1 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
3 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
4 ISA_2_5_MASKS_EMBEDDED.
6 2018-02-23 Jakub Jelinek <jakub@redhat.com>
8 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
9 p->max as pointers rather than using iterative_hash_expr.
11 2018-02-23 Carl Love <cel@us.ibm.com>
13 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
14 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
16 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
17 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
18 P8V_BUILTIN_VEC_VUNSIGNED2.
20 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
23 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
24 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
26 * lra-constraints.c (curr_insn_transform): Set up
27 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
29 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
30 * lra-eliminations.c (spill_pseudos): Ditto.
31 (process_insn_for_elimination): Ditto.
32 * lra-lives.c (reg_early_clobber_p): Use the new macros.
33 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
34 LRA_NON_CLOBBERED_ALT.
36 2018-02-22 Martin Sebor <msebor@redhat.com>
38 PR tree-optimization/84480
39 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
40 to maybe_diag_stxncpy_trunc. Call it.
41 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
42 from gimple_fold_builtin_strcpy. Print inlining stack.
43 (handle_builtin_stxncpy): Print inlining stack.
44 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
46 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
49 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
50 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
51 and -fcheck-pointer-bounds are used together.
52 (indirect_thunk_prefix): New enum.
53 (indirect_thunk_need_prefix): New function.
54 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
55 "_nt" instead of "_bnd" for NOTRACK prefix.
56 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
57 (output_indirect_thunk_function): Likewise.
59 (ix86_code_end): Update output_indirect_thunk_function calls.
60 (ix86_output_indirect_branch_via_reg): Replace
61 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
62 (ix86_output_indirect_branch_via_push): Likewise.
63 (ix86_output_function_return): Likewise.
64 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
65 incompatible with -fcf-protection=branch and
66 -fcheck-pointer-bounds.
68 2018-02-22 Steve Ellcey <sellcey@cavium.com>
71 * config/aarch64/aarch64.c (aarch64_print_address_internal):
72 Change gcc_assert call to output_operand_lossage.
74 2018-02-22 Steve Ellcey <sellcey@cavium.com>
76 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
78 2018-02-22 DJ Delorie <dj@redhat.com>
79 Sebastian Perta <sebastian.perta@renesas.com>
80 Oleg Endo <olegendo@gcc.gnu.org>
82 * config/rx/rx.c (rx_rtx_costs): New function.
83 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
85 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
87 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
89 2018-02-22 Martin Liska <mliska@suse.cz>
92 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
93 Add "native" as a possible value.
95 2018-02-22 Martin Liska <mliska@suse.cz>
98 * config/i386/i386.c (ix86_option_override_internal):
99 Add "native" as a possible value for -march and -mtune.
101 2018-02-22 Jakub Jelinek <jakub@redhat.com>
104 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
105 to all type variants.
107 PR tree-optimization/84503
108 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
109 width as info->bitpos + info->bitsize - start.
110 (merged_store_group::merge_overlapping): Simplify width computation.
111 (check_no_overlap): New function.
112 (imm_store_chain_info::try_coalesce_bswap): Compute expected
113 start + width and last_order of the group, fail if check_no_overlap
115 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
116 to group if check_no_overlap fails.
118 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
120 * config/rs6000/altivec.md: Delete contraint arguments to
121 define_expand, define_split, and define_peephole2, and in
122 define_insn_and_split if always unused.
123 * config/rs6000/darwin.md: Ditto.
124 * config/rs6000/dfp.md: Ditto.
125 * config/rs6000/rs6000.md: Ditto.
126 * config/rs6000/sync.md: Ditto.
127 * config/rs6000/vector.md: Ditto.
128 * config/rs6000/vsx.md: Ditto.
130 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
132 * config/rs6000/altivec.md: Write output control strings as braced
133 blocks instead of double-quoted strings.
134 * config/rs6000/darwin.md: Ditto.
135 * config/rs6000/rs6000.md: Ditto.
136 * config/rs6000/vector.md: Ditto.
137 * config/rs6000/vsx.md: Ditto.
139 2018-02-21 Jason Merrill <jason@redhat.com>
141 PR c++/84314 - ICE with templates and fastcall attribute.
142 * attribs.c (build_type_attribute_qual_variant): Remove assert.
144 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
146 * ipa-cp.c (determine_versionability): Fix comment typos.
148 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
151 * ipa-cp.c (determine_versionability): Do not version functions caling
154 2018-02-21 Martin Liska <mliska@suse.cz>
157 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
158 Add "native" as a possible value.
159 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
160 the macro when native cpu detection is available.
162 2018-02-21 Martin Liska <mliska@suse.cz>
165 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
166 Add "native" as a possible value.
167 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
168 when native cpu detection is available.
170 2018-02-21 Jakub Jelinek <jakub@redhat.com>
171 Martin Sebor <msebor@redhat.com>
173 PR tree-optimization/84478
174 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
176 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
177 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
178 support which is conservatively correct, for 2 only stay conservative
179 for maxlen. Formatting and comment capitalization fixes. Add STRICT
180 argument to the 2 argument get_range_strlen, adjust 6 arg
181 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
183 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
184 (gimple_fold_builtin_strlen): Pass true as last argument to
187 2018-02-20 Martin Sebor <msebor@redhat.com>
190 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
191 (builtin_memref::set_base_and_offset): Same. Handle inner references.
192 (builtin_memref::builtin_memref): Factor out parts into
193 set_base_and_offset and call it.
195 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
198 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
199 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
200 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
201 search at the associated MODE_INT.
203 2018-02-20 Jeff Law <law@redhat.com>
206 PR tree-optimization/81592
208 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
209 for range data rather than using global data.
210 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
211 range data rather than using global data.
212 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
213 pass it to children as needed.
214 (struct directive::fmtresult): Similarly.
215 (struct directive::set_width): Similarly.
216 (struct directive::set_precision): Similarly.
217 (format_integer, format_directive, parse_directive): Similarly.
218 (format_none): Accept unnamed vr_values parameter.
219 (format_percent, format_floating, format_character): Similarly.
220 (format_string, format_plain): Similarly.
221 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
222 the EVRP range analyzer for range data rather than using global data.
223 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
224 gimple-ssa-evrp-analyze.h
225 (class sprintf_dom_walker): Add after_dom_children member function.
226 Add evrp_range_analyzer member.
227 (sprintf_dom_walker::before_dom_children): Call into the EVRP
228 range analyzer as needed.
229 (sprintf_dom_walker::after_dom_children): New member function.
230 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
232 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
233 (evrp_range_analyzer::pop_to_marker): Likewise.
235 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
237 PR tree-optimization/84419
238 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
239 with the required type if its current type is compatible but
242 2018-02-20 Jakub Jelinek <jakub@redhat.com>
245 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
248 2018-02-20 Martin Liska <mliska@suse.cz>
251 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
252 possible values if we don't have a hint.
254 2018-02-20 Martin Liska <mliska@suse.cz>
258 * final.c (shorten_branches): Build align_tab array with one
260 * opts.c (finish_options): Add alignment option limit check.
261 (MAX_CODE_ALIGN): Likewise.
262 (MAX_CODE_ALIGN_VALUE): Likewise.
263 * doc/invoke.texi: Document maximum allowed option value for
264 all -falign-* options.
266 2018-02-19 Jakub Jelinek <jakub@redhat.com>
269 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
270 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
271 * var-tracking.c (emit_note_insn_var_location): Remove all references
272 to NOTE_INSN_CALL_ARG_LOCATION.
273 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
274 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
275 Use copy_rtx_if_shared.
276 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
277 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
278 (dwarf2out_var_location): Remove handling of
279 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
281 * final.c (final_scan_insn): Remove all references to
282 NOTE_INSN_CALL_ARG_LOCATION.
283 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
284 before dumping final insns.
285 * except.c (emit_note_eh_region_end): Remove all references to
286 NOTE_INSN_CALL_ARG_LOCATION.
287 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
288 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
289 * config/arc/arc.c (hwloop_optimize): Likewise.
290 * config/arm/arm.c (create_fix_barrier): Likewise.
291 * config/s390/s390.c (s390_chunkify_start): Likewise.
292 * config/sh/sh.c (find_barrier): Likewise.
293 * config/i386/i386.c (rest_of_insert_endbranch,
294 ix86_seh_fixup_eh_fallthru): Likewise.
295 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
296 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
297 * config/frv/frv.c (frv_function_prologue): Likewise.
298 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
300 (note_outside_basic_block_p): Remove all references to
301 NOTE_INSN_CALL_ARG_LOCATION.
302 * gengtype.c (adjust_field_rtx_def): Likewise.
303 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
305 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
306 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
309 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
312 PR tree-optimization/84452
313 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
314 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
317 2018-02-19 Martin Liska <mliska@suse.cz>
320 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
322 2018-02-19 Martin Liska <mliska@suse.cz>
323 Richard Sandiford <richard.sandiford@linaro.org>
325 PR tree-optimization/82491
326 * gimple-fold.c (get_base_constructor): Make earlier bail out
329 2018-02-19 Carl Love <cel@us.ibm.com>
331 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
332 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
334 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
337 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
339 * config/rl78/rl78.md (movdf): New define expand.
341 2018-02-19 Martin Liska <mliska@suse.cz>
344 * doc/invoke.texi: Fix typo.
345 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
347 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
349 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
350 handle rs6000_single_float and rs6000_double_float specially for
353 2018-02-16 Jeff Law <law@redhat.com>
355 * config/rx/rx.c (add_pop_cfi_notes): New function.;
358 2018-02-16 Jakub Jelinek <jakub@redhat.com>
361 * ipa-inline.c (inline_small_functions): Fix a typo.
363 2018-02-16 Nathan Sidwell <nathan@acm.org>
365 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
367 2018-02-16 Carl Love <cel@us.ibm.com>
369 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
370 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
371 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
372 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
373 expansion to P8V_BUILTIN_VEC_FLOAT2.
375 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
377 PR rtl-optimization/70023
378 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
379 src_regno into account.
381 2018-02-16 Carl Love <cel@us.ibm.com>
383 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
384 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
385 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
386 * config/rs6000/rs6000.c: Remove case statements for
387 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
388 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
389 and P9V_BUILTIN_VEC_VINSERT4B.
390 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
391 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
392 * config/rs6000/vsx.md:
393 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
396 2018-02-16 Carl Love <cel@us.ibm.com>
398 * config/rs6000/altivec.h: Add builtin names vec_extract4b
400 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
402 * config/rs6000/rs6000-c.c: Add the definitions for
403 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
404 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
405 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
406 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
407 definition for insert4b and define insn *insert3b_internal.
408 * doc/extend.texi: Add documentation for vec_extract4b.
410 2018-02-16 Nathan Sidwell <nathan@acm.org>
412 * doc/extend.texi (Backwards Compatibility): Mention friend
413 injection. Note for-scope is deprecated.
414 * doc/invoke.texi (-ffriend-injection): Deprecate.
416 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
418 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
419 that moved to I2, also allow destinations that are a paradoxical
420 subreg (instead of a normal reg).
422 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
425 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
428 2018-02-16 Richard Biener <rguenther@suse.de>
430 PR tree-optimization/84037
431 PR tree-optimization/84016
433 * config/i386/i386.c (ix86_builtin_vectorization_cost):
434 Adjust vec_construct for the fact we need additional higher latency
435 128bit inserts for AVX256 and AVX512 vector builds.
436 (ix86_add_stmt_cost): Scale vector construction cost for
439 2018-02-16 Richard Biener <rguenther@suse.de>
441 PR tree-optimization/84417
442 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
443 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
444 (non_rewritable_lvalue_p): Likewise, use poly-ints.
446 2018-02-16 Martin Liska <mliska@suse.cz>
449 * internal-fn.def (ASAN_CHECK): Set proper flags.
450 (ASAN_MARK): Likewise.
452 2018-02-16 Julia Koval <julia.koval@intel.com>
454 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
457 2018-02-16 Jakub Jelinek <jakub@redhat.com>
460 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
461 Use ++iter rather than iter++ for std::list iterators.
462 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
463 defer deleting them until all nodes in the forest are processed. Do
464 free even leaf nodes. Change to_process into auto_vec.
467 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
468 * vec.h (vec_default_construct): Use memset instead of placement new
469 if BROKEN_VALUE_INITIALIZATION is defined.
470 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
471 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
474 PR rtl-optimization/83723
475 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
476 * lra.c (lra_substitute_pseudo): Likewise. If true, use
477 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
479 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
481 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
483 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
485 PR rtl-optimization/81443
486 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
487 from inner REGs to paradoxical SUBREGs.
489 2018-02-16 Richard Biener <rguenther@suse.de>
491 PR tree-optimization/84399
492 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
493 For operands we can analyze at their definition make sure we can
494 analyze them at each use as well.
496 2018-02-16 Richard Biener <rguenther@suse.de>
498 PR tree-optimization/84190
499 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
500 volatile accesses if the decl isn't volatile.
502 2018-02-15 Jason Merrill <jason@redhat.com>
504 PR c++/84314 - ICE with templates and fastcall attribute.
505 * attribs.c (build_type_attribute_qual_variant): Don't clobber
506 TYPE_CANONICAL on an existing type.
508 2018-02-15 Jakub Jelinek <jakub@redhat.com>
510 PR tree-optimization/84383
511 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
512 dstoff nor call operand_equal_p if dstbase is NULL.
514 PR tree-optimization/84334
515 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
516 also a CONSTANT_CLASS_P, punt.
518 2018-02-14 Jim Wilson <jimw@sifive.com>
520 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
521 first SMALL_OPERAND check. New local min_second_step. Move assert
522 to where locals are set. Add TARGET_RVC support.
523 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
525 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
527 * doc/invoke.texi: Correct -Wformat-overflow code sample.
529 2018-02-14 Martin Sebor <msebor@redhat.com>
531 PR tree-optimization/83698
532 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
533 arrays constrain the offset range to their bounds.
534 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
535 (builtin_access::overlap): Avoid setting the size of overlap if it's
537 (maybe_diag_overlap): Also consider arrays when deciding what values
538 of offsets to include in diagnostics.
540 2018-02-14 Martin Sebor <msebor@redhat.com>
543 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
544 that correspond to the kind of a declaration.
546 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
549 * config/pa/pa.md: Load address of PIC label using the linkage table
550 if the label is nonlocal.
552 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
554 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
555 warning message if user requests -maltivec=be.
556 * doc/invoke.texi: Document deprecation of -maltivec=be.
558 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
561 * config/rs6000/rs6000-c.c: Update definitions for
562 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
563 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
565 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
568 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
569 add _get_ssp intrinsics. Remove argument from
570 __builtin_ia32_rdssp[d|q].
571 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
572 * config/i386/i386-builtin.def: Remove argument from
573 __builtin_ia32_rdssp[d|q].
574 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
575 ix86_expand_special_args_builtin for _rdssp[d|q].
576 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
577 Clear register before usage.
578 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
579 Add documentation for new _get_ssp and _inc_ssp intrinsics.
581 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
583 PR tree-optimization/84357
584 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
585 operand 1 of an ARRAY_REF too.
587 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
590 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
591 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
593 (set_of_reg): New struct.
594 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
595 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
596 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
598 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
599 Split into bitclr, bitset, bitinvert patterns if appropriate.
600 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
601 use rx_fuse_in_memory_bitop.
602 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
603 to named insn, correct maximum insn length.
605 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
608 * machmode.def: Define a complex mode for PARTIAL_INT.
609 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
611 * doc/rtl.texi: Document CSPImode.
612 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
614 (msp430_hard_regno_nregs_with_padding): Likewise.
616 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
619 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
621 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
623 PR rtl-optimization/84169
624 * combine.c (try_combine): New variable split_i2i3. Set it to true if
625 we generated a parallel as new i3 and we split that to new i2 and i3
626 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
627 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
628 those to i2, not i1. Partially rewrite this scan code.
630 2018-02-13 Jakub Jelinek <jakub@redhat.com>
633 * stor-layout.c (place_field): For variable length fields, adjust
634 offset_align afterwards not just based on the field's alignment,
635 but also on the size.
638 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
639 of exps and logs in the use_exp2 case.
641 2018-02-13 Jeff Law <law@redhat.com>
643 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
646 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
649 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
653 * common.opt (gas-loc-support, gas-locview-support): New.
654 (ginline-points, ginternal-reset-location-views): New.
655 * doc/invoke.texi: Document them. Use @itemx where intended.
656 (gvariable-location-views): Adjust.
657 * target.def (reset_location_view): New.
658 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
659 (TARGET_RESET_LOCATION_VIEW): New.
660 * doc/tm.texi: Rebuilt.
661 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
662 (dwarf2out_default_as_locview_support): New.
663 (output_asm_line_debug_info): Use option variables.
664 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
665 (output_loc_list): Likewise.
666 (add_high_low_attributes): Check option variables.
667 Don't output entry view attribute in strict mode.
668 (gen_inlined_subroutine_die): Check option variables.
669 (dwarf2out_inline_entry): Likewise.
670 (init_sections_and_labels): Likewise.
671 (dwarf2out_early_finish): Likewise.
672 (maybe_reset_location_view): New, from...
673 (dwarf2out_var_location): ... here. Call it.
674 * debug.h (dwarf2out_default_as_loc_support): Declare.
675 (dwarf2out_default_as_locview_support): Declare.
676 * hooks.c (hook_int_rtx_insn_0): New.
677 * hooks.h (hook_int_rtx_insn_0): Declare.
678 * toplev.c (process_options): Take -gas-loc-support and
679 -gas-locview-support from dwarf2out. Enable
680 -gvariable-location-views by default only with locview
681 assembler support. Enable -ginternal-reset-location-views by
682 default only if the target defines the corresponding hook.
683 Enable -ginline-points by default if location views are
684 enabled; force it disabled if statement frontiers are
686 * tree-inline.c (expand_call_inline): Check option variables.
687 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
689 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
691 PR tree-optimization/84321
692 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
693 handling. Also check whether the anti-range contains any values
694 that satisfy the mask; switch to a VR_RANGE if not.
696 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
699 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
701 2018-02-13 Martin Jambor <mjambor@suse.cz>
704 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
705 of call statements, also set location of a load to a temporary.
707 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
709 * config/rl78/rl78.c (add_vector_labels): New function.
710 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
711 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
712 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
713 which checks that no arguments are passed.
714 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
715 * doc/extend.texi: Documentation for the new attribute.
717 2018-02-13 Andreas Schwab <schwab@suse.de>
719 * config/riscv/linux.h (CPP_SPEC): Define.
721 2018-02-13 Jakub Jelinek <jakub@redhat.com>
724 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
725 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
726 OPTION_MASK_ISA_AES as first argument to def_builtin_const
727 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
728 instead of OPTION_MASK_ISA_PCLMUL as first argument to
729 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
730 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
731 temporarily for AES and PCLMUL builtins.
733 PR tree-optimization/84339
734 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
735 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
739 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
740 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
741 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
743 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
745 * omp-simd-clone.h: New file.
746 * omp-simd-clone.c: Include omp-simd-clone.h.
747 (expand_simd_clones): No longer static.
748 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
749 cgraph.h and omp-simd-clone.h.
750 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
751 (vect_recog_widen_shift_pattern): Formatting fix.
752 (vect_pattern_recog_1): Don't check optab for calls.
755 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
756 operands[2] into a REG before using gen_lowpart on it.
758 2018-02-12 Jeff Law <law@redhat.com>
761 * config/sh/sh.c (find_barrier): Consider a sibling call
764 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
765 successfully back substituting a reg.
767 2018-02-12 Richard Biener <rguenther@suse.de>
769 PR tree-optimization/84037
770 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
771 parameter, move visited init to caller.
772 (vect_slp_analyze_operations): Separate cost from validity
773 check, initialize visited once for all instances.
774 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
776 * tree-vect-stmts.c (vect_model_simple_cost): Make early
778 (vect_model_promotion_demotion_cost): Likewise.
779 (vectorizable_bswap): Guard cost modeling with !slp_node
780 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
782 (vectorizable_call): Likewise.
783 (vectorizable_conversion): Likewise.
784 (vectorizable_assignment): Likewise.
785 (vectorizable_shift): Likewise.
786 (vectorizable_operation): Likewise.
787 (vectorizable_store): Likewise.
788 (vectorizable_load): Likewise.
789 (vectorizable_condition): Likewise.
790 (vectorizable_comparison): Likewise.
792 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
795 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
796 (ASAN_MARK): Fix fnspec to account for return value, change pointer
797 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
799 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
802 * params.def (inline-min-speedup): Increase from 8 to 15.
803 (max-inline-insns-auto): Decrease from 40 to 30.
804 * ipa-split.c (consider_split): Add some buffer for function to
805 be considered inlining candidate.
806 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
809 2018-02-12 Richard Biener <rguenther@suse.de>
811 PR tree-optimization/84037
812 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
813 matched stmts if we cannot swap the non-matched ones.
815 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
817 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
818 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
819 _mm_maskz_scalef_round_ss): New intrinsics.
820 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
821 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
822 __builtin_ia32_scalefss_round): Remove.
823 (__builtin_ia32_scalefsd_mask_round,
824 __builtin_ia32_scalefss_mask_round): New intrinsics.
825 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
826 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
827 ((match_operand:VF_128 2 "<round_nimm_predicate>"
828 "<round_constraint>")): Changed to ...
829 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
830 "<round_scalar_constraint>")): ... this.
831 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
832 %0, %1, %2<round_op3>}"): Changed to ...
833 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
834 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
835 %2<round_scalar_mask_op3>}"): ... this.
836 * config/i386/subst.md (round_scalar_nimm_predicate): New.
838 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
840 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
841 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
842 (_mm_maskz_sqrt_round_ss): New intrinsics.
843 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
844 (__builtin_ia32_sqrtsd_mask_round)
845 (__builtin_ia32_sqrtss_mask_round): New builtins.
846 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
847 (__builtin_ia32_sqrtss_round): Remove.
848 (__builtin_ia32_sqrtsd_mask_round)
849 (__builtin_ia32_sqrtss_mask_round): New builtins.
850 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
851 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
852 ((match_operand:VF_128 1 "vector_operand"
853 "xBm,<round_constraint>")): Changed to ...
854 ((match_operand:VF_128 1 "vector_operand"
855 "xBm,<round_scalar_constraint>")): ... this.
856 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
857 %0, %2, %<iptr>1<round_op3>}): Changed to ...
858 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
859 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
860 %<iptr>1<round_scalar_mask_op3>}): ... this.
861 ((set_attr "prefix" "<round_prefix>")): Changed to ...
