018-02-08 Richard Biener <rguenther@suse.de>
1377 PR tree-optimization/84233
1378 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1379 changed flag instead of boguously re-using phi_inserted.
1381 2018-02-08 Martin Jambor <mjambor@suse.cz>
1383 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1384 static local variables.
1386 2018-02-08 Richard Biener <rguenther@suse.de>
1388 PR tree-optimization/84278
1389 * tree-vect-stmts.c (vectorizable_store): When looking for
1390 smaller vector types to perform grouped strided loads/stores
1391 make sure the mode is supported by the target.
1392 (vectorizable_load): Likewise.
1394 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1396 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1397 Increase LDP/STP opportunities by adding adjacent callee-saves.
1399 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1401 PR rtl-optimization/84068
1402 PR rtl-optimization/83459
1403 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1405 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1407 PR tree-optimization/84224
1408 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1409 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1412 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1415 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1416 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1417 count, remove one USE.
1419 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1421 * doc/install.texi (Configuration): Document the
1422 --with-long-double-format={ibm,ieee} PowerPC configuration
1426 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1427 Convert from define_expand to be define_insn_and_split. Rework
1428 float/double/_Float128 conversions to QI/HI/SImode to work with
1429 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1430 conversions to QI/HImode types did a store and then a load to
1431 truncate the value. For conversions to VSX registers, don't split
1432 the insn, instead emit the code directly. Use the code iterator
1433 any_fix to combine signed and unsigned conversions.
1434 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1435 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1436 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1437 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1438 (fix_<mode>di2_hw): Likewise.
1439 (fixuns_<mode>di2_hw): Likewise.
1440 (fix_<mode>si2_hw): Likewise.
1441 (fixuns_<mode>si2_hw): Likewise.
1442 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1443 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1444 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1445 fix<uns>_trunc<SFDF:mode>si2_p8.
1446 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1448 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1449 (fix<uns>_<mode>_mem): Likewise.
1450 (fctiw<u>z_<mode>_mem): Likewise.
1451 (fix<uns>_<mode>_mem): Likewise.
1452 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1453 the register allocator from doing a direct move to the GPRs to do
1454 a store, and instead use the ISA 3.0 store byte/half-word from
1455 vector register instruction. For IEEE 128-bit floating point,
1456 also optimize stores of 32-bit ints.
1457 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1459 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1461 * genextract.c (push_pathstr_operand): New function to support
1463 (walk_rtx): Call push_pathstr_operand.
1464 (print_path): Support [a-zA-Z].
1466 2018-02-07 Richard Biener <rguenther@suse.de>
1468 PR tree-optimization/84037
1469 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1470 (cse_and_gimplify_to_preheader): Declare.
1471 (vect_get_place_in_interleaving_chain): Likewise.
1472 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1474 (_loop_vec_info::~_loop_vec_info): Delete it.
1475 (cse_and_gimplify_to_preheader): New function.
1476 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1477 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1478 (vectorizable_load): Likewise. For grouped stores always base
1479 the IV on the first element.
1480 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1481 condition before gimplifying.
1483 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1485 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1486 *DIV_EXPR and *MOD_EXPR.
1488 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1491 * config/i386/i386.c (ix86_option_override_internal): Mask out
1492 the CF_SET bit when checking -fcf-protection.
1494 2018-02-07 Tom de Vries <tom@codesourcery.com>
1497 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1500 2018-02-07 Richard Biener <rguenther@suse.de>
1502 PR tree-optimization/84204
1503 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1506 PR tree-optimization/84205
1507 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1508 special-case isl_ast_op_zdiv_r.
1510 PR tree-optimization/84223
1511 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1512 Only add conditions from within the region.
1513 (gather_bbs::after_dom_children): Adjust.
1515 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1518 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1519 * config/avr/avr.md: Only post-reload split REG-REG moves if
1520 either register is REGERAL_REG_P.
1522 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1524 PR tree-optimization/84235
1525 * tree-ssa-scopedtables.c
1526 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1527 if the subtraction is performed in floating point type where NaNs are
1528 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1529 build 1. Formatting fix.
1531 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1534 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1535 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1536 and skip it regardless of bb boundaries. Use CALL_P macro,
1537 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1538 unnecessarily, formatting fix.
1540 2018-02-06 Michael Collison <michael.collison@arm.com>
1542 * config/arm/thumb2.md:
1543 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1544 (*thumb_mov_notscc): Ditto.
1546 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1549 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1550 unsigned_fix, not "s".
1552 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1554 * configure.ac (gcc_fn_eh_frame_ro): New function.
1555 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1556 correct .eh_frame permissions.
1557 * configure: Regenerate.
1559 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1561 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1564 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1566 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1567 Display warning message for -mno-speculate-indirect-jumps.
1569 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1571 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1573 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1575 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1577 PR tree-optimization/84225
1578 * tree-eh.c (find_trapping_overflow): Only call
1579 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1581 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1584 * config/i386/i386.c: Reimplement the check of possible options
1585 -mibt/-mshstk conbination. Change error messages.
1586 * doc/invoke.texi: Fix a typo: remove extra '='.
1588 2018-02-06 Marek Polacek <polacek@redhat.com>
1590 PR tree-optimization/84228
1591 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1593 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1596 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1597 emitted arch directives.
1598 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1599 __ARM_FEATURE_COPROC before changing architectures.
1601 2018-02-06 Richard Biener <rguenther@suse.de>
1603 * config/i386/i386.c (print_reg): Fix typo.
1604 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1606 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1608 * configure: Regenerate.
1610 2018-02-05 Martin Sebor <msebor@redhat.com>
1612 PR tree-optimization/83369
1613 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1616 2018-02-05 Martin Liska <mliska@suse.cz>
1618 * doc/invoke.texi: Cherry-pick upstream r323995.
1620 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1622 * ira.c (ira_init_register_move_cost): Adjust comment.
1624 2018-02-05 Martin Liska <mliska@suse.cz>
1626 PR gcov-profile/84137
1627 * doc/gcov.texi: Fix typo in documentation.
1629 2018-02-05 Martin Liska <mliska@suse.cz>
1631 PR gcov-profile/83879
1632 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1633 using dlopen functionality.
1635 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1637 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1638 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1639 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1640 _mm_maskz_range_round_ss): New intrinsics.
1641 (__builtin_ia32_rangesd128_round)
1642 (__builtin_ia32_rangess128_round): Remove.
1643 (__builtin_ia32_rangesd128_mask_round,
1644 __builtin_ia32_rangess128_mask_round): New builtins.
1645 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1646 __builtin_ia32_rangess128_round): Remove.
1647 (__builtin_ia32_rangesd128_mask_round,
1648 __builtin_ia32_rangess128_mask_round): New builtins.
1649 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1650 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1651 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1652 "<round_saeonly_constraint>")): Changed to ...
1653 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1654 "<round_saeonly_scalar_constraint>")): ... this.
1655 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1656 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1657 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1658 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1659 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1661 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1663 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1665 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1666 Remove all values except native, 8540 and 8548.
1668 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1670 * config/i386/i386.c (ix86_output_function_return): Pass
1671 INVALID_REGNUM, instead of -1, as invalid register number to
1672 indirect_thunk_name and output_indirect_thunk.
1674 2018-02-02 Julia Koval <julia.koval@intel.com>
1676 * config.gcc: Add -march=icelake.
1677 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1678 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1679 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1680 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1681 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1682 (processor_target_table): Add icelake.
1683 (ix86_option_override_internal): Handle new PTAs.
1684 (get_builtin_code_for_version): Handle icelake.
1685 (M_INTEL_COREI7_ICELAKE): New.
1686 (fold_builtin_cpu): Handle icelake.
1687 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1688 * doc/invoke.texi: Add -march=icelake.
1690 2018-02-02 Julia Koval <julia.koval@intel.com>
1692 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1693 to wide_int_bitmask.
1694 * wide-int-bitmask.h: New.
1696 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1699 * config/i386/i386.md: Replace Pmode with word_mode in
1700 builtin_setjmp_setup and builtin_longjmp to support x32.
1702 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1706 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1708 (rs6000_supported_cpu_names): New static variable.
1709 (linux_cpu_translation_table): Likewise.
1710 (elf_platform) <cpu>: Define new static variable and use it.
1711 Translate kernel AT_PLATFORM name to canonical name if needed.
1712 Error if platform name is unknown.
1714 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1717 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1719 2018-02-01 Jeff Law <law@redhat.com>
1722 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1723 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1724 the scratch if RELEASE_VIA_POP is false.
1725 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1726 If we have to save a temporary register, decrement SIZE appropriately.
1727 Pass new arguments to release_scratch_register_on_entry.
1728 (ix86_adjust_stack_and_probe): Likewise.
1729 (ix86_emit_probe_stack_range): Pass new arguments to
1730 release_scratch_register_on_entry.
1732 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1734 PR rtl-optimization/84157
1735 * combine.c (change_zero_ext): Use REG_P predicate in
1736 front of HARD_REGISTER_P predicate.
1738 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1740 * config/avr/avr.c (avr_option_override): Move disabling of
1741 -fdelete-null-pointer-checks to...
1742 * common/config/avr/avr-common.c (avr_option_optimization_table):
1745 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1747 PR tree-optimization/81635
1748 * tree-data-ref.c (split_constant_offset_1): For types that
1749 wrap on overflow, try to use range info to prove that wrapping
1752 2018-02-01 Renlin Li <renlin.li@arm.com>
1755 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1757 (aarch64_register_move_cost): Likewise.
1758 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1760 (REG_CLASS_NAMES): Likewise.
1761 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1762 TAILCALL_ADDR_REGS. Remove IP registers.
1763 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1765 2018-02-01 Richard Biener <rguenther@suse.de>
1767 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1768 for specifying RPO order and allow NULL for that.
1769 * domwalk.c (dom_walker::dom_walker): Likewise.
1770 (dom_walker::walk): Handle NULL RPO order.
1771 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1773 (rewrite_update_dom_walker): Likewise.
1774 (mark_def_dom_walker): Likewise.
1776 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1778 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1779 (aarch64_maybe_expand_sve_subreg_move): Declare.
1780 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1781 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1783 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1784 that are semantically a reverse operation.
1785 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1786 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1787 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1789 (aarch64_can_change_mode_class): For big-endian, forbid changes
1790 between two SVE modes if they have different element sizes.
1792 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1794 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1795 the TImode handling for big-endian targets.
1797 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1799 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1800 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1802 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1803 Remove BSWAP handing for big-endian targets and use the form of
1804 LD1RQ appropariate for the mode.
1806 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1808 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1809 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1812 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1815 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1816 check for operands that need to go through aarch64_sve_reload_be.
1818 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1820 PR tree-optimization/81661
1821 PR tree-optimization/84117
1822 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1823 * tree-eh.c: Include gimplify.h.
1824 (find_trapping_overflow, replace_trapping_overflow,
1825 rewrite_to_non_trapping_overflow): New functions.
1826 * tree-vect-loop.c: Include tree-eh.h.
1827 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1828 * tree-data-ref.c: Include tree-eh.h.
1829 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1831 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1833 PR rtl-optimization/84123
1834 * combine.c (change_zero_ext): Check if hard register satisfies
1835 can_change_dest_mode before calling gen_lowpart_SUBREG.
1837 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1840 * ira.c (ira_init_register_move_cost): Remove assert.
1842 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1844 PR rtl-optimization/84071
1845 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1846 * doc/tm.texi: Regenerate.
1848 2018-01-31 Richard Biener <rguenther@suse.de>
1850 PR tree-optimization/84132
1851 * tree-data-ref.c (analyze_miv_subscript): Properly
1852 check whether evolution_function_is_affine_multivariate_p
1853 before calling gcd_of_steps_may_divide_p.
1855 2018-01-31 Julia Koval <julia.koval@intel.com>
1858 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1859 * config/i386/i386.md (rdpid_rex64) New.
1860 (rdpid): Make 32bit only.
1862 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1865 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1866 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1868 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1871 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1873 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1875 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1877 PR rtl-optimization/84071
1878 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1879 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1881 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1883 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1884 (arc_attribute_table): Add 'aux' attribute.
1885 (arc_in_small_data_p): Consider aux like variables.
1886 (arc_is_aux_reg_p): New function.
1887 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1888 (arc_get_aux_arg): New function.
1889 (prepare_move_operands): Handle aux-register access.
1890 (arc_handle_aux_attribute): New function.
1891 * doc/extend.texi (ARC Variable attributes): Add subsection.
1893 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1895 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1896 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1897 (arc_attribute_table): Add 'uncached' attribute.
1898 (arc_print_operand): Print '.di' flag for uncached memory
1900 (arc_in_small_data_p): Do not consider for small data the uncached
1902 (arc_is_uncached_mem_p): New function.
1903 * config/arc/predicates.md (compact_store_memory_operand): Check
1904 for uncached memory accesses.
1905 (nonvol_nonimm_operand): Likewise.
1906 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1908 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1911 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1912 falign-loops=): Add Optimization flag.
1914 2018-01-30 Jeff Law <law@redhat.com>
1917 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1918 INT_REGISTERS_SAVED. Check it prior to calling
1919 get_scratch_register_on_entry.
1920 (ix86_adjust_stack_and_probe): Similarly.
1921 (ix86_emit_probe_stack_range): Similarly.
1922 (ix86_expand_prologue): Corresponding changes.
1924 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1927 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1928 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1930 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1933 * lra-constraints.c (curr_insn_transform): Process AND in the
1936 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1938 PR rtl-optimization/83986
1939 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1940 dependence against last_pending_memory_flush in addition to
1943 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1945 PR tree-optimization/81611
1946 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1949 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1952 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1955 2018-01-30 Richard Biener <rguenther@suse.de>
1956 Jakub Jelinek <jakub@redhat.com>
1958 PR tree-optimization/84111
1959 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1960 inner loops added during recursion, as they don't have up-to-date
1963 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1966 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1967 (can_inline_edge_by_limits_p): ... here.
1968 (can_early_inline_edge_p, check_callers,
1969 update_caller_keys, update_callee_keys, recursive_inlining,
1970 add_new_edges_to_heap, speculation_useful_p,
1971 inline_small_functions,
1972 inline_small_functions, flatten_function,
1973 inline_to_all_callers_1): Update.
1975 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1977 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1980 2018-01-30 Richard Biener <rguenther@suse.de>
1982 PR tree-optimization/83008
1983 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1984 invariant and constant vector uses in stmts when they need
1987 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1990 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1991 * configure: Regenerate.
1993 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1995 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1997 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1998 Use gen_rtx_REG rather than gen_lowpart.
2000 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2002 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2003 rather than 0 when creating partial subregs.
2005 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2007 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2010 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2013 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2014 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2015 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2016 flags. This restores the settings used before the 2017-07-24.
2017 Turning off pre increment/decrement/modify allows IVOPTS to
2018 optimize DF/SF loops where the index is an int.
2020 2018-01-29 Richard Biener <rguenther@suse.de>
2021 Kelvin Nilsen <kelvin@gcc.gnu.org>
2024 * tree-vect-stmts.c (vectorizable_call): Don't call
2025 targetm.vectorize_builtin_md_vectorized_function if callee is
2028 2018-01-22 Carl Love <cel@us.ibm.com>
2030 * doc/extend.tex: Fix typo in second arg in
2031 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2033 2018-01-29 Richard Biener <rguenther@suse.de>
2035 PR tree-optimization/84086
2036 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2037 (flush_ssaname_freelist): When SSA names were released reset
2038 the SCEV hash table.
2040 2018-01-29 Richard Biener <rguenther@suse.de>
2042 PR tree-optimization/84057
2043 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2044 removed paths when removing edges.
2046 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2048 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2049 -mfunction-return=@var{choice}.
2051 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2054 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2056 (layout::print_source_line): Likewise.
2057 (test_get_line_width_without_trailing_whitespace): Add test cases.
2059 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2062 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2065 2018-01-26 Jim Wilson <jimw@sifive.com>
2067 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2069 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2072 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2074 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2075 and CMP + SUB-immediate -> SUBS.
2077 2018-01-26 Martin Sebor <msebor@redhat.com>
2079 PR tree-optimization/83896
2080 * tree-ssa-strlen.c (get_string_len): Rename...
2081 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2082 Avoid assuming length is constant.
2083 (handle_char_store): Use HOST_WIDE_INT for string length.
2085 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2088 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2089 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2091 2018-01-26 Richard Biener <rguenther@suse.de>
2093 PR rtl-optimization/84003
2094 * dse.c (record_store): Only record redundant stores when
2095 the earlier store aliases at least all accesses the later one does.
2097 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2099 PR rtl-optimization/83985
2100 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2101 REG_CFA_RESTORE insns.
2102 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2103 REG_CFA_RESTORE insns here.
2106 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2107 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2109 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2111 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2112 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2113 (arc_init): Likewise.
2114 (arc_override_options): Likewise.
2115 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2117 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2119 * config/arc/arc.h (TARGET_DBNZ): Define.
2120 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2121 properly set the tune attribute.
2122 (dbnz): Use TARGET_DBNZ guard.
2123 * config/arc/arc.opt (mtune): Add core3 option.
2125 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2127 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2128 recognize new pic like addresses.
2129 (arc_delegitimize_address): Clean up.
2131 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2133 * config/arc/arc-arches.def: Option mrf16 valid for all
2135 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2136 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2137 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2138 * config/arc/arc-tables.opt: Regenerate.
2139 * config/arc/arc.c (arc_conditional_register_usage): Handle
2140 reduced register file case.
2141 (arc_file_start): Set must have build attributes.
2142 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2144 * config/arc/arc.opt (mrf16): Add new option.
2145 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2146 * config/arc/genmultilib.awk: Handle new mrf16 option.
2147 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2148 * config/arc/t-multilib: Regenerate.
2149 * doc/invoke.texi (ARC Options): Document mrf16 option.
2151 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2153 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2154 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2155 (arc_attribute_table): Add 'secure_call' attribute.
2156 (arc_print_operand): Print secure call operand.
2157 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2159 (arc_is_secure_call_p): New function. * config/arc/arc.md
2160 (call_i): Add support for sjli instruction.
2161 (call_value_i): Likewise.
2162 * config/arc/constraints.md (Csc): New constraint.
2164 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2165 John Eric Martin <John.Martin@emmicro-us.com>
2167 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2168 * config/arc/arc.c (_arc_jli_section): New struct.
2169 (arc_jli_section): New type.
2170 (rc_jli_sections): New static variable.
2171 (arc_handle_jli_attribute): New function.
2172 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2173 (arc_file_end): New function.
2174 (TARGET_ASM_FILE_END): Define.
2175 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2176 (arc_add_jli_section): New function.
2177 (jli_call_scan): Likewise.
2178 (arc_reorg): Call jli_call_scan.
2179 (arc_output_addsi): Remove 'S' from printing asm operand.
2180 (arc_is_jli_call_p): New function.
2181 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2183 (movhi_insn): Likewise.
2184 (movsi_insn): Likewise.
2185 (movsi_set_cc_insn): Likewise.
2186 (loadqi_update): Likewise.
2187 (load_zeroextendqisi_update): Likewise.
2188 (load_signextendqisi_update): Likewise.
2189 (loadhi_update): Likewise.
2190 (load_zeroextendhisi_update): Likewise.
2191 (load_signextendhisi_update): Likewise.
2192 (loadsi_update): Likewise.
2193 (loadsf_update): Likewise.
2194 (movsicc_insn): Likewise.
2195 (bset_insn): Likewise.
2196 (bxor_insn): Likewise.
2197 (bclr_insn): Likewise.
2198 (bmsk_insn): Likewise.
2199 (bicsi3_insn): Likewise.
2200 (cmpsi_cc_c_insn): Likewise.
2201 (movsi_ne): Likewise.
2202 (movsi_cond_exec): Likewise.
2203 (clrsbsi2): Likewise.
2214 (call_i): Remove 'S' asm letter, add jli instruction.
2215 (call_value_i): Likewise.
2216 * config/arc/arc.op (mjli-always): New option.
2217 * config/arc/constraints.md (Cji): New constraint.
2218 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2220 (subsf3_fpx): Likewise.
2221 (mulsf3_fpx): Likewise.
2222 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2224 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2226 * doc/invoke.texi (ARC): Document mjli-always option.
2228 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2230 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2231 avoid addition with 0 and use incw and decw where possible.
2233 2018-01-26 Richard Biener <rguenther@suse.de>
2235 PR tree-optimization/81082
2236 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2237 association if it requires casting to unsigned.
2238 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2239 from fold_plusminus_mult_expr to catch important cases late when
2240 range info is available.
2242 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2244 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2245 * configure.ac (hidden_linkonce): New test.
2246 * configure: Regenerate.
2247 * config.in: Regenerate.
2249 2018-01-26 Julia Koval <julia.koval@intel.com>
2251 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2252 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2253 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2254 _mm_mask_bitshuffle_epi64_mask): Fix type.
2255 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2256 USI_FTYPE_V4DI_V4DI_USI): Remove.
2257 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2258 __builtin_ia32_vpshufbitqmb256_mask,
2259 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2260 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2261 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2263 2018-01-26 Alan Modra <amodra@gmail.com>
2266 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2267 UNSPEC_VBPERMQ. Sort other unspecs.
2269 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2271 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2273 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2276 * predict.c (drop_profile): Do not push/pop cfun; update also
2278 (handle_missing_profiles): Fix logic looking for zero profiles.
2280 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2283 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2284 on functions with #pragma omp declare simd or functions with simd
2286 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2287 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2288 Remove trailing \n from warning_at calls.
2290 2018-01-25 Tom de Vries <tom@codesourcery.com>
2293 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2294 for neutered workers.
2296 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2299 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2300 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2302 2018-01-24 Jeff Law <law@redhat.com>
2305 * i386.c (get_probe_interval): Move to earlier point.
2306 (ix86_compute_frame_layout): If -fstack-clash-protection and
2307 the frame is larger than the probe interval, then use pushes
2308 to save registers rather than reg->mem moves.
2309 (ix86_expand_prologue): Remove conditional for int_registers_saved
2312 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2315 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2316 min/max for never referenced object.
2318 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2321 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2323 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2324 attributes from DECL_ATTRIBUTES (decl) without affecting
2325 DECL_ATTRIBUTES (current_function_decl).
2326 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2327 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2329 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2331 PR tree-optimization/83979
2332 * fold-const.c (fold_comparison): Use constant_boolean_node
2333 instead of boolean_{true,false}_node.
2335 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2337 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2340 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2342 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2343 Simplify the clause that sets the length attribute.
2344 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2345 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2346 clause that sets the length attribute.
2347 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2349 2018-01-24 Tom de Vries <tom@codesourcery.com>
2352 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2353 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2354 Add strict parameter.
2355 (prevent_branch_around_nothing): Insert dummy insn between branch to
2356 label and label with no ptx insn inbetween.
2357 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2359 2018-01-24 Tom de Vries <tom@codesourcery.com>
2362 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2363 for neutered threads in warp.
2364 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2366 2018-01-24 Richard Biener <rguenther@suse.de>
2368 PR tree-optimization/83176
2369 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2372 2018-01-24 Richard Biener <rguenther@suse.de>
2374 PR tree-optimization/82819
2375 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2376 code generating pluses that are no-ops in the target precision.
2378 2018-01-24 Richard Biener <rguenther@suse.de>
2381 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2383 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2385 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2386 to merge probabilities.
2387 * predict.c (probably_never_executed): Also mark as cold functions
2388 with global 0 profile and guessed local profile.
2389 * profile-count.c (profile_probability::combine_with_count): New
2391 * profile-count.h (profile_probability::operator*,
2392 profile_probability::operator*=, profile_probability::operator/,
2393 profile_probability::operator/=): Reduce precision to adjusted
2394 and set value to guessed on contradictory divisions.
2395 (profile_probability::combine_with_freq): Remove.
2396 (profile_probability::combine_wiht_count): Declare.
2397 (profile_count::force_nonzero):: Set to adjusted.
2398 (profile_count::probability_in):: Set quality to adjusted.
2399 * tree-ssa-tail-merge.c (replace_block_by): Use
2402 2018-01-23 Andrew Waterman <andrew@sifive.com>
2403 Jim Wilson <jimw@sifive.com>
2405 * config/riscv/riscv.c (riscv_stack_boundary): New.
2406 (riscv_option_override): Set riscv_stack_boundary. Handle
2407 riscv_preferred_stack_boundary_arg.
2408 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2409 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2410 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2411 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2412 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2413 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2415 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2418 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2419 of struct ix86_frame.
2420 (ix86_expand_epilogue): Likewise. Add a local variable for
2421 the reg_save_offset field in struct ix86_frame.
2423 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2425 PR tree-optimization/82604
2426 * tree-loop-distribution.c (enum partition_kind): New enum item
2427 PKIND_PARTIAL_MEMSET.
2428 (partition_builtin_p): Support above new enum item.
2429 (generate_code_for_partition): Ditto.
2430 (compute_access_range): Differentiate cases that equality can be
2431 proven at all loops, the innermost loops or no loops.
2432 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2433 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2434 (finalize_partitions, distribute_loop): Don't fuse partition of
2435 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2436 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2439 2018-01-23 Martin Liska <mliska@suse.cz>
2441 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2442 order to ignore the predictor.
2443 (PRED_POLYMORPHIC_CALL): Likewise.
2444 (PRED_RECURSIVE_CALL): Likewise.
2446 2018-01-23 Martin Liska <mliska@suse.cz>
2448 * tree-profile.c (tree_profiling): Print function header to
2449 aware reader which function we are working on.
2450 * value-prof.c (gimple_find_values_to_profile): Do not print
2451 not interesting value histograms.
2453 2018-01-23 Martin Liska <mliska@suse.cz>
2455 * profile-count.h (enum profile_quality): Add
2456 profile_uninitialized as the first value. Do not number values
2457 as they are zero based.
2458 (profile_count::verify): Update sanity check.
2459 (profile_probability::verify): Likewise.
2461 2018-01-23 Nathan Sidwell <nathan@acm.org>
2463 * doc/invoke.texi (ffor-scope): Deprecate.
2465 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2467 PR tree-optimization/83510
2468 * domwalk.c (set_all_edges_as_executable): New function.
2469 (dom_walker::dom_walker): Convert bool param
2470 "skip_unreachable_blocks" to enum reachability. Move setup of
2471 edge flags to set_all_edges_as_executable and only do it when
2472 reachability is REACHABLE_BLOCKS.
2473 * domwalk.h (enum dom_walker::reachability): New enum.
2474 (dom_walker::dom_walker): Convert bool param
2475 "skip_unreachable_blocks" to enum reachability.
2476 (set_all_edges_as_executable): New decl.
2477 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2478 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2480 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2481 but converting true to REACHABLE_BLOCKS.
2482 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2484 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2485 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2486 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2488 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2489 if check_all_array_refs will be called.
2491 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2493 * tree.c (selftest::test_location_wrappers): Add more test
2496 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2498 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2499 (selftest::test_bit_in_range): Likewise.
2501 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2504 * doc/sourcebuild.texi (vect_float): Say that the selector
2505 only describes the situation when -funsafe-math-optimizations is on.
2506 (vect_float_strict): Document.
2508 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2510 PR tree-optimization/83965
2511 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2512 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2513 instead of checking only for a reduction.
2514 (vect_recog_widen_sum_pattern): Likewise.
2516 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2518 * predict.c (probably_never_executed): Only use precise profile info.
2519 (compute_function_frequency): Skip after inlining hack since we now
2520 have quality checking.
2522 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2524 * profile-count.h (profile_probability::very_unlikely,
2525 profile_probability::unlikely, profile_probability::even): Set
2526 precision to guessed.
2528 2018-01-23 Richard Biener <rguenther@suse.de>
2530 PR tree-optimization/83963
2531 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2532 Properly terminate dominator walk when crossing the exit edge not
2533 when visiting its source block.
2535 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2538 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2539 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2541 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2543 PR tree-optimization/83957
2544 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2545 semicolon after for body surrounded by braces.
2547 PR tree-optimization/83081
2548 * profile-count.h (profile_probability::split): New method.
2549 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2550 Use profile_probability::split.
2551 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2552 when splitting a single conditional jump into 2.
2554 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2556 PR tree-optimization/69452
2557 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2560 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2562 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2563 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2564 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2566 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2568 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2569 * config/rl78/rl78.md (movdi): New define_expand.
2570 * config/rl78/rl78.c (rl78_split_movdi): New function.
2572 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2575 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2577 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2578 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2579 128-bit to produce an UNSPEC move to get the double word with the
2580 signbit and then a shift directly to do signbit.
2581 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2582 implementation with a new version that just does either a direct
2583 move or a regular move. Move memory interface to separate insns.
2584 Move insns so they are next to the expander.
2585 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2586 with signbit move. Split big and little endian case.
2587 (signbit<mode>2_dm_mem_le): Likewise.
2588 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2589 (signbit<mode>2_dm2): Likewise.
2591 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2593 * config/rl78/rl78.md (anddi3): New define_expand.
2595 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2597 * config/rl78/rl78.md (umindi3): New define_expand.
2599 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2601 * config/rl78/rl78.md (smindi3): New define_expand.
2603 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2605 * config/rl78/rl78.md (smaxdi3): New define_expand.
2607 2018-01-22 Carl Love <cel@us.ibm.com>
2609 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2610 LVX_V1TI): Add macro expansion.
2611 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2612 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2613 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2614 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2615 Change check to determine if the instruction is a byte reversing
2616 entry. Fix typo in comment.
2617 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2618 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2619 Add def_builtin calls for new builtins.
2620 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2621 Add define_insn expansion.
2623 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2625 * config/rl78/rl78.md (umaxdi3): New define_expand.
2627 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2629 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2630 for non-QImode registers.
2632 2018-01-22 Richard Biener <rguenther@suse.de>
2634 PR tree-optimization/83963
2635 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2636 including the loop exit block.
2637 (scop_detection::merge_sese): Likewise.
2638 (scop_detection::add_scop): Do it here instead.
2640 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2642 * doc/sourcebuild.texi (arm_softfloat): Document.
2644 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2647 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2648 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2649 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2651 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2652 David Edelsohn <dje.gcc@gmail.com>
2655 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2656 Change "crset eq" to "crset 2".
2657 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2658 (*call_indirect_aix<mode>_nospec): Likewise.
2659 (*call_value_indirect_aix<mode>_nospec): Likewise.
2660 (*call_indirect_elfv2<mode>_nospec): Likewise.
2661 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2662 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2663 change assembly output from . to $.
2664 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2665 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2666 (*tablejump<mode>_internal1_nospec): Likewise.
2668 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2671 * config/sh/sh_optimize_sett_clrt.cc:
2672 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2674 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2676 PR tree-optimization/83940
2677 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2678 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2679 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2680 use it to pass back the definition type.
2681 (vect_check_store_rhs): Likewise rhs_dt_out.
2682 (vect_build_gather_load_calls): Add a mask_dt argument and use
2683 it instead of a call to vect_is_simple_use.
2684 (vectorizable_store): Update calls to vect_check_load_store_mask
2685 and vect_check_store_rhs. Use the dt returned by the latter instead
2686 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2687 instead of calls to vect_is_simple_use. Pass the scalar rather
2688 than the vector operand to vect_is_simple_use when handling
2689 second and subsequent copies of an rhs value.
2690 (vectorizable_load): Update calls to vect_check_load_store_mask
2691 and vect_build_gather_load_calls. Use the cached mask_dt and
2692 gs_info.offset_dt instead of calls to vect_is_simple_use.
2694 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2697 * tree-emutls.c: Include gimplify.h.
2698 (lower_emutls_2): New function.
2699 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2700 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2701 it before further processing.
2704 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2705 UINTVAL (trueop1) instead of INTVAL (op1).
2707 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2711 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2712 INCOMING_FRAME_SP_OFFSET if not defined.
2713 (scan_trace): Add ENTRY argument. If true and
2714 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2715 emit a note to adjust the CFA offset.
2716 (create_cfi_notes): Adjust scan_trace callers.
2717 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2718 INCOMING_FRAME_SP_OFFSET in the CIE.
2719 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2720 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2722 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2723 * doc/tm.texi: Regenerated.
2725 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2727 PR rtl-optimization/83147
2728 * lra-constraints.c (remove_inheritance_pseudos): Use
2729 lra_substitute_pseudo_within_insn.
2731 2018-01-19 Tom de Vries <tom@codesourcery.com>
2732 Cesar Philippidis <cesar@codesourcery.com>
2735 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2737 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2740 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2741 spaces for function labels.
2743 2018-01-19 Martin Liska <mliska@suse.cz>
2745 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2746 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2747 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2748 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2749 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2750 (PRED_CONST_RETURN): Change from 69 to 65.
2751 (PRED_NULL_RETURN): Change from 91 to 71.
2752 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2753 (PRED_LOOP_GUARD): Change from 66 to 73.
2755 2018-01-19 Martin Liska <mliska@suse.cz>
2757 * predict.c (predict_insn_def): Add new assert.
2758 (struct branch_predictor): Change type to signed integer.
2759 (test_prediction_value_range): Amend test to cover
2761 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2762 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2763 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2764 (PRED_LOOP_IV_COMPARE): Likewise.
2765 * predict.h (PROB_UNINITIALIZED): Define new constant.
2767 2018-01-19 Martin Liska <mliska@suse.cz>
2769 * predict.c (dump_prediction): Add new format for
2770 analyze_brprob.py script which is enabled with -details
2772 * profile-count.h (precise_p): New function.
2774 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2776 PR tree-optimization/83922
2777 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2778 there are no statements that need masking.
2779 (vect_active_double_reduction_p): New function.
2780 (vect_analyze_loop_operations): Use it when handling phis that
2781 are not in the loop header.
2783 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2785 PR tree-optimization/83914
2786 * tree-vect-loop.c (vectorizable_induction): Don't convert
2787 init_expr or apply the peeling adjustment for inductions
2788 that are nested within the vectorized loop.
2790 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2792 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2795 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2799 * function.h (gimplify_parameters): Add gimple_seq * argument.
2800 * function.c: Include gimple.h and options.h.
2801 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2802 for the added local temporaries if needed.
2803 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2804 if there are any parameter cleanups, wrap whole body into a
2805 try/finally with the cleanups.
2807 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2810 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2811 Use GET_MODE_CLASS for scalar floating point.
2813 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2817 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2818 Fix call of call_cgraph_insertion_hooks.
2820 2018-01-18 Martin Sebor <msebor@redhat.com>
2822 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2824 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2827 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2830 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2833 * common.opt: (-ffile-prefix-map): New option.
2834 * opts.c (common_handle_option): Defer it.
2835 * opts-global.c (handle_common_deferred_options): Handle it.
2836 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2837 * file-prefix-map.h: New file.
2838 (remap_debug_filename, add_debug_prefix_map): ...here.
2839 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2840 * final.c (debug_prefix_map, add_debug_prefix_map
2841 remap_debug_filename): Move to...
2842 * file-prefix-map.c: New file.
2843 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2844 generalize, get rid of alloca(), use strrchr() instead of strchr().
2845 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2846 Implement in terms of add_prefix_map().
2847 (remap_macro_filename, remap_debug_filename): Implement in term of
2849 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2850 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2851 * dbxout.c: Include file-prefix-map.h.
2852 * varasm.c: Likewise.
2853 * vmsdbgout.c: Likewise.
2854 * xcoffout.c: Likewise.
2855 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2856 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2857 * doc/invoke.texi (-ffile-prefix-map): Document.
2858 (-fdebug-prefix-map): Update description.
2860 2018-01-18 Martin Liska <mliska@suse.cz>
2862 * config/i386/i386.c (indirect_thunk_name): Document that also
2864 (output_indirect_thunk): Document why both instructions
2865 (pause and lfence) are generated.
2867 2018-01-18 Richard Biener <rguenther@suse.de>
2869 PR tree-optimization/83887
2870 * graphite-scop-detection.c
2871 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2872 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2873 (scop_detection::merge_sese): Re-implement with a flood-fill
2874 algorithm that properly finds a SESE region if it exists.
2876 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2879 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2880 pointer_diff optimizations use view_convert instead of convert.
2882 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2884 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2885 Generate different code for -mno-speculate-indirect-jumps.
2886 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2887 (*call_indirect_aix<mode>): Disable for
2888 -mno-speculate-indirect-jumps.
2889 (*call_indirect_aix<mode>_nospec): New define_insn.
2890 (*call_value_indirect_aix<mode>): Disable for
2891 -mno-speculate-indirect-jumps.
2892 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2893 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2894 -mno-speculate-indirect-jumps.
2895 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2897 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2899 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2900 long double type, set the flags for noting the default long double
2901 type, even if we don't pass or return a long double type.
2903 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2906 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2909 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2911 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2912 support for merge[hl].
2913 (fold_mergehl_helper): New helper function.
2914 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2915 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2916 (altivec_vmrglw_direct): Add xxmrglw insn.
2918 2018-01-17 Andrew Waterman <andrew@sifive.com>
2920 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2921 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2923 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2926 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2927 call the lto_location_cache before reading the
2928 DECL_SOURCE_LOCATION of the types.
2930 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2931 Richard Sandiford <richard.sandiford@linaro.org>
2933 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2934 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2935 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2936 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2937 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2939 * config/aarch64/constraints.md (aarch64_movti_operand):
2941 * config/aarch64/predicates.md (Uti): Add new constraint.
2943 2018-01-17 Carl Love <cel@us.ibm.com>
2945 * config/rs6000/vsx.md (define_expand xl_len_r,
2946 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2947 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2949 (define_expand, define_insn): Move the shift left from the
2950 define_insn to the define_expand for lxvl and stxvl instructions.
2951 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2952 and XL_LEN_R definitions to PURE.
2954 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2956 * config/i386/i386.c (indirect_thunk_name): Declare regno
2957 as unsigned int. Compare regno with INVALID_REGNUM.
2958 (output_indirect_thunk): Ditto.
2959 (output_indirect_thunk_function): Ditto.
2960 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2961 in the call to output_indirect_thunk_function.
2963 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2966 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2967 rather than the size of inner_type to determine the stack slot size
2968 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2970 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2973 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2976 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2978 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2979 endian Linux systems to optionally enable multilibs for selecting
2980 the long double type if the user configured an explicit type.
2981 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2982 have no long double multilibs if not defined.
2983 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2984 warn if the user used -mabi={ieee,ibm}longdouble and we built
2985 multilibs for long double.
2986 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2987 appropriate multilib option.
2988 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2990 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2991 for building long double multilibs.
2992 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2994 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2996 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2999 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3001 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3004 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3007 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3010 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3012 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3013 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3015 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3017 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3018 different rtl trees depending on TARGET_64BIT.
3019 (rs6000_gen_lvx): Likewise.
3021 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3023 * config/visium/visium.md (nop): Tweak comment.
3024 (hazard_nop): Likewise.
3026 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3028 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3029 -mspeculate-indirect-jumps.
3030 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3031 for -mno-speculate-indirect-jumps.
3032 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3033 (*call_value_indirect_elfv2<mode>): Disable for
3034 -mno-speculate-indirect-jumps.
3035 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3036 (indirect_jump): Emit different RTL for
3037 -mno-speculate-indirect-jumps.
3038 (*indirect_jump<mode>): Disable for
3039 -mno-speculate-indirect-jumps.
3040 (*indirect_jump<mode>_nospec): New define_insn.
3041 (tablejump): Emit different RTL for
3042 -mno-speculate-indirect-jumps.
3043 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3044 (tablejumpsi_nospec): New define_expand.
3045 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3046 (tablejumpdi_nospec): New define_expand.
3047 (*tablejump<mode>_internal1): Disable for
3048 -mno-speculate-indirect-jumps.
3049 (*tablejump<mode>_internal1_nospec): New define_insn.
3050 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3053 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3055 * caller-save.c (insert_save): Drop unnecessary parameter. All
3058 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3059 Richard Biener <rguenth@suse.de>
3062 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3063 return early, inline manually is_gimple_sizepos. Make sure if we
3064 call gimplify_expr we don't end up with a gimple constant.
3065 * tree.c (variably_modified_type_p): Don't return true for
3066 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3067 * gimplify.h (is_gimple_sizepos): Remove.
3069 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3071 PR tree-optimization/83857
3072 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3073 vectorizable_live_operation for pure SLP statements.
3074 (vectorizable_live_operation): Handle PHIs.
3076 2018-01-16 Richard Biener <rguenther@suse.de>
3078 PR tree-optimization/83867
3079 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3080 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3082 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3085 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3086 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3087 If off is not INTEGER_CST, issue a may not be aligned warning
3088 rather than isn't aligned. Use isn%'t rather than isn't.
3089 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3091 <case MULT_EXPR>: Improve the case when bottom and one of the
3092 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3093 operand, in that case check if the other operand is multiple of
3094 bottom divided by the INTEGER_CST operand.
3096 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3099 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3100 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3101 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3102 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3103 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3104 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3105 (pa_function_arg_size): New function.
3107 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3109 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3110 in a separate statement.
3112 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3114 PR tree-optimization/83847
3115 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3116 group gathers and scatters.
3118 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3120 PR rtl-optimization/86620
3121 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3123 PR rtl-optimization/83213
3124 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3125 to last if both are JUMP_INSNs.
3127 PR tree-optimization/83843
3128 * gimple-ssa-store-merging.c
3129 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3130 store_immediate_info for bswap/nop orig_stores.
3132 2018-01-15 Andrew Waterman <andrew@sifive.com>
3134 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3136 <UDIV>: Increase cost if !TARGET_DIV.
3138 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3140 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3141 (define_attr "cr_logical_3op"): New.
3142 (cceq_ior_compare): Adjust.
3143 (cceq_ior_compare_complement): Adjust.
3144 (*cceq_rev_compare): Adjust.
3145 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3146 (is_cracked_insn): Adjust.
3147 (insn_must_be_first_in_group): Adjust.
3148 * config/rs6000/40x.md: Adjust.
3149 * config/rs6000/440.md: Adjust.
3150 * config/rs6000/476.md: Adjust.
3151 * config/rs6000/601.md: Adjust.
3152 * config/rs6000/603.md: Adjust.
3153 * config/rs6000/6xx.md: Adjust.
3154 * config/rs6000/7450.md: Adjust.
3155 * config/rs6000/7xx.md: Adjust.
3156 * config/rs6000/8540.md: Adjust.
3157 * config/rs6000/cell.md: Adjust.
3158 * config/rs6000/e300c2c3.md: Adjust.
3159 * config/rs6000/e500mc.md: Adjust.
3160 * config/rs6000/e500mc64.md: Adjust.
3161 * config/rs6000/e5500.md: Adjust.
3162 * config/rs6000/e6500.md: Adjust.
3163 * config/rs6000/mpc.md: Adjust.
3164 * config/rs6000/power4.md: Adjust.
3165 * config/rs6000/power5.md: Adjust.
3166 * config/rs6000/power6.md: Adjust.
3167 * config/rs6000/power7.md: Adjust.
3168 * config/rs6000/power8.md: Adjust.
3169 * config/rs6000/power9.md: Adjust.
3170 * config/rs6000/rs64.md: Adjust.
3171 * config/rs6000/titan.md: Adjust.
3173 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3175 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3176 ix86_indirect_branch_register logic.
3178 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3180 * config/i386/constraints.md (Bs): Update
3181 ix86_indirect_branch_register check. Don't check
3182 ix86_indirect_branch_register with GOT_memory_operand.
3184 * config/i386/predicates.md (GOT_memory_operand): Don't check
3185 ix86_indirect_branch_register here.
3186 (GOT32_symbol_operand): Likewise.
3188 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3190 * config/i386/predicates.md (constant_call_address_operand):
3191 Rewrite ix86_indirect_branch_register logic.
3192 (sibcall_insn_operand): Likewise.
3194 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3196 * config/i386/constraints.md (Bs): Replace
3197 ix86_indirect_branch_thunk_register with
3198 ix86_indirect_branch_register.
3200 * config/i386/i386.md (indirect_jump): Likewise.
3201 (tablejump): Likewise.