862 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
864 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
867 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
868 Cast vec_cmpeq result to correct type.
869 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
870 Cast vec_cmpgt result to correct type.
872 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
874 * final.c (final_scan_insn_1): Renamed from...
875 (final_scan_insn): ... this. New wrapper, to recover
876 seen from the outermost call in recursive ones.
877 * config/sparc/sparc.c (output_return): Drop seen from call.
878 (output_sibcall): Likewise.
879 * config/visium/visium.c (output_branch): Likewise.
881 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
883 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
886 2018-02-10 Alan Modra <amodra@gmail.com>
889 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
890 Specify LR as an input.
892 2018-02-10 Jakub Jelinek <jakub@redhat.com>
895 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
896 remove_member_access_dummy_vars): New functions.
897 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
898 lower_omp_1, execute_lower_omp): Use them.
900 PR rtl-optimization/84308
901 * shrink-wrap.c (spread_components): Release todo vector.
903 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
905 PR rtl-optimization/57193
906 * ira-color.c (struct allocno_color_data): Add member
907 conflict_allocno_hard_prefs.
908 (update_conflict_allocno_hard_prefs): New.
909 (bucket_allocno_compare_func): Add a preference based on
910 conflict_allocno_hard_prefs.
911 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
912 (color_allocnos): Remove a dead code. Initiate
913 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
915 2018-02-09 Jakub Jelinek <jakub@redhat.com>
918 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
919 constraint from =wa to wa. Avoid a subreg on the output operand,
920 instead use a pseudo and subreg it in a move.
921 (p9_xxbrd_<mode>): Changed to ...
922 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
923 (p9_xxbrd_v2df): New expander.
924 (p9_xxbrw_<mode>): Changed to ...
925 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
926 (p9_xxbrw_v4sf): New expander.
928 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
930 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
932 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
935 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
936 multiply in 32-bit mode.
937 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
938 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
941 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
943 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
944 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
945 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
946 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
948 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
951 * dwarf2out.c (is_trivial_indirect_ref): New function.
952 (dwarf2out_late_global_decl): Do not generate a location
953 attribute for variables that have a non-trivial DECL_VALUE_EXPR
954 and that are not defined in the current unit.
956 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
958 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
959 instead of a libcall for UNORDERED.
961 2018-02-09 Tamar Christina <tamar.christina@arm.com>
964 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
965 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
967 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
970 * config/s390/s390.c (s390_set_current_function): Invoke
971 s390_indirect_branch_settings also if fndecl didn't change.
973 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
975 * config/rs6000/rs6000.md (blockage): Set length to zero.
977 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
979 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
981 2018-02-09 Jakub Jelinek <jakub@redhat.com>
984 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
985 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
989 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
990 PARALLEL incoming that failed vt_get_decl_and_offset check.
993 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
994 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
996 (get_variable_section): For decls in named .bss* sections pass true as
997 second argument to bss_initializer_p.
999 2018-02-09 Marek Polacek <polacek@redhat.com>
1000 Jakub Jelinek <jakub@redhat.com>
1003 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1004 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1005 Sync some changes from cxx_fold_indirect_ref.
1007 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1009 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1011 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1012 (BLOCK_INLINE_ENTRY_LABEL): New.
1013 (dwarf2out_var_location): Disregard inline entry markers.
1014 (inline_entry_data): New struct.
1015 (inline_entry_data_hasher): New hashtable type.
1016 (inline_entry_data_hasher::hash): New.
1017 (inline_entry_data_hasher::equal): New.
1018 (inline_entry_data_table): New variable.
1019 (add_high_low_attributes): Add DW_AT_entry_pc and
1020 DW_AT_GNU_entry_view attributes if a pending entry is found
1021 in inline_entry_data_table. Add old entry_pc attribute only
1022 if debug nonbinding markers are disabled.
1023 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1024 markers are enabled.
1025 (block_within_block_p, dwarf2out_inline_entry): New.
1026 (dwarf2out_finish): Check that no entries remained in
1027 inline_entry_data_table.
1028 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1029 (final_scan_insn, notice_source_line): Likewise.
1030 (rest_of_clean_state): Skip inline entry markers.
1031 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1033 * gimple.c (gimple_build_debug_inline_entry): New.
1034 * gimple.h (enum gimple_debug_subcode): Add
1035 GIMPLE_DEBUG_INLINE_ENTRY.
1036 (gimple_build_debug_inline_entry): Declare.
1037 (gimple_debug_inline_entry_p): New.
1038 (gimple_debug_nonbind_marker_p): Adjust.
1039 * insn-notes.def (INLINE_ENTRY): New.
1040 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1041 inline entry marker notes.
1042 (print_insn): Likewise.
1043 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1044 (INSN_DEBUG_MARKER_KIND): Likewise.
1045 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1046 * tree-inline.c (expand_call_inline): Build and insert
1047 debug_inline_entry stmt.
1048 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1049 inline entry blocks early, if nonbind markers are enabled.
1050 (dump_scope_block): Dump fragment info.
1051 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1052 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1053 (gimple_build_debug_inline_entry): New.
1054 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1055 Enable/disable inline entry points too.
1056 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1057 (DEBUG_INSN): Describe inline entry markers.
1059 * common.opt (gvariable-location-views): New.
1060 (gvariable-location-views=incompat5): New.
1061 * config.in: Rebuilt.
1062 * configure: Rebuilt.
1063 * configure.ac: Test assembler for view support.
1064 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1065 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1066 * dwarf2out.c (var_loc_view): New typedef.
1067 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1068 (dwarf2out_locviews_in_attribute): New.
1069 (dwarf2out_locviews_in_loclist): New.
1070 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1071 (enum dw_line_info_opcode): Add LI_adv_address.
1072 (struct dw_line_info_table): Add view.
1073 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1074 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1075 (zero_view_p): New variable.
1076 (ZERO_VIEW_P): New macro.
1077 (output_asm_line_debug_info): New.
1078 (struct var_loc_node): Add view.
1079 (add_AT_view_list, AT_loc_list): New.
1080 (add_var_loc_to_decl): Add view param. Test it against last.
1081 (new_loc_list): Add view params. Record them.
1082 (AT_loc_list_ptr): Handle loc and view lists.
1083 (view_list_to_loc_list_val_node): New.
1084 (print_dw_val): Handle dw_val_class_view_list.
1085 (size_of_die): Likewise.
1086 (value_format): Likewise.
1087 (loc_list_has_views): New.
1088 (gen_llsym): Set vl_symbol too.
1089 (maybe_gen_llsym, skip_loc_list_entry): New.
1090 (dwarf2out_maybe_output_loclist_view_pair): New.
1091 (output_loc_list): Output view list or entries too.
1092 (output_view_list_offset): New.
1093 (output_die): Handle dw_val_class_view_list.
1094 (output_dwarf_version): New.
1095 (output_compilation_unit_header): Use it.
1096 (output_skeleton_debug_sections): Likewise.
1097 (output_rnglists, output_line_info): Likewise.
1098 (output_pubnames, output_aranges): Update version comments.
1099 (output_one_line_info_table): Output view numbers in asm comments.
1100 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1101 Call maybe_gen_llsym.
1102 (loc_list_from_tree_1): Adjust.
1103 (add_AT_location_description): Create view list attribute if
1104 needed, check it's absent otherwise.
1105 (convert_cfa_to_fb_loc_list): Adjust.
1106 (maybe_emit_file): Call output_asm_line_debug_info for test.
1107 (dwarf2out_var_location): Reset views as needed. Precompute
1108 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1109 attribute. Set view.
1110 (new_line_info_table): Reset next view.
1111 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1112 (dwarf2out_source_line): Likewise. Output view resets and labels to
1113 the assembler, or select appropriate line info opcodes.
1114 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1115 (optimize_string_length): Catch it. Adjust.
1116 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1117 dw_val_class_view_list, and remove it if no longer needed.
1118 (hash_loc_list): Hash view numbers.
1119 (loc_list_hasher::equal): Compare them.
1120 (optimize_location_lists): Check whether a view list symbol is
1121 needed, and whether the locview attribute is present, and
1122 whether they match. Remove the locview attribute if no longer
1124 (index_location_lists): Call skip_loc_list_entry for test.
1125 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1126 Use output_dwarf_version.
1127 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1128 (struct dw_val_node): Add val_view_list.
1129 * final.c (SEEN_NEXT_VIEW): New.
1130 (set_next_view_needed): New.
1131 (clear_next_view_needed): New.
1132 (maybe_output_next_view): New.
1133 (final_start_function): Rename to...
1134 (final_start_function_1): ... this. Take pointer to FIRST,
1135 add SEEN parameter. Emit param bindings in the initial view.
1136 (final_start_function): Reintroduce SEEN-less interface.
1137 (final): Rename to...
1138 (final_1): ... this. Take SEEN parameter. Output final pending
1139 next view at the end.
1140 (final): Reintroduce seen-less interface.
1141 (final_scan_insn): Output pending next view before switching
1142 sections or ending a block. Mark the next view as needed when
1143 outputting variable locations. Notify debug backend of section
1144 changes, and of location view changes.
1145 (rest_of_handle_final): Adjust.
1146 * toplev.c (process_options): Autodetect value for debug variable
1147 location views option. Warn on incompat5 without -gdwarf-5.
1148 * doc/invoke.texi (gvariable-location-views): New.
1149 (gvariable-location-views=incompat5): New.
1150 (gno-variable-location-views): New.
1152 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1154 PR tree-optimization/84136
1155 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1156 that the result of find_edge is non-NULL.
1158 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1161 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1162 storing integer register in SImode. Fix cost of 256 and 512
1163 byte aligned SSE register store.
1165 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1167 * config/i386/i386.c (ix86_multiplication_cost): Fix
1168 multiplication cost for TARGET_AVX512DQ.
1170 2018-02-08 Marek Polacek <polacek@redhat.com>
1172 PR tree-optimization/84238
1173 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1176 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1178 PR tree-optimization/84265
1179 * tree-vect-stmts.c (vectorizable_store): Don't treat
1180 VMAT_CONTIGUOUS accesses as grouped.
1181 (vectorizable_load): Likewise.
1183 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1185 PR tree-optimization/81635
1186 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1187 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1188 (test_round_for_mask): New functions.
1189 (wide_int_cc_tests): Call test_round_for_mask.
1190 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1191 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1192 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1193 range returned by get_range_info.
1195 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1198 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1199 * symtab.c: Include builtins.h
1200 (symtab_node::output_to_lto_symbol_table_p): Move here
1201 from lto-streamer-out.c:output_symbol_p.
1202 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1203 (output_symbol_p): Move all logic to symtab.c
1204 (produce_symtab): Update.
1206 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1208 * config/s390/s390-opts.h (enum indirect_branch): Define.
1209 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1210 (s390_indirect_branch_via_thunk)
1211 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1212 (enum s390_indirect_branch_type): Define.
1213 * config/s390/s390.c (struct s390_frame_layout, struct
1214 machine_function): Remove.
1215 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1216 (indirect_branch_table_label_no, indirect_branch_table_name):
1218 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1219 (enum s390_indirect_branch_option): Define.
1220 (s390_return_addr_from_memory): New function.
1221 (s390_handle_string_attribute): New function.
1222 (s390_attribute_table): Add new attribute handler.
1223 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1224 (s390_indirect_branch_via_thunk): New function.
1225 (s390_indirect_branch_via_inline_thunk): New function.
1226 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1227 sibling call optimization for non z10 compiles.
1228 (s390_emit_call): Force indirect branch target to be a single
1229 register. Add r1 clobber for non-z10 compiles.
1230 (s390_emit_epilogue): Emit return jump via return_use expander.
1231 (s390_reorg): Handle JUMP_INSNs as execute targets.
1232 (s390_option_override_internal): Perform validity checks for the
1233 new command line options.
1234 (s390_indirect_branch_attrvalue): New function.
1235 (s390_indirect_branch_settings): New function.
1236 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1237 (s390_output_indirect_thunk_function): New function.
1238 (s390_code_end): Implement target hook.
1239 (s390_case_values_threshold): Implement target hook.
1240 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1242 * config/s390/s390.h (struct s390_frame_layout)
1243 (struct machine_function): Move here from s390.c.
1244 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1245 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1246 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1247 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1248 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1249 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1250 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1251 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1252 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1253 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1254 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1255 (mnemonic attribute): Add values which aren't recognized
1257 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1258 pattern for branch conversion. Fix mnemonic attribute.
1259 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1260 indirect branch via thunk if requested.
1261 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1262 ("*indirect_jump"): Disable for branch conversion using out of
1264 ("indirect_jump_via_thunk<mode>_z10")
1265 ("indirect_jump_via_thunk<mode>")
1266 ("indirect_jump_via_inlinethunk<mode>_z10")
1267 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1268 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1269 ("casesi_jump_via_inlinethunk<mode>_z10")
1270 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1271 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1272 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1273 ("*indirect2_jump"): Disable for branch conversion.
1274 ("casesi_jump"): Turn into expander and expand patterns for branch
1276 ("return_use"): New expander.
1277 ("*return"): Emit return via thunk and rename it to ...
1278 ("*return<mode>"): ... this one.
1279 * config/s390/s390.opt: Add new options and and enum for the
1282 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1284 * lra-constraints.c (match_reload): Unconditionally use
1285 gen_lowpart_SUBREG, rather than selecting between that
1286 and equivalent gen_rtx_SUBREG code.
1288 2018-02-08 Richard Biener <rguenther@suse.de>
1290 PR tree-optimization/84233
1291 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1292 changed flag instead of boguously re-using phi_inserted.
1294 2018-02-08 Martin Jambor <mjambor@suse.cz>
1296 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1297 static local variables.
1299 2018-02-08 Richard Biener <rguenther@suse.de>
1301 PR tree-optimization/84278
1302 * tree-vect-stmts.c (vectorizable_store): When looking for
1303 smaller vector types to perform grouped strided loads/stores
1304 make sure the mode is supported by the target.
1305 (vectorizable_load): Likewise.
1307 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1309 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1310 Increase LDP/STP opportunities by adding adjacent callee-saves.
1312 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1314 PR rtl-optimization/84068
1315 PR rtl-optimization/83459
1316 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1318 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1320 PR tree-optimization/84224
1321 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1322 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1325 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1328 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1329 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1330 count, remove one USE.
1332 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1334 * doc/install.texi (Configuration): Document the
1335 --with-long-double-format={ibm,ieee} PowerPC configuration
1339 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1340 Convert from define_expand to be define_insn_and_split. Rework
1341 float/double/_Float128 conversions to QI/HI/SImode to work with
1342 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1343 conversions to QI/HImode types did a store and then a load to
1344 truncate the value. For conversions to VSX registers, don't split
1345 the insn, instead emit the code directly. Use the code iterator
1346 any_fix to combine signed and unsigned conversions.
1347 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1348 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1349 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1350 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1351 (fix_<mode>di2_hw): Likewise.
1352 (fixuns_<mode>di2_hw): Likewise.
1353 (fix_<mode>si2_hw): Likewise.
1354 (fixuns_<mode>si2_hw): Likewise.
1355 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1356 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1357 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1358 fix<uns>_trunc<SFDF:mode>si2_p8.
1359 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1361 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1362 (fix<uns>_<mode>_mem): Likewise.
1363 (fctiw<u>z_<mode>_mem): Likewise.
1364 (fix<uns>_<mode>_mem): Likewise.
1365 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1366 the register allocator from doing a direct move to the GPRs to do
1367 a store, and instead use the ISA 3.0 store byte/half-word from
1368 vector register instruction. For IEEE 128-bit floating point,
1369 also optimize stores of 32-bit ints.
1370 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1372 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1374 * genextract.c (push_pathstr_operand): New function to support
1376 (walk_rtx): Call push_pathstr_operand.
1377 (print_path): Support [a-zA-Z].
1379 2018-02-07 Richard Biener <rguenther@suse.de>
1381 PR tree-optimization/84037
1382 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1383 (cse_and_gimplify_to_preheader): Declare.
1384 (vect_get_place_in_interleaving_chain): Likewise.
1385 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1387 (_loop_vec_info::~_loop_vec_info): Delete it.
1388 (cse_and_gimplify_to_preheader): New function.
1389 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1390 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1391 (vectorizable_load): Likewise. For grouped stores always base
1392 the IV on the first element.
1393 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1394 condition before gimplifying.
1396 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1398 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1399 *DIV_EXPR and *MOD_EXPR.
1401 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1404 * config/i386/i386.c (ix86_option_override_internal): Mask out
1405 the CF_SET bit when checking -fcf-protection.
1407 2018-02-07 Tom de Vries <tom@codesourcery.com>
1410 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1413 2018-02-07 Richard Biener <rguenther@suse.de>
1415 PR tree-optimization/84204
1416 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1419 PR tree-optimization/84205
1420 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1421 special-case isl_ast_op_zdiv_r.
1423 PR tree-optimization/84223
1424 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1425 Only add conditions from within the region.
1426 (gather_bbs::after_dom_children): Adjust.
1428 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1431 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1432 * config/avr/avr.md: Only post-reload split REG-REG moves if
1433 either register is REGERAL_REG_P.
1435 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1437 PR tree-optimization/84235
1438 * tree-ssa-scopedtables.c
1439 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1440 if the subtraction is performed in floating point type where NaNs are
1441 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1442 build 1. Formatting fix.
1444 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1447 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1448 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1449 and skip it regardless of bb boundaries. Use CALL_P macro,
1450 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1451 unnecessarily, formatting fix.
1453 2018-02-06 Michael Collison <michael.collison@arm.com>
1455 * config/arm/thumb2.md:
1456 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1457 (*thumb_mov_notscc): Ditto.
1459 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1462 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1463 unsigned_fix, not "s".
1465 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1467 * configure.ac (gcc_fn_eh_frame_ro): New function.
1468 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1469 correct .eh_frame permissions.
1470 * configure: Regenerate.
1472 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1474 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1477 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1479 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1480 Display warning message for -mno-speculate-indirect-jumps.
1482 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1484 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1486 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1488 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1490 PR tree-optimization/84225
1491 * tree-eh.c (find_trapping_overflow): Only call
1492 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1494 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1497 * config/i386/i386.c: Reimplement the check of possible options
1498 -mibt/-mshstk conbination. Change error messages.
1499 * doc/invoke.texi: Fix a typo: remove extra '='.
1501 2018-02-06 Marek Polacek <polacek@redhat.com>
1503 PR tree-optimization/84228
1504 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1506 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1509 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1510 emitted arch directives.
1511 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1512 __ARM_FEATURE_COPROC before changing architectures.
1514 2018-02-06 Richard Biener <rguenther@suse.de>
1516 * config/i386/i386.c (print_reg): Fix typo.
1517 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1519 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1521 * configure: Regenerate.
1523 2018-02-05 Martin Sebor <msebor@redhat.com>
1525 PR tree-optimization/83369
1526 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1529 2018-02-05 Martin Liska <mliska@suse.cz>
1531 * doc/invoke.texi: Cherry-pick upstream r323995.
1533 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1535 * ira.c (ira_init_register_move_cost): Adjust comment.
1537 2018-02-05 Martin Liska <mliska@suse.cz>
1539 PR gcov-profile/84137
1540 * doc/gcov.texi: Fix typo in documentation.
1542 2018-02-05 Martin Liska <mliska@suse.cz>
1544 PR gcov-profile/83879
1545 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1546 using dlopen functionality.
1548 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1550 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1551 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1552 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1553 _mm_maskz_range_round_ss): New intrinsics.
1554 (__builtin_ia32_rangesd128_round)
1555 (__builtin_ia32_rangess128_round): Remove.
1556 (__builtin_ia32_rangesd128_mask_round,
1557 __builtin_ia32_rangess128_mask_round): New builtins.
1558 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1559 __builtin_ia32_rangess128_round): Remove.
1560 (__builtin_ia32_rangesd128_mask_round,
1561 __builtin_ia32_rangess128_mask_round): New builtins.
1562 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1563 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1564 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1565 "<round_saeonly_constraint>")): Changed to ...
1566 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1567 "<round_saeonly_scalar_constraint>")): ... this.
1568 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1569 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1570 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1571 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1572 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1574 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1576 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1578 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1579 Remove all values except native, 8540 and 8548.
1581 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1583 * config/i386/i386.c (ix86_output_function_return): Pass
1584 INVALID_REGNUM, instead of -1, as invalid register number to
1585 indirect_thunk_name and output_indirect_thunk.
1587 2018-02-02 Julia Koval <julia.koval@intel.com>
1589 * config.gcc: Add -march=icelake.
1590 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1591 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1592 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1593 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1594 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1595 (processor_target_table): Add icelake.
1596 (ix86_option_override_internal): Handle new PTAs.
1597 (get_builtin_code_for_version): Handle icelake.
1598 (M_INTEL_COREI7_ICELAKE): New.
1599 (fold_builtin_cpu): Handle icelake.
1600 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1601 * doc/invoke.texi: Add -march=icelake.
1603 2018-02-02 Julia Koval <julia.koval@intel.com>
1605 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1606 to wide_int_bitmask.
1607 * wide-int-bitmask.h: New.
1609 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1612 * config/i386/i386.md: Replace Pmode with word_mode in
1613 builtin_setjmp_setup and builtin_longjmp to support x32.
1615 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1619 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1621 (rs6000_supported_cpu_names): New static variable.
1622 (linux_cpu_translation_table): Likewise.
1623 (elf_platform) <cpu>: Define new static variable and use it.
1624 Translate kernel AT_PLATFORM name to canonical name if needed.
1625 Error if platform name is unknown.
1627 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1630 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1632 2018-02-01 Jeff Law <law@redhat.com>
1635 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1636 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1637 the scratch if RELEASE_VIA_POP is false.
1638 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1639 If we have to save a temporary register, decrement SIZE appropriately.
1640 Pass new arguments to release_scratch_register_on_entry.
1641 (ix86_adjust_stack_and_probe): Likewise.
1642 (ix86_emit_probe_stack_range): Pass new arguments to
1643 release_scratch_register_on_entry.
1645 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1647 PR rtl-optimization/84157
1648 * combine.c (change_zero_ext): Use REG_P predicate in
1649 front of HARD_REGISTER_P predicate.
1651 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1653 * config/avr/avr.c (avr_option_override): Move disabling of
1654 -fdelete-null-pointer-checks to...
1655 * common/config/avr/avr-common.c (avr_option_optimization_table):
1658 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1660 PR tree-optimization/81635
1661 * tree-data-ref.c (split_constant_offset_1): For types that
1662 wrap on overflow, try to use range info to prove that wrapping
1665 2018-02-01 Renlin Li <renlin.li@arm.com>
1668 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1670 (aarch64_register_move_cost): Likewise.