3202 (*sibcall_memory): Likewise.
3203 (*sibcall_value_memory): Likewise.
3204 Peepholes of indirect call and jump via memory: Likewise.
3205 * config/i386/i386.opt: Likewise.
3206 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3207 (GOT_memory_operand): Likewise.
3208 (call_insn_operand): Likewise.
3209 (sibcall_insn_operand): Likewise.
3210 (GOT32_symbol_operand): Likewise.
3212 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3215 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3216 type rather than type addr's type points to.
3217 (expand_omp_atomic_mutex): Likewise.
3218 (expand_omp_atomic): Likewise.
3220 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3223 * config/i386/i386.c (output_indirect_thunk_function): Use
3224 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3225 for __x86_return_thunk.
3227 2018-01-15 Richard Biener <rguenther@suse.de>
3230 * expmed.c (extract_bit_field_1): Fix typo.
3232 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3235 * config/arm/iterators.md (VF): New mode iterator.
3236 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3237 Remove integer-related logic from pattern.
3238 (neon_vabd<mode>_3): Likewise.
3240 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3243 * common.opt (fstrict-overflow): No longer an alias.
3244 (fwrapv-pointer): New option.
3245 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3246 also for pointer types based on flag_wrapv_pointer.
3247 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3248 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3249 opts->x_flag_wrapv got set.
3250 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3251 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3252 POINTER_TYPE_OVERFLOW_UNDEFINED.
3253 * match.pd: Likewise in address comparison pattern.
3254 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3256 2018-01-15 Richard Biener <rguenther@suse.de>
3259 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3260 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3261 Reset type names to their identifier if their TYPE_DECL doesn't
3262 have linkage (and thus is used for ODR and devirt).
3263 (save_debug_info_for_decl): Remove.
3264 (save_debug_info_for_type): Likewise.
3265 (add_tree_to_fld_list): Adjust.
3266 * tree-pretty-print.c (dump_generic_node): Make dumping of
3267 type names more robust.
3269 2018-01-15 Richard Biener <rguenther@suse.de>
3271 * BASE-VER: Bump to 8.0.1.
3273 2018-01-14 Martin Sebor <msebor@redhat.com>
3276 * builtins.c (check_access): Avoid warning when the no-warning bit
3279 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3281 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3282 * ira-color (allocno_hard_regs_compare): Likewise.
3284 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3287 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3288 Use .pushsection/.popsection.
3290 2018-01-14 Martin Sebor <msebor@redhat.com>
3293 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3295 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3297 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3298 entry from extra_headers.
3299 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3300 extra_headers, make the list bitwise identical to the i?86-*-* one.
3302 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3304 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3305 -mcmodel=large with -mindirect-branch=thunk,
3306 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3307 -mfunction-return=thunk-extern.
3308 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3309 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3310 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3312 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3314 * config/i386/i386.c (print_reg): Print the name of the full
3315 integer register without '%'.
3316 (ix86_print_operand): Handle 'V'.
3317 * doc/extend.texi: Document 'V' modifier.
3319 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3321 * config/i386/constraints.md (Bs): Disallow memory operand for
3322 -mindirect-branch-register.
3324 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3325 (GOT_memory_operand): Likewise.
3326 (call_insn_operand): Likewise.
3327 (sibcall_insn_operand): Likewise.
3328 (GOT32_symbol_operand): Likewise.
3329 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3330 for -mindirect-branch-register.
3331 (tablejump): Likewise.
3332 (*sibcall_memory): Likewise.
3333 (*sibcall_value_memory): Likewise.
3334 Disallow peepholes of indirect call and jump via memory for
3335 -mindirect-branch-register.
3336 (*call_pop): Replace m with Bw.
3337 (*call_value_pop): Likewise.
3338 (*sibcall_pop_memory): Replace m with Bs.
3339 * config/i386/i386.opt (mindirect-branch-register): New option.
3340 * doc/invoke.texi: Document -mindirect-branch-register option.
3342 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3344 * config/i386/i386-protos.h (ix86_output_function_return): New.
3345 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3346 set function_return_type.
3347 (indirect_thunk_name): Add ret_p to indicate thunk for function
3349 (output_indirect_thunk_function): Pass false to
3350 indirect_thunk_name.
3351 (ix86_output_indirect_branch_via_reg): Likewise.
3352 (ix86_output_indirect_branch_via_push): Likewise.
3353 (output_indirect_thunk_function): Create alias for function
3354 return thunk if regno < 0.
3355 (ix86_output_function_return): New function.
3356 (ix86_handle_fndecl_attribute): Handle function_return.
3357 (ix86_attribute_table): Add function_return.
3358 * config/i386/i386.h (machine_function): Add
3359 function_return_type.
3360 * config/i386/i386.md (simple_return_internal): Use
3361 ix86_output_function_return.
3362 (simple_return_internal_long): Likewise.
3363 * config/i386/i386.opt (mfunction-return=): New option.
3364 (indirect_branch): Mention -mfunction-return=.
3365 * doc/extend.texi: Document function_return function attribute.
3366 * doc/invoke.texi: Document -mfunction-return= option.
3368 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3370 * config/i386/i386-opts.h (indirect_branch): New.
3371 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3372 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3373 with local indirect jump when converting indirect call and jump.
3374 (ix86_set_indirect_branch_type): New.
3375 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3376 (indirectlabelno): New.
3377 (indirect_thunk_needed): Likewise.
3378 (indirect_thunk_bnd_needed): Likewise.
3379 (indirect_thunks_used): Likewise.
3380 (indirect_thunks_bnd_used): Likewise.
3381 (INDIRECT_LABEL): Likewise.
3382 (indirect_thunk_name): Likewise.
3383 (output_indirect_thunk): Likewise.
3384 (output_indirect_thunk_function): Likewise.
3385 (ix86_output_indirect_branch_via_reg): Likewise.
3386 (ix86_output_indirect_branch_via_push): Likewise.
3387 (ix86_output_indirect_branch): Likewise.
3388 (ix86_output_indirect_jmp): Likewise.
3389 (ix86_code_end): Call output_indirect_thunk_function if needed.
3390 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3392 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3393 (ix86_attribute_table): Add indirect_branch.
3394 * config/i386/i386.h (machine_function): Add indirect_branch_type
3395 and has_local_indirect_jump.
3396 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3398 (tablejump): Likewise.
3399 (*indirect_jump): Use ix86_output_indirect_jmp.
3400 (*tablejump_1): Likewise.
3401 (simple_return_indirect_internal): Likewise.
3402 * config/i386/i386.opt (mindirect-branch=): New option.
3403 (indirect_branch): New.
3406 (thunk-inline): Likewise.
3407 (thunk-extern): Likewise.
3408 * doc/extend.texi: Document indirect_branch function attribute.
3409 * doc/invoke.texi: Document -mindirect-branch= option.
3411 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3414 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3416 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3418 * ipa-inline.c (want_inline_small_function_p): Return false if
3419 inlining has already failed with CIF_FINAL_ERROR.
3420 (update_caller_keys): Call want_inline_small_function_p before
3422 (update_callee_keys): Likewise.
3424 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3426 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3428 (rs6000_quadword_masked_address_p): Likewise.
3429 (quad_aligned_load_p): Likewise.
3430 (quad_aligned_store_p): Likewise.
3431 (const_load_sequence_p): Add comment to describe the outer-most loop.
3432 (mimic_memory_attributes_and_flags): New function.
3433 (rs6000_gen_stvx): Likewise.
3434 (replace_swapped_aligned_store): Likewise.
3435 (rs6000_gen_lvx): Likewise.
3436 (replace_swapped_aligned_load): Likewise.
3437 (replace_swapped_load_constant): Capitalize argument name in
3438 comment describing this function.
3439 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3440 and stores that access quad-word aligned addresses and replace
3441 with stvx or lvx instructions when appropriate.
3442 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3443 New function prototype.
3444 (rs6000_quadword_masked_address_p): Likewise.
3445 (rs6000_gen_lvx): Likewise.
3446 (rs6000_gen_stvx): Likewise.
3447 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3448 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3449 when memory address is aligned.
3450 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3451 this split to select lvx instruction when memory address is aligned.
3452 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3453 instruction when memory address is aligned.
3454 (*vsx_le_perm_load_v16qi): Likewise.
3455 (four unnamed splitters): Modify to select the stvx instruction
3456 when memory is aligned.
3458 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3460 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3461 which appears in the queue multiple times.
3463 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3464 Alan Hayward <alan.hayward@arm.com>
3465 David Sherwood <david.sherwood@arm.com>
3467 * tree-vectorizer.h (vec_lower_bound): New structure.
3468 (_loop_vec_info): Add check_nonzero and lower_bounds.
3469 (LOOP_VINFO_CHECK_NONZERO): New macro.
3470 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3471 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3472 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3473 fields. Make seg_len the distance travelled, not including the
3475 (dr_direction_indicator): Declare.
3476 (dr_zero_step_indicator): Likewise.
3477 (dr_known_forward_stride_p): Likewise.
3478 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3480 (runtime_alias_check_p): Allow runtime alias checks with
3482 (operator ==): Compare access_size and align.
3483 (prune_runtime_alias_test_list): Rework for new distinction between
3484 the access_size and seg_len.
3485 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3487 (get_segment_min_max): New function.
3488 (create_intersect_range_checks): Use it.
3489 (dr_step_indicator): New function.
3490 (dr_direction_indicator): Likewise.
3491 (dr_zero_step_indicator): Likewise.
3492 (dr_known_forward_stride_p): Likewise.
3493 * tree-loop-distribution.c (data_ref_segment_size): Return
3494 DR_STEP * (niters - 1).
3495 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3497 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3498 (vect_preserves_scalar_order_p): New function, split out from...
3499 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3500 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3501 (vect_vfa_access_size): New function.
3502 (vect_vfa_align): Likewise.
3503 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3504 (dump_lower_bound): New function.
3505 (vect_check_lower_bound): Likewise.
3506 (vect_small_gap_p): Likewise.
3507 (vectorizable_with_step_bound_p): Likewise.
3508 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3509 depencies if the vectorization factor is 1. Convert the checks
3510 for nonzero steps into checks on the bounds of DR_STEP. Try using
3511 a bunds check for variable steps if the minimum required step is
3512 relatively small. Update calls to the dr_with_seg_len
3513 constructor and to vect_compile_time_alias.
3514 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3516 (vect_loop_versioning): Call it.
3517 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3519 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3521 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3522 Alan Hayward <alan.hayward@arm.com>
3523 David Sherwood <david.sherwood@arm.com>
3525 * doc/sourcebuild.texi (vect_scatter_store): Document.
3526 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3528 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3530 * genopinit.c (main): Add supports_vec_scatter_store and
3531 supports_vec_scatter_store_cached to target_optabs.
3532 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3533 IFN_MASK_SCATTER_STORE.
3534 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3536 * internal-fn.h (internal_store_fn_p): Declare.
3537 (internal_fn_stored_value_index): Likewise.
3538 * internal-fn.c (scatter_store_direct): New macro.
3539 (expand_scatter_store_optab_fn): New function.
3540 (direct_scatter_store_optab_supported_p): New macro.
3541 (internal_store_fn_p): New function.
3542 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3543 IFN_MASK_SCATTER_STORE.
3544 (internal_fn_mask_index): Likewise.
3545 (internal_fn_stored_value_index): New function.
3546 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3548 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3549 * optabs-query.c (supports_vec_scatter_store_p): New function.
3550 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3551 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3552 true for scatter stores.
3553 (vect_gather_scatter_fn_p): Handle scatter stores too.
3554 (vect_check_gather_scatter): Consider using scatter stores if
3555 supports_vec_scatter_store_p.
3556 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3558 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3559 internal_fn_stored_value_index.
3560 (check_load_store_masking): Handle scatter stores too.
3561 (vect_get_store_rhs): Make public.
3562 (vectorizable_call): Use internal_store_fn_p.
3563 (vectorizable_store): Handle scatter store internal functions.
3564 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3565 when deciding whether the end of the group has been reached.
3566 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3567 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3568 (mask_scatter_store<mode>): New insns.
3570 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3571 Alan Hayward <alan.hayward@arm.com>
3572 David Sherwood <david.sherwood@arm.com>
3574 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3575 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3576 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3578 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3579 Use vect_truncate_gather_scatter_offset if we can't treat the
3580 operation as a normal gather load or scatter store.
3581 (get_group_load_store_type): Take the gather_scatter_info
3582 as argument. Try using a gather load or scatter store for
3583 single-element groups.
3584 (get_load_store_type): Update calls to get_group_load_store_type
3585 and vect_use_strided_gather_scatters_p.
3587 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3588 Alan Hayward <alan.hayward@arm.com>
3589 David Sherwood <david.sherwood@arm.com>
3591 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3592 optional tree argument.
3593 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3595 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3596 but continue to use the current value as a fallback.
3597 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3598 to compare the updates.
3599 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3600 (get_load_store_type): Use it when handling a strided access.
3601 (vect_get_strided_load_store_ops): New function.
3602 (vect_get_data_ptr_increment): Likewise.
3603 (vectorizable_load): Handle strided gather loads. Always pass
3604 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3606 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3607 Alan Hayward <alan.hayward@arm.com>
3608 David Sherwood <david.sherwood@arm.com>
3610 * doc/md.texi (gather_load@var{m}): Document.
3611 (mask_gather_load@var{m}): Likewise.
3612 * genopinit.c (main): Add supports_vec_gather_load and
3613 supports_vec_gather_load_cached to target_optabs.
3614 * optabs-tree.c (init_tree_optimization_optabs): Use
3615 ggc_cleared_alloc to allocate target_optabs.
3616 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3617 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3619 * internal-fn.h (internal_load_fn_p): Declare.
3620 (internal_gather_scatter_fn_p): Likewise.
3621 (internal_fn_mask_index): Likewise.
3622 (internal_gather_scatter_fn_supported_p): Likewise.
3623 * internal-fn.c (gather_load_direct): New macro.
3624 (expand_gather_load_optab_fn): New function.
3625 (direct_gather_load_optab_supported_p): New macro.
3626 (direct_internal_fn_optab): New function.
3627 (internal_load_fn_p): Likewise.
3628 (internal_gather_scatter_fn_p): Likewise.
3629 (internal_fn_mask_index): Likewise.
3630 (internal_gather_scatter_fn_supported_p): Likewise.
3631 * optabs-query.c (supports_at_least_one_mode_p): New function.
3632 (supports_vec_gather_load_p): Likewise.
3633 * optabs-query.h (supports_vec_gather_load_p): Declare.
3634 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3635 and memory_type field.
3636 (NUM_PATTERNS): Bump to 15.
3637 * tree-vect-data-refs.c: Include internal-fn.h.
3638 (vect_gather_scatter_fn_p): New function.
3639 (vect_describe_gather_scatter_call): Likewise.
3640 (vect_check_gather_scatter): Try using internal functions for
3641 gather loads. Recognize existing calls to a gather load function.
3642 (vect_analyze_data_refs): Consider using gather loads if
3643 supports_vec_gather_load_p.
3644 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3645 (vect_get_gather_scatter_offset_type): Likewise.
3646 (vect_convert_mask_for_vectype): Likewise.
3647 (vect_add_conversion_to_patterm): Likewise.
3648 (vect_try_gather_scatter_pattern): Likewise.
3649 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3650 (vect_vect_recog_func_ptrs): Add it.
3651 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3652 internal_fn_mask_index and internal_gather_scatter_fn_p.
3653 (check_load_store_masking): Take the gather_scatter_info as an
3654 argument and handle gather loads.
3655 (vect_get_gather_scatter_ops): New function.
3656 (vectorizable_call): Check internal_load_fn_p.
3657 (vectorizable_load): Likewise. Handle gather load internal
3659 (vectorizable_store): Update call to check_load_store_masking.
3660 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3661 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3662 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3663 (aarch64_gather_scale_operand_d): New predicates.
3664 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3665 (mask_gather_load<mode>): New insns.
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 * optabs.def (fold_left_plus_optab): New optab.
3672 * doc/md.texi (fold_left_plus_@var{m}): Document.
3673 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3674 * internal-fn.c (fold_left_direct): Define.
3675 (expand_fold_left_optab_fn): Likewise.
3676 (direct_fold_left_optab_supported_p): Likewise.
3677 * fold-const-call.c (fold_const_fold_left): New function.
3678 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3679 * tree-parloops.c (valid_reduction_p): New function.
3680 (gather_scalar_reductions): Use it.
3681 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3682 (vect_finish_replace_stmt): Declare.
3683 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3684 (needs_fold_left_reduction_p): New function, split out from...
3685 (vect_is_simple_reduction): ...here. Accept reductions that
3686 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3687 (vect_force_simple_reduction): Also store the reduction type in
3688 the assignment's STMT_VINFO_REDUC_TYPE.
3689 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3690 (merge_with_identity): New function.
3691 (vect_expand_fold_left): Likewise.
3692 (vectorize_fold_left_reduction): Likewise.
3693 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3694 scalar phi in place for it. Check for target support and reject
3695 cases that would reassociate the operation. Defer the transform
3696 phase to vectorize_fold_left_reduction.
3697 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3698 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3699 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3701 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3703 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3704 call to ifc_temp_var.
3706 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3707 Alan Hayward <alan.hayward@arm.com>
3708 David Sherwood <david.sherwood@arm.com>
3710 * target.def (legitimize_address_displacement): Take the original
3711 offset as a poly_int.
3712 * targhooks.h (default_legitimize_address_displacement): Update
3714 * targhooks.c (default_legitimize_address_displacement): Likewise.
3715 * doc/tm.texi: Regenerate.
3716 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3717 as an argument, moving assert of ad->disp == ad->disp_term to...
3718 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3719 Try calling targetm.legitimize_address_displacement before expanding
3720 the address rather than afterwards, and adjust for the new interface.
3721 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3722 Match the new hook interface. Handle SVE addresses.
3723 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3726 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3728 * Makefile.in (OBJS): Add early-remat.o.
3729 * target.def (select_early_remat_modes): New hook.
3730 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3731 * doc/tm.texi: Regenerate.
3732 * targhooks.h (default_select_early_remat_modes): Declare.
3733 * targhooks.c (default_select_early_remat_modes): New function.
3734 * timevar.def (TV_EARLY_REMAT): New timevar.
3735 * passes.def (pass_early_remat): New pass.
3736 * tree-pass.h (make_pass_early_remat): Declare.
3737 * early-remat.c: New file.
3738 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3740 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3742 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3743 Alan Hayward <alan.hayward@arm.com>
3744 David Sherwood <david.sherwood@arm.com>
3746 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3747 vfm1 with a bound_epilog parameter.
3748 (vect_do_peeling): Update calls accordingly, and move the prologue
3749 call earlier in the function. Treat the base bound_epilog as 0 for
3750 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3751 this base when peeling for gaps.
3752 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3753 with fully-masked loops.
3754 (vect_estimate_min_profitable_iters): Handle the single peeled
3755 iteration in that case.
3757 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3758 Alan Hayward <alan.hayward@arm.com>
3759 David Sherwood <david.sherwood@arm.com>
3761 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3762 single-element interleaving even if the size is not a power of 2.
3763 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3764 accesses for single-element interleaving if the group size is
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 * doc/md.texi (fold_extract_last_@var{m}): Document.
3772 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3773 * optabs.def (fold_extract_last_optab): New optab.
3774 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3775 * internal-fn.c (fold_extract_direct): New macro.
3776 (expand_fold_extract_optab_fn): Likewise.
3777 (direct_fold_extract_optab_supported_p): Likewise.
3778 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3779 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3780 EXTRACT_LAST_REDUCTION.
3781 (get_initial_def_for_reduction): Do not create an initial vector
3782 for EXTRACT_LAST_REDUCTION reductions.
3783 (vectorizable_reduction): Leave the scalar phi in place for
3784 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3785 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3786 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3787 transform phase to vectorizable_condition.
3788 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3790 (vect_finish_stmt_generation): ...here.
3791 (vect_finish_replace_stmt): New function.
3792 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3793 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3795 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3797 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3798 Alan Hayward <alan.hayward@arm.com>
3799 David Sherwood <david.sherwood@arm.com>
3801 * doc/md.texi (extract_last_@var{m}): Document.
3802 * optabs.def (extract_last_optab): New optab.
3803 * internal-fn.def (EXTRACT_LAST): New internal function.
3804 * internal-fn.c (cond_unary_direct): New macro.
3805 (expand_cond_unary_optab_fn): Likewise.
3806 (direct_cond_unary_optab_supported_p): Likewise.
3807 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3808 loops using EXTRACT_LAST.
3809 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3810 (extract_last_<mode>): ...this optab.
3811 (vec_extract<mode><Vel>): Update accordingly.
3813 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3814 Alan Hayward <alan.hayward@arm.com>
3815 David Sherwood <david.sherwood@arm.com>
3817 * target.def (empty_mask_is_expensive): New hook.
3818 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3819 * doc/tm.texi: Regenerate.
3820 * targhooks.h (default_empty_mask_is_expensive): Declare.
3821 * targhooks.c (default_empty_mask_is_expensive): New function.
3822 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3823 if the target says that empty masks are expensive.
3824 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3826 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
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 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3833 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3834 (vect_use_loop_mask_for_alignment_p): New function.
3835 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3836 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3837 niters_skip argument. Make sure that the first niters_skip elements
3838 of the first iteration are inactive.
3839 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3840 Update call to vect_set_loop_masks_directly.
3841 (get_misalign_in_elems): New function, split out from...
3842 (vect_gen_prolog_loop_niters): ...here.
3843 (vect_update_init_of_dr): Take a code argument that specifies whether
3844 the adjustment should be added or subtracted.
3845 (vect_update_init_of_drs): Likewise.
3846 (vect_prepare_for_masked_peels): New function.
3847 (vect_do_peeling): Skip prologue peeling if we're using a mask
3848 instead. Update call to vect_update_inits_of_drs.
3849 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3851 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3852 alignment. Do not include the number of peeled iterations in
3853 the minimum threshold in that case.