1671 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1673 (REG_CLASS_NAMES): Likewise.
1674 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1675 TAILCALL_ADDR_REGS. Remove IP registers.
1676 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1678 2018-02-01 Richard Biener <rguenther@suse.de>
1680 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1681 for specifying RPO order and allow NULL for that.
1682 * domwalk.c (dom_walker::dom_walker): Likewise.
1683 (dom_walker::walk): Handle NULL RPO order.
1684 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1686 (rewrite_update_dom_walker): Likewise.
1687 (mark_def_dom_walker): Likewise.
1689 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1691 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1692 (aarch64_maybe_expand_sve_subreg_move): Declare.
1693 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1694 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1696 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1697 that are semantically a reverse operation.
1698 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1699 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1700 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1702 (aarch64_can_change_mode_class): For big-endian, forbid changes
1703 between two SVE modes if they have different element sizes.
1705 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1707 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1708 the TImode handling for big-endian targets.
1710 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1712 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1713 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1715 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1716 Remove BSWAP handing for big-endian targets and use the form of
1717 LD1RQ appropariate for the mode.
1719 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1721 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1722 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1725 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1728 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1729 check for operands that need to go through aarch64_sve_reload_be.
1731 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1733 PR tree-optimization/81661
1734 PR tree-optimization/84117
1735 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1736 * tree-eh.c: Include gimplify.h.
1737 (find_trapping_overflow, replace_trapping_overflow,
1738 rewrite_to_non_trapping_overflow): New functions.
1739 * tree-vect-loop.c: Include tree-eh.h.
1740 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1741 * tree-data-ref.c: Include tree-eh.h.
1742 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1744 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1746 PR rtl-optimization/84123
1747 * combine.c (change_zero_ext): Check if hard register satisfies
1748 can_change_dest_mode before calling gen_lowpart_SUBREG.
1750 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1753 * ira.c (ira_init_register_move_cost): Remove assert.
1755 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1757 PR rtl-optimization/84071
1758 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1759 * doc/tm.texi: Regenerate.
1761 2018-01-31 Richard Biener <rguenther@suse.de>
1763 PR tree-optimization/84132
1764 * tree-data-ref.c (analyze_miv_subscript): Properly
1765 check whether evolution_function_is_affine_multivariate_p
1766 before calling gcd_of_steps_may_divide_p.
1768 2018-01-31 Julia Koval <julia.koval@intel.com>
1771 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1772 * config/i386/i386.md (rdpid_rex64) New.
1773 (rdpid): Make 32bit only.
1775 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1778 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1779 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1781 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1784 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1786 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1788 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1790 PR rtl-optimization/84071
1791 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1792 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1794 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1796 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1797 (arc_attribute_table): Add 'aux' attribute.
1798 (arc_in_small_data_p): Consider aux like variables.
1799 (arc_is_aux_reg_p): New function.
1800 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1801 (arc_get_aux_arg): New function.
1802 (prepare_move_operands): Handle aux-register access.
1803 (arc_handle_aux_attribute): New function.
1804 * doc/extend.texi (ARC Variable attributes): Add subsection.
1806 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1808 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1809 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1810 (arc_attribute_table): Add 'uncached' attribute.
1811 (arc_print_operand): Print '.di' flag for uncached memory
1813 (arc_in_small_data_p): Do not consider for small data the uncached
1815 (arc_is_uncached_mem_p): New function.
1816 * config/arc/predicates.md (compact_store_memory_operand): Check
1817 for uncached memory accesses.
1818 (nonvol_nonimm_operand): Likewise.
1819 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1821 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1824 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1825 falign-loops=): Add Optimization flag.
1827 2018-01-30 Jeff Law <law@redhat.com>
1830 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1831 INT_REGISTERS_SAVED. Check it prior to calling
1832 get_scratch_register_on_entry.
1833 (ix86_adjust_stack_and_probe): Similarly.
1834 (ix86_emit_probe_stack_range): Similarly.
1835 (ix86_expand_prologue): Corresponding changes.
1837 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1840 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1841 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1843 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1846 * lra-constraints.c (curr_insn_transform): Process AND in the
1849 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1851 PR rtl-optimization/83986
1852 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1853 dependence against last_pending_memory_flush in addition to
1856 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1858 PR tree-optimization/81611
1859 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1862 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1865 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1868 2018-01-30 Richard Biener <rguenther@suse.de>
1869 Jakub Jelinek <jakub@redhat.com>
1871 PR tree-optimization/84111
1872 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1873 inner loops added during recursion, as they don't have up-to-date
1876 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1879 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1880 (can_inline_edge_by_limits_p): ... here.
1881 (can_early_inline_edge_p, check_callers,
1882 update_caller_keys, update_callee_keys, recursive_inlining,
1883 add_new_edges_to_heap, speculation_useful_p,
1884 inline_small_functions,
1885 inline_small_functions, flatten_function,
1886 inline_to_all_callers_1): Update.
1888 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1890 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1893 2018-01-30 Richard Biener <rguenther@suse.de>
1895 PR tree-optimization/83008
1896 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1897 invariant and constant vector uses in stmts when they need
1900 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1903 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1904 * configure: Regenerate.
1906 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1908 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1910 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1911 Use gen_rtx_REG rather than gen_lowpart.
1913 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1915 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1916 rather than 0 when creating partial subregs.
1918 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1920 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1923 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1926 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1927 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1928 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1929 flags. This restores the settings used before the 2017-07-24.
1930 Turning off pre increment/decrement/modify allows IVOPTS to
1931 optimize DF/SF loops where the index is an int.
1933 2018-01-29 Richard Biener <rguenther@suse.de>
1934 Kelvin Nilsen <kelvin@gcc.gnu.org>
1937 * tree-vect-stmts.c (vectorizable_call): Don't call
1938 targetm.vectorize_builtin_md_vectorized_function if callee is
1941 2018-01-22 Carl Love <cel@us.ibm.com>
1943 * doc/extend.tex: Fix typo in second arg in
1944 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1946 2018-01-29 Richard Biener <rguenther@suse.de>
1948 PR tree-optimization/84086
1949 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1950 (flush_ssaname_freelist): When SSA names were released reset
1951 the SCEV hash table.
1953 2018-01-29 Richard Biener <rguenther@suse.de>
1955 PR tree-optimization/84057
1956 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1957 removed paths when removing edges.
1959 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1961 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1962 -mfunction-return=@var{choice}.
1964 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1967 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1969 (layout::print_source_line): Likewise.
1970 (test_get_line_width_without_trailing_whitespace): Add test cases.
1972 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1975 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1978 2018-01-26 Jim Wilson <jimw@sifive.com>
1980 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1982 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1985 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1987 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1988 and CMP + SUB-immediate -> SUBS.
1990 2018-01-26 Martin Sebor <msebor@redhat.com>
1992 PR tree-optimization/83896
1993 * tree-ssa-strlen.c (get_string_len): Rename...
1994 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1995 Avoid assuming length is constant.
1996 (handle_char_store): Use HOST_WIDE_INT for string length.
1998 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2001 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2002 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2004 2018-01-26 Richard Biener <rguenther@suse.de>
2006 PR rtl-optimization/84003
2007 * dse.c (record_store): Only record redundant stores when
2008 the earlier store aliases at least all accesses the later one does.
2010 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2012 PR rtl-optimization/83985
2013 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2014 REG_CFA_RESTORE insns.
2015 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2016 REG_CFA_RESTORE insns here.
2019 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2020 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2022 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2024 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2025 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2026 (arc_init): Likewise.
2027 (arc_override_options): Likewise.
2028 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2030 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2032 * config/arc/arc.h (TARGET_DBNZ): Define.
2033 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2034 properly set the tune attribute.
2035 (dbnz): Use TARGET_DBNZ guard.
2036 * config/arc/arc.opt (mtune): Add core3 option.
2038 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2040 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2041 recognize new pic like addresses.
2042 (arc_delegitimize_address): Clean up.
2044 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2046 * config/arc/arc-arches.def: Option mrf16 valid for all
2048 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2049 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2050 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2051 * config/arc/arc-tables.opt: Regenerate.
2052 * config/arc/arc.c (arc_conditional_register_usage): Handle
2053 reduced register file case.
2054 (arc_file_start): Set must have build attributes.
2055 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2057 * config/arc/arc.opt (mrf16): Add new option.
2058 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2059 * config/arc/genmultilib.awk: Handle new mrf16 option.
2060 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2061 * config/arc/t-multilib: Regenerate.
2062 * doc/invoke.texi (ARC Options): Document mrf16 option.
2064 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2066 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2067 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2068 (arc_attribute_table): Add 'secure_call' attribute.
2069 (arc_print_operand): Print secure call operand.
2070 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2072 (arc_is_secure_call_p): New function. * config/arc/arc.md
2073 (call_i): Add support for sjli instruction.
2074 (call_value_i): Likewise.
2075 * config/arc/constraints.md (Csc): New constraint.
2077 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2078 John Eric Martin <John.Martin@emmicro-us.com>
2080 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2081 * config/arc/arc.c (_arc_jli_section): New struct.
2082 (arc_jli_section): New type.
2083 (rc_jli_sections): New static variable.
2084 (arc_handle_jli_attribute): New function.
2085 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2086 (arc_file_end): New function.
2087 (TARGET_ASM_FILE_END): Define.
2088 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2089 (arc_add_jli_section): New function.
2090 (jli_call_scan): Likewise.
2091 (arc_reorg): Call jli_call_scan.
2092 (arc_output_addsi): Remove 'S' from printing asm operand.
2093 (arc_is_jli_call_p): New function.
2094 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2096 (movhi_insn): Likewise.
2097 (movsi_insn): Likewise.
2098 (movsi_set_cc_insn): Likewise.
2099 (loadqi_update): Likewise.
2100 (load_zeroextendqisi_update): Likewise.
2101 (load_signextendqisi_update): Likewise.
2102 (loadhi_update): Likewise.
2103 (load_zeroextendhisi_update): Likewise.
2104 (load_signextendhisi_update): Likewise.
2105 (loadsi_update): Likewise.
2106 (loadsf_update): Likewise.
2107 (movsicc_insn): Likewise.
2108 (bset_insn): Likewise.
2109 (bxor_insn): Likewise.
2110 (bclr_insn): Likewise.
2111 (bmsk_insn): Likewise.
2112 (bicsi3_insn): Likewise.
2113 (cmpsi_cc_c_insn): Likewise.
2114 (movsi_ne): Likewise.
2115 (movsi_cond_exec): Likewise.
2116 (clrsbsi2): Likewise.
2127 (call_i): Remove 'S' asm letter, add jli instruction.
2128 (call_value_i): Likewise.
2129 * config/arc/arc.op (mjli-always): New option.
2130 * config/arc/constraints.md (Cji): New constraint.
2131 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2133 (subsf3_fpx): Likewise.
2134 (mulsf3_fpx): Likewise.
2135 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2137 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2139 * doc/invoke.texi (ARC): Document mjli-always option.
2141 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2143 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2144 avoid addition with 0 and use incw and decw where possible.
2146 2018-01-26 Richard Biener <rguenther@suse.de>
2148 PR tree-optimization/81082
2149 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2150 association if it requires casting to unsigned.
2151 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2152 from fold_plusminus_mult_expr to catch important cases late when
2153 range info is available.
2155 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2157 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2158 * configure.ac (hidden_linkonce): New test.
2159 * configure: Regenerate.
2160 * config.in: Regenerate.
2162 2018-01-26 Julia Koval <julia.koval@intel.com>
2164 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2165 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2166 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2167 _mm_mask_bitshuffle_epi64_mask): Fix type.
2168 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2169 USI_FTYPE_V4DI_V4DI_USI): Remove.
2170 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2171 __builtin_ia32_vpshufbitqmb256_mask,
2172 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2173 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2174 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2176 2018-01-26 Alan Modra <amodra@gmail.com>
2179 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2180 UNSPEC_VBPERMQ. Sort other unspecs.
2182 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2184 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2186 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2189 * predict.c (drop_profile): Do not push/pop cfun; update also
2191 (handle_missing_profiles): Fix logic looking for zero profiles.
2193 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2196 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2197 on functions with #pragma omp declare simd or functions with simd
2199 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2200 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2201 Remove trailing \n from warning_at calls.
2203 2018-01-25 Tom de Vries <tom@codesourcery.com>
2206 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2207 for neutered workers.
2209 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2212 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2213 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2215 2018-01-24 Jeff Law <law@redhat.com>
2218 * i386.c (get_probe_interval): Move to earlier point.
2219 (ix86_compute_frame_layout): If -fstack-clash-protection and
2220 the frame is larger than the probe interval, then use pushes
2221 to save registers rather than reg->mem moves.
2222 (ix86_expand_prologue): Remove conditional for int_registers_saved
2225 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2228 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2229 min/max for never referenced object.
2231 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2234 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2236 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2237 attributes from DECL_ATTRIBUTES (decl) without affecting
2238 DECL_ATTRIBUTES (current_function_decl).
2239 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2240 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2242 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2244 PR tree-optimization/83979
2245 * fold-const.c (fold_comparison): Use constant_boolean_node
2246 instead of boolean_{true,false}_node.
2248 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2250 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2253 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2255 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2256 Simplify the clause that sets the length attribute.
2257 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2258 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2259 clause that sets the length attribute.
2260 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2262 2018-01-24 Tom de Vries <tom@codesourcery.com>
2265 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2266 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2267 Add strict parameter.
2268 (prevent_branch_around_nothing): Insert dummy insn between branch to
2269 label and label with no ptx insn inbetween.
2270 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2272 2018-01-24 Tom de Vries <tom@codesourcery.com>
2275 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2276 for neutered threads in warp.
2277 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2279 2018-01-24 Richard Biener <rguenther@suse.de>
2281 PR tree-optimization/83176
2282 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2285 2018-01-24 Richard Biener <rguenther@suse.de>
2287 PR tree-optimization/82819
2288 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2289 code generating pluses that are no-ops in the target precision.
2291 2018-01-24 Richard Biener <rguenther@suse.de>
2294 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2296 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2298 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2299 to merge probabilities.
2300 * predict.c (probably_never_executed): Also mark as cold functions
2301 with global 0 profile and guessed local profile.
2302 * profile-count.c (profile_probability::combine_with_count): New
2304 * profile-count.h (profile_probability::operator*,
2305 profile_probability::operator*=, profile_probability::operator/,
2306 profile_probability::operator/=): Reduce precision to adjusted
2307 and set value to guessed on contradictory divisions.
2308 (profile_probability::combine_with_freq): Remove.
2309 (profile_probability::combine_wiht_count): Declare.
2310 (profile_count::force_nonzero):: Set to adjusted.
2311 (profile_count::probability_in):: Set quality to adjusted.
2312 * tree-ssa-tail-merge.c (replace_block_by): Use
2315 2018-01-23 Andrew Waterman <andrew@sifive.com>
2316 Jim Wilson <jimw@sifive.com>
2318 * config/riscv/riscv.c (riscv_stack_boundary): New.
2319 (riscv_option_override): Set riscv_stack_boundary. Handle
2320 riscv_preferred_stack_boundary_arg.
2321 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2322 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2323 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2324 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2325 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2326 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2328 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2331 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2332 of struct ix86_frame.
2333 (ix86_expand_epilogue): Likewise. Add a local variable for
2334 the reg_save_offset field in struct ix86_frame.
2336 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2338 PR tree-optimization/82604
2339 * tree-loop-distribution.c (enum partition_kind): New enum item
2340 PKIND_PARTIAL_MEMSET.
2341 (partition_builtin_p): Support above new enum item.
2342 (generate_code_for_partition): Ditto.
2343 (compute_access_range): Differentiate cases that equality can be
2344 proven at all loops, the innermost loops or no loops.
2345 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2346 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2347 (finalize_partitions, distribute_loop): Don't fuse partition of
2348 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2349 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2352 2018-01-23 Martin Liska <mliska@suse.cz>
2354 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2355 order to ignore the predictor.
2356 (PRED_POLYMORPHIC_CALL): Likewise.
2357 (PRED_RECURSIVE_CALL): Likewise.
2359 2018-01-23 Martin Liska <mliska@suse.cz>
2361 * tree-profile.c (tree_profiling): Print function header to
2362 aware reader which function we are working on.
2363 * value-prof.c (gimple_find_values_to_profile): Do not print
2364 not interesting value histograms.
2366 2018-01-23 Martin Liska <mliska@suse.cz>
2368 * profile-count.h (enum profile_quality): Add
2369 profile_uninitialized as the first value. Do not number values
2370 as they are zero based.
2371 (profile_count::verify): Update sanity check.
2372 (profile_probability::verify): Likewise.
2374 2018-01-23 Nathan Sidwell <nathan@acm.org>
2376 * doc/invoke.texi (ffor-scope): Deprecate.
2378 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2380 PR tree-optimization/83510
2381 * domwalk.c (set_all_edges_as_executable): New function.
2382 (dom_walker::dom_walker): Convert bool param
2383 "skip_unreachable_blocks" to enum reachability. Move setup of
2384 edge flags to set_all_edges_as_executable and only do it when
2385 reachability is REACHABLE_BLOCKS.
2386 * domwalk.h (enum dom_walker::reachability): New enum.
2387 (dom_walker::dom_walker): Convert bool param
2388 "skip_unreachable_blocks" to enum reachability.
2389 (set_all_edges_as_executable): New decl.
2390 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2391 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2393 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2394 but converting true to REACHABLE_BLOCKS.
2395 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2397 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2398 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2399 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2401 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2402 if check_all_array_refs will be called.
2404 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2406 * tree.c (selftest::test_location_wrappers): Add more test
2409 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2411 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2412 (selftest::test_bit_in_range): Likewise.
2414 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2417 * doc/sourcebuild.texi (vect_float): Say that the selector
2418 only describes the situation when -funsafe-math-optimizations is on.
2419 (vect_float_strict): Document.
2421 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2423 PR tree-optimization/83965
2424 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2425 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2426 instead of checking only for a reduction.
2427 (vect_recog_widen_sum_pattern): Likewise.
2429 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2431 * predict.c (probably_never_executed): Only use precise profile info.
2432 (compute_function_frequency): Skip after inlining hack since we now
2433 have quality checking.
2435 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2437 * profile-count.h (profile_probability::very_unlikely,
2438 profile_probability::unlikely, profile_probability::even): Set
2439 precision to guessed.
2441 2018-01-23 Richard Biener <rguenther@suse.de>
2443 PR tree-optimization/83963
2444 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2445 Properly terminate dominator walk when crossing the exit edge not
2446 when visiting its source block.
2448 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2451 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2452 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2454 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2456 PR tree-optimization/83957
2457 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2458 semicolon after for body surrounded by braces.
2460 PR tree-optimization/83081
2461 * profile-count.h (profile_probability::split): New method.
2462 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2463 Use profile_probability::split.
2464 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2465 when splitting a single conditional jump into 2.
2467 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2469 PR tree-optimization/69452
2470 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2473 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2475 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2476 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2477 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2479 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2481 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2482 * config/rl78/rl78.md (movdi): New define_expand.
2483 * config/rl78/rl78.c (rl78_split_movdi): New function.
2485 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2488 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2490 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2491 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2492 128-bit to produce an UNSPEC move to get the double word with the
2493 signbit and then a shift directly to do signbit.
2494 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2495 implementation with a new version that just does either a direct
2496 move or a regular move. Move memory interface to separate insns.
2497 Move insns so they are next to the expander.
2498 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2499 with signbit move. Split big and little endian case.
2500 (signbit<mode>2_dm_mem_le): Likewise.
2501 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2502 (signbit<mode>2_dm2): Likewise.
2504 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2506 * config/rl78/rl78.md (anddi3): New define_expand.
2508 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2510 * config/rl78/rl78.md (umindi3): New define_expand.
2512 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2514 * config/rl78/rl78.md (smindi3): New define_expand.
2516 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2518 * config/rl78/rl78.md (smaxdi3): New define_expand.
2520 2018-01-22 Carl Love <cel@us.ibm.com>
2522 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2523 LVX_V1TI): Add macro expansion.
2524 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2525 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2526 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2527 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2528 Change check to determine if the instruction is a byte reversing
2529 entry. Fix typo in comment.
2530 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2531 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2532 Add def_builtin calls for new builtins.
2533 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2534 Add define_insn expansion.
2536 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2538 * config/rl78/rl78.md (umaxdi3): New define_expand.
2540 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2542 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2543 for non-QImode registers.
2545 2018-01-22 Richard Biener <rguenther@suse.de>
2547 PR tree-optimization/83963
2548 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2549 including the loop exit block.
2550 (scop_detection::merge_sese): Likewise.
2551 (scop_detection::add_scop): Do it here instead.
2553 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2555 * doc/sourcebuild.texi (arm_softfloat): Document.
2557 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2560 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2561 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2562 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2564 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2565 David Edelsohn <dje.gcc@gmail.com>
2568 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2569 Change "crset eq" to "crset 2".
2570 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2571 (*call_indirect_aix<mode>_nospec): Likewise.
2572 (*call_value_indirect_aix<mode>_nospec): Likewise.
2573 (*call_indirect_elfv2<mode>_nospec): Likewise.
2574 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2575 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2576 change assembly output from . to $.
2577 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2578 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2579 (*tablejump<mode>_internal1_nospec): Likewise.
2581 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2584 * config/sh/sh_optimize_sett_clrt.cc:
2585 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2587 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2589 PR tree-optimization/83940
2590 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2591 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2592 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2593 use it to pass back the definition type.
2594 (vect_check_store_rhs): Likewise rhs_dt_out.
2595 (vect_build_gather_load_calls): Add a mask_dt argument and use
2596 it instead of a call to vect_is_simple_use.
2597 (vectorizable_store): Update calls to vect_check_load_store_mask
2598 and vect_check_store_rhs. Use the dt returned by the latter instead
2599 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2600 instead of calls to vect_is_simple_use. Pass the scalar rather
2601 than the vector operand to vect_is_simple_use when handling
2602 second and subsequent copies of an rhs value.
2603 (vectorizable_load): Update calls to vect_check_load_store_mask
2604 and vect_build_gather_load_calls. Use the cached mask_dt and
2605 gs_info.offset_dt instead of calls to vect_is_simple_use.
2607 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2610 * tree-emutls.c: Include gimplify.h.
2611 (lower_emutls_2): New function.
2612 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2613 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2614 it before further processing.
2617 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2618 UINTVAL (trueop1) instead of INTVAL (op1).
2620 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2624 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2625 INCOMING_FRAME_SP_OFFSET if not defined.
2626 (scan_trace): Add ENTRY argument. If true and
2627 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2628 emit a note to adjust the CFA offset.