3854 (vectorizable_induction): Adjust the start value down by
3855 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3856 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3857 Take the number of skipped iterations into account when calculating
3859 * tree-vect-stmts.c (vect_gen_while_not): New function.
3861 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3862 Alan Hayward <alan.hayward@arm.com>
3863 David Sherwood <david.sherwood@arm.com>
3865 * doc/sourcebuild.texi (vect_fully_masked): Document.
3866 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3868 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3870 (vect_analyze_loop_2): ...here. Don't check the vectorization
3871 factor against the number of loop iterations if the loop is
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 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3879 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3880 (dump_groups): Update accordingly.
3881 (iv_use::mem_type): New member variable.
3882 (address_p): New function.
3883 (record_use): Add a mem_type argument and initialize the new
3885 (record_group_use): Add a mem_type argument. Use address_p.
3886 Remove obsolete null checks of base_object. Update call to record_use.
3887 (find_interesting_uses_op): Update call to record_group_use.
3888 (find_interesting_uses_cond): Likewise.
3889 (find_interesting_uses_address): Likewise.
3890 (get_mem_type_for_internal_fn): New function.
3891 (find_address_like_use): Likewise.
3892 (find_interesting_uses_stmt): Try find_address_like_use before
3893 calling find_interesting_uses_op.
3894 (addr_offset_valid_p): Use the iv mem_type field as the type
3895 of the addressed memory.
3896 (add_autoinc_candidates): Likewise.
3897 (get_address_cost): Likewise.
3898 (split_small_address_groups_p): Use address_p.
3899 (split_address_groups): Likewise.
3900 (add_iv_candidate_for_use): Likewise.
3901 (autoinc_possible_for_pair): Likewise.
3902 (rewrite_groups): Likewise.
3903 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3904 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3905 (get_alias_ptr_type_for_ptr_address): New function.
3906 (rewrite_use_address): Rewrite address uses in calls that were
3907 identified by find_address_like_use.
3909 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3910 Alan Hayward <alan.hayward@arm.com>
3911 David Sherwood <david.sherwood@arm.com>
3913 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3915 * gimple-expr.h (is_gimple_addressable: Likewise.
3916 * gimple-expr.c (is_gimple_address): Likewise.
3917 * internal-fn.c (expand_call_mem_ref): New function.
3918 (expand_mask_load_optab_fn): Use it.
3919 (expand_mask_store_optab_fn): Likewise.
3921 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3922 Alan Hayward <alan.hayward@arm.com>
3923 David Sherwood <david.sherwood@arm.com>
3925 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3926 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3927 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3928 (cond_umax@var{mode}): Document.
3929 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3930 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3931 (cond_umin_optab, cond_umax_optab): New optabs.
3932 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3933 (COND_IOR, COND_XOR): New internal functions.
3934 * internal-fn.h (get_conditional_internal_fn): Declare.
3935 * internal-fn.c (cond_binary_direct): New macro.
3936 (expand_cond_binary_optab_fn): Likewise.
3937 (direct_cond_binary_optab_supported_p): Likewise.
3938 (get_conditional_internal_fn): New function.
3939 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3940 Cope with reduction statements that are vectorized as calls rather
3942 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3943 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3944 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3945 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3946 (UNSPEC_COND_EOR): New unspecs.
3947 (optab): Add mappings for them.
3948 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3949 (sve_int_op, sve_fp_op): New int attributes.
3951 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3952 Alan Hayward <alan.hayward@arm.com>
3953 David Sherwood <david.sherwood@arm.com>
3955 * optabs.def (while_ult_optab): New optab.
3956 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3957 * internal-fn.def (WHILE_ULT): New internal function.
3958 * internal-fn.h (direct_internal_fn_supported_p): New override
3959 that takes two types as argument.
3960 * internal-fn.c (while_direct): New macro.
3961 (expand_while_optab_fn): New function.
3962 (convert_optab_supported_p): Likewise.
3963 (direct_while_optab_supported_p): New macro.
3964 * wide-int.h (wi::udiv_ceil): New function.
3965 * tree-vectorizer.h (rgroup_masks): New structure.
3966 (vec_loop_masks): New typedef.
3967 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3969 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3970 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3971 (vect_max_vf): New function.
3972 (slpeel_make_loop_iterate_ntimes): Delete.
3973 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3974 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3975 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3976 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3977 internal-fn.h, stor-layout.h and optabs-query.h.
3978 (vect_set_loop_mask): New function.
3979 (add_preheader_seq): Likewise.
3980 (add_header_seq): Likewise.
3981 (interleave_supported_p): Likewise.
3982 (vect_maybe_permute_loop_masks): Likewise.
3983 (vect_set_loop_masks_directly): Likewise.
3984 (vect_set_loop_condition_masked): Likewise.
3985 (vect_set_loop_condition_unmasked): New function, split out from
3986 slpeel_make_loop_iterate_ntimes.
3987 (slpeel_make_loop_iterate_ntimes): Rename to..
3988 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3989 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3990 (vect_do_peeling): Update call accordingly.
3991 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3993 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3994 mask_compare_type, can_fully_mask_p and fully_masked_p.
3995 (release_vec_loop_masks): New function.
3996 (_loop_vec_info): Use it to free the loop masks.
3997 (can_produce_all_loop_masks_p): New function.
3998 (vect_get_max_nscalars_per_iter): Likewise.
3999 (vect_verify_full_masking): Likewise.
4000 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4001 retries, and free the mask rgroups before retrying. Check loop-wide
4002 reasons for disallowing fully-masked loops. Make the final decision
4003 about whether use a fully-masked loop or not.
4004 (vect_estimate_min_profitable_iters): Do not assume that peeling
4005 for the number of iterations will be needed for fully-masked loops.
4006 (vectorizable_reduction): Disable fully-masked loops.
4007 (vectorizable_live_operation): Likewise.
4008 (vect_halve_mask_nunits): New function.
4009 (vect_double_mask_nunits): Likewise.
4010 (vect_record_loop_mask): Likewise.
4011 (vect_get_loop_mask): Likewise.
4012 (vect_transform_loop): Handle the case in which the final loop
4013 iteration might handle a partial vector. Call vect_set_loop_condition
4014 instead of slpeel_make_loop_iterate_ntimes.
4015 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4016 (check_load_store_masking): New function.
4017 (prepare_load_store_mask): Likewise.
4018 (vectorizable_store): Handle fully-masked loops.
4019 (vectorizable_load): Likewise.
4020 (supportable_widening_operation): Use vect_halve_mask_nunits for
4022 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4023 (vect_gen_while): New function.
4024 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4025 (aarch64_uqdec<mode>): New insn.
4027 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4028 Alan Hayward <alan.hayward@arm.com>
4029 David Sherwood <david.sherwood@arm.com>
4031 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4032 (reduc_xor_scal_optab): New optabs.
4033 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4034 (reduc_xor_scal_@var{m}): Document.
4035 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4036 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4038 * fold-const-call.c (fold_const_call): Handle them.
4039 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4040 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4041 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4042 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4043 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4044 (UNSPEC_XORV): New unspecs.
4045 (optab): Add entries for them.
4046 (BITWISEV): New int iterator.
4047 (bit_reduc_op): New int attributes.
4049 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4050 Alan Hayward <alan.hayward@arm.com>
4051 David Sherwood <david.sherwood@arm.com>
4053 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4054 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4055 * optabs.def (vec_shl_insert_optab): New optab.
4056 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4057 (duplicate_and_interleave): Likewise.
4058 * tree-vect-loop.c: Include internal-fn.h.
4059 (neutral_op_for_slp_reduction): New function, split out from
4060 get_initial_defs_for_reduction.
4061 (get_initial_def_for_reduction): Handle option 2 for variable-length
4062 vectors by loading the neutral value into a vector and then shifting
4063 the initial value into element 0.
4064 (get_initial_defs_for_reduction): Replace the code argument with
4065 the neutral value calculated by neutral_op_for_slp_reduction.
4066 Use gimple_build_vector for constant-length vectors.
4067 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4068 but the first group_size elements have a neutral value.
4069 Use duplicate_and_interleave otherwise.
4070 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4071 Update call to get_initial_defs_for_reduction. Handle SLP
4072 reductions for variable-length vectors by creating one vector
4073 result for each scalar result, with the elements associated
4074 with other scalar results stubbed out with the neutral value.
4075 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4076 Require IFN_VEC_SHL_INSERT for double reductions on
4077 variable-length vectors, or SLP reductions that have
4078 a neutral value. Require can_duplicate_and_interleave_p
4079 support for variable-length unchained SLP reductions if there
4080 is no neutral value, such as for MIN/MAX reductions. Also require
4081 the number of vector elements to be a multiple of the number of
4082 SLP statements when doing variable-length unchained SLP reductions.
4083 Update call to vect_create_epilog_for_reduction.
4084 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4085 and remove initial values.
4086 (duplicate_and_interleave): Make public.
4087 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4088 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4090 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4091 Alan Hayward <alan.hayward@arm.com>
4092 David Sherwood <david.sherwood@arm.com>
4094 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4095 (can_duplicate_and_interleave_p): New function.
4096 (vect_get_and_check_slp_defs): Take the vector of statements
4097 rather than just the current one. Remove excess parentheses.
4098 Restriction rejectinon of vect_constant_def and vect_external_def
4099 for variable-length vectors to boolean types, or types for which
4100 can_duplicate_and_interleave_p is false.
4101 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4102 (duplicate_and_interleave): New function.
4103 (vect_get_constant_vectors): Use gimple_build_vector for
4104 constant-length vectors and suitable variable-length constant
4105 vectors. Use duplicate_and_interleave for other variable-length
4106 vectors. Don't defer the update when inserting new statements.
4108 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4109 Alan Hayward <alan.hayward@arm.com>
4110 David Sherwood <david.sherwood@arm.com>
4112 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4113 min_profitable_iters doesn't go negative.
4115 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4116 Alan Hayward <alan.hayward@arm.com>
4117 David Sherwood <david.sherwood@arm.com>
4119 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4120 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4121 * optabs.def (vec_mask_load_lanes_optab): New optab.
4122 (vec_mask_store_lanes_optab): Likewise.
4123 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4124 (MASK_STORE_LANES): Likewise.
4125 * internal-fn.c (mask_load_lanes_direct): New macro.
4126 (mask_store_lanes_direct): Likewise.
4127 (expand_mask_load_optab_fn): Handle masked operations.
4128 (expand_mask_load_lanes_optab_fn): New macro.
4129 (expand_mask_store_optab_fn): Handle masked operations.
4130 (expand_mask_store_lanes_optab_fn): New macro.
4131 (direct_mask_load_lanes_optab_supported_p): Likewise.
4132 (direct_mask_store_lanes_optab_supported_p): Likewise.
4133 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4135 (vect_load_lanes_supported): Likewise.
4136 * tree-vect-data-refs.c (strip_conversion): New function.
4137 (can_group_stmts_p): Likewise.
4138 (vect_analyze_data_ref_accesses): Use it instead of checking
4139 for a pair of assignments.
4140 (vect_store_lanes_supported): Take a masked_p parameter.
4141 (vect_load_lanes_supported): Likewise.
4142 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4143 vect_store_lanes_supported and vect_load_lanes_supported.
4144 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4145 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4146 parameter. Don't allow gaps for masked accesses.
4147 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4148 and vect_load_lanes_supported.
4149 (get_load_store_type): Take a masked_p parameter and update
4150 call to get_group_load_store_type.
4151 (vectorizable_store): Update call to get_load_store_type.
4152 Handle IFN_MASK_STORE_LANES.
4153 (vectorizable_load): Update call to get_load_store_type.
4154 Handle IFN_MASK_LOAD_LANES.
4156 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4157 Alan Hayward <alan.hayward@arm.com>
4158 David Sherwood <david.sherwood@arm.com>
4160 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4162 * config/aarch64/aarch64-protos.h
4163 (aarch64_sve_struct_memory_operand_p): Declare.
4164 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4165 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4166 (VPRED, vpred): Handle SVE structure modes.
4167 * config/aarch64/constraints.md (Utx): New constraint.
4168 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4169 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4170 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4171 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4172 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4173 structure modes. Split into pieces after RA.
4174 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4175 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4177 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4178 SVE structure modes.
4179 (aarch64_classify_address): Likewise.
4180 (sizetochar): Move earlier in file.
4181 (aarch64_print_operand): Handle SVE register lists.
4182 (aarch64_array_mode): New function.
4183 (aarch64_sve_struct_memory_operand_p): Likewise.
4184 (TARGET_ARRAY_MODE): Redefine.
4186 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4187 Alan Hayward <alan.hayward@arm.com>
4188 David Sherwood <david.sherwood@arm.com>
4190 * target.def (array_mode): New target hook.
4191 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4192 * doc/tm.texi: Regenerate.
4193 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4194 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4195 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4197 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4200 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4201 Alan Hayward <alan.hayward@arm.com>
4202 David Sherwood <david.sherwood@arm.com>
4204 * fold-const.c (fold_binary_loc): Check the argument types
4205 rather than the result type when testing for a vector operation.
4207 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4209 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4210 * doc/tm.texi: Regenerate.
4212 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4213 Alan Hayward <alan.hayward@arm.com>
4214 David Sherwood <david.sherwood@arm.com>
4216 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4217 (sve): Document new AArch64 extension.
4218 * doc/md.texi (w): Extend the description of the AArch64
4219 constraint to include SVE vectors.
4220 (Upl, Upa): Document new AArch64 predicate constraints.
4221 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4223 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4224 (msve-vector-bits=): New option.
4225 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4226 SVE when these are disabled.
4227 (sve): New extension.
4228 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4229 modes. Adjust their number of units based on aarch64_sve_vg.
4230 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4231 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4232 aarch64_addr_query_type.
4233 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4234 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4235 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4236 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4237 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4238 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4239 (aarch64_simd_imm_zero_p): Delete.
4240 (aarch64_check_zero_based_sve_index_immediate): Declare.
4241 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4242 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4243 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4244 (aarch64_sve_float_mul_immediate_p): Likewise.
4245 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4247 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4248 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4249 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4250 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4251 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4252 (aarch64_regmode_natural_size): Likewise.
4253 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4254 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4256 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4257 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4258 for VG and the SVE predicate registers.
4259 (V_ALIASES): Add a "z"-prefixed alias.
4260 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4261 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4262 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4263 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4264 (REG_CLASS_NAMES): Add entries for them.
4265 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4266 and the predicate registers.
4267 (aarch64_sve_vg): Declare.
4268 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4269 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4270 (REGMODE_NATURAL_SIZE): Define.
4271 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4273 * config/aarch64/aarch64.c: Include cfgrtl.h.
4274 (simd_immediate_info): Add a constructor for series vectors,
4275 and an associated step field.
4276 (aarch64_sve_vg): New variable.
4277 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4278 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4279 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4280 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4281 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4282 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4283 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4284 (aarch64_get_mask_mode): New functions.
4285 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4286 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4287 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4288 predicate modes and predicate registers. Explicitly restrict
4289 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4290 to store a vector mode if it is recognized by
4291 aarch64_classify_vector_mode.
4292 (aarch64_regmode_natural_size): New function.
4293 (aarch64_hard_regno_caller_save_mode): Return the original mode
4295 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4296 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4297 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4298 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4300 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4301 does not overlap dest if the function is frame-related. Handle
4303 (aarch64_split_add_offset): New function.
4304 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4305 them aarch64_add_offset.
4306 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4307 and update call to aarch64_sub_sp.
4308 (aarch64_add_cfa_expression): New function.
4309 (aarch64_expand_prologue): Pass extra temporary registers to the
4310 functions above. Handle the case in which we need to emit new
4311 DW_CFA_expressions for registers that were originally saved
4312 relative to the stack pointer, but now have to be expressed
4313 relative to the frame pointer.
4314 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4316 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4317 IP0 and IP1 values for SVE frames.
4318 (aarch64_expand_vec_series): New function.
4319 (aarch64_expand_sve_widened_duplicate): Likewise.
4320 (aarch64_expand_sve_const_vector): Likewise.
4321 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4322 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4323 into the register, rather than emitting a SET directly.
4324 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4325 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4326 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4327 (offset_9bit_signed_scaled_p): New functions.
4328 (aarch64_replicate_bitmask_imm): New function.
4329 (aarch64_bitmask_imm): Use it.
4330 (aarch64_cannot_force_const_mem): Reject expressions involving
4331 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4332 (aarch64_classify_index): Handle SVE indices, by requiring
4333 a plain register index with a scale that matches the element size.
4334 (aarch64_classify_address): Handle SVE addresses. Assert that
4335 the mode of the address is VOIDmode or an integer mode.
4336 Update call to aarch64_classify_symbol.
4337 (aarch64_classify_symbolic_expression): Update call to
4338 aarch64_classify_symbol.
4339 (aarch64_const_vec_all_in_range_p): New function.
4340 (aarch64_print_vector_float_operand): Likewise.
4341 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4342 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4343 and the FP immediates 1.0 and 0.5.
4344 (aarch64_print_address_internal): Handle SVE addresses.
4345 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4346 (aarch64_regno_regclass): Handle predicate registers.
4347 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4349 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4350 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4351 (aarch64_convert_sve_vector_bits): New function.
4352 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4353 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4355 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4356 Handle SVE vector and predicate modes. Accept VL-based constants
4357 that need only one temporary register, and VL offsets that require
4358 no temporary registers.
4359 (aarch64_conditional_register_usage): Mark the predicate registers
4360 as fixed if SVE isn't available.
4361 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4362 Return true for SVE vector and predicate modes.
4363 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4364 rather than an unsigned int. Handle SVE modes.
4365 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4367 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4369 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4370 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4371 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4372 (aarch64_sve_float_mul_immediate_p): New functions.
4373 (aarch64_sve_valid_immediate): New function.
4374 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4375 Explicitly reject structure modes. Check for INDEX constants.
4376 Handle PTRUE and PFALSE constants.
4377 (aarch64_check_zero_based_sve_index_immediate): New function.
4378 (aarch64_simd_imm_zero_p): Delete.
4379 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4380 vector modes. Accept constants in the range of CNT[BHWD].
4381 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4382 ask for an Advanced SIMD mode.
4383 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4384 (aarch64_simd_vector_alignment): Handle SVE predicates.
4385 (aarch64_vectorize_preferred_vector_alignment): New function.
4386 (aarch64_simd_vector_alignment_reachable): Use it instead of
4388 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4389 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4391 (MAX_VECT_LEN): Delete.
4392 (expand_vec_perm_d): Add a vec_flags field.
4393 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4394 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4395 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4397 (aarch64_evpc_rev): Rename to...
4398 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4399 (aarch64_evpc_rev_global): New function.
4400 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4401 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4403 (aarch64_evpc_sve_tbl): New function.
4404 (aarch64_expand_vec_perm_const_1): Update after rename of
4405 aarch64_evpc_rev. Handle SVE permutes too, trying
4406 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4407 than aarch64_evpc_tbl.
4408 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4409 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4410 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4411 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4412 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4413 (aarch64_expand_sve_vcond): New functions.
4414 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4415 of aarch64_vector_mode_p.
4416 (aarch64_dwarf_poly_indeterminate_value): New function.
4417 (aarch64_compute_pressure_classes): Likewise.
4418 (aarch64_can_change_mode_class): Likewise.
4419 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4420 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4421 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4422 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4423 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4424 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4425 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4426 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4428 (Dn, Dl, Dr): Accept const as well as const_vector.
4429 (Dz): Likewise. Compare against CONST0_RTX.
4430 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4431 of "vector" where appropriate.
4432 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4433 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4434 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4435 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4436 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4437 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4438 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4439 (v_int_equiv): Extend to SVE modes.
4440 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4442 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4443 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4444 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4445 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4446 (SVE_COND_FP_CMP): New int iterators.
4447 (perm_hilo): Handle the new unpack unspecs.
4448 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4450 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4451 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4452 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4453 (aarch64_equality_operator, aarch64_constant_vector_operand)
4454 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4455 (aarch64_sve_nonimmediate_operand): Likewise.
4456 (aarch64_sve_general_operand): Likewise.
4457 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4458 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4459 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4460 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4461 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4462 (aarch64_sve_float_arith_immediate): Likewise.
4463 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4464 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4465 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4466 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4467 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4468 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4469 (aarch64_sve_float_arith_operand): Likewise.
4470 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4471 (aarch64_sve_float_mul_operand): Likewise.
4472 (aarch64_sve_vec_perm_operand): Likewise.
4473 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4474 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4475 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4476 as well as const_vector.
4477 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4478 in file. Use CONST0_RTX and CONSTM1_RTX.
4479 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4480 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4481 Use aarch64_simd_imm_zero.
4482 * config/aarch64/aarch64-sve.md: New file.
4483 * config/aarch64/aarch64.md: Include it.
4484 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4485 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4486 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4487 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4488 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4489 (sve): New attribute.
4490 (enabled): Disable instructions with the sve attribute unless
4492 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4493 aarch64_expand_mov_immediate.
4494 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4495 CNT[BHSD] immediates.
4496 (movti): Split CONST_POLY_INT moves into two halves.
4497 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4498 Split additions that need a temporary here if the destination
4499 is the stack pointer.
4500 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4501 (*add<mode>3_poly_1): New instruction.
4502 (set_clobber_cc): New expander.
4504 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4506 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4507 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4508 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4509 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4510 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4511 Change innermode from fixed_mode_size to machine_mode.
4512 (simplify_subreg): Update call accordingly. Handle a constant-sized
4513 subreg of a variable-length CONST_VECTOR.
4515 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4516 Alan Hayward <alan.hayward@arm.com>
4517 David Sherwood <david.sherwood@arm.com>
4519 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4520 (add_offset_to_base): New function, split out from...
4521 (create_mem_ref): ...here. When handling a scale other than 1,
4522 check first whether the address is valid without the offset.
4523 Add it into the base if so, leaving the index and scale as-is.
4525 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4528 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4529 fold_for_warn before checking if arg2 is INTEGER_CST.
4531 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4533 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4534 (store_multiple_operation): Delete.
4535 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4536 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4537 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4538 guarded by TARGET_STRING.
4539 (rs6000_output_load_multiple): Delete.
4540 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4541 OPTION_MASK_STRING / TARGET_STRING handling.
4542 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4543 (const rs6000_opt_masks) <"string">: Change mask to 0.
4544 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4545 (MASK_STRING): Delete.
4546 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4548 (load_multiple): Delete.
4555 (store_multiple): Delete.
4562 (movmemsi_8reg): Delete.
4563 (corresponding unnamed define_insn): Delete.
4564 (movmemsi_6reg): Delete.
4565 (corresponding unnamed define_insn): Delete.
4566 (movmemsi_4reg): Delete.
4567 (corresponding unnamed define_insn): Delete.
4568 (movmemsi_2reg): Delete.
4569 (corresponding unnamed define_insn): Delete.
4570 (movmemsi_1reg): Delete.
4571 (corresponding unnamed define_insn): Delete.
4572 * config/rs6000/rs6000.opt (mno-string): New.
4573 (mstring): Replace by deprecation warning stub.
4574 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4576 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4578 * regrename.c (regrename_do_replace): If replacing the same
4579 reg multiple times, try to reuse last created gen_raw_REG.
4582 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4583 main to workaround a bug in GDB.
4585 2018-01-12 Tom de Vries <tom@codesourcery.com>
4588 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4590 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4592 PR rtl-optimization/80481
4593 * ira-color.c (get_cap_member): New function.
4594 (allocnos_conflict_by_live_ranges_p): Use it.
4595 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4596 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4598 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4601 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4602 (*saddl_se_1): Ditto.
4604 (*ssubl_se_1): Ditto.
4606 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4608 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4609 rather than wi::to_widest for DR_INITs.
4610 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4611 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4612 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4614 (vect_analyze_group_access_1): Note that here.
4616 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4618 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4619 polynomial type sizes.
4621 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4623 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4624 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4625 (gimple_add_tmp_var): Likewise.
4627 2018-01-12 Martin Liska <mliska@suse.cz>
4629 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4630 (gimple_alloc_sizes): Likewise.
4631 (dump_gimple_statistics): Use PRIu64 in printf format.
4632 * gimple.h: Change uint64_t to int.
4634 2018-01-12 Martin Liska <mliska@suse.cz>
4636 * tree-core.h: Use uint64_t instead of int.
4637 * tree.c (tree_node_counts): Likewise.
4638 (tree_node_sizes): Likewise.
4639 (dump_tree_statistics): Use PRIu64 in printf format.
4641 2018-01-12 Martin Liska <mliska@suse.cz>
4643 * Makefile.in: As qsort_chk is implemented in vec.c, add
4644 vec.o to linkage of gencfn-macros.
4645 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4646 passing the info to record_node_allocation_statistics.
4647 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4649 * ggc-common.c (struct ggc_usage): Add operator== and use
4650 it in operator< and compare function.
4651 * mem-stats.h (struct mem_usage): Likewise.
4652 * vec.c (struct vec_usage): Remove operator< and compare
4653 function. Can be simply inherited.
4655 2018-01-12 Martin Jambor <mjambor@suse.cz>
4658 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4659 * tree-ssa-math-opts.c: Include domwalk.h.
4660 (convert_mult_to_fma_1): New function.
4661 (fma_transformation_info): New type.
4662 (fma_deferring_state): Likewise.
4663 (cancel_fma_deferring): New function.
4664 (result_of_phi): Likewise.
4665 (last_fma_candidate_feeds_initial_phi): Likewise.
4666 (convert_mult_to_fma): Added deferring logic, split actual
4667 transformation to convert_mult_to_fma_1.
4668 (math_opts_dom_walker): New type.
4669 (math_opts_dom_walker::after_dom_children): New method, body moved
4670 here from pass_optimize_widening_mul::execute, added deferring logic
4672 (pass_optimize_widening_mul::execute): Moved most of code to
4673 math_opts_dom_walker::after_dom_children.
4674 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4675 * config/i386/i386.c (ix86_option_override_internal): Added
4676 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4678 2018-01-12 Richard Biener <rguenther@suse.de>
4681 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4682 inline instance vars.
4684 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4687 * config/rx/rx.c (rx_is_restricted_memory_address):
4690 2018-01-12 Richard Biener <rguenther@suse.de>
4692 PR tree-optimization/80846
4693 * target.def (split_reduction): New target hook.
4694 * targhooks.c (default_split_reduction): New function.
4695 * targhooks.h (default_split_reduction): Declare.
4696 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4697 target requests first reduce vectors by combining low and high
4699 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4700 (get_vectype_for_scalar_type_and_size): Export.
4701 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4702 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4703 * doc/tm.texi: Regenerate.
4704 * config/i386/i386.c (ix86_split_reduction): Implement
4705 TARGET_VECTORIZE_SPLIT_REDUCTION.
4707 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4710 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4711 in PIC mode except for TARGET_VXWORKS_RTP.
4712 * config/sparc/sparc.c: Include cfgrtl.h.
4713 (TARGET_INIT_PIC_REG): Define.
4714 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4715 (sparc_pic_register_p): New predicate.
4716 (sparc_legitimate_address_p): Use it.
4717 (sparc_legitimize_pic_address): Likewise.
4718 (sparc_delegitimize_address): Likewise.
4719 (sparc_mode_dependent_address_p): Likewise.
4720 (gen_load_pcrel_sym): Remove 4th parameter.
4721 (load_got_register): Adjust call to above. Remove obsolete stuff.
4722 (sparc_expand_prologue): Do not call load_got_register here.
4723 (sparc_flat_expand_prologue): Likewise.
4724 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4725 (sparc_use_pseudo_pic_reg): New function.
4726 (sparc_init_pic_reg): Likewise.
4727 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4728 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4730 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4732 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4733 Add item for branch_cost.
4735 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4737 PR rtl-optimization/83565
4738 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4739 not extend the result to a larger mode for rotate operations.
4740 (num_sign_bit_copies1): Likewise.
4742 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4745 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4747 Use values-Xc.o for -pedantic.
4748 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4750 2018-01-12 Martin Liska <mliska@suse.cz>
4753 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4755 (possible_polymorphic_call_targets): Use it.
4756 (ipa_devirt): Likewise.
4758 2018-01-12 Martin Liska <mliska@suse.cz>
4760 * profile-count.h (enum profile_quality): Use 0 as invalid
4761 enum value of profile_quality.
4763 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4765 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4766 -mext-string options.
4768 2018-01-12 Richard Biener <rguenther@suse.de>
4770 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4771 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4772 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4774 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4776 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4778 * configure.ac (--with-long-double-format): Add support for the
4779 configuration option to change the default long double format on
4781 * config.gcc (powerpc*-linux*-*): Likewise.
4782 * configure: Regenerate.
4783 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4784 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4785 used without modification.
4787 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4789 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4790 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4791 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4792 MISC_BUILTIN_SPEC_BARRIER.
4793 (rs6000_init_builtins): Likewise.
4794 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4796 (speculation_barrier): New define_insn.
4797 * doc/extend.texi: Document __builtin_speculation_barrier.
4799 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4802 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4803 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4804 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4806 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4807 integral modes instead of "ss" and "sd".
4808 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4809 vectors with 32-bit and 64-bit elements.
4810 (vecdupssescalarmodesuffix): New mode attribute.
4811 (vec_dup<mode>): Use it.
4813 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4816 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4817 frame if argument is passed on stack.
4819 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4822 * ree.c (combine_reaching_defs): Optimize also
4823 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4824 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4826 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4829 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4831 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4834 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4835 after they are computed.
4837 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4839 PR tree-optimization/83695
4840 * gimple-loop-linterchange.cc
4841 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4842 reset cached scev information after interchange.
4843 (pass_linterchange::execute): Remove call to scev_reset_htab.
4845 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4847 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4848 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4849 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4850 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4851 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4852 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4853 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4854 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4855 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4856 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4857 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4858 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4859 (V_lane_reg): Likewise.
4860 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4862 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4863 (vfmal_lane_low<mode>_intrinsic,
4864 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4865 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4866 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4867 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4868 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4869 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4871 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4873 * config/arm/arm-cpus.in (fp16fml): New feature.
4874 (ALL_SIMD): Add fp16fml.
4875 (armv8.2-a): Add fp16fml as an option.
4876 (armv8.3-a): Likewise.
4877 (armv8.4-a): Add fp16fml as part of fp16.
4878 * config/arm/arm.h (TARGET_FP16FML): Define.
4879 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4881 * config/arm/arm-modes.def (V2HF): Define.
4882 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4883 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4884 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4885 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4886 vfmsl_low, vfmsl_high): New set of builtins.
4887 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4888 (vfml_op): New code attribute.
4889 (VFMLHALVES): New int iterator.
4890 (VFML, VFMLSEL): New mode attributes.
4891 (V_reg): Define mapping for V2HF.
4892 (V_hi, V_lo): New mode attributes.
4893 (VF_constraint): Likewise.
4894 (vfml_half, vfml_half_selector): New int attributes.
4895 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4897 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4898 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4900 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4901 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4902 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4903 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4905 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4906 Document new effective target and option set.
4908 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4910 * config/arm/arm-cpus.in (armv8_4): New feature.
4911 (ARMv8_4a): New fgroup.
4912 (armv8.4-a): New arch.
4913 * config/arm/arm-tables.opt: Regenerate.
4914 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4915 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4916 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4917 Add matching rules for -march=armv8.4-a and extensions.
4918 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4920 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4923 * config/rx/rx.md (BW): New mode attribute.
4924 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4926 2018-01-11 Richard Biener <rguenther@suse.de>
4928 PR tree-optimization/83435
4929 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4930 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4931 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4933 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4934 Alan Hayward <alan.hayward@arm.com>
4935 David Sherwood <david.sherwood@arm.com>
4937 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4939 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4940 (aarch64_print_address_internal): Use it to check for a zero offset.
4942 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4943 Alan Hayward <alan.hayward@arm.com>
4944 David Sherwood <david.sherwood@arm.com>
4946 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4947 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4948 Return a poly_int64 rather than a HOST_WIDE_INT.
4949 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4950 rather than a HOST_WIDE_INT.
4951 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4952 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4953 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4954 final_offset from HOST_WIDE_INT to poly_int64.
4955 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4956 to_constant when getting the number of units in an Advanced SIMD
4958 (aarch64_builtin_vectorized_function): Check for a constant number
4960 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4962 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4963 attribute instead of GET_MODE_NUNITS.
4964 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4965 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4966 GET_MODE_SIZE for fixed-size registers.
4967 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4968 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4969 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4970 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4971 (aarch64_print_operand, aarch64_print_address_internal)
4972 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4973 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4974 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4975 Handle polynomial GET_MODE_SIZE.
4976 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4977 wider than SImode without modification.
4978 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4979 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4980 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4981 passing and returning SVE modes.
4982 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4983 rather than GEN_INT.
4984 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4985 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4986 (aarch64_allocate_and_probe_stack_space): Likewise.
4987 (aarch64_layout_frame): Cope with polynomial offsets.
4988 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4989 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4991 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4992 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4993 poly_int64 rather than a HOST_WIDE_INT.
4994 (aarch64_get_separate_components, aarch64_process_components)
4995 (aarch64_expand_prologue, aarch64_expand_epilogue)
4996 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4997 (aarch64_anchor_offset): New function, split out from...
4998 (aarch64_legitimize_address): ...here.
4999 (aarch64_builtin_vectorization_cost): Handle polynomial
5000 TYPE_VECTOR_SUBPARTS.
5001 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5003 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5004 number of elements from the PARALLEL rather than the mode.
5005 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5006 rather than GET_MODE_BITSIZE.
5007 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5008 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5009 (aarch64_expand_vec_perm_const_1): Handle polynomial
5010 d->perm.length () and d->perm elements.
5011 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5012 Apply to_constant to d->perm elements.
5013 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5014 polynomial CONST_VECTOR_NUNITS.
5015 (aarch64_move_pointer): Take amount as a poly_int64 rather
5017 (aarch64_progress_pointer): Avoid temporary variable.
5018 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5019 the mode attribute instead of GET_MODE.
5021 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5022 Alan Hayward <alan.hayward@arm.com>
5023 David Sherwood <david.sherwood@arm.com>
5025 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5026 x exists before using it.
5027 (aarch64_add_constant_internal): Rename to...
5028 (aarch64_add_offset_1): ...this. Replace regnum with separate
5029 src and dest rtxes. Handle the case in which they're different,
5030 including when the offset is zero. Replace scratchreg with an rtx.
5031 Use 2 additions if there is no spare register into which we can
5032 move a 16-bit constant.
5033 (aarch64_add_constant): Delete.
5034 (aarch64_add_offset): Replace reg with separate src and dest
5035 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5036 Use aarch64_add_offset_1.
5037 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5038 an rtx rather than an int. Take the delta as a poly_int64
5039 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5040 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5041 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5042 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5043 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5045 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5046 aarch64_add_constant.
5048 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5050 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5051 Use scalar_float_mode.
5053 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5055 * config/aarch64/aarch64-simd.md
5056 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5057 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5058 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5059 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5060 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5061 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5062 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5063 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5064 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5065 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5067 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5070 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5071 targ_options->x_arm_arch_string is non NULL.
5073 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5075 * config/aarch64/aarch64.h
5076 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5078 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5081 * expmed.c (emit_store_flag_force): Swap if const op0
5082 and change VOIDmode to mode of op0.
5084 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5086 PR rtl-optimization/83761
5087 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5088 than bytes to mode_for_size.
5090 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5093 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5094 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5097 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5100 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5101 when in layout mode.
5102 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5103 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5106 2018-01-10 Michael Collison <michael.collison@arm.com>
5108 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5109 * config/aarch64/aarch64-option-extension.def: Add
5110 AARCH64_OPT_EXTENSION of 'fp16fml'.
5111 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5112 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5113 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5114 * config/aarch64/constraints.md (Ui7): New constraint.
5115 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5116 (VFMLA_SEL_W): Ditto.
5119 (VFMLA16_LOW): New int iterator.
5120 (VFMLA16_HIGH): Ditto.
5121 (UNSPEC_FMLAL): New unspec.
5122 (UNSPEC_FMLSL): Ditto.
5123 (UNSPEC_FMLAL2): Ditto.
5124 (UNSPEC_FMLSL2): Ditto.
5125 (f16mac): New code attribute.
5126 * config/aarch64/aarch64-simd-builtins.def
5127 (aarch64_fmlal_lowv2sf): Ditto.
5128 (aarch64_fmlsl_lowv2sf): Ditto.
5129 (aarch64_fmlalq_lowv4sf): Ditto.
5130 (aarch64_fmlslq_lowv4sf): Ditto.
5131 (aarch64_fmlal_highv2sf): Ditto.
5132 (aarch64_fmlsl_highv2sf): Ditto.
5133 (aarch64_fmlalq_highv4sf): Ditto.
5134 (aarch64_fmlslq_highv4sf): Ditto.
5135 (aarch64_fmlal_lane_lowv2sf): Ditto.
5136 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5137 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5138 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5139 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5140 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5141 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5142 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5143 (aarch64_fmlal_lane_highv2sf): Ditto.
5144 (aarch64_fmlsl_lane_highv2sf): Ditto.
5145 (aarch64_fmlal_laneq_highv2sf): Ditto.
5146 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5147 (aarch64_fmlalq_lane_highv4sf): Ditto.
5148 (aarch64_fmlsl_lane_highv4sf): Ditto.
5149 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5150 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5151 * config/aarch64/aarch64-simd.md:
5152 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5153 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5154 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5155 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5156 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5157 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5158 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5159 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5160 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5161 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5162 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5163 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5164 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5165 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5166 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5167 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5168 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5169 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5170 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5171 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5172 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5173 (vfmlsl_low_u32): Ditto.
5174 (vfmlalq_low_u32): Ditto.
5175 (vfmlslq_low_u32): Ditto.
5176 (vfmlal_high_u32): Ditto.
5177 (vfmlsl_high_u32): Ditto.
5178 (vfmlalq_high_u32): Ditto.
5179 (vfmlslq_high_u32): Ditto.
5180 (vfmlal_lane_low_u32): Ditto.
5181 (vfmlsl_lane_low_u32): Ditto.
5182 (vfmlal_laneq_low_u32): Ditto.
5183 (vfmlsl_laneq_low_u32): Ditto.
5184 (vfmlalq_lane_low_u32): Ditto.
5185 (vfmlslq_lane_low_u32): Ditto.
5186 (vfmlalq_laneq_low_u32): Ditto.
5187 (vfmlslq_laneq_low_u32): Ditto.
5188 (vfmlal_lane_high_u32): Ditto.
5189 (vfmlsl_lane_high_u32): Ditto.
5190 (vfmlal_laneq_high_u32): Ditto.
5191 (vfmlsl_laneq_high_u32): Ditto.
5192 (vfmlalq_lane_high_u32): Ditto.
5193 (vfmlslq_lane_high_u32): Ditto.
5194 (vfmlalq_laneq_high_u32): Ditto.
5195 (vfmlslq_laneq_high_u32): Ditto.
5196 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5197 (AARCH64_FL_FOR_ARCH8_4): New.
5198 (AARCH64_ISA_F16FML): New ISA flag.
5199 (TARGET_F16FML): New feature flag for fp16fml.
5200 (doc/invoke.texi): Document new fp16fml option.
5202 2018-01-10 Michael Collison <michael.collison@arm.com>
5204 * config/aarch64/aarch64-builtins.c:
5205 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5206 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5207 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5208 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5209 (AARCH64_ISA_SHA3): New ISA flag.
5210 (TARGET_SHA3): New feature flag for sha3.
5211 * config/aarch64/iterators.md (sha512_op): New int attribute.
5212 (CRYPTO_SHA512): New int iterator.
5213 (UNSPEC_SHA512H): New unspec.
5214 (UNSPEC_SHA512H2): Ditto.
5215 (UNSPEC_SHA512SU0): Ditto.
5216 (UNSPEC_SHA512SU1): Ditto.
5217 * config/aarch64/aarch64-simd-builtins.def
5218 (aarch64_crypto_sha512hqv2di): New builtin.
5219 (aarch64_crypto_sha512h2qv2di): Ditto.
5220 (aarch64_crypto_sha512su0qv2di): Ditto.
5221 (aarch64_crypto_sha512su1qv2di): Ditto.
5222 (aarch64_eor3qv8hi): Ditto.
5223 (aarch64_rax1qv2di): Ditto.
5224 (aarch64_xarqv2di): Ditto.
5225 (aarch64_bcaxqv8hi): Ditto.
5226 * config/aarch64/aarch64-simd.md:
5227 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5228 (aarch64_crypto_sha512su0qv2di): Ditto.
5229 (aarch64_crypto_sha512su1qv2di): Ditto.
5230 (aarch64_eor3qv8hi): Ditto.
5231 (aarch64_rax1qv2di): Ditto.
5232 (aarch64_xarqv2di): Ditto.
5233 (aarch64_bcaxqv8hi): Ditto.
5234 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5235 (vsha512h2q_u64): Ditto.
5236 (vsha512su0q_u64): Ditto.
5237 (vsha512su1q_u64): Ditto.
5238 (veor3q_u16): Ditto.
5239 (vrax1q_u64): Ditto.
5241 (vbcaxq_u16): Ditto.
5242 * config/arm/types.md (crypto_sha512): New type attribute.
5243 (crypto_sha3): Ditto.
5244 (doc/invoke.texi): Document new sha3 option.
5246 2018-01-10 Michael Collison <michael.collison@arm.com>
5248 * config/aarch64/aarch64-builtins.c:
5249 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5250 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5251 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5252 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5253 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5254 (AARCH64_ISA_SM4): New ISA flag.
5255 (TARGET_SM4): New feature flag for sm4.
5256 * config/aarch64/aarch64-simd-builtins.def
5257 (aarch64_sm3ss1qv4si): Ditto.
5258 (aarch64_sm3tt1aq4si): Ditto.
5259 (aarch64_sm3tt1bq4si): Ditto.
5260 (aarch64_sm3tt2aq4si): Ditto.
5261 (aarch64_sm3tt2bq4si): Ditto.
5262 (aarch64_sm3partw1qv4si): Ditto.
5263 (aarch64_sm3partw2qv4si): Ditto.
5264 (aarch64_sm4eqv4si): Ditto.
5265 (aarch64_sm4ekeyqv4si): Ditto.
5266 * config/aarch64/aarch64-simd.md:
5267 (aarch64_sm3ss1qv4si): Ditto.
5268 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5269 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5270 (aarch64_sm4eqv4si): Ditto.
5271 (aarch64_sm4ekeyqv4si): Ditto.
5272 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5273 (sm3part_op): Ditto.
5274 (CRYPTO_SM3TT): Ditto.
5275 (CRYPTO_SM3PART): Ditto.
5276 (UNSPEC_SM3SS1): New unspec.
5277 (UNSPEC_SM3TT1A): Ditto.
5278 (UNSPEC_SM3TT1B): Ditto.
5279 (UNSPEC_SM3TT2A): Ditto.
5280 (UNSPEC_SM3TT2B): Ditto.
5281 (UNSPEC_SM3PARTW1): Ditto.
5282 (UNSPEC_SM3PARTW2): Ditto.
5283 (UNSPEC_SM4E): Ditto.
5284 (UNSPEC_SM4EKEY): Ditto.
5285 * config/aarch64/constraints.md (Ui2): New constraint.
5286 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5287 * config/arm/types.md (crypto_sm3): New type attribute.
5288 (crypto_sm4): Ditto.
5289 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5290 (vsm3tt1aq_u32): Ditto.
5291 (vsm3tt1bq_u32): Ditto.
5292 (vsm3tt2aq_u32): Ditto.
5293 (vsm3tt2bq_u32): Ditto.
5294 (vsm3partw1q_u32): Ditto.
5295 (vsm3partw2q_u32): Ditto.
5296 (vsm4eq_u32): Ditto.
5297 (vsm4ekeyq_u32): Ditto.
5298 (doc/invoke.texi): Document new sm4 option.
5300 2018-01-10 Michael Collison <michael.collison@arm.com>
5302 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5303 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5304 (AARCH64_FL_FOR_ARCH8_4): New.