2629 (create_cfi_notes): Adjust scan_trace callers.
2630 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2631 INCOMING_FRAME_SP_OFFSET in the CIE.
2632 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2633 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2635 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2636 * doc/tm.texi: Regenerated.
2638 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2640 PR rtl-optimization/83147
2641 * lra-constraints.c (remove_inheritance_pseudos): Use
2642 lra_substitute_pseudo_within_insn.
2644 2018-01-19 Tom de Vries <tom@codesourcery.com>
2645 Cesar Philippidis <cesar@codesourcery.com>
2648 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2650 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2653 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2654 spaces for function labels.
2656 2018-01-19 Martin Liska <mliska@suse.cz>
2658 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2659 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2660 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2661 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2662 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2663 (PRED_CONST_RETURN): Change from 69 to 65.
2664 (PRED_NULL_RETURN): Change from 91 to 71.
2665 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2666 (PRED_LOOP_GUARD): Change from 66 to 73.
2668 2018-01-19 Martin Liska <mliska@suse.cz>
2670 * predict.c (predict_insn_def): Add new assert.
2671 (struct branch_predictor): Change type to signed integer.
2672 (test_prediction_value_range): Amend test to cover
2674 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2675 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2676 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2677 (PRED_LOOP_IV_COMPARE): Likewise.
2678 * predict.h (PROB_UNINITIALIZED): Define new constant.
2680 2018-01-19 Martin Liska <mliska@suse.cz>
2682 * predict.c (dump_prediction): Add new format for
2683 analyze_brprob.py script which is enabled with -details
2685 * profile-count.h (precise_p): New function.
2687 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2689 PR tree-optimization/83922
2690 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2691 there are no statements that need masking.
2692 (vect_active_double_reduction_p): New function.
2693 (vect_analyze_loop_operations): Use it when handling phis that
2694 are not in the loop header.
2696 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2698 PR tree-optimization/83914
2699 * tree-vect-loop.c (vectorizable_induction): Don't convert
2700 init_expr or apply the peeling adjustment for inductions
2701 that are nested within the vectorized loop.
2703 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2705 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2708 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2712 * function.h (gimplify_parameters): Add gimple_seq * argument.
2713 * function.c: Include gimple.h and options.h.
2714 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2715 for the added local temporaries if needed.
2716 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2717 if there are any parameter cleanups, wrap whole body into a
2718 try/finally with the cleanups.
2720 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2723 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2724 Use GET_MODE_CLASS for scalar floating point.
2726 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2730 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2731 Fix call of call_cgraph_insertion_hooks.
2733 2018-01-18 Martin Sebor <msebor@redhat.com>
2735 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2737 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2740 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2743 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2746 * common.opt: (-ffile-prefix-map): New option.
2747 * opts.c (common_handle_option): Defer it.
2748 * opts-global.c (handle_common_deferred_options): Handle it.
2749 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2750 * file-prefix-map.h: New file.
2751 (remap_debug_filename, add_debug_prefix_map): ...here.
2752 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2753 * final.c (debug_prefix_map, add_debug_prefix_map
2754 remap_debug_filename): Move to...
2755 * file-prefix-map.c: New file.
2756 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2757 generalize, get rid of alloca(), use strrchr() instead of strchr().
2758 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2759 Implement in terms of add_prefix_map().
2760 (remap_macro_filename, remap_debug_filename): Implement in term of
2762 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2763 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2764 * dbxout.c: Include file-prefix-map.h.
2765 * varasm.c: Likewise.
2766 * vmsdbgout.c: Likewise.
2767 * xcoffout.c: Likewise.
2768 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2769 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2770 * doc/invoke.texi (-ffile-prefix-map): Document.
2771 (-fdebug-prefix-map): Update description.
2773 2018-01-18 Martin Liska <mliska@suse.cz>
2775 * config/i386/i386.c (indirect_thunk_name): Document that also
2777 (output_indirect_thunk): Document why both instructions
2778 (pause and lfence) are generated.
2780 2018-01-18 Richard Biener <rguenther@suse.de>
2782 PR tree-optimization/83887
2783 * graphite-scop-detection.c
2784 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2785 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2786 (scop_detection::merge_sese): Re-implement with a flood-fill
2787 algorithm that properly finds a SESE region if it exists.
2789 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2792 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2793 pointer_diff optimizations use view_convert instead of convert.
2795 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2797 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2798 Generate different code for -mno-speculate-indirect-jumps.
2799 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2800 (*call_indirect_aix<mode>): Disable for
2801 -mno-speculate-indirect-jumps.
2802 (*call_indirect_aix<mode>_nospec): New define_insn.
2803 (*call_value_indirect_aix<mode>): Disable for
2804 -mno-speculate-indirect-jumps.
2805 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2806 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2807 -mno-speculate-indirect-jumps.
2808 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2810 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2812 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2813 long double type, set the flags for noting the default long double
2814 type, even if we don't pass or return a long double type.
2816 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2819 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2822 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2824 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2825 support for merge[hl].
2826 (fold_mergehl_helper): New helper function.
2827 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2828 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2829 (altivec_vmrglw_direct): Add xxmrglw insn.
2831 2018-01-17 Andrew Waterman <andrew@sifive.com>
2833 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2834 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2836 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2839 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2840 call the lto_location_cache before reading the
2841 DECL_SOURCE_LOCATION of the types.
2843 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2844 Richard Sandiford <richard.sandiford@linaro.org>
2846 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2847 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2848 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2849 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2850 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2852 * config/aarch64/constraints.md (aarch64_movti_operand):
2854 * config/aarch64/predicates.md (Uti): Add new constraint.
2856 2018-01-17 Carl Love <cel@us.ibm.com>
2858 * config/rs6000/vsx.md (define_expand xl_len_r,
2859 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2860 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2862 (define_expand, define_insn): Move the shift left from the
2863 define_insn to the define_expand for lxvl and stxvl instructions.
2864 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2865 and XL_LEN_R definitions to PURE.
2867 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2869 * config/i386/i386.c (indirect_thunk_name): Declare regno
2870 as unsigned int. Compare regno with INVALID_REGNUM.
2871 (output_indirect_thunk): Ditto.
2872 (output_indirect_thunk_function): Ditto.
2873 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2874 in the call to output_indirect_thunk_function.
2876 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2879 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2880 rather than the size of inner_type to determine the stack slot size
2881 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2883 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2886 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2889 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2891 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2892 endian Linux systems to optionally enable multilibs for selecting
2893 the long double type if the user configured an explicit type.
2894 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2895 have no long double multilibs if not defined.
2896 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2897 warn if the user used -mabi={ieee,ibm}longdouble and we built
2898 multilibs for long double.
2899 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2900 appropriate multilib option.
2901 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2903 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2904 for building long double multilibs.
2905 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2907 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2909 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2912 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2914 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2917 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2920 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2923 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2925 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2926 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2928 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2930 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2931 different rtl trees depending on TARGET_64BIT.
2932 (rs6000_gen_lvx): Likewise.
2934 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2936 * config/visium/visium.md (nop): Tweak comment.
2937 (hazard_nop): Likewise.
2939 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2941 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2942 -mspeculate-indirect-jumps.
2943 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2944 for -mno-speculate-indirect-jumps.
2945 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2946 (*call_value_indirect_elfv2<mode>): Disable for
2947 -mno-speculate-indirect-jumps.
2948 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2949 (indirect_jump): Emit different RTL for
2950 -mno-speculate-indirect-jumps.
2951 (*indirect_jump<mode>): Disable for
2952 -mno-speculate-indirect-jumps.
2953 (*indirect_jump<mode>_nospec): New define_insn.
2954 (tablejump): Emit different RTL for
2955 -mno-speculate-indirect-jumps.
2956 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2957 (tablejumpsi_nospec): New define_expand.
2958 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2959 (tablejumpdi_nospec): New define_expand.
2960 (*tablejump<mode>_internal1): Disable for
2961 -mno-speculate-indirect-jumps.
2962 (*tablejump<mode>_internal1_nospec): New define_insn.
2963 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2966 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2968 * caller-save.c (insert_save): Drop unnecessary parameter. All
2971 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2972 Richard Biener <rguenth@suse.de>
2975 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2976 return early, inline manually is_gimple_sizepos. Make sure if we
2977 call gimplify_expr we don't end up with a gimple constant.
2978 * tree.c (variably_modified_type_p): Don't return true for
2979 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2980 * gimplify.h (is_gimple_sizepos): Remove.
2982 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2984 PR tree-optimization/83857
2985 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2986 vectorizable_live_operation for pure SLP statements.
2987 (vectorizable_live_operation): Handle PHIs.
2989 2018-01-16 Richard Biener <rguenther@suse.de>
2991 PR tree-optimization/83867
2992 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2993 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2995 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2998 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2999 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3000 If off is not INTEGER_CST, issue a may not be aligned warning
3001 rather than isn't aligned. Use isn%'t rather than isn't.
3002 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3004 <case MULT_EXPR>: Improve the case when bottom and one of the
3005 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3006 operand, in that case check if the other operand is multiple of
3007 bottom divided by the INTEGER_CST operand.
3009 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3012 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3013 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3014 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3015 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3016 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3017 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3018 (pa_function_arg_size): New function.
3020 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3022 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3023 in a separate statement.
3025 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3027 PR tree-optimization/83847
3028 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3029 group gathers and scatters.
3031 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3033 PR rtl-optimization/86620
3034 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3036 PR rtl-optimization/83213
3037 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3038 to last if both are JUMP_INSNs.
3040 PR tree-optimization/83843
3041 * gimple-ssa-store-merging.c
3042 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3043 store_immediate_info for bswap/nop orig_stores.
3045 2018-01-15 Andrew Waterman <andrew@sifive.com>
3047 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3049 <UDIV>: Increase cost if !TARGET_DIV.
3051 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3053 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3054 (define_attr "cr_logical_3op"): New.
3055 (cceq_ior_compare): Adjust.
3056 (cceq_ior_compare_complement): Adjust.
3057 (*cceq_rev_compare): Adjust.
3058 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3059 (is_cracked_insn): Adjust.
3060 (insn_must_be_first_in_group): Adjust.
3061 * config/rs6000/40x.md: Adjust.
3062 * config/rs6000/440.md: Adjust.
3063 * config/rs6000/476.md: Adjust.
3064 * config/rs6000/601.md: Adjust.
3065 * config/rs6000/603.md: Adjust.
3066 * config/rs6000/6xx.md: Adjust.
3067 * config/rs6000/7450.md: Adjust.
3068 * config/rs6000/7xx.md: Adjust.
3069 * config/rs6000/8540.md: Adjust.
3070 * config/rs6000/cell.md: Adjust.
3071 * config/rs6000/e300c2c3.md: Adjust.
3072 * config/rs6000/e500mc.md: Adjust.
3073 * config/rs6000/e500mc64.md: Adjust.
3074 * config/rs6000/e5500.md: Adjust.
3075 * config/rs6000/e6500.md: Adjust.
3076 * config/rs6000/mpc.md: Adjust.
3077 * config/rs6000/power4.md: Adjust.
3078 * config/rs6000/power5.md: Adjust.
3079 * config/rs6000/power6.md: Adjust.
3080 * config/rs6000/power7.md: Adjust.
3081 * config/rs6000/power8.md: Adjust.
3082 * config/rs6000/power9.md: Adjust.
3083 * config/rs6000/rs64.md: Adjust.
3084 * config/rs6000/titan.md: Adjust.
3086 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3088 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3089 ix86_indirect_branch_register logic.
3091 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3093 * config/i386/constraints.md (Bs): Update
3094 ix86_indirect_branch_register check. Don't check
3095 ix86_indirect_branch_register with GOT_memory_operand.
3097 * config/i386/predicates.md (GOT_memory_operand): Don't check
3098 ix86_indirect_branch_register here.
3099 (GOT32_symbol_operand): Likewise.
3101 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3103 * config/i386/predicates.md (constant_call_address_operand):
3104 Rewrite ix86_indirect_branch_register logic.
3105 (sibcall_insn_operand): Likewise.
3107 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3109 * config/i386/constraints.md (Bs): Replace
3110 ix86_indirect_branch_thunk_register with
3111 ix86_indirect_branch_register.
3113 * config/i386/i386.md (indirect_jump): Likewise.
3114 (tablejump): Likewise.
3115 (*sibcall_memory): Likewise.
3116 (*sibcall_value_memory): Likewise.
3117 Peepholes of indirect call and jump via memory: Likewise.
3118 * config/i386/i386.opt: Likewise.
3119 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3120 (GOT_memory_operand): Likewise.
3121 (call_insn_operand): Likewise.
3122 (sibcall_insn_operand): Likewise.
3123 (GOT32_symbol_operand): Likewise.
3125 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3128 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3129 type rather than type addr's type points to.
3130 (expand_omp_atomic_mutex): Likewise.
3131 (expand_omp_atomic): Likewise.
3133 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3136 * config/i386/i386.c (output_indirect_thunk_function): Use
3137 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3138 for __x86_return_thunk.
3140 2018-01-15 Richard Biener <rguenther@suse.de>
3143 * expmed.c (extract_bit_field_1): Fix typo.
3145 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3148 * config/arm/iterators.md (VF): New mode iterator.
3149 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3150 Remove integer-related logic from pattern.
3151 (neon_vabd<mode>_3): Likewise.
3153 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3156 * common.opt (fstrict-overflow): No longer an alias.
3157 (fwrapv-pointer): New option.
3158 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3159 also for pointer types based on flag_wrapv_pointer.
3160 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3161 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3162 opts->x_flag_wrapv got set.
3163 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3164 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3165 POINTER_TYPE_OVERFLOW_UNDEFINED.
3166 * match.pd: Likewise in address comparison pattern.
3167 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3169 2018-01-15 Richard Biener <rguenther@suse.de>
3172 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3173 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3174 Reset type names to their identifier if their TYPE_DECL doesn't
3175 have linkage (and thus is used for ODR and devirt).
3176 (save_debug_info_for_decl): Remove.
3177 (save_debug_info_for_type): Likewise.
3178 (add_tree_to_fld_list): Adjust.
3179 * tree-pretty-print.c (dump_generic_node): Make dumping of
3180 type names more robust.
3182 2018-01-15 Richard Biener <rguenther@suse.de>
3184 * BASE-VER: Bump to 8.0.1.
3186 2018-01-14 Martin Sebor <msebor@redhat.com>
3189 * builtins.c (check_access): Avoid warning when the no-warning bit
3192 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3194 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3195 * ira-color (allocno_hard_regs_compare): Likewise.
3197 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3200 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3201 Use .pushsection/.popsection.
3203 2018-01-14 Martin Sebor <msebor@redhat.com>
3206 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3208 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3210 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3211 entry from extra_headers.
3212 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3213 extra_headers, make the list bitwise identical to the i?86-*-* one.
3215 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3217 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3218 -mcmodel=large with -mindirect-branch=thunk,
3219 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3220 -mfunction-return=thunk-extern.
3221 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3222 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3223 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3225 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3227 * config/i386/i386.c (print_reg): Print the name of the full
3228 integer register without '%'.
3229 (ix86_print_operand): Handle 'V'.
3230 * doc/extend.texi: Document 'V' modifier.
3232 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3234 * config/i386/constraints.md (Bs): Disallow memory operand for
3235 -mindirect-branch-register.
3237 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3238 (GOT_memory_operand): Likewise.
3239 (call_insn_operand): Likewise.
3240 (sibcall_insn_operand): Likewise.
3241 (GOT32_symbol_operand): Likewise.
3242 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3243 for -mindirect-branch-register.
3244 (tablejump): Likewise.
3245 (*sibcall_memory): Likewise.
3246 (*sibcall_value_memory): Likewise.
3247 Disallow peepholes of indirect call and jump via memory for
3248 -mindirect-branch-register.
3249 (*call_pop): Replace m with Bw.
3250 (*call_value_pop): Likewise.
3251 (*sibcall_pop_memory): Replace m with Bs.
3252 * config/i386/i386.opt (mindirect-branch-register): New option.
3253 * doc/invoke.texi: Document -mindirect-branch-register option.
3255 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3257 * config/i386/i386-protos.h (ix86_output_function_return): New.
3258 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3259 set function_return_type.
3260 (indirect_thunk_name): Add ret_p to indicate thunk for function
3262 (output_indirect_thunk_function): Pass false to
3263 indirect_thunk_name.
3264 (ix86_output_indirect_branch_via_reg): Likewise.
3265 (ix86_output_indirect_branch_via_push): Likewise.
3266 (output_indirect_thunk_function): Create alias for function
3267 return thunk if regno < 0.
3268 (ix86_output_function_return): New function.
3269 (ix86_handle_fndecl_attribute): Handle function_return.
3270 (ix86_attribute_table): Add function_return.
3271 * config/i386/i386.h (machine_function): Add
3272 function_return_type.
3273 * config/i386/i386.md (simple_return_internal): Use
3274 ix86_output_function_return.
3275 (simple_return_internal_long): Likewise.
3276 * config/i386/i386.opt (mfunction-return=): New option.
3277 (indirect_branch): Mention -mfunction-return=.
3278 * doc/extend.texi: Document function_return function attribute.
3279 * doc/invoke.texi: Document -mfunction-return= option.
3281 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3283 * config/i386/i386-opts.h (indirect_branch): New.
3284 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3285 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3286 with local indirect jump when converting indirect call and jump.
3287 (ix86_set_indirect_branch_type): New.
3288 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3289 (indirectlabelno): New.
3290 (indirect_thunk_needed): Likewise.
3291 (indirect_thunk_bnd_needed): Likewise.
3292 (indirect_thunks_used): Likewise.
3293 (indirect_thunks_bnd_used): Likewise.
3294 (INDIRECT_LABEL): Likewise.
3295 (indirect_thunk_name): Likewise.
3296 (output_indirect_thunk): Likewise.
3297 (output_indirect_thunk_function): Likewise.
3298 (ix86_output_indirect_branch_via_reg): Likewise.
3299 (ix86_output_indirect_branch_via_push): Likewise.
3300 (ix86_output_indirect_branch): Likewise.
3301 (ix86_output_indirect_jmp): Likewise.
3302 (ix86_code_end): Call output_indirect_thunk_function if needed.
3303 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3305 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3306 (ix86_attribute_table): Add indirect_branch.
3307 * config/i386/i386.h (machine_function): Add indirect_branch_type
3308 and has_local_indirect_jump.
3309 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3311 (tablejump): Likewise.
3312 (*indirect_jump): Use ix86_output_indirect_jmp.
3313 (*tablejump_1): Likewise.
3314 (simple_return_indirect_internal): Likewise.
3315 * config/i386/i386.opt (mindirect-branch=): New option.
3316 (indirect_branch): New.
3319 (thunk-inline): Likewise.
3320 (thunk-extern): Likewise.
3321 * doc/extend.texi: Document indirect_branch function attribute.
3322 * doc/invoke.texi: Document -mindirect-branch= option.
3324 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3327 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3329 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3331 * ipa-inline.c (want_inline_small_function_p): Return false if
3332 inlining has already failed with CIF_FINAL_ERROR.
3333 (update_caller_keys): Call want_inline_small_function_p before
3335 (update_callee_keys): Likewise.
3337 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3339 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3341 (rs6000_quadword_masked_address_p): Likewise.
3342 (quad_aligned_load_p): Likewise.
3343 (quad_aligned_store_p): Likewise.
3344 (const_load_sequence_p): Add comment to describe the outer-most loop.
3345 (mimic_memory_attributes_and_flags): New function.
3346 (rs6000_gen_stvx): Likewise.
3347 (replace_swapped_aligned_store): Likewise.
3348 (rs6000_gen_lvx): Likewise.
3349 (replace_swapped_aligned_load): Likewise.
3350 (replace_swapped_load_constant): Capitalize argument name in
3351 comment describing this function.
3352 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3353 and stores that access quad-word aligned addresses and replace
3354 with stvx or lvx instructions when appropriate.
3355 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3356 New function prototype.
3357 (rs6000_quadword_masked_address_p): Likewise.
3358 (rs6000_gen_lvx): Likewise.
3359 (rs6000_gen_stvx): Likewise.
3360 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3361 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3362 when memory address is aligned.
3363 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3364 this split to select lvx instruction when memory address is aligned.
3365 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3366 instruction when memory address is aligned.
3367 (*vsx_le_perm_load_v16qi): Likewise.
3368 (four unnamed splitters): Modify to select the stvx instruction
3369 when memory is aligned.
3371 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3373 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3374 which appears in the queue multiple times.
3376 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3377 Alan Hayward <alan.hayward@arm.com>
3378 David Sherwood <david.sherwood@arm.com>
3380 * tree-vectorizer.h (vec_lower_bound): New structure.
3381 (_loop_vec_info): Add check_nonzero and lower_bounds.
3382 (LOOP_VINFO_CHECK_NONZERO): New macro.
3383 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3384 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3385 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3386 fields. Make seg_len the distance travelled, not including the
3388 (dr_direction_indicator): Declare.
3389 (dr_zero_step_indicator): Likewise.
3390 (dr_known_forward_stride_p): Likewise.
3391 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3393 (runtime_alias_check_p): Allow runtime alias checks with
3395 (operator ==): Compare access_size and align.
3396 (prune_runtime_alias_test_list): Rework for new distinction between
3397 the access_size and seg_len.
3398 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3400 (get_segment_min_max): New function.
3401 (create_intersect_range_checks): Use it.
3402 (dr_step_indicator): New function.
3403 (dr_direction_indicator): Likewise.
3404 (dr_zero_step_indicator): Likewise.
3405 (dr_known_forward_stride_p): Likewise.
3406 * tree-loop-distribution.c (data_ref_segment_size): Return
3407 DR_STEP * (niters - 1).
3408 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3410 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3411 (vect_preserves_scalar_order_p): New function, split out from...
3412 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3413 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3414 (vect_vfa_access_size): New function.
3415 (vect_vfa_align): Likewise.
3416 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3417 (dump_lower_bound): New function.
3418 (vect_check_lower_bound): Likewise.
3419 (vect_small_gap_p): Likewise.
3420 (vectorizable_with_step_bound_p): Likewise.
3421 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3422 depencies if the vectorization factor is 1. Convert the checks
3423 for nonzero steps into checks on the bounds of DR_STEP. Try using
3424 a bunds check for variable steps if the minimum required step is
3425 relatively small. Update calls to the dr_with_seg_len
3426 constructor and to vect_compile_time_alias.
3427 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3429 (vect_loop_versioning): Call it.
3430 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3432 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3434 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3435 Alan Hayward <alan.hayward@arm.com>
3436 David Sherwood <david.sherwood@arm.com>
3438 * doc/sourcebuild.texi (vect_scatter_store): Document.
3439 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3441 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3443 * genopinit.c (main): Add supports_vec_scatter_store and
3444 supports_vec_scatter_store_cached to target_optabs.
3445 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3446 IFN_MASK_SCATTER_STORE.
3447 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3449 * internal-fn.h (internal_store_fn_p): Declare.
3450 (internal_fn_stored_value_index): Likewise.
3451 * internal-fn.c (scatter_store_direct): New macro.
3452 (expand_scatter_store_optab_fn): New function.
3453 (direct_scatter_store_optab_supported_p): New macro.
3454 (internal_store_fn_p): New function.
3455 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3456 IFN_MASK_SCATTER_STORE.
3457 (internal_fn_mask_index): Likewise.
3458 (internal_fn_stored_value_index): New function.
3459 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3461 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3462 * optabs-query.c (supports_vec_scatter_store_p): New function.
3463 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3464 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3465 true for scatter stores.
3466 (vect_gather_scatter_fn_p): Handle scatter stores too.
3467 (vect_check_gather_scatter): Consider using scatter stores if
3468 supports_vec_scatter_store_p.
3469 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3471 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3472 internal_fn_stored_value_index.
3473 (check_load_store_masking): Handle scatter stores too.
3474 (vect_get_store_rhs): Make public.
3475 (vectorizable_call): Use internal_store_fn_p.
3476 (vectorizable_store): Handle scatter store internal functions.
3477 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3478 when deciding whether the end of the group has been reached.
3479 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3480 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3481 (mask_scatter_store<mode>): New insns.
3483 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3484 Alan Hayward <alan.hayward@arm.com>
3485 David Sherwood <david.sherwood@arm.com>
3487 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3488 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3489 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3491 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3492 Use vect_truncate_gather_scatter_offset if we can't treat the
3493 operation as a normal gather load or scatter store.
3494 (get_group_load_store_type): Take the gather_scatter_info
3495 as argument. Try using a gather load or scatter store for
3496 single-element groups.
3497 (get_load_store_type): Update calls to get_group_load_store_type
3498 and vect_use_strided_gather_scatters_p.
3500 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3501 Alan Hayward <alan.hayward@arm.com>
3502 David Sherwood <david.sherwood@arm.com>
3504 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3505 optional tree argument.
3506 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3508 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3509 but continue to use the current value as a fallback.
3510 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3511 to compare the updates.
3512 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3513 (get_load_store_type): Use it when handling a strided access.
3514 (vect_get_strided_load_store_ops): New function.
3515 (vect_get_data_ptr_increment): Likewise.
3516 (vectorizable_load): Handle strided gather loads. Always pass
3517 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3519 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3520 Alan Hayward <alan.hayward@arm.com>
3521 David Sherwood <david.sherwood@arm.com>
3523 * doc/md.texi (gather_load@var{m}): Document.
3524 (mask_gather_load@var{m}): Likewise.
3525 * genopinit.c (main): Add supports_vec_gather_load and
3526 supports_vec_gather_load_cached to target_optabs.
3527 * optabs-tree.c (init_tree_optimization_optabs): Use
3528 ggc_cleared_alloc to allocate target_optabs.
3529 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3530 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3532 * internal-fn.h (internal_load_fn_p): Declare.
3533 (internal_gather_scatter_fn_p): Likewise.
3534 (internal_fn_mask_index): Likewise.
3535 (internal_gather_scatter_fn_supported_p): Likewise.
3536 * internal-fn.c (gather_load_direct): New macro.
3537 (expand_gather_load_optab_fn): New function.
3538 (direct_gather_load_optab_supported_p): New macro.
3539 (direct_internal_fn_optab): New function.
3540 (internal_load_fn_p): Likewise.
3541 (internal_gather_scatter_fn_p): Likewise.
3542 (internal_fn_mask_index): Likewise.
3543 (internal_gather_scatter_fn_supported_p): Likewise.
3544 * optabs-query.c (supports_at_least_one_mode_p): New function.
3545 (supports_vec_gather_load_p): Likewise.
3546 * optabs-query.h (supports_vec_gather_load_p): Declare.
3547 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3548 and memory_type field.
3549 (NUM_PATTERNS): Bump to 15.
3550 * tree-vect-data-refs.c: Include internal-fn.h.
3551 (vect_gather_scatter_fn_p): New function.
3552 (vect_describe_gather_scatter_call): Likewise.
3553 (vect_check_gather_scatter): Try using internal functions for
3554 gather loads. Recognize existing calls to a gather load function.
3555 (vect_analyze_data_refs): Consider using gather loads if
3556 supports_vec_gather_load_p.
3557 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3558 (vect_get_gather_scatter_offset_type): Likewise.
3559 (vect_convert_mask_for_vectype): Likewise.
3560 (vect_add_conversion_to_patterm): Likewise.
3561 (vect_try_gather_scatter_pattern): Likewise.
3562 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3563 (vect_vect_recog_func_ptrs): Add it.
3564 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3565 internal_fn_mask_index and internal_gather_scatter_fn_p.
3566 (check_load_store_masking): Take the gather_scatter_info as an
3567 argument and handle gather loads.
3568 (vect_get_gather_scatter_ops): New function.
3569 (vectorizable_call): Check internal_load_fn_p.
3570 (vectorizable_load): Likewise. Handle gather load internal
3572 (vectorizable_store): Update call to check_load_store_masking.
3573 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3574 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3575 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3576 (aarch64_gather_scale_operand_d): New predicates.
3577 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3578 (mask_gather_load<mode>): New insns.
3580 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3581 Alan Hayward <alan.hayward@arm.com>
3582 David Sherwood <david.sherwood@arm.com>
3584 * optabs.def (fold_left_plus_optab): New optab.
3585 * doc/md.texi (fold_left_plus_@var{m}): Document.
3586 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3587 * internal-fn.c (fold_left_direct): Define.
3588 (expand_fold_left_optab_fn): Likewise.
3589 (direct_fold_left_optab_supported_p): Likewise.
3590 * fold-const-call.c (fold_const_fold_left): New function.
3591 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3592 * tree-parloops.c (valid_reduction_p): New function.
3593 (gather_scalar_reductions): Use it.
3594 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3595 (vect_finish_replace_stmt): Declare.
3596 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3597 (needs_fold_left_reduction_p): New function, split out from...
3598 (vect_is_simple_reduction): ...here. Accept reductions that
3599 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3600 (vect_force_simple_reduction): Also store the reduction type in
3601 the assignment's STMT_VINFO_REDUC_TYPE.
3602 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3603 (merge_with_identity): New function.
3604 (vect_expand_fold_left): Likewise.
3605 (vectorize_fold_left_reduction): Likewise.
3606 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3607 scalar phi in place for it. Check for target support and reject
3608 cases that would reassociate the operation. Defer the transform
3609 phase to vectorize_fold_left_reduction.
3610 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3611 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3612 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3614 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3616 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3617 call to ifc_temp_var.
3619 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3620 Alan Hayward <alan.hayward@arm.com>
3621 David Sherwood <david.sherwood@arm.com>
3623 * target.def (legitimize_address_displacement): Take the original
3624 offset as a poly_int.
3625 * targhooks.h (default_legitimize_address_displacement): Update
3627 * targhooks.c (default_legitimize_address_displacement): Likewise.
3628 * doc/tm.texi: Regenerate.
3629 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3630 as an argument, moving assert of ad->disp == ad->disp_term to...
3631 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3632 Try calling targetm.legitimize_address_displacement before expanding
3633 the address rather than afterwards, and adjust for the new interface.
3634 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3635 Match the new hook interface. Handle SVE addresses.
3636 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3639 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3641 * Makefile.in (OBJS): Add early-remat.o.
3642 * target.def (select_early_remat_modes): New hook.
3643 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3644 * doc/tm.texi: Regenerate.
3645 * targhooks.h (default_select_early_remat_modes): Declare.
3646 * targhooks.c (default_select_early_remat_modes): New function.
3647 * timevar.def (TV_EARLY_REMAT): New timevar.
3648 * passes.def (pass_early_remat): New pass.
3649 * tree-pass.h (make_pass_early_remat): Declare.
3650 * early-remat.c: New file.
3651 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3653 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3655 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3656 Alan Hayward <alan.hayward@arm.com>
3657 David Sherwood <david.sherwood@arm.com>
3659 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3660 vfm1 with a bound_epilog parameter.
3661 (vect_do_peeling): Update calls accordingly, and move the prologue
3662 call earlier in the function. Treat the base bound_epilog as 0 for
3663 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3664 this base when peeling for gaps.
3665 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3666 with fully-masked loops.
3667 (vect_estimate_min_profitable_iters): Handle the single peeled
3668 iteration in that case.
3670 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3671 Alan Hayward <alan.hayward@arm.com>
3672 David Sherwood <david.sherwood@arm.com>
3674 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3675 single-element interleaving even if the size is not a power of 2.
3676 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3677 accesses for single-element interleaving if the group size is
3680 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3681 Alan Hayward <alan.hayward@arm.com>
3682 David Sherwood <david.sherwood@arm.com>
3684 * doc/md.texi (fold_extract_last_@var{m}): Document.
3685 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3686 * optabs.def (fold_extract_last_optab): New optab.
3687 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3688 * internal-fn.c (fold_extract_direct): New macro.
3689 (expand_fold_extract_optab_fn): Likewise.
3690 (direct_fold_extract_optab_supported_p): Likewise.
3691 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3692 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3693 EXTRACT_LAST_REDUCTION.
3694 (get_initial_def_for_reduction): Do not create an initial vector
3695 for EXTRACT_LAST_REDUCTION reductions.
3696 (vectorizable_reduction): Leave the scalar phi in place for
3697 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3698 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3699 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3700 transform phase to vectorizable_condition.
3701 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3703 (vect_finish_stmt_generation): ...here.
3704 (vect_finish_replace_stmt): New function.
3705 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3706 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3708 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3710 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3711 Alan Hayward <alan.hayward@arm.com>
3712 David Sherwood <david.sherwood@arm.com>
3714 * doc/md.texi (extract_last_@var{m}): Document.
3715 * optabs.def (extract_last_optab): New optab.
3716 * internal-fn.def (EXTRACT_LAST): New internal function.
3717 * internal-fn.c (cond_unary_direct): New macro.
3718 (expand_cond_unary_optab_fn): Likewise.
3719 (direct_cond_unary_optab_supported_p): Likewise.
3720 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3721 loops using EXTRACT_LAST.
3722 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3723 (extract_last_<mode>): ...this optab.
3724 (vec_extract<mode><Vel>): Update accordingly.
3726 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3727 Alan Hayward <alan.hayward@arm.com>
3728 David Sherwood <david.sherwood@arm.com>
3730 * target.def (empty_mask_is_expensive): New hook.
3731 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3732 * doc/tm.texi: Regenerate.
3733 * targhooks.h (default_empty_mask_is_expensive): Declare.
3734 * targhooks.c (default_empty_mask_is_expensive): New function.
3735 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3736 if the target says that empty masks are expensive.
3737 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3739 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3741 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3742 Alan Hayward <alan.hayward@arm.com>
3743 David Sherwood <david.sherwood@arm.com>
3745 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3746 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3747 (vect_use_loop_mask_for_alignment_p): New function.
3748 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3749 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3750 niters_skip argument. Make sure that the first niters_skip elements
3751 of the first iteration are inactive.
3752 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3753 Update call to vect_set_loop_masks_directly.
3754 (get_misalign_in_elems): New function, split out from...
3755 (vect_gen_prolog_loop_niters): ...here.
3756 (vect_update_init_of_dr): Take a code argument that specifies whether
3757 the adjustment should be added or subtracted.
3758 (vect_update_init_of_drs): Likewise.
3759 (vect_prepare_for_masked_peels): New function.
3760 (vect_do_peeling): Skip prologue peeling if we're using a mask
3761 instead. Update call to vect_update_inits_of_drs.
3762 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3764 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3765 alignment. Do not include the number of peeled iterations in
3766 the minimum threshold in that case.
3767 (vectorizable_induction): Adjust the start value down by
3768 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3769 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3770 Take the number of skipped iterations into account when calculating
3772 * tree-vect-stmts.c (vect_gen_while_not): New function.
3774 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3775 Alan Hayward <alan.hayward@arm.com>
3776 David Sherwood <david.sherwood@arm.com>
3778 * doc/sourcebuild.texi (vect_fully_masked): Document.
3779 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3781 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3783 (vect_analyze_loop_2): ...here. Don't check the vectorization
3784 factor against the number of loop iterations if the loop is
3787 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3788 Alan Hayward <alan.hayward@arm.com>
3789 David Sherwood <david.sherwood@arm.com>
3791 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3792 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3793 (dump_groups): Update accordingly.
3794 (iv_use::mem_type): New member variable.
3795 (address_p): New function.
3796 (record_use): Add a mem_type argument and initialize the new
3798 (record_group_use): Add a mem_type argument. Use address_p.
3799 Remove obsolete null checks of base_object. Update call to record_use.
3800 (find_interesting_uses_op): Update call to record_group_use.
3801 (find_interesting_uses_cond): Likewise.
3802 (find_interesting_uses_address): Likewise.
3803 (get_mem_type_for_internal_fn): New function.
3804 (find_address_like_use): Likewise.
3805 (find_interesting_uses_stmt): Try find_address_like_use before
3806 calling find_interesting_uses_op.
3807 (addr_offset_valid_p): Use the iv mem_type field as the type
3808 of the addressed memory.
3809 (add_autoinc_candidates): Likewise.
3810 (get_address_cost): Likewise.
3811 (split_small_address_groups_p): Use address_p.
3812 (split_address_groups): Likewise.
3813 (add_iv_candidate_for_use): Likewise.
3814 (autoinc_possible_for_pair): Likewise.
3815 (rewrite_groups): Likewise.
3816 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3817 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3818 (get_alias_ptr_type_for_ptr_address): New function.
3819 (rewrite_use_address): Rewrite address uses in calls that were
3820 identified by find_address_like_use.
3822 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3823 Alan Hayward <alan.hayward@arm.com>
3824 David Sherwood <david.sherwood@arm.com>
3826 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3828 * gimple-expr.h (is_gimple_addressable: Likewise.
3829 * gimple-expr.c (is_gimple_address): Likewise.
3830 * internal-fn.c (expand_call_mem_ref): New function.
3831 (expand_mask_load_optab_fn): Use it.
3832 (expand_mask_store_optab_fn): Likewise.
3834 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3835 Alan Hayward <alan.hayward@arm.com>
3836 David Sherwood <david.sherwood@arm.com>
3838 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3839 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3840 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3841 (cond_umax@var{mode}): Document.
3842 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3843 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3844 (cond_umin_optab, cond_umax_optab): New optabs.
3845 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3846 (COND_IOR, COND_XOR): New internal functions.
3847 * internal-fn.h (get_conditional_internal_fn): Declare.
3848 * internal-fn.c (cond_binary_direct): New macro.
3849 (expand_cond_binary_optab_fn): Likewise.
3850 (direct_cond_binary_optab_supported_p): Likewise.
3851 (get_conditional_internal_fn): New function.
3852 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3853 Cope with reduction statements that are vectorized as calls rather
3855 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3856 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3857 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3858 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3859 (UNSPEC_COND_EOR): New unspecs.
3860 (optab): Add mappings for them.
3861 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3862 (sve_int_op, sve_fp_op): New int attributes.
3864 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3865 Alan Hayward <alan.hayward@arm.com>
3866 David Sherwood <david.sherwood@arm.com>
3868 * optabs.def (while_ult_optab): New optab.
3869 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3870 * internal-fn.def (WHILE_ULT): New internal function.
3871 * internal-fn.h (direct_internal_fn_supported_p): New override
3872 that takes two types as argument.
3873 * internal-fn.c (while_direct): New macro.
3874 (expand_while_optab_fn): New function.
3875 (convert_optab_supported_p): Likewise.
3876 (direct_while_optab_supported_p): New macro.
3877 * wide-int.h (wi::udiv_ceil): New function.
3878 * tree-vectorizer.h (rgroup_masks): New structure.
3879 (vec_loop_masks): New typedef.
3880 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3882 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3883 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3884 (vect_max_vf): New function.
3885 (slpeel_make_loop_iterate_ntimes): Delete.
3886 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3887 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3888 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3889 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3890 internal-fn.h, stor-layout.h and optabs-query.h.
3891 (vect_set_loop_mask): New function.
3892 (add_preheader_seq): Likewise.
3893 (add_header_seq): Likewise.
3894 (interleave_supported_p): Likewise.
3895 (vect_maybe_permute_loop_masks): Likewise.
3896 (vect_set_loop_masks_directly): Likewise.
3897 (vect_set_loop_condition_masked): Likewise.
3898 (vect_set_loop_condition_unmasked): New function, split out from
3899 slpeel_make_loop_iterate_ntimes.
3900 (slpeel_make_loop_iterate_ntimes): Rename to..
3901 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3902 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3903 (vect_do_peeling): Update call accordingly.
3904 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3906 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3907 mask_compare_type, can_fully_mask_p and fully_masked_p.
3908 (release_vec_loop_masks): New function.
3909 (_loop_vec_info): Use it to free the loop masks.
3910 (can_produce_all_loop_masks_p): New function.
3911 (vect_get_max_nscalars_per_iter): Likewise.
3912 (vect_verify_full_masking): Likewise.
3913 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3914 retries, and free the mask rgroups before retrying. Check loop-wide
3915 reasons for disallowing fully-masked loops. Make the final decision
3916 about whether use a fully-masked loop or not.
3917 (vect_estimate_min_profitable_iters): Do not assume that peeling
3918 for the number of iterations will be needed for fully-masked loops.
3919 (vectorizable_reduction): Disable fully-masked loops.
3920 (vectorizable_live_operation): Likewise.
3921 (vect_halve_mask_nunits): New function.
3922 (vect_double_mask_nunits): Likewise.
3923 (vect_record_loop_mask): Likewise.
3924 (vect_get_loop_mask): Likewise.
3925 (vect_transform_loop): Handle the case in which the final loop
3926 iteration might handle a partial vector. Call vect_set_loop_condition
3927 instead of slpeel_make_loop_iterate_ntimes.
3928 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3929 (check_load_store_masking): New function.
3930 (prepare_load_store_mask): Likewise.
3931 (vectorizable_store): Handle fully-masked loops.
3932 (vectorizable_load): Likewise.
3933 (supportable_widening_operation): Use vect_halve_mask_nunits for
3935 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3936 (vect_gen_while): New function.
3937 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3938 (aarch64_uqdec<mode>): New insn.
3940 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3941 Alan Hayward <alan.hayward@arm.com>
3942 David Sherwood <david.sherwood@arm.com>
3944 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3945 (reduc_xor_scal_optab): New optabs.
3946 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3947 (reduc_xor_scal_@var{m}): Document.
3948 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3949 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3951 * fold-const-call.c (fold_const_call): Handle them.
3952 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3953 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3954 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3955 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3956 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3957 (UNSPEC_XORV): New unspecs.
3958 (optab): Add entries for them.
3959 (BITWISEV): New int iterator.
3960 (bit_reduc_op): New int attributes.
3962 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3963 Alan Hayward <alan.hayward@arm.com>
3964 David Sherwood <david.sherwood@arm.com>
3966 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3967 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3968 * optabs.def (vec_shl_insert_optab): New optab.
3969 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3970 (duplicate_and_interleave): Likewise.
3971 * tree-vect-loop.c: Include internal-fn.h.
3972 (neutral_op_for_slp_reduction): New function, split out from
3973 get_initial_defs_for_reduction.
3974 (get_initial_def_for_reduction): Handle option 2 for variable-length
3975 vectors by loading the neutral value into a vector and then shifting
3976 the initial value into element 0.
3977 (get_initial_defs_for_reduction): Replace the code argument with
3978 the neutral value calculated by neutral_op_for_slp_reduction.
3979 Use gimple_build_vector for constant-length vectors.
3980 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3981 but the first group_size elements have a neutral value.
3982 Use duplicate_and_interleave otherwise.
3983 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3984 Update call to get_initial_defs_for_reduction. Handle SLP
3985 reductions for variable-length vectors by creating one vector
3986 result for each scalar result, with the elements associated
3987 with other scalar results stubbed out with the neutral value.
3988 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3989 Require IFN_VEC_SHL_INSERT for double reductions on
3990 variable-length vectors, or SLP reductions that have
3991 a neutral value. Require can_duplicate_and_interleave_p
3992 support for variable-length unchained SLP reductions if there
3993 is no neutral value, such as for MIN/MAX reductions. Also require
3994 the number of vector elements to be a multiple of the number of
3995 SLP statements when doing variable-length unchained SLP reductions.
3996 Update call to vect_create_epilog_for_reduction.
3997 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3998 and remove initial values.
3999 (duplicate_and_interleave): Make public.
4000 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4001 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4003 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4004 Alan Hayward <alan.hayward@arm.com>
4005 David Sherwood <david.sherwood@arm.com>
4007 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4008 (can_duplicate_and_interleave_p): New function.
4009 (vect_get_and_check_slp_defs): Take the vector of statements
4010 rather than just the current one. Remove excess parentheses.
4011 Restriction rejectinon of vect_constant_def and vect_external_def
4012 for variable-length vectors to boolean types, or types for which
4013 can_duplicate_and_interleave_p is false.
4014 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4015 (duplicate_and_interleave): New function.
4016 (vect_get_constant_vectors): Use gimple_build_vector for
4017 constant-length vectors and suitable variable-length constant
4018 vectors. Use duplicate_and_interleave for other variable-length
4019 vectors. Don't defer the update when inserting new statements.
4021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4022 Alan Hayward <alan.hayward@arm.com>
4023 David Sherwood <david.sherwood@arm.com>
4025 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4026 min_profitable_iters doesn't go negative.
4028 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4029 Alan Hayward <alan.hayward@arm.com>
4030 David Sherwood <david.sherwood@arm.com>
4032 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4033 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4034 * optabs.def (vec_mask_load_lanes_optab): New optab.
4035 (vec_mask_store_lanes_optab): Likewise.
4036 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4037 (MASK_STORE_LANES): Likewise.
4038 * internal-fn.c (mask_load_lanes_direct): New macro.
4039 (mask_store_lanes_direct): Likewise.
4040 (expand_mask_load_optab_fn): Handle masked operations.
4041 (expand_mask_load_lanes_optab_fn): New macro.
4042 (expand_mask_store_optab_fn): Handle masked operations.
4043 (expand_mask_store_lanes_optab_fn): New macro.