5305 (AARCH64_FL_V8_4): New flag.
5306 (doc/invoke.texi): Document new armv8.4-a option.
5308 2018-01-10 Michael Collison <michael.collison@arm.com>
5310 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5311 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5312 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5313 * config/aarch64/aarch64-option-extension.def: Add
5314 AARCH64_OPT_EXTENSION of 'sha2'.
5315 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5316 (crypto): Disable sha2 and aes if crypto disabled.
5317 (crypto): Enable aes and sha2 if enabled.
5318 (simd): Disable sha2 and aes if simd disabled.
5319 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5321 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5322 (TARGET_SHA2): New feature flag for sha2.
5323 (TARGET_AES): New feature flag for aes.
5324 * config/aarch64/aarch64-simd.md:
5325 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5326 conditional on TARGET_AES.
5327 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5328 (aarch64_crypto_sha1hsi): Make pattern conditional
5330 (aarch64_crypto_sha1hv4si): Ditto.
5331 (aarch64_be_crypto_sha1hv4si): Ditto.
5332 (aarch64_crypto_sha1su1v4si): Ditto.
5333 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5334 (aarch64_crypto_sha1su0v4si): Ditto.
5335 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5336 (aarch64_crypto_sha256su0v4si): Ditto.
5337 (aarch64_crypto_sha256su1v4si): Ditto.
5338 (doc/invoke.texi): Document new aes and sha2 options.
5340 2018-01-10 Martin Sebor <msebor@redhat.com>
5342 PR tree-optimization/83781
5343 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5346 2018-01-11 Martin Sebor <msebor@gmail.com>
5347 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5349 PR tree-optimization/83501
5350 PR tree-optimization/81703
5352 * tree-ssa-strlen.c (get_string_cst): Rename...
5353 (get_string_len): ...to this. Handle global constants.
5354 (handle_char_store): Adjust.
5356 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5357 Jim Wilson <jimw@sifive.com>
5359 * config/riscv/riscv-protos.h (riscv_output_return): New.
5360 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5361 (riscv_attribute_table, riscv_output_return),
5362 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5363 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5364 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5366 (riscv_expand_prologue): Add early return for naked function.
5367 (riscv_expand_epilogue): Likewise.
5368 (riscv_function_ok_for_sibcall): Return false for naked function.
5369 (riscv_set_current_function): New.
5370 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5371 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5372 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5373 * doc/extend.texi (RISC-V Function Attributes): New.
5375 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5377 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5378 check for 128-bit long double before checking TCmode.
5379 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5380 128-bit long doubles before checking TFmode or TCmode.
5381 (FLOAT128_IBM_P): Likewise.
5383 2018-01-10 Martin Sebor <msebor@redhat.com>
5385 PR tree-optimization/83671
5386 * builtins.c (c_strlen): Unconditionally return zero for the empty
5388 Use -Warray-bounds for warnings.
5389 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5390 for non-constant array indices with COMPONENT_REF, arrays of
5391 arrays, and pointers to arrays.
5392 (gimple_fold_builtin_strlen): Determine and set length range for
5393 non-constant character arrays.
5395 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5398 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5401 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5403 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5405 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5408 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5409 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5410 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5411 indexed_or_indirect_operand predicate.
5412 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5413 (*vsx_le_perm_load_v8hi): Likewise.
5414 (*vsx_le_perm_load_v16qi): Likewise.
5415 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5416 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5417 (*vsx_le_perm_store_v8hi): Likewise.
5418 (*vsx_le_perm_store_v16qi): Likewise.
5419 (eight unnamed splitters): Likewise.
5421 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5423 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5424 * config/rs6000/emmintrin.h: Likewise.
5425 * config/rs6000/mmintrin.h: Likewise.
5426 * config/rs6000/xmmintrin.h: Likewise.
5428 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5431 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5433 * tree.c (tree_nop_conversion): Return true for location wrapper
5435 (maybe_wrap_with_location): New function.
5436 (selftest::check_strip_nops): New function.
5437 (selftest::test_location_wrappers): New function.
5438 (selftest::tree_c_tests): Call it.
5439 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5440 (maybe_wrap_with_location): New decl.
5441 (EXPR_LOCATION_WRAPPER_P): New macro.
5442 (location_wrapper_p): New inline function.
5443 (tree_strip_any_location_wrapper): New inline function.
5445 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5448 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5449 stack_realign_offset for the largest alignment of stack slot
5451 (ix86_find_max_used_stack_alignment): New function.
5452 (ix86_finalize_stack_frame_flags): Use it. Set
5453 max_used_stack_alignment if we don't realign stack.
5454 * config/i386/i386.h (machine_function): Add
5455 max_used_stack_alignment.
5457 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5459 * config/arm/arm.opt (-mbranch-cost): New option.
5460 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5463 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5466 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5467 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5469 2018-01-10 Richard Biener <rguenther@suse.de>
5472 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5473 early out so it also covers the case where we have a non-NULL
5476 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5478 PR tree-optimization/83753
5479 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5480 for non-strided grouped accesses if the number of elements is 1.
5482 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5485 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5486 * i386.h (TARGET_USE_GATHER): Define.
5487 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5489 2018-01-10 Martin Liska <mliska@suse.cz>
5492 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5493 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5495 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5496 CLEANUP_NO_PARTITIONING is not set.
5498 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5500 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5501 for vectors, as a partial revert of r254296.
5502 * rtl.h (const_vec_p): Delete.
5503 (const_vec_duplicate_p): Don't test for vector CONSTs.
5504 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5505 * expmed.c (make_tree): Likewise.
5508 * common.md (E, F): Use CONSTANT_P instead of checking for
5510 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5511 checking for CONST_VECTOR.
5513 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5516 * predict.c (force_edge_cold): Handle in more sane way edges
5519 2018-01-09 Carl Love <cel@us.ibm.com>
5521 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5523 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5524 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5525 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5526 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5527 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5528 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5529 * config/rs6000/rs6000-protos.h: Add extern defition for
5530 rs6000_generate_float2_double_code.
5531 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5533 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5534 (float2_v2df): Add define_expand.
5536 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5539 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5540 op_mode in the force_to_mode call.
5542 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5544 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5545 instead of checking each element individually.
5546 (aarch64_evpc_uzp): Likewise.
5547 (aarch64_evpc_zip): Likewise.
5548 (aarch64_evpc_ext): Likewise.
5549 (aarch64_evpc_rev): Likewise.
5550 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5551 instead of checking each element individually. Return true without
5553 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5554 whether all selected elements come from the same input, instead of
5555 checking each element individually. Remove calls to gen_rtx_REG,
5556 start_sequence and end_sequence and instead assert that no rtl is
5559 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5561 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5562 order of HIGH and CONST checks.
5564 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5566 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5567 if the destination isn't an SSA_NAME.
5569 2018-01-09 Richard Biener <rguenther@suse.de>
5571 PR tree-optimization/83668
5572 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5574 (canonicalize_loop_form): ... here, renamed from ...
5575 (canonicalize_loop_closed_ssa_form): ... this and amended to
5576 swap successor edges for loop exit blocks to make us use
5577 the RPO order we need for initial schedule generation.
5579 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5581 PR tree-optimization/64811
5582 * match.pd: When optimizing comparisons with Inf, avoid
5583 introducing or losing exceptions from comparisons with NaN.
5585 2018-01-09 Martin Liska <mliska@suse.cz>
5588 * asan.c (shadow_mem_size): Add gcc_assert.
5590 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5592 Don't save registers in main().
5595 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5596 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5597 * config/avr/avr.c (avr_set_current_function): Don't error if
5598 naked, OS_task or OS_main are specified at the same time.
5599 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5601 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5603 * common/config/avr/avr-common.c (avr_option_optimization_table):
5604 Switch on -mmain-is-OS_task for optimizing compilations.
5606 2018-01-09 Richard Biener <rguenther@suse.de>
5608 PR tree-optimization/83572
5609 * graphite.c: Include cfganal.h.
5610 (graphite_transform_loops): Connect infinite loops to exit
5611 and remove fake edges at the end.
5613 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5615 * ipa-inline.c (edge_badness): Revert accidental checkin.
5617 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5620 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5621 symbols; not inline clones.
5623 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5626 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5627 hard registers. Formatting fixes.
5629 PR preprocessor/83722
5630 * gcc.c (try_generate_repro): Pass
5631 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5632 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5635 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5636 Kito Cheng <kito.cheng@gmail.com>
5638 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5639 (riscv_leaf_function_p): Delete.
5640 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5642 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5644 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5646 (do_ifelse): New function.
5647 (do_isel): New function.
5648 (do_sub3): New function.
5649 (do_add3): New function.
5650 (do_load_mask_compare): New function.
5651 (do_overlap_load_compare): New function.
5652 (expand_compare_loop): New function.
5653 (expand_block_compare): Call expand_compare_loop() when appropriate.
5654 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5656 (-mblock-compare-inline-loop-limit): New option.
5658 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5661 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5662 Reverse order of second and third operands in first alternative.
5663 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5664 of first and second elements in UNSPEC_VPERMR vector.
5665 (altivec_expand_vec_perm_le): Likewise.
5667 2018-01-08 Jeff Law <law@redhat.com>
5669 PR rtl-optimizatin/81308
5670 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5671 (process_switch): If group_case_labels makes a change, then set
5673 (pass_convert_switch::execute): If a switch is converted, then
5674 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5676 PR rtl-optimization/81308
5677 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5680 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5682 PR target/83663 - Revert r255946
5683 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5684 generation for cases where splatting a value is not useful.
5685 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5686 across a vec_duplicate and a paradoxical subreg forming a vector
5687 mode to a vec_concat.
5689 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5691 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5692 -march=armv8.3-a variants.
5693 * config/arm/t-multilib: Likewise.
5694 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5696 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5698 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5700 (cceq_ior_compare_complement): Give it a name so I can use it, and
5701 change boolean_or_operator predicate to boolean_operator so it can
5702 be used to generate a crand.
5703 (eqne): New code iterator.
5704 (bd/bd_neg): New code_attrs.
5705 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5706 a single define_insn.
5707 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5708 decrement (bdnzt/bdnzf/bdzt/bdzf).
5709 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5710 with the new names of the branch decrement patterns, and added the
5711 names of the branch decrement conditional patterns.
5713 2018-01-08 Richard Biener <rguenther@suse.de>
5715 PR tree-optimization/83563
5716 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5719 2018-01-08 Richard Biener <rguenther@suse.de>
5722 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5724 2018-01-08 Richard Biener <rguenther@suse.de>
5726 PR tree-optimization/83685
5727 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5728 references to abnormals.
5730 2018-01-08 Richard Biener <rguenther@suse.de>
5733 * dwarf2out.c (output_indirect_strings): Handle empty
5734 skeleton_debug_str_hash.
5735 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5737 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5739 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5740 (emit_store_direct): Likewise.
5741 (arc_trampoline_adjust_address): Likewise.
5742 (arc_asm_trampoline_template): New function.
5743 (arc_initialize_trampoline): Use asm_trampoline_template.
5744 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5745 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5746 * config/arc/arc.md (flush_icache): Delete pattern.
5748 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5750 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5751 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5754 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5757 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5758 by not USED_FOR_TARGET.
5759 (make_pass_resolve_sw_modes): Likewise.
5761 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5763 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5766 2018-01-08 Richard Biener <rguenther@suse.de>
5769 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5771 2018-01-08 Richard Biener <rguenther@suse.de>
5774 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5776 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5779 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5780 basic blocks with a small number of successors.
5781 (convert_control_dep_chain_into_preds): Improve handling of
5783 (dump_predicates): Split apart into...
5784 (dump_pred_chain): ...here...
5785 (dump_pred_info): ...and here.
5786 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5787 (can_chain_union_be_invalidated_p): Improve check for invalidation
5789 (uninit_uses_cannot_happen): Avoid unnecessary if
5790 convert_control_dep_chain_into_preds yielded nothing.
5792 2018-01-06 Martin Sebor <msebor@redhat.com>
5794 PR tree-optimization/83640
5795 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5796 subtracting negative offset from size.
5797 (builtin_access::overlap): Adjust offset bounds of the access to fall
5798 within the size of the object if possible.
5800 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5802 PR rtl-optimization/83699
5803 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5804 extract_bit_field_as_subreg to cases in which the extracted
5805 value is also a vector.
5807 * lra-constraints.c (process_alt_operands): Test for the equivalence
5808 substitutions when detecting a possible reload cycle.
5810 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5813 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5814 by default if flag_selective_schedling{,2}. Formatting fixes.
5816 PR rtl-optimization/83682
5817 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5818 if it has non-VECTOR_MODE element mode.
5819 (vec_duplicate_p): Likewise.
5822 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5823 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5825 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5828 * config/i386/i386-builtin.def
5829 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5830 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5831 Require also OPTION_MASK_ISA_AVX512F in addition to
5832 OPTION_MASK_ISA_GFNI.
5833 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5834 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5835 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5836 to OPTION_MASK_ISA_GFNI.
5837 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5838 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5839 OPTION_MASK_ISA_AVX512BW.
5840 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5841 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5842 addition to OPTION_MASK_ISA_GFNI.
5843 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5844 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5845 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5846 to OPTION_MASK_ISA_GFNI.
5847 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5848 a requirement for all ISAs rather than any of them with a few
5850 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5852 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5853 bitmasks to be enabled with 3 exceptions, instead of requiring any
5854 enabled ISA with lots of exceptions.
5855 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5856 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5857 Change avx512bw in isa attribute to avx512f.
5858 * config/i386/sgxintrin.h: Add license boilerplate.
5859 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5860 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5861 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5862 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5864 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5865 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5866 temporarily sse2 rather than sse if not enabled already.
5869 * config/i386/sse.md (VI248_VLBW): Rename to ...
5870 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5871 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5872 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5873 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5874 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5875 mode iterator instead of VI248_VLBW.
5877 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5879 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5880 (record_modified): Skip clobbers; add debug output.
5881 (param_change_prob): Use sreal frequencies.
5883 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5885 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5886 punt for user-aligned variables.
5888 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5890 * tree-chrec.c (chrec_contains_symbols): Return true for
5893 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5896 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5897 of (x|y) == x for BICS pattern.
5899 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5901 PR tree-optimization/83605
5902 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5903 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5906 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5908 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5909 * config/epiphany/rtems.h: New file.
5911 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5912 Uros Bizjak <ubizjak@gmail.com>
5915 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5916 QIreg_operand instead of register_operand predicate.
5917 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5918 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5919 comments instead of -fmitigate[-_]rop.
5921 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5924 * cgraphunit.c (symbol_table::compile): Switch to text_section
5925 before calling assembly_start debug hook.
5926 * run-rtl-passes.c (run_rtl_passes): Likewise.
5929 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5931 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5932 range_int_cst_p rather than !symbolic_range_p before calling
5933 extract_range_from_multiplicative_op_1.
5935 2018-01-04 Jeff Law <law@redhat.com>
5937 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5938 redundant test in assertion.
5940 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5942 * doc/rtl.texi: Document machine_mode wrapper classes.
5944 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5946 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5949 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5951 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5952 the VEC_PERM_EXPR fold to fail.
5954 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5957 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5958 to switched_sections.
5960 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5963 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5966 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5969 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5970 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5972 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5975 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5976 is BLKmode and bitpos not zero or mode change is needed.
5978 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5981 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5984 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5987 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5988 instead of MULT rtx. Update all corresponding splitters.
5990 (*ssub<modesuffix>): Ditto.
5992 (*cmp_sadd_di): Update split patterns.
5993 (*cmp_sadd_si): Ditto.
5994 (*cmp_sadd_sidi): Ditto.
5995 (*cmp_ssub_di): Ditto.
5996 (*cmp_ssub_si): Ditto.
5997 (*cmp_ssub_sidi): Ditto.
5998 * config/alpha/predicates.md (const23_operand): New predicate.
5999 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6000 Look for ASHIFT, not MULT inner operand.
6001 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6003 2018-01-04 Martin Liska <mliska@suse.cz>
6005 PR gcov-profile/83669
6006 * gcov.c (output_intermediate_file): Add version to intermediate
6008 * doc/gcov.texi: Document new field 'version' in intermediate
6009 file format. Fix location of '-k' option of gcov command.
6011 2018-01-04 Martin Liska <mliska@suse.cz>
6014 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6016 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6018 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6020 2018-01-03 Martin Sebor <msebor@redhat.com>
6022 PR tree-optimization/83655
6023 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6024 checking calls with invalid arguments.
6026 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6028 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6029 (vectorizable_mask_load_store): Delete.
6030 (vectorizable_call): Return false for masked loads and stores.
6031 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6032 instead of gimple_assign_rhs1.
6033 (vectorizable_load): Handle IFN_MASK_LOAD.
6034 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6036 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6038 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6040 (vectorizable_mask_load_store): ...here.
6041 (vectorizable_load): ...and here.
6043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6045 * tree-vect-stmts.c (vect_build_all_ones_mask)
6046 (vect_build_zero_merge_argument): New functions, split out from...
6047 (vectorizable_load): ...here.
6049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6051 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6053 (vectorizable_mask_load_store): ...here.
6054 (vectorizable_store): ...and here.
6056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6058 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6060 (vectorizable_mask_load_store): ...here.
6062 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6064 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6065 (vect_model_store_cost): Take a vec_load_store_type instead of a
6067 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6068 (vect_model_store_cost): Take a vec_load_store_type instead of a
6070 (vectorizable_mask_load_store): Update accordingly.
6071 (vectorizable_store): Likewise.
6072 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6074 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6076 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6077 IFN_MASK_LOAD calls here rather than...
6078 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6080 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6081 Alan Hayward <alan.hayward@arm.com>
6082 David Sherwood <david.sherwood@arm.com>
6084 * expmed.c (extract_bit_field_1): For vector extracts,
6085 fall back to extract_bit_field_as_subreg if vec_extract
6088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6089 Alan Hayward <alan.hayward@arm.com>
6090 David Sherwood <david.sherwood@arm.com>
6092 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6093 they are variable or constant sized.
6094 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6095 slots for constant-sized data.
6097 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6098 Alan Hayward <alan.hayward@arm.com>
6099 David Sherwood <david.sherwood@arm.com>
6101 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6102 handling COND_EXPRs with boolean comparisons, try to find a better
6103 basis for the mask type than the boolean itself.
6105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6107 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6108 is calculated and how it can be overridden.
6109 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6110 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6112 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6116 Alan Hayward <alan.hayward@arm.com>
6117 David Sherwood <david.sherwood@arm.com>
6119 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6120 Remove the mode argument.
6121 (aarch64_simd_valid_immediate): Remove the mode and inverse
6123 * config/aarch64/iterators.md (bitsize): New iterator.
6124 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6125 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6126 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6127 aarch64_simd_valid_immediate.
6128 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6129 (aarch64_reg_or_bic_imm): Likewise.
6130 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6131 with an insn_type enum and msl with a modifier_type enum.
6132 Replace element_width with a scalar_mode. Change the shift
6133 to unsigned int. Add constructors for scalar_float_mode and
6134 scalar_int_mode elements.
6135 (aarch64_vect_float_const_representable_p): Delete.
6136 (aarch64_can_const_movi_rtx_p)
6137 (aarch64_simd_scalar_immediate_valid_for_move)
6138 (aarch64_simd_make_constant): Update call to
6139 aarch64_simd_valid_immediate.
6140 (aarch64_advsimd_valid_immediate_hs): New function.
6141 (aarch64_advsimd_valid_immediate): Likewise.
6142 (aarch64_simd_valid_immediate): Remove mode and inverse
6143 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6144 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6145 and aarch64_float_const_representable_p on the result.
6146 (aarch64_output_simd_mov_immediate): Remove mode argument.
6147 Update call to aarch64_simd_valid_immediate and use of
6148 simd_immediate_info.
6149 (aarch64_output_scalar_simd_mov_immediate): Update call
6152 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6153 Alan Hayward <alan.hayward@arm.com>
6154 David Sherwood <david.sherwood@arm.com>
6156 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6157 (mode_nunits): Likewise CONST_MODE_NUNITS.
6158 * machmode.def (ADJUST_NUNITS): Document.
6159 * genmodes.c (mode_data::need_nunits_adj): New field.
6160 (blank_mode): Update accordingly.
6161 (adj_nunits): New variable.
6162 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6164 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6165 listed in adj_nunits.
6166 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6167 listed in adj_nunits. Don't emit case statements for such modes.
6168 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6169 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6170 nothing if adj_nunits is nonnull.
6171 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6172 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6173 (emit_mode_fbit): Update use of print_maybe_const_decl.
6174 (emit_move_size): Likewise. Treat the array as non-const
6176 (emit_mode_adjustments): Handle adj_nunits.
6178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6180 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6181 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6182 (VECTOR_MODES): Use it.
6183 (make_vector_modes): Take the prefix as an argument.
6185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6186 Alan Hayward <alan.hayward@arm.com>
6187 David Sherwood <david.sherwood@arm.com>
6189 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6190 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6191 for MODE_VECTOR_BOOL.
6192 * machmode.def (VECTOR_BOOL_MODE): Document.
6193 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6194 (make_vector_bool_mode): New function.
6195 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6197 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6198 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6200 * stor-layout.c (int_mode_for_mode): Likewise.
6201 * tree.c (build_vector_type_for_mode): Likewise.
6202 * varasm.c (output_constant_pool_2): Likewise.
6203 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6204 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6205 for MODE_VECTOR_BOOL.
6206 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6207 of mode class checks.
6208 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6209 instead of a list of mode class checks.
6210 (expand_vector_scalar_condition): Likewise.
6211 (type_for_widest_vector_mode): Handle BImode as an inner mode.
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_size): Change from unsigned short to
6219 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6220 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6221 or if measurement_type is not polynomial.
6222 (fixed_size_mode::includes_p): Check for constant-sized modes.
6223 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6224 return a poly_uint16 rather than an unsigned short.
6225 (emit_mode_size): Change the type of mode_size from unsigned short
6226 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6227 (emit_mode_adjustments): Cope with polynomial vector sizes.
6228 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6230 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6232 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6233 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6234 * caller-save.c (setup_save_areas): Likewise.
6235 (replace_reg_with_saved_mem): Likewise.
6236 * calls.c (emit_library_call_value_1): Likewise.
6237 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6238 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6239 (gen_lowpart_for_combine): Likewise.
6240 * convert.c (convert_to_integer_1): Likewise.
6241 * cse.c (equiv_constant, cse_insn): Likewise.
6242 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6243 (cselib_subst_to_values): Likewise.
6244 * dce.c (word_dce_process_block): Likewise.
6245 * df-problems.c (df_word_lr_mark_ref): Likewise.
6246 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6247 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6248 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6249 (rtl_for_decl_location): Likewise.
6250 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6251 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6252 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6253 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6254 (expand_expr_real_1): Likewise.
6255 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6256 (pad_below): Likewise.
6257 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6258 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6259 * ira.c (get_subreg_tracking_sizes): Likewise.
6260 * ira-build.c (ira_create_allocno_objects): Likewise.
6261 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6262 (ira_sort_regnos_for_alter_reg): Likewise.
6263 * ira-costs.c (record_operand_costs): Likewise.
6264 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6265 (resolve_simple_move): Likewise.
6266 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6267 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6268 (lra_constraints): Likewise.
6269 (CONST_POOL_OK_P): Reject variable-sized modes.
6270 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6271 (add_pseudo_to_slot, lra_spill): Likewise.
6272 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6273 * optabs-query.c (get_best_extraction_insn): Likewise.
6274 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6275 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6276 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6277 * recog.c (offsettable_address_addr_space_p): Likewise.
6278 * regcprop.c (maybe_mode_change): Likewise.
6279 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6280 * regrename.c (build_def_use): Likewise.
6281 * regstat.c (dump_reg_info): Likewise.
6282 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6283 (find_reloads, find_reloads_subreg_address): Likewise.
6284 * reload1.c (eliminate_regs_1): Likewise.
6285 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6286 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6287 (simplify_binary_operation_1, simplify_subreg): Likewise.
6288 * targhooks.c (default_function_arg_padding): Likewise.
6289 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6290 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6291 (verify_gimple_assign_ternary): Likewise.
6292 * tree-inline.c (estimate_move_cost): Likewise.
6293 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6294 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6295 (get_address_cost_ainc): Likewise.
6296 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6297 (vect_supportable_dr_alignment): Likewise.
6298 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6299 (vectorizable_reduction): Likewise.
6300 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6301 (vectorizable_operation, vectorizable_load): Likewise.
6302 * tree.c (build_same_sized_truth_vector_type): Likewise.
6303 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6304 * var-tracking.c (emit_note_insn_var_location): Likewise.
6305 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6306 (ADDR_VEC_ALIGN): Likewise.
6308 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6309 Alan Hayward <alan.hayward@arm.com>
6310 David Sherwood <david.sherwood@arm.com>
6312 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6314 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6315 or if measurement_type is polynomial.
6316 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6317 * combine.c (make_extraction): Likewise.
6318 * dse.c (find_shift_sequence): Likewise.
6319 * dwarf2out.c (mem_loc_descriptor): Likewise.
6320 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6321 (extract_bit_field, extract_low_bits): Likewise.
6322 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6323 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6324 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6325 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6326 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6327 * reload.c (find_reloads): Likewise.
6328 * reload1.c (alter_reg): Likewise.
6329 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6330 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6331 * tree-if-conv.c (predicate_mem_writes): Likewise.
6332 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6333 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6334 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6335 * valtrack.c (dead_debug_insert_temp): Likewise.
6336 * varasm.c (mergeable_constant_section): Likewise.
6337 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6340 Alan Hayward <alan.hayward@arm.com>
6341 David Sherwood <david.sherwood@arm.com>
6343 * expr.c (expand_assignment): Cope with polynomial mode sizes
6344 when assigning to a CONCAT.
6346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6347 Alan Hayward <alan.hayward@arm.com>
6348 David Sherwood <david.sherwood@arm.com>
6350 * machmode.h (mode_precision): Change from unsigned short to
6352 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6354 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6355 or if measurement_type is not polynomial.
6356 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6357 in which the mode is already known to be a scalar_int_mode.
6358 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6359 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6361 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6362 for GET_MODE_PRECISION.
6363 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6364 for GET_MODE_PRECISION.
6365 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6367 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6368 (expand_field_assignment, make_extraction): Likewise.
6369 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6370 (get_last_value): Likewise.
6371 * convert.c (convert_to_integer_1): Likewise.
6372 * cse.c (cse_insn): Likewise.
6373 * expr.c (expand_expr_real_1): Likewise.
6374 * lra-constraints.c (simplify_operand_subreg): Likewise.
6375 * optabs-query.c (can_atomic_load_p): Likewise.
6376 * optabs.c (expand_atomic_load): Likewise.
6377 (expand_atomic_store): Likewise.
6378 * ree.c (combine_reaching_defs): Likewise.
6379 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6380 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6381 * tree.h (type_has_mode_precision_p): Likewise.
6382 * ubsan.c (instrument_si_overflow): Likewise.
6384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6385 Alan Hayward <alan.hayward@arm.com>
6386 David Sherwood <david.sherwood@arm.com>
6388 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6389 polynomial numbers of units.
6390 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6391 (valid_vector_subparts_p): New function.
6392 (build_vector_type): Remove temporary shim and take the number
6393 of units as a poly_uint64 rather than an int.
6394 (build_opaque_vector_type): Take the number of units as a
6395 poly_uint64 rather than an int.
6396 * tree.c (build_vector_from_ctor): Handle polynomial
6397 TYPE_VECTOR_SUBPARTS.
6398 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6399 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6400 (build_vector_from_val): If the number of units is variable,
6401 use build_vec_duplicate_cst for constant operands and
6402 VEC_DUPLICATE_EXPR otherwise.
6403 (make_vector_type): Remove temporary is_constant ().
6404 (build_vector_type, build_opaque_vector_type): Take the number of
6405 units as a poly_uint64 rather than an int.
6406 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6408 * cfgexpand.c (expand_debug_expr): Likewise.
6409 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6410 (store_constructor, expand_expr_real_1): Likewise.
6411 (const_scalar_mask_from_tree): Likewise.
6412 * fold-const-call.c (fold_const_reduction): Likewise.
6413 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6414 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6415 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6416 (fold_relational_const): Likewise.
6417 (native_interpret_vector): Likewise. Change the size from an
6418 int to an unsigned int.
6419 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6420 TYPE_VECTOR_SUBPARTS.
6421 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6422 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6423 duplicating a non-constant operand into a variable-length vector.
6424 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6425 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6426 * ipa-icf.c (sem_variable::equals): Likewise.
6427 * match.pd: Likewise.
6428 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6429 * print-tree.c (print_node): Likewise.
6430 * stor-layout.c (layout_type): Likewise.
6431 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6432 * tree-cfg.c (verify_gimple_comparison): Likewise.
6433 (verify_gimple_assign_binary): Likewise.
6434 (verify_gimple_assign_ternary): Likewise.
6435 (verify_gimple_assign_single): Likewise.
6436 * tree-pretty-print.c (dump_generic_node): Likewise.
6437 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6438 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6439 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6440 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6441 (vect_shift_permute_load_chain): Likewise.
6442 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6443 (expand_vector_condition, optimize_vector_constructor): Likewise.
6444 (lower_vec_perm, get_compute_type): Likewise.
6445 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6446 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6447 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6448 (vect_recog_mask_conversion_pattern): Likewise.
6449 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6450 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6451 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6452 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6453 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6454 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6455 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6456 (supportable_widening_operation): Likewise.
6457 (supportable_narrowing_operation): Likewise.
6458 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6460 * varasm.c (output_constant): Likewise.
6462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6463 Alan Hayward <alan.hayward@arm.com>
6464 David Sherwood <david.sherwood@arm.com>
6466 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6467 so that both the length == 3 and length != 3 cases set up their
6468 own permute vectors. Add comments explaining why we know the
6469 number of elements is constant.
6470 (vect_permute_load_chain): Likewise.
6472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6473 Alan Hayward <alan.hayward@arm.com>
6474 David Sherwood <david.sherwood@arm.com>
6476 * machmode.h (mode_nunits): Change from unsigned char to
6478 (ONLY_FIXED_SIZE_MODES): New macro.
6479 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6480 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6481 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6483 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6484 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6485 or if measurement_type is not polynomial.
6486 * genmodes.c (ZERO_COEFFS): New macro.
6487 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6489 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6490 Use ZERO_COEFFS when emitting initializers.
6491 * data-streamer.h (bp_pack_poly_value): New function.
6492 (bp_unpack_poly_value): Likewise.
6493 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6494 for GET_MODE_NUNITS.
6495 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6496 for GET_MODE_NUNITS.
6497 * tree.c (make_vector_type): Remove temporary shim and make
6498 the real function take the number of units as a poly_uint64
6500 (build_vector_type_for_mode): Handle polynomial nunits.
6501 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6502 * emit-rtl.c (const_vec_series_p_1): Likewise.
6503 (gen_rtx_CONST_VECTOR): Likewise.
6504 * fold-const.c (test_vec_duplicate_folding): Likewise.
6505 * genrecog.c (validate_pattern): Likewise.
6506 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6507 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6508 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6509 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6510 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6511 * rtlanal.c (subreg_get_info): Likewise.
6512 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6513 (vect_grouped_load_supported): Likewise.
6514 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6515 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6516 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6517 (simplify_const_unary_operation, simplify_binary_operation_1)
6518 (simplify_const_binary_operation, simplify_ternary_operation)
6519 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6520 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6521 instead of CONST_VECTOR_NUNITS.
6522 * varasm.c (output_constant_pool_2): Likewise.
6523 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6524 explicit-encoded elements in the XVEC for variable-length vectors.
6526 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6528 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6530 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6531 Alan Hayward <alan.hayward@arm.com>
6532 David Sherwood <david.sherwood@arm.com>
6534 * coretypes.h (fixed_size_mode): Declare.
6535 (fixed_size_mode_pod): New typedef.
6536 * builtins.h (target_builtins::x_apply_args_mode)
6537 (target_builtins::x_apply_result_mode): Change type to
6538 fixed_size_mode_pod.
6539 * builtins.c (apply_args_size, apply_result_size, result_vector)
6540 (expand_builtin_apply_args_1, expand_builtin_apply)
6541 (expand_builtin_return): Update accordingly.
6543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6545 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6546 * cselib.c (cselib_hash_rtx): Likewise.
6547 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6548 CONST_VECTOR encoding.
6550 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6551 Jeff Law <law@redhat.com>
6554 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6555 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6556 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6557 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6560 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6561 explicitly probe *sp in a noreturn function if there were any callee
6562 register saves or frame pointer is needed.
6564 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6567 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6568 BLKmode for ternary, binary or unary expressions.
6571 * var-tracking.c (delete_vta_debug_insn): New inline function.
6572 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6573 insns from get_insns () to NULL instead of each bb separately.
6574 Use delete_vta_debug_insn. No longer static.
6575 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6576 delete_vta_debug_insns callers.
6577 * rtl.h (delete_vta_debug_insns): Declare.
6578 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6579 instead of variable_tracking_main.
6581 2018-01-03 Martin Sebor <msebor@redhat.com>
6583 PR tree-optimization/83603
6584 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6585 arguments past the endof the argument list in functions declared
6586 without a prototype.
6587 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6588 Avoid checking when arguments are null.
6590 2018-01-03 Martin Sebor <msebor@redhat.com>
6593 * doc/extend.texi (attribute const): Fix a typo.
6594 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6595 issuing -Wsuggest-attribute for void functions.
6597 2018-01-03 Martin Sebor <msebor@redhat.com>
6599 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6600 offset_int::from instead of wide_int::to_shwi.
6601 (maybe_diag_overlap): Remove assertion.
6602 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6603 * gimple-ssa-sprintf.c (format_directive): Same.
6604 (parse_directive): Same.
6605 (sprintf_dom_walker::compute_format_length): Same.
6606 (try_substitute_return_value): Same.
6608 2018-01-03 Jeff Law <law@redhat.com>
6611 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6612 non-constant residual for zero at runtime and avoid probing in
6613 that case. Reorganize code for trailing problem to mirror handling
6616 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6618 PR tree-optimization/83501
6619 * tree-ssa-strlen.c (get_string_cst): New.
6620 (handle_char_store): Call get_string_cst.
6622 2018-01-03 Martin Liska <mliska@suse.cz>
6624 PR tree-optimization/83593
6625 * tree-ssa-strlen.c: Include tree-cfg.h.
6626 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6627 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6628 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6630 (strlen_dom_walker::before_dom_children): Call
6631 gimple_purge_dead_eh_edges. Dump tranformation with details
6633 (strlen_dom_walker::before_dom_children): Update call by adding
6634 new argument cleanup_eh.
6635 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6637 2018-01-03 Martin Liska <mliska@suse.cz>
6640 * cif-code.def (VARIADIC_THUNK): New enum value.
6641 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6644 2018-01-03 Jan Beulich <jbeulich@suse.com>
6646 * sse.md (mov<mode>_internal): Tighten condition for when to use
6647 vmovdqu<ssescalarsize> for TI and OI modes.
6649 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6651 Update copyright years.
6653 2018-01-03 Martin Liska <mliska@suse.cz>
6656 * ipa-visibility.c (function_and_variable_visibility): Skip
6657 functions with noipa attribure.
6659 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6661 * gcc.c (process_command): Update copyright notice dates.
6662 * gcov-dump.c (print_version): Ditto.
6663 * gcov.c (print_version): Ditto.
6664 * gcov-tool.c (print_version): Ditto.
6665 * gengtype.c (create_file): Ditto.
6666 * doc/cpp.texi: Bump @copying's copyright year.
6667 * doc/cppinternals.texi: Ditto.
6668 * doc/gcc.texi: Ditto.
6669 * doc/gccint.texi: Ditto.
6670 * doc/gcov.texi: Ditto.
6671 * doc/install.texi: Ditto.
6672 * doc/invoke.texi: Ditto.
6674 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6676 * vector-builder.h (vector_builder::m_full_nelts): Change from
6677 unsigned int to poly_uint64.
6678 (vector_builder::full_nelts): Update prototype accordingly.
6679 (vector_builder::new_vector): Likewise.
6680 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6681 (vector_builder::operator ==): Likewise.
6682 (vector_builder::finalize): Likewise.
6683 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6684 Take the number of elements as a poly_uint64 rather than an
6686 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6687 from unsigned int to poly_uint64.
6688 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6689 (vec_perm_indices::new_vector): Likewise.
6690 (vec_perm_indices::length): Likewise.
6691 (vec_perm_indices::nelts_per_input): Likewise.
6692 (vec_perm_indices::input_nelts): Likewise.
6693 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6694 number of elements per input as a poly_uint64 rather than an
6695 unsigned int. Use the original encoding for variable-length
6696 vectors, rather than clamping each individual element.
6697 For the second and subsequent elements in each pattern,
6698 clamp the step and base before clamping their sum.
6699 (vec_perm_indices::series_p): Handle polynomial element counts.
6700 (vec_perm_indices::all_in_range_p): Likewise.
6701 (vec_perm_indices_to_tree): Likewise.
6702 (vec_perm_indices_to_rtx): Likewise.
6703 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6704 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6705 (tree_vector_builder::new_binary_operation): Handle polynomial
6706 element counts. Return false if we need to know the number
6707 of elements at compile time.
6708 * fold-const.c (fold_vec_perm): Punt if the number of elements
6709 isn't known at compile time.
6711 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6713 * vec-perm-indices.h (vec_perm_builder): Change element type
6714 from HOST_WIDE_INT to poly_int64.
6715 (vec_perm_indices::element_type): Update accordingly.
6716 (vec_perm_indices::clamp): Handle polynomial element_types.
6717 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6718 (vec_perm_indices::all_in_range_p): Likewise.
6719 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6721 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6722 polynomial vec_perm_indices element types.
6723 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6724 * fold-const.c (fold_vec_perm): Likewise.
6725 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6726 * tree-vect-generic.c (lower_vec_perm): Likewise.
6727 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6728 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6729 element type to HOST_WIDE_INT.
6731 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6732 Alan Hayward <alan.hayward@arm.com>
6733 David Sherwood <david.sherwood@arm.com>
6735 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6736 rather than an int. Use plus_constant.
6737 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6738 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6740 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6741 Alan Hayward <alan.hayward@arm.com>
6742 David Sherwood <david.sherwood@arm.com>
6744 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6745 a HOST_WIDE_INT to a poly_int64.
6747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6748 Alan Hayward <alan.hayward@arm.com>
6749 David Sherwood <david.sherwood@arm.com>
6751 * calls.c (load_register_parameters): Cope with polynomial
6752 mode sizes. Require a constant size for BLKmode parameters
6753 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6754 forces a parameter to be padded at the lsb end in order to
6755 fill a complete number of words, require the parameter size
6756 to be ordered wrt UNITS_PER_WORD.
6758 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6759 Alan Hayward <alan.hayward@arm.com>
6760 David Sherwood <david.sherwood@arm.com>
6762 * reload1.c (spill_stack_slot_width): Change element type
6763 from unsigned int to poly_uint64_pod.
6764 (alter_reg): Treat mode sizes as polynomial.
6766 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6767 Alan Hayward <alan.hayward@arm.com>
6768 David Sherwood <david.sherwood@arm.com>
6770 * reload.c (complex_word_subreg_p): New function.
6771 (reload_inner_reg_of_subreg, push_reload): Use it.
6773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6774 Alan Hayward <alan.hayward@arm.com>
6775 David Sherwood <david.sherwood@arm.com>
6777 * lra-constraints.c (process_alt_operands): Reject matched
6778 operands whose sizes aren't ordered.
6779 (match_reload): Refer to this check here.
6781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6782 Alan Hayward <alan.hayward@arm.com>
6783 David Sherwood <david.sherwood@arm.com>
6785 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6786 that the mode size is in the set {1, 2, 4, 8, 16}.
6788 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6789 Alan Hayward <alan.hayward@arm.com>
6790 David Sherwood <david.sherwood@arm.com>
6792 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6793 Use plus_constant instead of gen_rtx_PLUS.
6795 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6796 Alan Hayward <alan.hayward@arm.com>
6797 David Sherwood <david.sherwood@arm.com>
6799 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6800 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6801 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6802 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6803 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6804 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6805 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6806 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6807 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6808 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6810 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6811 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6812 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6813 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6814 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6815 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6816 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6817 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6818 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6819 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6821 * expr.c (emit_move_resolve_push): Treat the input and result
6822 of PUSH_ROUNDING as a poly_int64.
6823 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6824 (emit_push_insn): Likewise.
6825 * lra-eliminations.c (mark_not_eliminable): Likewise.
6826 * recog.c (push_operand): Likewise.
6827 * reload1.c (elimination_effects): Likewise.
6828 * rtlanal.c (nonzero_bits1): Likewise.
6829 * calls.c (store_one_arg): Likewise. Require the padding to be
6830 known at compile time.
6832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6833 Alan Hayward <alan.hayward@arm.com>
6834 David Sherwood <david.sherwood@arm.com>
6836 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6837 Use plus_constant instead of gen_rtx_PLUS.
6839 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6840 Alan Hayward <alan.hayward@arm.com>
6841 David Sherwood <david.sherwood@arm.com>
6843 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6846 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6847 Alan Hayward <alan.hayward@arm.com>
6848 David Sherwood <david.sherwood@arm.com>
6850 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6851 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6852 via stack temporaries. Treat the mode size as polynomial too.
6854 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6855 Alan Hayward <alan.hayward@arm.com>
6856 David Sherwood <david.sherwood@arm.com>
6858 * expr.c (expand_expr_real_2): When handling conversions involving
6859 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6860 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6861 as a poly_uint64 too.
6863 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6864 Alan Hayward <alan.hayward@arm.com>
6865 David Sherwood <david.sherwood@arm.com>
6867 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6870 Alan Hayward <alan.hayward@arm.com>
6871 David Sherwood <david.sherwood@arm.com>
6873 * combine.c (can_change_dest_mode): Handle polynomial
6874 REGMODE_NATURAL_SIZE.
6875 * expmed.c (store_bit_field_1): Likewise.
6876 * expr.c (store_constructor): Likewise.
6877 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6878 and polynomial REGMODE_NATURAL_SIZE.
6879 (gen_lowpart_common): Likewise.
6880 * reginfo.c (record_subregs_of_mode): Likewise.
6881 * rtlanal.c (read_modify_subreg_p): Likewise.
6883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6884 Alan Hayward <alan.hayward@arm.com>
6885 David Sherwood <david.sherwood@arm.com>
6887 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6888 numbers of elements.
6890 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6891 Alan Hayward <alan.hayward@arm.com>
6892 David Sherwood <david.sherwood@arm.com>
6894 * match.pd: Cope with polynomial numbers of vector elements.
6896 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6897 Alan Hayward <alan.hayward@arm.com>
6898 David Sherwood <david.sherwood@arm.com>
6900 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6901 in a POINTER_PLUS_EXPR.
6903 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6904 Alan Hayward <alan.hayward@arm.com>
6905 David Sherwood <david.sherwood@arm.com>
6907 * omp-simd-clone.c (simd_clone_subparts): New function.
6908 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6909 (ipa_simd_modify_function_body): Likewise.
6911 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6912 Alan Hayward <alan.hayward@arm.com>
6913 David Sherwood <david.sherwood@arm.com>
6915 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6916 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6917 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6918 (expand_vector_condition, vector_element): Likewise.
6919 (subparts_gt): New function.
6920 (get_compute_type): Use subparts_gt.
6921 (count_type_subparts): Delete.
6922 (expand_vector_operations_1): Use subparts_gt instead of
6923 count_type_subparts.
6925 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6926 Alan Hayward <alan.hayward@arm.com>
6927 David Sherwood <david.sherwood@arm.com>
6929 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6930 (vect_compile_time_alias): ...this new function. Do the calculation
6931 on poly_ints rather than trees.
6932 (vect_prune_runtime_alias_test_list): Update call accordingly.