4044 (direct_mask_load_lanes_optab_supported_p): Likewise.
4045 (direct_mask_store_lanes_optab_supported_p): Likewise.
4046 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4048 (vect_load_lanes_supported): Likewise.
4049 * tree-vect-data-refs.c (strip_conversion): New function.
4050 (can_group_stmts_p): Likewise.
4051 (vect_analyze_data_ref_accesses): Use it instead of checking
4052 for a pair of assignments.
4053 (vect_store_lanes_supported): Take a masked_p parameter.
4054 (vect_load_lanes_supported): Likewise.
4055 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4056 vect_store_lanes_supported and vect_load_lanes_supported.
4057 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4058 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4059 parameter. Don't allow gaps for masked accesses.
4060 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4061 and vect_load_lanes_supported.
4062 (get_load_store_type): Take a masked_p parameter and update
4063 call to get_group_load_store_type.
4064 (vectorizable_store): Update call to get_load_store_type.
4065 Handle IFN_MASK_STORE_LANES.
4066 (vectorizable_load): Update call to get_load_store_type.
4067 Handle IFN_MASK_LOAD_LANES.
4069 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4070 Alan Hayward <alan.hayward@arm.com>
4071 David Sherwood <david.sherwood@arm.com>
4073 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4075 * config/aarch64/aarch64-protos.h
4076 (aarch64_sve_struct_memory_operand_p): Declare.
4077 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4078 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4079 (VPRED, vpred): Handle SVE structure modes.
4080 * config/aarch64/constraints.md (Utx): New constraint.
4081 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4082 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4083 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4084 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4085 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4086 structure modes. Split into pieces after RA.
4087 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4088 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4090 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4091 SVE structure modes.
4092 (aarch64_classify_address): Likewise.
4093 (sizetochar): Move earlier in file.
4094 (aarch64_print_operand): Handle SVE register lists.
4095 (aarch64_array_mode): New function.
4096 (aarch64_sve_struct_memory_operand_p): Likewise.
4097 (TARGET_ARRAY_MODE): Redefine.
4099 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4100 Alan Hayward <alan.hayward@arm.com>
4101 David Sherwood <david.sherwood@arm.com>
4103 * target.def (array_mode): New target hook.
4104 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4105 * doc/tm.texi: Regenerate.
4106 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4107 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4108 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4110 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4113 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4114 Alan Hayward <alan.hayward@arm.com>
4115 David Sherwood <david.sherwood@arm.com>
4117 * fold-const.c (fold_binary_loc): Check the argument types
4118 rather than the result type when testing for a vector operation.
4120 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4122 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4123 * doc/tm.texi: Regenerate.
4125 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4126 Alan Hayward <alan.hayward@arm.com>
4127 David Sherwood <david.sherwood@arm.com>
4129 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4130 (sve): Document new AArch64 extension.
4131 * doc/md.texi (w): Extend the description of the AArch64
4132 constraint to include SVE vectors.
4133 (Upl, Upa): Document new AArch64 predicate constraints.
4134 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4136 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4137 (msve-vector-bits=): New option.
4138 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4139 SVE when these are disabled.
4140 (sve): New extension.
4141 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4142 modes. Adjust their number of units based on aarch64_sve_vg.
4143 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4144 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4145 aarch64_addr_query_type.
4146 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4147 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4148 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4149 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4150 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4151 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4152 (aarch64_simd_imm_zero_p): Delete.
4153 (aarch64_check_zero_based_sve_index_immediate): Declare.
4154 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4155 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4156 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4157 (aarch64_sve_float_mul_immediate_p): Likewise.
4158 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4160 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4161 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4162 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4163 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4164 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4165 (aarch64_regmode_natural_size): Likewise.
4166 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4167 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4169 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4170 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4171 for VG and the SVE predicate registers.
4172 (V_ALIASES): Add a "z"-prefixed alias.
4173 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4174 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4175 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4176 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4177 (REG_CLASS_NAMES): Add entries for them.
4178 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4179 and the predicate registers.
4180 (aarch64_sve_vg): Declare.
4181 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4182 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4183 (REGMODE_NATURAL_SIZE): Define.
4184 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4186 * config/aarch64/aarch64.c: Include cfgrtl.h.
4187 (simd_immediate_info): Add a constructor for series vectors,
4188 and an associated step field.
4189 (aarch64_sve_vg): New variable.
4190 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4191 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4192 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4193 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4194 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4195 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4196 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4197 (aarch64_get_mask_mode): New functions.
4198 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4199 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4200 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4201 predicate modes and predicate registers. Explicitly restrict
4202 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4203 to store a vector mode if it is recognized by
4204 aarch64_classify_vector_mode.
4205 (aarch64_regmode_natural_size): New function.
4206 (aarch64_hard_regno_caller_save_mode): Return the original mode
4208 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4209 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4210 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4211 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4213 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4214 does not overlap dest if the function is frame-related. Handle
4216 (aarch64_split_add_offset): New function.
4217 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4218 them aarch64_add_offset.
4219 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4220 and update call to aarch64_sub_sp.
4221 (aarch64_add_cfa_expression): New function.
4222 (aarch64_expand_prologue): Pass extra temporary registers to the
4223 functions above. Handle the case in which we need to emit new
4224 DW_CFA_expressions for registers that were originally saved
4225 relative to the stack pointer, but now have to be expressed
4226 relative to the frame pointer.
4227 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4229 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4230 IP0 and IP1 values for SVE frames.
4231 (aarch64_expand_vec_series): New function.
4232 (aarch64_expand_sve_widened_duplicate): Likewise.
4233 (aarch64_expand_sve_const_vector): Likewise.
4234 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4235 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4236 into the register, rather than emitting a SET directly.
4237 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4238 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4239 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4240 (offset_9bit_signed_scaled_p): New functions.
4241 (aarch64_replicate_bitmask_imm): New function.
4242 (aarch64_bitmask_imm): Use it.
4243 (aarch64_cannot_force_const_mem): Reject expressions involving
4244 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4245 (aarch64_classify_index): Handle SVE indices, by requiring
4246 a plain register index with a scale that matches the element size.
4247 (aarch64_classify_address): Handle SVE addresses. Assert that
4248 the mode of the address is VOIDmode or an integer mode.
4249 Update call to aarch64_classify_symbol.
4250 (aarch64_classify_symbolic_expression): Update call to
4251 aarch64_classify_symbol.
4252 (aarch64_const_vec_all_in_range_p): New function.
4253 (aarch64_print_vector_float_operand): Likewise.
4254 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4255 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4256 and the FP immediates 1.0 and 0.5.
4257 (aarch64_print_address_internal): Handle SVE addresses.
4258 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4259 (aarch64_regno_regclass): Handle predicate registers.
4260 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4262 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4263 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4264 (aarch64_convert_sve_vector_bits): New function.
4265 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4266 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4268 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4269 Handle SVE vector and predicate modes. Accept VL-based constants
4270 that need only one temporary register, and VL offsets that require
4271 no temporary registers.
4272 (aarch64_conditional_register_usage): Mark the predicate registers
4273 as fixed if SVE isn't available.
4274 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4275 Return true for SVE vector and predicate modes.
4276 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4277 rather than an unsigned int. Handle SVE modes.
4278 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4280 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4282 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4283 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4284 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4285 (aarch64_sve_float_mul_immediate_p): New functions.
4286 (aarch64_sve_valid_immediate): New function.
4287 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4288 Explicitly reject structure modes. Check for INDEX constants.
4289 Handle PTRUE and PFALSE constants.
4290 (aarch64_check_zero_based_sve_index_immediate): New function.
4291 (aarch64_simd_imm_zero_p): Delete.
4292 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4293 vector modes. Accept constants in the range of CNT[BHWD].
4294 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4295 ask for an Advanced SIMD mode.
4296 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4297 (aarch64_simd_vector_alignment): Handle SVE predicates.
4298 (aarch64_vectorize_preferred_vector_alignment): New function.
4299 (aarch64_simd_vector_alignment_reachable): Use it instead of
4301 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4302 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4304 (MAX_VECT_LEN): Delete.
4305 (expand_vec_perm_d): Add a vec_flags field.
4306 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4307 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4308 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4310 (aarch64_evpc_rev): Rename to...
4311 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4312 (aarch64_evpc_rev_global): New function.
4313 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4314 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4316 (aarch64_evpc_sve_tbl): New function.
4317 (aarch64_expand_vec_perm_const_1): Update after rename of
4318 aarch64_evpc_rev. Handle SVE permutes too, trying
4319 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4320 than aarch64_evpc_tbl.
4321 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4322 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4323 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4324 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4325 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4326 (aarch64_expand_sve_vcond): New functions.
4327 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4328 of aarch64_vector_mode_p.
4329 (aarch64_dwarf_poly_indeterminate_value): New function.
4330 (aarch64_compute_pressure_classes): Likewise.
4331 (aarch64_can_change_mode_class): Likewise.
4332 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4333 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4334 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4335 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4336 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4337 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4338 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4339 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4341 (Dn, Dl, Dr): Accept const as well as const_vector.
4342 (Dz): Likewise. Compare against CONST0_RTX.
4343 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4344 of "vector" where appropriate.
4345 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4346 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4347 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4348 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4349 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4350 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4351 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4352 (v_int_equiv): Extend to SVE modes.
4353 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4355 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4356 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4357 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4358 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4359 (SVE_COND_FP_CMP): New int iterators.
4360 (perm_hilo): Handle the new unpack unspecs.
4361 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4363 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4364 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4365 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4366 (aarch64_equality_operator, aarch64_constant_vector_operand)
4367 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4368 (aarch64_sve_nonimmediate_operand): Likewise.
4369 (aarch64_sve_general_operand): Likewise.
4370 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4371 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4372 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4373 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4374 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4375 (aarch64_sve_float_arith_immediate): Likewise.
4376 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4377 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4378 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4379 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4380 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4381 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4382 (aarch64_sve_float_arith_operand): Likewise.
4383 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4384 (aarch64_sve_float_mul_operand): Likewise.
4385 (aarch64_sve_vec_perm_operand): Likewise.
4386 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4387 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4388 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4389 as well as const_vector.
4390 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4391 in file. Use CONST0_RTX and CONSTM1_RTX.
4392 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4393 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4394 Use aarch64_simd_imm_zero.
4395 * config/aarch64/aarch64-sve.md: New file.
4396 * config/aarch64/aarch64.md: Include it.
4397 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4398 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4399 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4400 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4401 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4402 (sve): New attribute.
4403 (enabled): Disable instructions with the sve attribute unless
4405 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4406 aarch64_expand_mov_immediate.
4407 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4408 CNT[BHSD] immediates.
4409 (movti): Split CONST_POLY_INT moves into two halves.
4410 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4411 Split additions that need a temporary here if the destination
4412 is the stack pointer.
4413 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4414 (*add<mode>3_poly_1): New instruction.
4415 (set_clobber_cc): New expander.
4417 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4419 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4420 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4421 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4422 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4423 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4424 Change innermode from fixed_mode_size to machine_mode.
4425 (simplify_subreg): Update call accordingly. Handle a constant-sized
4426 subreg of a variable-length CONST_VECTOR.
4428 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4429 Alan Hayward <alan.hayward@arm.com>
4430 David Sherwood <david.sherwood@arm.com>
4432 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4433 (add_offset_to_base): New function, split out from...
4434 (create_mem_ref): ...here. When handling a scale other than 1,
4435 check first whether the address is valid without the offset.
4436 Add it into the base if so, leaving the index and scale as-is.
4438 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4441 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4442 fold_for_warn before checking if arg2 is INTEGER_CST.
4444 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4446 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4447 (store_multiple_operation): Delete.
4448 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4449 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4450 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4451 guarded by TARGET_STRING.
4452 (rs6000_output_load_multiple): Delete.
4453 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4454 OPTION_MASK_STRING / TARGET_STRING handling.
4455 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4456 (const rs6000_opt_masks) <"string">: Change mask to 0.
4457 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4458 (MASK_STRING): Delete.
4459 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4461 (load_multiple): Delete.
4468 (store_multiple): Delete.
4475 (movmemsi_8reg): Delete.
4476 (corresponding unnamed define_insn): Delete.
4477 (movmemsi_6reg): Delete.
4478 (corresponding unnamed define_insn): Delete.
4479 (movmemsi_4reg): Delete.
4480 (corresponding unnamed define_insn): Delete.
4481 (movmemsi_2reg): Delete.
4482 (corresponding unnamed define_insn): Delete.
4483 (movmemsi_1reg): Delete.
4484 (corresponding unnamed define_insn): Delete.
4485 * config/rs6000/rs6000.opt (mno-string): New.
4486 (mstring): Replace by deprecation warning stub.
4487 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4489 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4491 * regrename.c (regrename_do_replace): If replacing the same
4492 reg multiple times, try to reuse last created gen_raw_REG.
4495 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4496 main to workaround a bug in GDB.
4498 2018-01-12 Tom de Vries <tom@codesourcery.com>
4501 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4503 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4505 PR rtl-optimization/80481
4506 * ira-color.c (get_cap_member): New function.
4507 (allocnos_conflict_by_live_ranges_p): Use it.
4508 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4509 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4511 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4514 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4515 (*saddl_se_1): Ditto.
4517 (*ssubl_se_1): Ditto.
4519 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4521 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4522 rather than wi::to_widest for DR_INITs.
4523 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4524 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4525 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4527 (vect_analyze_group_access_1): Note that here.
4529 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4531 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4532 polynomial type sizes.
4534 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4536 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4537 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4538 (gimple_add_tmp_var): Likewise.
4540 2018-01-12 Martin Liska <mliska@suse.cz>
4542 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4543 (gimple_alloc_sizes): Likewise.
4544 (dump_gimple_statistics): Use PRIu64 in printf format.
4545 * gimple.h: Change uint64_t to int.
4547 2018-01-12 Martin Liska <mliska@suse.cz>
4549 * tree-core.h: Use uint64_t instead of int.
4550 * tree.c (tree_node_counts): Likewise.
4551 (tree_node_sizes): Likewise.
4552 (dump_tree_statistics): Use PRIu64 in printf format.
4554 2018-01-12 Martin Liska <mliska@suse.cz>
4556 * Makefile.in: As qsort_chk is implemented in vec.c, add
4557 vec.o to linkage of gencfn-macros.
4558 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4559 passing the info to record_node_allocation_statistics.
4560 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4562 * ggc-common.c (struct ggc_usage): Add operator== and use
4563 it in operator< and compare function.
4564 * mem-stats.h (struct mem_usage): Likewise.
4565 * vec.c (struct vec_usage): Remove operator< and compare
4566 function. Can be simply inherited.
4568 2018-01-12 Martin Jambor <mjambor@suse.cz>
4571 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4572 * tree-ssa-math-opts.c: Include domwalk.h.
4573 (convert_mult_to_fma_1): New function.
4574 (fma_transformation_info): New type.
4575 (fma_deferring_state): Likewise.
4576 (cancel_fma_deferring): New function.
4577 (result_of_phi): Likewise.
4578 (last_fma_candidate_feeds_initial_phi): Likewise.
4579 (convert_mult_to_fma): Added deferring logic, split actual
4580 transformation to convert_mult_to_fma_1.
4581 (math_opts_dom_walker): New type.
4582 (math_opts_dom_walker::after_dom_children): New method, body moved
4583 here from pass_optimize_widening_mul::execute, added deferring logic
4585 (pass_optimize_widening_mul::execute): Moved most of code to
4586 math_opts_dom_walker::after_dom_children.
4587 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4588 * config/i386/i386.c (ix86_option_override_internal): Added
4589 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4591 2018-01-12 Richard Biener <rguenther@suse.de>
4594 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4595 inline instance vars.
4597 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4600 * config/rx/rx.c (rx_is_restricted_memory_address):
4603 2018-01-12 Richard Biener <rguenther@suse.de>
4605 PR tree-optimization/80846
4606 * target.def (split_reduction): New target hook.
4607 * targhooks.c (default_split_reduction): New function.
4608 * targhooks.h (default_split_reduction): Declare.
4609 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4610 target requests first reduce vectors by combining low and high
4612 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4613 (get_vectype_for_scalar_type_and_size): Export.
4614 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4615 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4616 * doc/tm.texi: Regenerate.
4617 * config/i386/i386.c (ix86_split_reduction): Implement
4618 TARGET_VECTORIZE_SPLIT_REDUCTION.
4620 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4623 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4624 in PIC mode except for TARGET_VXWORKS_RTP.
4625 * config/sparc/sparc.c: Include cfgrtl.h.
4626 (TARGET_INIT_PIC_REG): Define.
4627 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4628 (sparc_pic_register_p): New predicate.
4629 (sparc_legitimate_address_p): Use it.
4630 (sparc_legitimize_pic_address): Likewise.
4631 (sparc_delegitimize_address): Likewise.
4632 (sparc_mode_dependent_address_p): Likewise.
4633 (gen_load_pcrel_sym): Remove 4th parameter.
4634 (load_got_register): Adjust call to above. Remove obsolete stuff.
4635 (sparc_expand_prologue): Do not call load_got_register here.
4636 (sparc_flat_expand_prologue): Likewise.
4637 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4638 (sparc_use_pseudo_pic_reg): New function.
4639 (sparc_init_pic_reg): Likewise.
4640 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4641 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4643 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4645 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4646 Add item for branch_cost.
4648 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4650 PR rtl-optimization/83565
4651 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4652 not extend the result to a larger mode for rotate operations.
4653 (num_sign_bit_copies1): Likewise.
4655 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4658 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4660 Use values-Xc.o for -pedantic.
4661 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4663 2018-01-12 Martin Liska <mliska@suse.cz>
4666 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4668 (possible_polymorphic_call_targets): Use it.
4669 (ipa_devirt): Likewise.
4671 2018-01-12 Martin Liska <mliska@suse.cz>
4673 * profile-count.h (enum profile_quality): Use 0 as invalid
4674 enum value of profile_quality.
4676 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4678 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4679 -mext-string options.
4681 2018-01-12 Richard Biener <rguenther@suse.de>
4683 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4684 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4685 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4687 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4689 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4691 * configure.ac (--with-long-double-format): Add support for the
4692 configuration option to change the default long double format on
4694 * config.gcc (powerpc*-linux*-*): Likewise.
4695 * configure: Regenerate.
4696 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4697 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4698 used without modification.
4700 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4702 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4703 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4704 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4705 MISC_BUILTIN_SPEC_BARRIER.
4706 (rs6000_init_builtins): Likewise.
4707 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4709 (speculation_barrier): New define_insn.
4710 * doc/extend.texi: Document __builtin_speculation_barrier.
4712 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4715 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4716 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4717 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4719 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4720 integral modes instead of "ss" and "sd".
4721 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4722 vectors with 32-bit and 64-bit elements.
4723 (vecdupssescalarmodesuffix): New mode attribute.
4724 (vec_dup<mode>): Use it.
4726 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4729 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4730 frame if argument is passed on stack.
4732 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4735 * ree.c (combine_reaching_defs): Optimize also
4736 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4737 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4739 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4742 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4744 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4747 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4748 after they are computed.
4750 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4752 PR tree-optimization/83695
4753 * gimple-loop-linterchange.cc
4754 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4755 reset cached scev information after interchange.
4756 (pass_linterchange::execute): Remove call to scev_reset_htab.
4758 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4760 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4761 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4762 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4763 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4764 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4765 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4766 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4767 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4768 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4769 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4770 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4771 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4772 (V_lane_reg): Likewise.
4773 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4775 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4776 (vfmal_lane_low<mode>_intrinsic,
4777 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4778 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4779 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4780 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4781 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4782 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4784 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4786 * config/arm/arm-cpus.in (fp16fml): New feature.
4787 (ALL_SIMD): Add fp16fml.
4788 (armv8.2-a): Add fp16fml as an option.
4789 (armv8.3-a): Likewise.
4790 (armv8.4-a): Add fp16fml as part of fp16.
4791 * config/arm/arm.h (TARGET_FP16FML): Define.
4792 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4794 * config/arm/arm-modes.def (V2HF): Define.
4795 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4796 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4797 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4798 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4799 vfmsl_low, vfmsl_high): New set of builtins.
4800 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4801 (vfml_op): New code attribute.
4802 (VFMLHALVES): New int iterator.
4803 (VFML, VFMLSEL): New mode attributes.
4804 (V_reg): Define mapping for V2HF.
4805 (V_hi, V_lo): New mode attributes.
4806 (VF_constraint): Likewise.
4807 (vfml_half, vfml_half_selector): New int attributes.
4808 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4810 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4811 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4813 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4814 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4815 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4816 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4818 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4819 Document new effective target and option set.
4821 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4823 * config/arm/arm-cpus.in (armv8_4): New feature.
4824 (ARMv8_4a): New fgroup.
4825 (armv8.4-a): New arch.
4826 * config/arm/arm-tables.opt: Regenerate.
4827 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4828 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4829 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4830 Add matching rules for -march=armv8.4-a and extensions.
4831 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4833 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4836 * config/rx/rx.md (BW): New mode attribute.
4837 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4839 2018-01-11 Richard Biener <rguenther@suse.de>
4841 PR tree-optimization/83435
4842 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4843 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4844 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4846 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4847 Alan Hayward <alan.hayward@arm.com>
4848 David Sherwood <david.sherwood@arm.com>
4850 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4852 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4853 (aarch64_print_address_internal): Use it to check for a zero offset.
4855 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4856 Alan Hayward <alan.hayward@arm.com>
4857 David Sherwood <david.sherwood@arm.com>
4859 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4860 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4861 Return a poly_int64 rather than a HOST_WIDE_INT.
4862 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4863 rather than a HOST_WIDE_INT.
4864 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4865 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4866 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4867 final_offset from HOST_WIDE_INT to poly_int64.
4868 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4869 to_constant when getting the number of units in an Advanced SIMD
4871 (aarch64_builtin_vectorized_function): Check for a constant number
4873 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4875 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4876 attribute instead of GET_MODE_NUNITS.
4877 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4878 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4879 GET_MODE_SIZE for fixed-size registers.
4880 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4881 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4882 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4883 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4884 (aarch64_print_operand, aarch64_print_address_internal)
4885 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4886 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4887 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4888 Handle polynomial GET_MODE_SIZE.
4889 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4890 wider than SImode without modification.
4891 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4892 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4893 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4894 passing and returning SVE modes.
4895 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4896 rather than GEN_INT.
4897 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4898 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4899 (aarch64_allocate_and_probe_stack_space): Likewise.
4900 (aarch64_layout_frame): Cope with polynomial offsets.
4901 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4902 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4904 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4905 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4906 poly_int64 rather than a HOST_WIDE_INT.
4907 (aarch64_get_separate_components, aarch64_process_components)
4908 (aarch64_expand_prologue, aarch64_expand_epilogue)
4909 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4910 (aarch64_anchor_offset): New function, split out from...
4911 (aarch64_legitimize_address): ...here.
4912 (aarch64_builtin_vectorization_cost): Handle polynomial
4913 TYPE_VECTOR_SUBPARTS.
4914 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4916 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4917 number of elements from the PARALLEL rather than the mode.
4918 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4919 rather than GET_MODE_BITSIZE.
4920 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4921 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4922 (aarch64_expand_vec_perm_const_1): Handle polynomial
4923 d->perm.length () and d->perm elements.
4924 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4925 Apply to_constant to d->perm elements.
4926 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4927 polynomial CONST_VECTOR_NUNITS.
4928 (aarch64_move_pointer): Take amount as a poly_int64 rather
4930 (aarch64_progress_pointer): Avoid temporary variable.
4931 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4932 the mode attribute instead of GET_MODE.
4934 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4935 Alan Hayward <alan.hayward@arm.com>
4936 David Sherwood <david.sherwood@arm.com>
4938 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4939 x exists before using it.
4940 (aarch64_add_constant_internal): Rename to...
4941 (aarch64_add_offset_1): ...this. Replace regnum with separate
4942 src and dest rtxes. Handle the case in which they're different,
4943 including when the offset is zero. Replace scratchreg with an rtx.
4944 Use 2 additions if there is no spare register into which we can
4945 move a 16-bit constant.
4946 (aarch64_add_constant): Delete.
4947 (aarch64_add_offset): Replace reg with separate src and dest
4948 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4949 Use aarch64_add_offset_1.
4950 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4951 an rtx rather than an int. Take the delta as a poly_int64
4952 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4953 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4954 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4955 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4956 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4958 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4959 aarch64_add_constant.
4961 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4963 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4964 Use scalar_float_mode.
4966 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4968 * config/aarch64/aarch64-simd.md
4969 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4970 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4971 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4972 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4973 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4974 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4975 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4976 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4977 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4978 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4980 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4983 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4984 targ_options->x_arm_arch_string is non NULL.
4986 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4988 * config/aarch64/aarch64.h
4989 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4991 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4994 * expmed.c (emit_store_flag_force): Swap if const op0
4995 and change VOIDmode to mode of op0.
4997 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4999 PR rtl-optimization/83761
5000 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5001 than bytes to mode_for_size.
5003 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5006 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5007 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5010 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5013 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5014 when in layout mode.
5015 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5016 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5019 2018-01-10 Michael Collison <michael.collison@arm.com>
5021 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5022 * config/aarch64/aarch64-option-extension.def: Add
5023 AARCH64_OPT_EXTENSION of 'fp16fml'.
5024 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5025 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5026 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5027 * config/aarch64/constraints.md (Ui7): New constraint.
5028 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5029 (VFMLA_SEL_W): Ditto.
5032 (VFMLA16_LOW): New int iterator.
5033 (VFMLA16_HIGH): Ditto.
5034 (UNSPEC_FMLAL): New unspec.
5035 (UNSPEC_FMLSL): Ditto.
5036 (UNSPEC_FMLAL2): Ditto.
5037 (UNSPEC_FMLSL2): Ditto.
5038 (f16mac): New code attribute.
5039 * config/aarch64/aarch64-simd-builtins.def
5040 (aarch64_fmlal_lowv2sf): Ditto.
5041 (aarch64_fmlsl_lowv2sf): Ditto.
5042 (aarch64_fmlalq_lowv4sf): Ditto.
5043 (aarch64_fmlslq_lowv4sf): Ditto.
5044 (aarch64_fmlal_highv2sf): Ditto.
5045 (aarch64_fmlsl_highv2sf): Ditto.
5046 (aarch64_fmlalq_highv4sf): Ditto.
5047 (aarch64_fmlslq_highv4sf): Ditto.
5048 (aarch64_fmlal_lane_lowv2sf): Ditto.
5049 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5050 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5051 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5052 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5053 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5054 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5055 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5056 (aarch64_fmlal_lane_highv2sf): Ditto.
5057 (aarch64_fmlsl_lane_highv2sf): Ditto.
5058 (aarch64_fmlal_laneq_highv2sf): Ditto.
5059 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5060 (aarch64_fmlalq_lane_highv4sf): Ditto.
5061 (aarch64_fmlsl_lane_highv4sf): Ditto.
5062 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5063 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5064 * config/aarch64/aarch64-simd.md:
5065 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5066 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5067 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5068 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5069 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5070 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5071 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5072 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5073 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5074 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5075 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5076 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5077 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5078 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5079 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5080 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5081 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5082 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5083 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5084 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5085 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5086 (vfmlsl_low_u32): Ditto.
5087 (vfmlalq_low_u32): Ditto.
5088 (vfmlslq_low_u32): Ditto.
5089 (vfmlal_high_u32): Ditto.
5090 (vfmlsl_high_u32): Ditto.
5091 (vfmlalq_high_u32): Ditto.
5092 (vfmlslq_high_u32): Ditto.
5093 (vfmlal_lane_low_u32): Ditto.
5094 (vfmlsl_lane_low_u32): Ditto.
5095 (vfmlal_laneq_low_u32): Ditto.
5096 (vfmlsl_laneq_low_u32): Ditto.
5097 (vfmlalq_lane_low_u32): Ditto.
5098 (vfmlslq_lane_low_u32): Ditto.
5099 (vfmlalq_laneq_low_u32): Ditto.
5100 (vfmlslq_laneq_low_u32): Ditto.
5101 (vfmlal_lane_high_u32): Ditto.
5102 (vfmlsl_lane_high_u32): Ditto.
5103 (vfmlal_laneq_high_u32): Ditto.
5104 (vfmlsl_laneq_high_u32): Ditto.
5105 (vfmlalq_lane_high_u32): Ditto.
5106 (vfmlslq_lane_high_u32): Ditto.
5107 (vfmlalq_laneq_high_u32): Ditto.
5108 (vfmlslq_laneq_high_u32): Ditto.
5109 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5110 (AARCH64_FL_FOR_ARCH8_4): New.
5111 (AARCH64_ISA_F16FML): New ISA flag.
5112 (TARGET_F16FML): New feature flag for fp16fml.
5113 (doc/invoke.texi): Document new fp16fml option.
5115 2018-01-10 Michael Collison <michael.collison@arm.com>
5117 * config/aarch64/aarch64-builtins.c:
5118 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5119 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5120 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5121 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5122 (AARCH64_ISA_SHA3): New ISA flag.
5123 (TARGET_SHA3): New feature flag for sha3.
5124 * config/aarch64/iterators.md (sha512_op): New int attribute.
5125 (CRYPTO_SHA512): New int iterator.
5126 (UNSPEC_SHA512H): New unspec.
5127 (UNSPEC_SHA512H2): Ditto.
5128 (UNSPEC_SHA512SU0): Ditto.
5129 (UNSPEC_SHA512SU1): Ditto.
5130 * config/aarch64/aarch64-simd-builtins.def
5131 (aarch64_crypto_sha512hqv2di): New builtin.
5132 (aarch64_crypto_sha512h2qv2di): Ditto.
5133 (aarch64_crypto_sha512su0qv2di): Ditto.
5134 (aarch64_crypto_sha512su1qv2di): Ditto.
5135 (aarch64_eor3qv8hi): Ditto.
5136 (aarch64_rax1qv2di): Ditto.
5137 (aarch64_xarqv2di): Ditto.
5138 (aarch64_bcaxqv8hi): Ditto.
5139 * config/aarch64/aarch64-simd.md:
5140 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5141 (aarch64_crypto_sha512su0qv2di): Ditto.
5142 (aarch64_crypto_sha512su1qv2di): Ditto.
5143 (aarch64_eor3qv8hi): Ditto.
5144 (aarch64_rax1qv2di): Ditto.
5145 (aarch64_xarqv2di): Ditto.
5146 (aarch64_bcaxqv8hi): Ditto.
5147 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5148 (vsha512h2q_u64): Ditto.
5149 (vsha512su0q_u64): Ditto.
5150 (vsha512su1q_u64): Ditto.
5151 (veor3q_u16): Ditto.
5152 (vrax1q_u64): Ditto.
5154 (vbcaxq_u16): Ditto.
5155 * config/arm/types.md (crypto_sha512): New type attribute.
5156 (crypto_sha3): Ditto.
5157 (doc/invoke.texi): Document new sha3 option.
5159 2018-01-10 Michael Collison <michael.collison@arm.com>
5161 * config/aarch64/aarch64-builtins.c:
5162 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5163 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5164 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5165 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5166 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5167 (AARCH64_ISA_SM4): New ISA flag.
5168 (TARGET_SM4): New feature flag for sm4.
5169 * config/aarch64/aarch64-simd-builtins.def
5170 (aarch64_sm3ss1qv4si): Ditto.
5171 (aarch64_sm3tt1aq4si): Ditto.
5172 (aarch64_sm3tt1bq4si): Ditto.
5173 (aarch64_sm3tt2aq4si): Ditto.
5174 (aarch64_sm3tt2bq4si): Ditto.
5175 (aarch64_sm3partw1qv4si): Ditto.
5176 (aarch64_sm3partw2qv4si): Ditto.
5177 (aarch64_sm4eqv4si): Ditto.
5178 (aarch64_sm4ekeyqv4si): Ditto.
5179 * config/aarch64/aarch64-simd.md:
5180 (aarch64_sm3ss1qv4si): Ditto.
5181 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5182 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5183 (aarch64_sm4eqv4si): Ditto.
5184 (aarch64_sm4ekeyqv4si): Ditto.
5185 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5186 (sm3part_op): Ditto.
5187 (CRYPTO_SM3TT): Ditto.
5188 (CRYPTO_SM3PART): Ditto.
5189 (UNSPEC_SM3SS1): New unspec.
5190 (UNSPEC_SM3TT1A): Ditto.
5191 (UNSPEC_SM3TT1B): Ditto.
5192 (UNSPEC_SM3TT2A): Ditto.
5193 (UNSPEC_SM3TT2B): Ditto.
5194 (UNSPEC_SM3PARTW1): Ditto.
5195 (UNSPEC_SM3PARTW2): Ditto.
5196 (UNSPEC_SM4E): Ditto.
5197 (UNSPEC_SM4EKEY): Ditto.
5198 * config/aarch64/constraints.md (Ui2): New constraint.
5199 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5200 * config/arm/types.md (crypto_sm3): New type attribute.
5201 (crypto_sm4): Ditto.
5202 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5203 (vsm3tt1aq_u32): Ditto.
5204 (vsm3tt1bq_u32): Ditto.
5205 (vsm3tt2aq_u32): Ditto.
5206 (vsm3tt2bq_u32): Ditto.
5207 (vsm3partw1q_u32): Ditto.
5208 (vsm3partw2q_u32): Ditto.
5209 (vsm4eq_u32): Ditto.
5210 (vsm4ekeyq_u32): Ditto.
5211 (doc/invoke.texi): Document new sm4 option.
5213 2018-01-10 Michael Collison <michael.collison@arm.com>
5215 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5216 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5217 (AARCH64_FL_FOR_ARCH8_4): New.
5218 (AARCH64_FL_V8_4): New flag.
5219 (doc/invoke.texi): Document new armv8.4-a option.
5221 2018-01-10 Michael Collison <michael.collison@arm.com>
5223 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5224 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5225 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5226 * config/aarch64/aarch64-option-extension.def: Add
5227 AARCH64_OPT_EXTENSION of 'sha2'.
5228 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5229 (crypto): Disable sha2 and aes if crypto disabled.
5230 (crypto): Enable aes and sha2 if enabled.
5231 (simd): Disable sha2 and aes if simd disabled.
5232 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5234 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5235 (TARGET_SHA2): New feature flag for sha2.
5236 (TARGET_AES): New feature flag for aes.
5237 * config/aarch64/aarch64-simd.md:
5238 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5239 conditional on TARGET_AES.
5240 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5241 (aarch64_crypto_sha1hsi): Make pattern conditional
5243 (aarch64_crypto_sha1hv4si): Ditto.
5244 (aarch64_be_crypto_sha1hv4si): Ditto.
5245 (aarch64_crypto_sha1su1v4si): Ditto.
5246 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5247 (aarch64_crypto_sha1su0v4si): Ditto.
5248 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5249 (aarch64_crypto_sha256su0v4si): Ditto.
5250 (aarch64_crypto_sha256su1v4si): Ditto.
5251 (doc/invoke.texi): Document new aes and sha2 options.
5253 2018-01-10 Martin Sebor <msebor@redhat.com>
5255 PR tree-optimization/83781
5256 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5259 2018-01-11 Martin Sebor <msebor@gmail.com>
5260 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5262 PR tree-optimization/83501
5263 PR tree-optimization/81703
5265 * tree-ssa-strlen.c (get_string_cst): Rename...
5266 (get_string_len): ...to this. Handle global constants.
5267 (handle_char_store): Adjust.
5269 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5270 Jim Wilson <jimw@sifive.com>
5272 * config/riscv/riscv-protos.h (riscv_output_return): New.
5273 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5274 (riscv_attribute_table, riscv_output_return),
5275 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5276 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5277 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5279 (riscv_expand_prologue): Add early return for naked function.
5280 (riscv_expand_epilogue): Likewise.
5281 (riscv_function_ok_for_sibcall): Return false for naked function.
5282 (riscv_set_current_function): New.
5283 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5284 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5285 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5286 * doc/extend.texi (RISC-V Function Attributes): New.
5288 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5290 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5291 check for 128-bit long double before checking TCmode.
5292 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5293 128-bit long doubles before checking TFmode or TCmode.
5294 (FLOAT128_IBM_P): Likewise.
5296 2018-01-10 Martin Sebor <msebor@redhat.com>
5298 PR tree-optimization/83671
5299 * builtins.c (c_strlen): Unconditionally return zero for the empty
5301 Use -Warray-bounds for warnings.
5302 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5303 for non-constant array indices with COMPONENT_REF, arrays of
5304 arrays, and pointers to arrays.
5305 (gimple_fold_builtin_strlen): Determine and set length range for
5306 non-constant character arrays.
5308 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5311 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5314 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5316 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5318 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5321 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5322 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5323 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5324 indexed_or_indirect_operand predicate.
5325 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5326 (*vsx_le_perm_load_v8hi): Likewise.
5327 (*vsx_le_perm_load_v16qi): Likewise.
5328 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5329 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5330 (*vsx_le_perm_store_v8hi): Likewise.
5331 (*vsx_le_perm_store_v16qi): Likewise.
5332 (eight unnamed splitters): Likewise.
5334 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5336 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5337 * config/rs6000/emmintrin.h: Likewise.
5338 * config/rs6000/mmintrin.h: Likewise.
5339 * config/rs6000/xmmintrin.h: Likewise.
5341 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5344 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5346 * tree.c (tree_nop_conversion): Return true for location wrapper
5348 (maybe_wrap_with_location): New function.
5349 (selftest::check_strip_nops): New function.
5350 (selftest::test_location_wrappers): New function.
5351 (selftest::tree_c_tests): Call it.
5352 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5353 (maybe_wrap_with_location): New decl.
5354 (EXPR_LOCATION_WRAPPER_P): New macro.
5355 (location_wrapper_p): New inline function.
5356 (tree_strip_any_location_wrapper): New inline function.
5358 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5361 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5362 stack_realign_offset for the largest alignment of stack slot
5364 (ix86_find_max_used_stack_alignment): New function.
5365 (ix86_finalize_stack_frame_flags): Use it. Set
5366 max_used_stack_alignment if we don't realign stack.
5367 * config/i386/i386.h (machine_function): Add
5368 max_used_stack_alignment.
5370 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5372 * config/arm/arm.opt (-mbranch-cost): New option.
5373 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5376 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5379 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5380 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5382 2018-01-10 Richard Biener <rguenther@suse.de>
5385 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5386 early out so it also covers the case where we have a non-NULL
5389 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5391 PR tree-optimization/83753
5392 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5393 for non-strided grouped accesses if the number of elements is 1.
5395 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5398 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5399 * i386.h (TARGET_USE_GATHER): Define.
5400 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5402 2018-01-10 Martin Liska <mliska@suse.cz>
5405 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5406 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5408 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5409 CLEANUP_NO_PARTITIONING is not set.
5411 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5413 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5414 for vectors, as a partial revert of r254296.
5415 * rtl.h (const_vec_p): Delete.
5416 (const_vec_duplicate_p): Don't test for vector CONSTs.
5417 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5418 * expmed.c (make_tree): Likewise.
5421 * common.md (E, F): Use CONSTANT_P instead of checking for
5423 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5424 checking for CONST_VECTOR.
5426 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5429 * predict.c (force_edge_cold): Handle in more sane way edges
5432 2018-01-09 Carl Love <cel@us.ibm.com>
5434 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5436 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5437 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5438 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5439 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5440 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5441 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5442 * config/rs6000/rs6000-protos.h: Add extern defition for
5443 rs6000_generate_float2_double_code.
5444 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5446 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5447 (float2_v2df): Add define_expand.
5449 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5452 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5453 op_mode in the force_to_mode call.
5455 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5457 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5458 instead of checking each element individually.
5459 (aarch64_evpc_uzp): Likewise.
5460 (aarch64_evpc_zip): Likewise.
5461 (aarch64_evpc_ext): Likewise.
5462 (aarch64_evpc_rev): Likewise.
5463 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5464 instead of checking each element individually. Return true without
5466 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5467 whether all selected elements come from the same input, instead of
5468 checking each element individually. Remove calls to gen_rtx_REG,
5469 start_sequence and end_sequence and instead assert that no rtl is
5472 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5474 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5475 order of HIGH and CONST checks.
5477 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5479 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5480 if the destination isn't an SSA_NAME.
5482 2018-01-09 Richard Biener <rguenther@suse.de>
5484 PR tree-optimization/83668
5485 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5487 (canonicalize_loop_form): ... here, renamed from ...
5488 (canonicalize_loop_closed_ssa_form): ... this and amended to
5489 swap successor edges for loop exit blocks to make us use
5490 the RPO order we need for initial schedule generation.
5492 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5494 PR tree-optimization/64811
5495 * match.pd: When optimizing comparisons with Inf, avoid
5496 introducing or losing exceptions from comparisons with NaN.
5498 2018-01-09 Martin Liska <mliska@suse.cz>
5501 * asan.c (shadow_mem_size): Add gcc_assert.
5503 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5505 Don't save registers in main().
5508 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5509 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5510 * config/avr/avr.c (avr_set_current_function): Don't error if
5511 naked, OS_task or OS_main are specified at the same time.
5512 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5514 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5516 * common/config/avr/avr-common.c (avr_option_optimization_table):
5517 Switch on -mmain-is-OS_task for optimizing compilations.
5519 2018-01-09 Richard Biener <rguenther@suse.de>
5521 PR tree-optimization/83572
5522 * graphite.c: Include cfganal.h.
5523 (graphite_transform_loops): Connect infinite loops to exit
5524 and remove fake edges at the end.
5526 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5528 * ipa-inline.c (edge_badness): Revert accidental checkin.
5530 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5533 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5534 symbols; not inline clones.
5536 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5539 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5540 hard registers. Formatting fixes.
5542 PR preprocessor/83722
5543 * gcc.c (try_generate_repro): Pass
5544 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5545 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5548 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5549 Kito Cheng <kito.cheng@gmail.com>
5551 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5552 (riscv_leaf_function_p): Delete.
5553 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5555 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5557 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5559 (do_ifelse): New function.
5560 (do_isel): New function.
5561 (do_sub3): New function.
5562 (do_add3): New function.
5563 (do_load_mask_compare): New function.
5564 (do_overlap_load_compare): New function.
5565 (expand_compare_loop): New function.
5566 (expand_block_compare): Call expand_compare_loop() when appropriate.
5567 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5569 (-mblock-compare-inline-loop-limit): New option.
5571 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5574 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5575 Reverse order of second and third operands in first alternative.
5576 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5577 of first and second elements in UNSPEC_VPERMR vector.
5578 (altivec_expand_vec_perm_le): Likewise.
5580 2018-01-08 Jeff Law <law@redhat.com>
5582 PR rtl-optimizatin/81308
5583 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5584 (process_switch): If group_case_labels makes a change, then set
5586 (pass_convert_switch::execute): If a switch is converted, then
5587 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5589 PR rtl-optimization/81308
5590 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5593 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5595 PR target/83663 - Revert r255946
5596 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5597 generation for cases where splatting a value is not useful.
5598 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5599 across a vec_duplicate and a paradoxical subreg forming a vector
5600 mode to a vec_concat.
5602 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5604 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5605 -march=armv8.3-a variants.
5606 * config/arm/t-multilib: Likewise.
5607 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5609 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5611 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5613 (cceq_ior_compare_complement): Give it a name so I can use it, and
5614 change boolean_or_operator predicate to boolean_operator so it can
5615 be used to generate a crand.
5616 (eqne): New code iterator.
5617 (bd/bd_neg): New code_attrs.
5618 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5619 a single define_insn.
5620 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5621 decrement (bdnzt/bdnzf/bdzt/bdzf).
5622 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5623 with the new names of the branch decrement patterns, and added the
5624 names of the branch decrement conditional patterns.
5626 2018-01-08 Richard Biener <rguenther@suse.de>
5628 PR tree-optimization/83563
5629 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5632 2018-01-08 Richard Biener <rguenther@suse.de>
5635 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5637 2018-01-08 Richard Biener <rguenther@suse.de>
5639 PR tree-optimization/83685
5640 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5641 references to abnormals.
5643 2018-01-08 Richard Biener <rguenther@suse.de>
5646 * dwarf2out.c (output_indirect_strings): Handle empty
5647 skeleton_debug_str_hash.
5648 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5650 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5652 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5653 (emit_store_direct): Likewise.
5654 (arc_trampoline_adjust_address): Likewise.
5655 (arc_asm_trampoline_template): New function.
5656 (arc_initialize_trampoline): Use asm_trampoline_template.
5657 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5658 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5659 * config/arc/arc.md (flush_icache): Delete pattern.
5661 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5663 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5664 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5667 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5670 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5671 by not USED_FOR_TARGET.
5672 (make_pass_resolve_sw_modes): Likewise.
5674 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5676 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5679 2018-01-08 Richard Biener <rguenther@suse.de>
5682 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5684 2018-01-08 Richard Biener <rguenther@suse.de>
5687 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5689 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5692 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5693 basic blocks with a small number of successors.