6934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6935 Alan Hayward <alan.hayward@arm.com>
6936 David Sherwood <david.sherwood@arm.com>
6938 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6940 (vect_schedule_slp_instance): Likewise.
6942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6943 Alan Hayward <alan.hayward@arm.com>
6944 David Sherwood <david.sherwood@arm.com>
6946 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6947 constant and extern definitions for variable-length vectors.
6948 (vect_get_constant_vectors): Note that the number of units
6949 is known to be constant.
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 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6956 of units as polynomial. Choose between WIDE and NARROW based
6959 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6960 Alan Hayward <alan.hayward@arm.com>
6961 David Sherwood <david.sherwood@arm.com>
6963 * tree-vect-stmts.c (simd_clone_subparts): New function.
6964 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6967 Alan Hayward <alan.hayward@arm.com>
6968 David Sherwood <david.sherwood@arm.com>
6970 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6971 vectors as polynomial. Use build_index_vector for
6974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6975 Alan Hayward <alan.hayward@arm.com>
6976 David Sherwood <david.sherwood@arm.com>
6978 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6979 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6980 for variable-length vectors.
6981 (vectorizable_mask_load_store): Treat the number of units as
6982 polynomial, asserting that it is constant if the condition has
6983 already been enforced.
6984 (vectorizable_store, vectorizable_load): Likewise.
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-loop.c (vectorizable_live_operation): Treat the number
6991 of units as polynomial. Punt if we can't tell at compile time
6992 which vector contains the final result.
6994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6995 Alan Hayward <alan.hayward@arm.com>
6996 David Sherwood <david.sherwood@arm.com>
6998 * tree-vect-loop.c (vectorizable_induction): Treat the number
6999 of units as polynomial. Punt on SLP inductions. Use an integer
7000 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7001 cast of such a series for variable-length floating-point
7004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7005 Alan Hayward <alan.hayward@arm.com>
7006 David Sherwood <david.sherwood@arm.com>
7008 * tree.h (build_index_vector): Declare.
7009 * tree.c (build_index_vector): New function.
7010 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7011 of units as polynomial, forcibly converting it to a constant if
7012 vectorizable_reduction has already enforced the condition.
7013 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7014 to create a {1,2,3,...} vector.
7015 (vectorizable_reduction): Treat the number of units as polynomial.
7016 Choose vectype_in based on the largest scalar element size rather
7017 than the smallest number of units. Enforce the restrictions
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-data-refs.c (vector_alignment_reachable_p): Treat the
7025 number of units as polynomial.
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 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7032 * target.def (autovectorize_vector_sizes): Return the vector sizes
7033 by pointer, using vector_sizes rather than a bitmask.
7034 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7035 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7036 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7038 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7039 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7040 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7041 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7042 * omp-general.c (omp_max_vf): Likewise.
7043 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7044 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7045 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7046 * tree-vect-slp.c (vect_slp_bb): Likewise.
7047 * doc/tm.texi: Regenerate.
7048 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7050 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7051 the vector size as a poly_uint64 rather than an unsigned int.
7052 (current_vector_size): Change from an unsigned int to a poly_uint64.
7053 (get_vectype_for_scalar_type): Update accordingly.
7054 * tree.h (build_truth_vector_type): Take the size and number of
7055 units as a poly_uint64 rather than an unsigned int.
7056 (build_vector_type): Add a temporary overload that takes
7057 the number of units as a poly_uint64 rather than an unsigned int.
7058 * tree.c (make_vector_type): Likewise.
7059 (build_truth_vector_type): Take the number of units as a poly_uint64
7060 rather than an unsigned int.
7062 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7063 Alan Hayward <alan.hayward@arm.com>
7064 David Sherwood <david.sherwood@arm.com>
7066 * target.def (get_mask_mode): Take the number of units and length
7067 as poly_uint64s rather than unsigned ints.
7068 * targhooks.h (default_get_mask_mode): Update accordingly.
7069 * targhooks.c (default_get_mask_mode): Likewise.
7070 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7071 * doc/tm.texi: Regenerate.
7073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7074 Alan Hayward <alan.hayward@arm.com>
7075 David Sherwood <david.sherwood@arm.com>
7077 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7078 * omp-general.c (omp_max_vf): Likewise.
7079 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7080 (expand_omp_simd): Handle polynomial safelen.
7081 * omp-low.c (omplow_simd_context): Add a default constructor.
7082 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7083 (lower_rec_simd_input_clauses): Update accordingly.
7084 (lower_rec_input_clauses): Likewise.
7086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7087 Alan Hayward <alan.hayward@arm.com>
7088 David Sherwood <david.sherwood@arm.com>
7090 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7091 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7092 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7093 (vect_analyze_slp_cost): Likewise.
7094 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7095 (vect_model_load_cost): Likewise.
7097 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7098 Alan Hayward <alan.hayward@arm.com>
7099 David Sherwood <david.sherwood@arm.com>
7101 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7102 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7103 from an unsigned int * to a poly_uint64_pod *.
7104 (calculate_unrolling_factor): New function.
7105 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7108 Alan Hayward <alan.hayward@arm.com>
7109 David Sherwood <david.sherwood@arm.com>
7111 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7112 from an unsigned int to a poly_uint64.
7113 (_loop_vec_info::slp_unrolling_factor): Likewise.
7114 (_loop_vec_info::vectorization_factor): Change from an int
7116 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7117 (vect_get_num_vectors): New function.
7118 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7119 (vect_get_num_copies): Use vect_get_num_vectors.
7120 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7121 to an unsigned int *.
7122 (vect_analyze_data_refs): Change min_vf from an int * to a
7124 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7125 than an unsigned HOST_WIDE_INT.
7126 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7127 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7128 to an unsigned int *.
7129 (vect_analyze_data_ref_dependences): Likewise.
7130 (vect_compute_data_ref_alignment): Handle polynomial vf.
7131 (vect_enhance_data_refs_alignment): Likewise.
7132 (vect_prune_runtime_alias_test_list): Likewise.
7133 (vect_shift_permute_load_chain): Likewise.
7134 (vect_supportable_dr_alignment): Likewise.
7135 (dependence_distance_ge_vf): Take the vectorization factor as a
7136 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7137 (vect_analyze_data_refs): Change min_vf from an int * to a
7139 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7140 vfm1 as a poly_uint64 rather than an int. Make the same change
7141 for the returned bound_scalar.
7142 (vect_gen_vector_loop_niters): Handle polynomial vf.
7143 (vect_do_peeling): Likewise. Update call to
7144 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7145 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7147 * tree-vect-loop.c (vect_determine_vectorization_factor)
7148 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7149 (vect_get_known_peeling_cost): Likewise.
7150 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7151 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7152 (vect_transform_loop): Likewise. Use the lowest possible VF when
7153 updating the upper bounds of the loop.
7154 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7156 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7157 polynomial unroll factors.
7158 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7159 (vect_make_slp_decision): Likewise.
7160 (vect_supported_load_permutation_p): Likewise, and polynomial
7162 (vect_analyze_slp_cost): Handle polynomial vf.
7163 (vect_slp_analyze_node_operations): Likewise.
7164 (vect_slp_analyze_bb_1): Likewise.
7165 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7166 than an unsigned HOST_WIDE_INT.
7167 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7168 (vectorizable_load): Handle polynomial vf.
7169 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7171 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7173 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7174 Alan Hayward <alan.hayward@arm.com>
7175 David Sherwood <david.sherwood@arm.com>
7177 * match.pd: Handle bit operations involving three constants
7178 and try to fold one pair.
7180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7182 * tree-vect-loop-manip.c: Include gimple-fold.h.
7183 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7184 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7185 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7186 Add a path that uses a step of VF instead of 1, but disable it
7188 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7189 and niters_no_overflow parameters. Update calls to
7190 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7191 Create a new SSA name if the latter choses to use a ste other
7192 than zero, and return it via niters_vector_mult_vf_var.
7193 * tree-vect-loop.c (vect_transform_loop): Update calls to
7194 vect_do_peeling, vect_gen_vector_loop_niters and
7195 slpeel_make_loop_iterate_ntimes.
7196 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7197 (vect_gen_vector_loop_niters): Update declarations after above changes.
7199 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7201 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7202 128-bit round to integer instructions.
7203 (ceil<mode>2): Likewise.
7204 (btrunc<mode>2): Likewise.
7205 (round<mode>2): Likewise.
7207 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7209 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7210 unaligned VSX load/store on P8/P9.
7211 (expand_block_clear): Allow the use of unaligned VSX
7212 load/store on P8/P9.
7214 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7216 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7218 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7219 swap associated with both a load and a store.
7221 2018-01-02 Andrew Waterman <andrew@sifive.com>
7223 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7224 * config/riscv/riscv.md (clear_cache): Use it.
7226 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7228 * web.c: Remove out-of-date comment.
7230 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7232 * expr.c (fixup_args_size_notes): Check that any existing
7233 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7234 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7235 (emit_single_push_insn): ...here.
7237 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7239 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7240 (const_vector_encoded_nelts): New function.
7241 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7242 (const_vector_int_elt, const_vector_elt): Declare.
7243 * emit-rtl.c (const_vector_int_elt_1): New function.
7244 (const_vector_elt): Likewise.
7245 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7246 of CONST_VECTOR_ELT.
7248 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7250 * expr.c: Include rtx-vector-builder.h.
7251 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7252 directly on the tree encoding.
7253 (const_vector_from_tree): Likewise.
7254 * optabs.c: Include rtx-vector-builder.h.
7255 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7256 sequence of "u" values.
7257 * vec-perm-indices.c: Include rtx-vector-builder.h.
7258 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7259 directly on the vec_perm_indices encoding.
7261 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7263 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7264 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7265 * rtx-vector-builder.h: New file.
7266 * rtx-vector-builder.c: Likewise.
7267 * rtl.h (rtx_def::u2): Add a const_vector field.
7268 (CONST_VECTOR_NPATTERNS): New macro.
7269 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7270 (CONST_VECTOR_DUPLICATE_P): Likewise.
7271 (CONST_VECTOR_STEPPED_P): Likewise.
7272 (CONST_VECTOR_ENCODED_ELT): Likewise.
7273 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7274 (unwrap_const_vec_duplicate): Likewise.
7275 (const_vec_series_p): Check for a non-duplicated vector encoding.
7276 Say that the function only returns true for integer vectors.
7277 * emit-rtl.c: Include rtx-vector-builder.h.
7278 (gen_const_vec_duplicate_1): Delete.
7279 (gen_const_vector): Call gen_const_vec_duplicate instead of
7280 gen_const_vec_duplicate_1.
7281 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7282 (gen_const_vec_duplicate): Use rtx_vector_builder.
7283 (gen_const_vec_series): Likewise.
7284 (gen_rtx_CONST_VECTOR): Likewise.
7285 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7286 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7287 Build a new vector rather than modifying a CONST_VECTOR in-place.
7288 (handle_special_swappables): Update call accordingly.
7289 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7290 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7291 Build a new vector rather than modifying a CONST_VECTOR in-place.
7292 (handle_special_swappables): Update call accordingly.
7294 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7296 * simplify-rtx.c (simplify_const_binary_operation): Use
7297 CONST_VECTOR_ELT instead of XVECEXP.
7299 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7301 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7302 the selector elements to be different from the data elements
7303 if the selector is a VECTOR_CST.
7304 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7305 ssizetype for the selector.
7307 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7309 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7310 before testing each element individually.
7311 * tree-vect-generic.c (lower_vec_perm): Likewise.
7313 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7315 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7316 * selftest-run-tests.c (selftest::run_tests): Call it.
7317 * vector-builder.h (vector_builder::operator ==): New function.
7318 (vector_builder::operator !=): Likewise.
7319 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7320 (vec_perm_indices::all_from_input_p): New function.
7321 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7322 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7323 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7324 instead of reading the VECTOR_CST directly. Detect whether both
7325 vector inputs are the same before constructing the vec_perm_indices,
7326 and update the number of inputs argument accordingly. Use the
7327 utility functions added above. Only construct sel2 if we need to.
7329 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7331 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7332 the broadcast of the low byte.
7333 (expand_mult_highpart): Use an explicit encoding for the permutes.
7334 * optabs-query.c (can_mult_highpart_p): Likewise.
7335 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7336 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7337 (vectorizable_bswap): Likewise.
7338 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7339 explicit encoding for the power-of-2 permutes.
7340 (vect_permute_store_chain): Likewise.
7341 (vect_grouped_load_supported): Likewise.
7342 (vect_permute_load_chain): Likewise.
7344 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7346 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7347 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7348 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7349 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7350 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7351 (vect_gen_perm_mask_any): Likewise.
7353 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7355 * int-vector-builder.h: New file.
7356 * vec-perm-indices.h: Include int-vector-builder.h.
7357 (vec_perm_indices): Redefine as an int_vector_builder.
7358 (auto_vec_perm_indices): Delete.
7359 (vec_perm_builder): Redefine as a stand-alone class.
7360 (vec_perm_indices::vec_perm_indices): New function.
7361 (vec_perm_indices::clamp): Likewise.
7362 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7363 (vec_perm_indices::new_vector): New function.
7364 (vec_perm_indices::new_expanded_vector): Update for new
7365 vec_perm_indices class.
7366 (vec_perm_indices::rotate_inputs): New function.
7367 (vec_perm_indices::all_in_range_p): Operate directly on the
7368 encoded form, without computing elided elements.
7369 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7370 encoding. Update for new vec_perm_indices class.
7371 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7372 the given vec_perm_builder.
7373 (expand_vec_perm_var): Update vec_perm_builder constructor.
7374 (expand_mult_highpart): Use vec_perm_builder instead of
7375 auto_vec_perm_indices.
7376 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7377 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7378 or double series encoding as appropriate.
7379 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7380 vec_perm_indices instead of auto_vec_perm_indices.
7381 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7382 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7383 (vect_permute_store_chain): Likewise.
7384 (vect_grouped_load_supported): Likewise.
7385 (vect_permute_load_chain): Likewise.
7386 (vect_shift_permute_load_chain): Likewise.
7387 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7388 (vect_transform_slp_perm_load): Likewise.
7389 (vect_schedule_slp_instance): Likewise.
7390 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7391 (vectorizable_mask_load_store): Likewise.
7392 (vectorizable_bswap): Likewise.
7393 (vectorizable_store): Likewise.
7394 (vectorizable_load): Likewise.
7395 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7396 vec_perm_indices instead of auto_vec_perm_indices. Use
7397 tree_to_vec_perm_builder to read the vector from a tree.
7398 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7399 vec_perm_builder instead of a vec_perm_indices.
7400 (have_whole_vector_shift): Use vec_perm_builder and
7401 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7402 truncation to calc_vec_perm_mask_for_shift.
7403 (vect_create_epilog_for_reduction): Likewise.
7404 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7405 from auto_vec_perm_indices to vec_perm_indices.
7406 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7407 instead of changing individual elements.
7408 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7409 the vector in d.perm.
7410 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7411 from auto_vec_perm_indices to vec_perm_indices.
7412 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7413 instead of changing individual elements.
7414 (arm_vectorize_vec_perm_const): Use new_vector to install
7415 the vector in d.perm.
7416 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7417 Update vec_perm_builder constructor.
7418 (rs6000_expand_interleave): Likewise.
7419 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7420 (rs6000_expand_interleave): Likewise.
7422 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7424 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7425 to qimode could truncate the indices.
7426 * optabs.c (expand_vec_perm_var): Likewise.
7428 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7430 * Makefile.in (OBJS): Add vec-perm-indices.o.
7431 * vec-perm-indices.h: New file.
7432 * vec-perm-indices.c: Likewise.
7433 * target.h (vec_perm_indices): Replace with a forward class
7435 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7436 * optabs.h: Include vec-perm-indices.h.
7437 (expand_vec_perm): Delete.
7438 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7439 (expand_vec_perm_const): Declare.
7440 * target.def (vec_perm_const_ok): Replace with...
7441 (vec_perm_const): ...this new hook.
7442 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7443 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7444 * doc/tm.texi: Regenerate.
7445 * optabs.def (vec_perm_const): Delete.
7446 * doc/md.texi (vec_perm_const): Likewise.
7447 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7448 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7449 expand_vec_perm for constant permutation vectors. Assert that
7450 the mode of variable permutation vectors is the integer equivalent
7451 of the mode that is being permuted.
7452 * optabs-query.h (selector_fits_mode_p): Declare.
7453 * optabs-query.c: Include vec-perm-indices.h.
7454 (selector_fits_mode_p): New function.
7455 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7456 is defined, instead of checking whether the vec_perm_const_optab
7457 exists. Use targetm.vectorize.vec_perm_const instead of
7458 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7459 fit in the vector mode before using a variable permute.
7460 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7461 vec_perm_indices instead of an rtx.
7462 (expand_vec_perm): Replace with...
7463 (expand_vec_perm_const): ...this new function. Take the selector
7464 as a vec_perm_indices rather than an rtx. Also take the mode of
7465 the selector. Update call to shift_amt_for_vec_perm_mask.
7466 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7467 Use vec_perm_indices::new_expanded_vector to expand the original
7468 selector into bytes. Check whether the indices fit in the vector
7469 mode before using a variable permute.
7470 (expand_vec_perm_var): Make global.
7471 (expand_mult_highpart): Use expand_vec_perm_const.
7472 * fold-const.c: Includes vec-perm-indices.h.
7473 * tree-ssa-forwprop.c: Likewise.
7474 * tree-vect-data-refs.c: Likewise.
7475 * tree-vect-generic.c: Likewise.
7476 * tree-vect-loop.c: Likewise.
7477 * tree-vect-slp.c: Likewise.
7478 * tree-vect-stmts.c: Likewise.
7479 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7481 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7482 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7483 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7484 (aarch64_vectorize_vec_perm_const): ...this new function.
7485 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7486 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7487 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7488 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7489 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7490 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7491 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7493 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7494 check for NEON modes.
7495 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7496 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7497 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7498 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7500 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7501 the old VEC_PERM_CONST conditions.
7502 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7503 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7504 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7505 (ia64_vectorize_vec_perm_const_ok): Merge into...
7506 (ia64_vectorize_vec_perm_const): ...this new function.
7507 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7508 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7509 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7510 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7511 * config/mips/mips.c (mips_expand_vec_perm_const)
7512 (mips_vectorize_vec_perm_const_ok): Merge into...
7513 (mips_vectorize_vec_perm_const): ...this new function.
7514 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7515 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7516 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7517 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7518 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7519 (rs6000_expand_vec_perm_const): Delete.
7520 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7522 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7523 (altivec_expand_vec_perm_const_le): Take each operand individually.
7524 Operate on constant selectors rather than rtxes.
7525 (altivec_expand_vec_perm_const): Likewise. Update call to
7526 altivec_expand_vec_perm_const_le.
7527 (rs6000_expand_vec_perm_const): Delete.
7528 (rs6000_vectorize_vec_perm_const_ok): Delete.
7529 (rs6000_vectorize_vec_perm_const): New function.
7530 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7531 an element count and rtx array.
7532 (rs6000_expand_extract_even): Update call accordingly.
7533 (rs6000_expand_interleave): Likewise.
7534 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7535 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7536 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7537 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7538 (rs6000_expand_vec_perm_const): Delete.
7539 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7540 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7541 (altivec_expand_vec_perm_const_le): Take each operand individually.
7542 Operate on constant selectors rather than rtxes.
7543 (altivec_expand_vec_perm_const): Likewise. Update call to
7544 altivec_expand_vec_perm_const_le.
7545 (rs6000_expand_vec_perm_const): Delete.
7546 (rs6000_vectorize_vec_perm_const_ok): Delete.
7547 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7548 reference to the SPE evmerge intructions.
7549 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7550 an element count and rtx array.
7551 (rs6000_expand_extract_even): Update call accordingly.
7552 (rs6000_expand_interleave): Likewise.
7553 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7554 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7556 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7558 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7560 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7561 vector mode and that that mode matches the mode of the data
7563 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7564 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7565 directly using expand_vec_perm_1 when forcing selectors into
7567 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7569 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7571 * optabs-query.h (can_vec_perm_p): Delete.
7572 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7573 * optabs-query.c (can_vec_perm_p): Split into...
7574 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7575 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7576 particular selector is valid.
7577 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7578 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7579 (vect_grouped_load_supported): Likewise.
7580 (vect_shift_permute_load_chain): Likewise.
7581 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7582 (vect_transform_slp_perm_load): Likewise.
7583 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7584 (vectorizable_bswap): Likewise.
7585 (vect_gen_perm_mask_checked): Likewise.
7586 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7587 implementations of variable permutation vectors into account
7588 when deciding which selector to use.
7589 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7590 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7591 with a false third argument.
7592 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7593 to test whether the constant selector is valid and can_vec_perm_var_p
7594 to test whether a variable selector is valid.
7596 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7598 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7599 * optabs-query.c (can_vec_perm_p): Likewise.
7600 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7601 instead of vec_perm_indices.
7602 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7603 (vect_gen_perm_mask_checked): Likewise,
7604 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7605 (vect_gen_perm_mask_checked): Likewise,
7607 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7609 * optabs-query.h (qimode_for_vec_perm): Declare.
7610 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7611 (qimode_for_vec_perm): ...this new function.
7612 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7614 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7616 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7617 does not have a conditional at the top.
7619 2018-01-02 Richard Biener <rguenther@suse.de>
7621 * ipa-inline.c (big_speedup_p): Fix expression.
7623 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7626 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7629 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7633 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7634 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7635 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7636 cond_taken_branch_cost 3->4.
7638 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7640 PR tree-optimization/83581
7641 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7642 TODO_cleanup_cfg if any changes have been made.
7645 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7646 convert_modes if target mode has the right side, but different mode
7650 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7651 last argument when extracting from CONCAT. If either from_real or
7652 from_imag is NULL, use expansion through memory. If result is not
7653 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7654 the parts directly to inner mode, if even that fails, use expansion
7658 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7659 check for bswap in mode rather than HImode and use that in expand_unop
7662 Copyright (C) 2018 Free Software Foundation, Inc.
7664 Copying and distribution of this file, with or without modification,
7665 are permitted in any medium without royalty provided the copyright
7666 notice and this notice are preserved.