5694 (convert_control_dep_chain_into_preds): Improve handling of
5696 (dump_predicates): Split apart into...
5697 (dump_pred_chain): ...here...
5698 (dump_pred_info): ...and here.
5699 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5700 (can_chain_union_be_invalidated_p): Improve check for invalidation
5702 (uninit_uses_cannot_happen): Avoid unnecessary if
5703 convert_control_dep_chain_into_preds yielded nothing.
5705 2018-01-06 Martin Sebor <msebor@redhat.com>
5707 PR tree-optimization/83640
5708 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5709 subtracting negative offset from size.
5710 (builtin_access::overlap): Adjust offset bounds of the access to fall
5711 within the size of the object if possible.
5713 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5715 PR rtl-optimization/83699
5716 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5717 extract_bit_field_as_subreg to cases in which the extracted
5718 value is also a vector.
5720 * lra-constraints.c (process_alt_operands): Test for the equivalence
5721 substitutions when detecting a possible reload cycle.
5723 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5726 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5727 by default if flag_selective_schedling{,2}. Formatting fixes.
5729 PR rtl-optimization/83682
5730 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5731 if it has non-VECTOR_MODE element mode.
5732 (vec_duplicate_p): Likewise.
5735 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5736 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5738 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5741 * config/i386/i386-builtin.def
5742 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5743 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5744 Require also OPTION_MASK_ISA_AVX512F in addition to
5745 OPTION_MASK_ISA_GFNI.
5746 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5747 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5748 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5749 to OPTION_MASK_ISA_GFNI.
5750 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5751 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5752 OPTION_MASK_ISA_AVX512BW.
5753 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5754 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5755 addition to OPTION_MASK_ISA_GFNI.
5756 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5757 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5758 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5759 to OPTION_MASK_ISA_GFNI.
5760 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5761 a requirement for all ISAs rather than any of them with a few
5763 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5765 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5766 bitmasks to be enabled with 3 exceptions, instead of requiring any
5767 enabled ISA with lots of exceptions.
5768 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5769 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5770 Change avx512bw in isa attribute to avx512f.
5771 * config/i386/sgxintrin.h: Add license boilerplate.
5772 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5773 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5774 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5775 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5777 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5778 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5779 temporarily sse2 rather than sse if not enabled already.
5782 * config/i386/sse.md (VI248_VLBW): Rename to ...
5783 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5784 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5785 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5786 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5787 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5788 mode iterator instead of VI248_VLBW.
5790 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5792 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5793 (record_modified): Skip clobbers; add debug output.
5794 (param_change_prob): Use sreal frequencies.
5796 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5798 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5799 punt for user-aligned variables.
5801 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5803 * tree-chrec.c (chrec_contains_symbols): Return true for
5806 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5809 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5810 of (x|y) == x for BICS pattern.
5812 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5814 PR tree-optimization/83605
5815 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5816 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5819 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5821 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5822 * config/epiphany/rtems.h: New file.
5824 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5825 Uros Bizjak <ubizjak@gmail.com>
5828 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5829 QIreg_operand instead of register_operand predicate.
5830 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5831 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5832 comments instead of -fmitigate[-_]rop.
5834 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5837 * cgraphunit.c (symbol_table::compile): Switch to text_section
5838 before calling assembly_start debug hook.
5839 * run-rtl-passes.c (run_rtl_passes): Likewise.
5842 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5844 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5845 range_int_cst_p rather than !symbolic_range_p before calling
5846 extract_range_from_multiplicative_op_1.
5848 2018-01-04 Jeff Law <law@redhat.com>
5850 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5851 redundant test in assertion.
5853 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5855 * doc/rtl.texi: Document machine_mode wrapper classes.
5857 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5859 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5862 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5864 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5865 the VEC_PERM_EXPR fold to fail.
5867 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5870 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5871 to switched_sections.
5873 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5876 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5879 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5882 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5883 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5885 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5888 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5889 is BLKmode and bitpos not zero or mode change is needed.
5891 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5894 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5897 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5900 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5901 instead of MULT rtx. Update all corresponding splitters.
5903 (*ssub<modesuffix>): Ditto.
5905 (*cmp_sadd_di): Update split patterns.
5906 (*cmp_sadd_si): Ditto.
5907 (*cmp_sadd_sidi): Ditto.
5908 (*cmp_ssub_di): Ditto.
5909 (*cmp_ssub_si): Ditto.
5910 (*cmp_ssub_sidi): Ditto.
5911 * config/alpha/predicates.md (const23_operand): New predicate.
5912 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5913 Look for ASHIFT, not MULT inner operand.
5914 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5916 2018-01-04 Martin Liska <mliska@suse.cz>
5918 PR gcov-profile/83669
5919 * gcov.c (output_intermediate_file): Add version to intermediate
5921 * doc/gcov.texi: Document new field 'version' in intermediate
5922 file format. Fix location of '-k' option of gcov command.
5924 2018-01-04 Martin Liska <mliska@suse.cz>
5927 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5929 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5931 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5933 2018-01-03 Martin Sebor <msebor@redhat.com>
5935 PR tree-optimization/83655
5936 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5937 checking calls with invalid arguments.
5939 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5941 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5942 (vectorizable_mask_load_store): Delete.
5943 (vectorizable_call): Return false for masked loads and stores.
5944 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5945 instead of gimple_assign_rhs1.
5946 (vectorizable_load): Handle IFN_MASK_LOAD.
5947 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5949 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5951 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5953 (vectorizable_mask_load_store): ...here.
5954 (vectorizable_load): ...and here.
5956 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5958 * tree-vect-stmts.c (vect_build_all_ones_mask)
5959 (vect_build_zero_merge_argument): New functions, split out from...
5960 (vectorizable_load): ...here.
5962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5964 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5966 (vectorizable_mask_load_store): ...here.
5967 (vectorizable_store): ...and here.
5969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5971 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5973 (vectorizable_mask_load_store): ...here.
5975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5977 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5978 (vect_model_store_cost): Take a vec_load_store_type instead of a
5980 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5981 (vect_model_store_cost): Take a vec_load_store_type instead of a
5983 (vectorizable_mask_load_store): Update accordingly.
5984 (vectorizable_store): Likewise.
5985 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5987 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5989 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5990 IFN_MASK_LOAD calls here rather than...
5991 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5993 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5994 Alan Hayward <alan.hayward@arm.com>
5995 David Sherwood <david.sherwood@arm.com>
5997 * expmed.c (extract_bit_field_1): For vector extracts,
5998 fall back to extract_bit_field_as_subreg if vec_extract
6001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6002 Alan Hayward <alan.hayward@arm.com>
6003 David Sherwood <david.sherwood@arm.com>
6005 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6006 they are variable or constant sized.
6007 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6008 slots for constant-sized data.
6010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6011 Alan Hayward <alan.hayward@arm.com>
6012 David Sherwood <david.sherwood@arm.com>
6014 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6015 handling COND_EXPRs with boolean comparisons, try to find a better
6016 basis for the mask type than the boolean itself.
6018 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6020 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6021 is calculated and how it can be overridden.
6022 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6023 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6025 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6029 Alan Hayward <alan.hayward@arm.com>
6030 David Sherwood <david.sherwood@arm.com>
6032 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6033 Remove the mode argument.
6034 (aarch64_simd_valid_immediate): Remove the mode and inverse
6036 * config/aarch64/iterators.md (bitsize): New iterator.
6037 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6038 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6039 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6040 aarch64_simd_valid_immediate.
6041 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6042 (aarch64_reg_or_bic_imm): Likewise.
6043 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6044 with an insn_type enum and msl with a modifier_type enum.
6045 Replace element_width with a scalar_mode. Change the shift
6046 to unsigned int. Add constructors for scalar_float_mode and
6047 scalar_int_mode elements.
6048 (aarch64_vect_float_const_representable_p): Delete.
6049 (aarch64_can_const_movi_rtx_p)
6050 (aarch64_simd_scalar_immediate_valid_for_move)
6051 (aarch64_simd_make_constant): Update call to
6052 aarch64_simd_valid_immediate.
6053 (aarch64_advsimd_valid_immediate_hs): New function.
6054 (aarch64_advsimd_valid_immediate): Likewise.
6055 (aarch64_simd_valid_immediate): Remove mode and inverse
6056 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6057 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6058 and aarch64_float_const_representable_p on the result.
6059 (aarch64_output_simd_mov_immediate): Remove mode argument.
6060 Update call to aarch64_simd_valid_immediate and use of
6061 simd_immediate_info.
6062 (aarch64_output_scalar_simd_mov_immediate): Update call
6065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6066 Alan Hayward <alan.hayward@arm.com>
6067 David Sherwood <david.sherwood@arm.com>
6069 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6070 (mode_nunits): Likewise CONST_MODE_NUNITS.
6071 * machmode.def (ADJUST_NUNITS): Document.
6072 * genmodes.c (mode_data::need_nunits_adj): New field.
6073 (blank_mode): Update accordingly.
6074 (adj_nunits): New variable.
6075 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6077 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6078 listed in adj_nunits.
6079 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6080 listed in adj_nunits. Don't emit case statements for such modes.
6081 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6082 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6083 nothing if adj_nunits is nonnull.
6084 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6085 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6086 (emit_mode_fbit): Update use of print_maybe_const_decl.
6087 (emit_move_size): Likewise. Treat the array as non-const
6089 (emit_mode_adjustments): Handle adj_nunits.
6091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6093 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6094 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6095 (VECTOR_MODES): Use it.
6096 (make_vector_modes): Take the prefix as an argument.
6098 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6099 Alan Hayward <alan.hayward@arm.com>
6100 David Sherwood <david.sherwood@arm.com>
6102 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6103 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6104 for MODE_VECTOR_BOOL.
6105 * machmode.def (VECTOR_BOOL_MODE): Document.
6106 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6107 (make_vector_bool_mode): New function.
6108 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6110 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6111 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6113 * stor-layout.c (int_mode_for_mode): Likewise.
6114 * tree.c (build_vector_type_for_mode): Likewise.
6115 * varasm.c (output_constant_pool_2): Likewise.
6116 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6117 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6118 for MODE_VECTOR_BOOL.
6119 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6120 of mode class checks.
6121 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6122 instead of a list of mode class checks.
6123 (expand_vector_scalar_condition): Likewise.
6124 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6127 Alan Hayward <alan.hayward@arm.com>
6128 David Sherwood <david.sherwood@arm.com>
6130 * machmode.h (mode_size): Change from unsigned short to
6132 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6133 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6134 or if measurement_type is not polynomial.
6135 (fixed_size_mode::includes_p): Check for constant-sized modes.
6136 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6137 return a poly_uint16 rather than an unsigned short.
6138 (emit_mode_size): Change the type of mode_size from unsigned short
6139 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6140 (emit_mode_adjustments): Cope with polynomial vector sizes.
6141 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6143 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6145 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6146 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6147 * caller-save.c (setup_save_areas): Likewise.
6148 (replace_reg_with_saved_mem): Likewise.
6149 * calls.c (emit_library_call_value_1): Likewise.
6150 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6151 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6152 (gen_lowpart_for_combine): Likewise.
6153 * convert.c (convert_to_integer_1): Likewise.
6154 * cse.c (equiv_constant, cse_insn): Likewise.
6155 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6156 (cselib_subst_to_values): Likewise.
6157 * dce.c (word_dce_process_block): Likewise.
6158 * df-problems.c (df_word_lr_mark_ref): Likewise.
6159 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6160 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6161 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6162 (rtl_for_decl_location): Likewise.
6163 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6164 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6165 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6166 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6167 (expand_expr_real_1): Likewise.
6168 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6169 (pad_below): Likewise.
6170 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6171 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6172 * ira.c (get_subreg_tracking_sizes): Likewise.
6173 * ira-build.c (ira_create_allocno_objects): Likewise.
6174 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6175 (ira_sort_regnos_for_alter_reg): Likewise.
6176 * ira-costs.c (record_operand_costs): Likewise.
6177 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6178 (resolve_simple_move): Likewise.
6179 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6180 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6181 (lra_constraints): Likewise.
6182 (CONST_POOL_OK_P): Reject variable-sized modes.
6183 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6184 (add_pseudo_to_slot, lra_spill): Likewise.
6185 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6186 * optabs-query.c (get_best_extraction_insn): Likewise.
6187 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6188 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6189 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6190 * recog.c (offsettable_address_addr_space_p): Likewise.
6191 * regcprop.c (maybe_mode_change): Likewise.
6192 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6193 * regrename.c (build_def_use): Likewise.
6194 * regstat.c (dump_reg_info): Likewise.
6195 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6196 (find_reloads, find_reloads_subreg_address): Likewise.
6197 * reload1.c (eliminate_regs_1): Likewise.
6198 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6199 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6200 (simplify_binary_operation_1, simplify_subreg): Likewise.
6201 * targhooks.c (default_function_arg_padding): Likewise.
6202 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6203 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6204 (verify_gimple_assign_ternary): Likewise.
6205 * tree-inline.c (estimate_move_cost): Likewise.
6206 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6207 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6208 (get_address_cost_ainc): Likewise.
6209 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6210 (vect_supportable_dr_alignment): Likewise.
6211 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6212 (vectorizable_reduction): Likewise.
6213 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6214 (vectorizable_operation, vectorizable_load): Likewise.
6215 * tree.c (build_same_sized_truth_vector_type): Likewise.
6216 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6217 * var-tracking.c (emit_note_insn_var_location): Likewise.
6218 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6219 (ADDR_VEC_ALIGN): Likewise.
6221 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6222 Alan Hayward <alan.hayward@arm.com>
6223 David Sherwood <david.sherwood@arm.com>
6225 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6227 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6228 or if measurement_type is polynomial.
6229 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6230 * combine.c (make_extraction): Likewise.
6231 * dse.c (find_shift_sequence): Likewise.
6232 * dwarf2out.c (mem_loc_descriptor): Likewise.
6233 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6234 (extract_bit_field, extract_low_bits): Likewise.
6235 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6236 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6237 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6238 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6239 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6240 * reload.c (find_reloads): Likewise.
6241 * reload1.c (alter_reg): Likewise.
6242 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6243 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6244 * tree-if-conv.c (predicate_mem_writes): Likewise.
6245 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6246 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6247 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6248 * valtrack.c (dead_debug_insert_temp): Likewise.
6249 * varasm.c (mergeable_constant_section): Likewise.
6250 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6253 Alan Hayward <alan.hayward@arm.com>
6254 David Sherwood <david.sherwood@arm.com>
6256 * expr.c (expand_assignment): Cope with polynomial mode sizes
6257 when assigning to a CONCAT.
6259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6260 Alan Hayward <alan.hayward@arm.com>
6261 David Sherwood <david.sherwood@arm.com>
6263 * machmode.h (mode_precision): Change from unsigned short to
6265 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6267 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6268 or if measurement_type is not polynomial.
6269 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6270 in which the mode is already known to be a scalar_int_mode.
6271 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6272 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6274 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6275 for GET_MODE_PRECISION.
6276 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6277 for GET_MODE_PRECISION.
6278 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6280 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6281 (expand_field_assignment, make_extraction): Likewise.
6282 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6283 (get_last_value): Likewise.
6284 * convert.c (convert_to_integer_1): Likewise.
6285 * cse.c (cse_insn): Likewise.
6286 * expr.c (expand_expr_real_1): Likewise.
6287 * lra-constraints.c (simplify_operand_subreg): Likewise.
6288 * optabs-query.c (can_atomic_load_p): Likewise.
6289 * optabs.c (expand_atomic_load): Likewise.
6290 (expand_atomic_store): Likewise.
6291 * ree.c (combine_reaching_defs): Likewise.
6292 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6293 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6294 * tree.h (type_has_mode_precision_p): Likewise.
6295 * ubsan.c (instrument_si_overflow): Likewise.
6297 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6298 Alan Hayward <alan.hayward@arm.com>
6299 David Sherwood <david.sherwood@arm.com>
6301 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6302 polynomial numbers of units.
6303 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6304 (valid_vector_subparts_p): New function.
6305 (build_vector_type): Remove temporary shim and take the number
6306 of units as a poly_uint64 rather than an int.
6307 (build_opaque_vector_type): Take the number of units as a
6308 poly_uint64 rather than an int.
6309 * tree.c (build_vector_from_ctor): Handle polynomial
6310 TYPE_VECTOR_SUBPARTS.
6311 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6312 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6313 (build_vector_from_val): If the number of units is variable,
6314 use build_vec_duplicate_cst for constant operands and
6315 VEC_DUPLICATE_EXPR otherwise.
6316 (make_vector_type): Remove temporary is_constant ().
6317 (build_vector_type, build_opaque_vector_type): Take the number of
6318 units as a poly_uint64 rather than an int.
6319 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6321 * cfgexpand.c (expand_debug_expr): Likewise.
6322 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6323 (store_constructor, expand_expr_real_1): Likewise.
6324 (const_scalar_mask_from_tree): Likewise.
6325 * fold-const-call.c (fold_const_reduction): Likewise.
6326 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6327 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6328 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6329 (fold_relational_const): Likewise.
6330 (native_interpret_vector): Likewise. Change the size from an
6331 int to an unsigned int.
6332 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6333 TYPE_VECTOR_SUBPARTS.
6334 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6335 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6336 duplicating a non-constant operand into a variable-length vector.
6337 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6338 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6339 * ipa-icf.c (sem_variable::equals): Likewise.
6340 * match.pd: Likewise.
6341 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6342 * print-tree.c (print_node): Likewise.
6343 * stor-layout.c (layout_type): Likewise.
6344 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6345 * tree-cfg.c (verify_gimple_comparison): Likewise.
6346 (verify_gimple_assign_binary): Likewise.
6347 (verify_gimple_assign_ternary): Likewise.
6348 (verify_gimple_assign_single): Likewise.
6349 * tree-pretty-print.c (dump_generic_node): Likewise.
6350 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6351 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6352 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6353 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6354 (vect_shift_permute_load_chain): Likewise.
6355 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6356 (expand_vector_condition, optimize_vector_constructor): Likewise.
6357 (lower_vec_perm, get_compute_type): Likewise.
6358 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6359 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6360 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6361 (vect_recog_mask_conversion_pattern): Likewise.
6362 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6363 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6364 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6365 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6366 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6367 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6368 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6369 (supportable_widening_operation): Likewise.
6370 (supportable_narrowing_operation): Likewise.
6371 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6373 * varasm.c (output_constant): Likewise.
6375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6376 Alan Hayward <alan.hayward@arm.com>
6377 David Sherwood <david.sherwood@arm.com>
6379 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6380 so that both the length == 3 and length != 3 cases set up their
6381 own permute vectors. Add comments explaining why we know the
6382 number of elements is constant.
6383 (vect_permute_load_chain): Likewise.
6385 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6386 Alan Hayward <alan.hayward@arm.com>
6387 David Sherwood <david.sherwood@arm.com>
6389 * machmode.h (mode_nunits): Change from unsigned char to
6391 (ONLY_FIXED_SIZE_MODES): New macro.
6392 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6393 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6394 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6396 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6397 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6398 or if measurement_type is not polynomial.
6399 * genmodes.c (ZERO_COEFFS): New macro.
6400 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6402 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6403 Use ZERO_COEFFS when emitting initializers.
6404 * data-streamer.h (bp_pack_poly_value): New function.
6405 (bp_unpack_poly_value): Likewise.
6406 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6407 for GET_MODE_NUNITS.
6408 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6409 for GET_MODE_NUNITS.
6410 * tree.c (make_vector_type): Remove temporary shim and make
6411 the real function take the number of units as a poly_uint64
6413 (build_vector_type_for_mode): Handle polynomial nunits.
6414 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6415 * emit-rtl.c (const_vec_series_p_1): Likewise.
6416 (gen_rtx_CONST_VECTOR): Likewise.
6417 * fold-const.c (test_vec_duplicate_folding): Likewise.
6418 * genrecog.c (validate_pattern): Likewise.
6419 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6420 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6421 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6422 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6423 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6424 * rtlanal.c (subreg_get_info): Likewise.
6425 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6426 (vect_grouped_load_supported): Likewise.
6427 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6428 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6429 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6430 (simplify_const_unary_operation, simplify_binary_operation_1)
6431 (simplify_const_binary_operation, simplify_ternary_operation)
6432 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6433 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6434 instead of CONST_VECTOR_NUNITS.
6435 * varasm.c (output_constant_pool_2): Likewise.
6436 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6437 explicit-encoded elements in the XVEC for variable-length vectors.
6439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6441 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6443 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6444 Alan Hayward <alan.hayward@arm.com>
6445 David Sherwood <david.sherwood@arm.com>
6447 * coretypes.h (fixed_size_mode): Declare.
6448 (fixed_size_mode_pod): New typedef.
6449 * builtins.h (target_builtins::x_apply_args_mode)
6450 (target_builtins::x_apply_result_mode): Change type to
6451 fixed_size_mode_pod.
6452 * builtins.c (apply_args_size, apply_result_size, result_vector)
6453 (expand_builtin_apply_args_1, expand_builtin_apply)
6454 (expand_builtin_return): Update accordingly.
6456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6458 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6459 * cselib.c (cselib_hash_rtx): Likewise.
6460 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6461 CONST_VECTOR encoding.
6463 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6464 Jeff Law <law@redhat.com>
6467 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6468 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6469 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6470 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6473 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6474 explicitly probe *sp in a noreturn function if there were any callee
6475 register saves or frame pointer is needed.
6477 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6480 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6481 BLKmode for ternary, binary or unary expressions.
6484 * var-tracking.c (delete_vta_debug_insn): New inline function.
6485 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6486 insns from get_insns () to NULL instead of each bb separately.
6487 Use delete_vta_debug_insn. No longer static.
6488 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6489 delete_vta_debug_insns callers.
6490 * rtl.h (delete_vta_debug_insns): Declare.
6491 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6492 instead of variable_tracking_main.
6494 2018-01-03 Martin Sebor <msebor@redhat.com>
6496 PR tree-optimization/83603
6497 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6498 arguments past the endof the argument list in functions declared
6499 without a prototype.
6500 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6501 Avoid checking when arguments are null.
6503 2018-01-03 Martin Sebor <msebor@redhat.com>
6506 * doc/extend.texi (attribute const): Fix a typo.
6507 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6508 issuing -Wsuggest-attribute for void functions.
6510 2018-01-03 Martin Sebor <msebor@redhat.com>
6512 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6513 offset_int::from instead of wide_int::to_shwi.
6514 (maybe_diag_overlap): Remove assertion.
6515 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6516 * gimple-ssa-sprintf.c (format_directive): Same.
6517 (parse_directive): Same.
6518 (sprintf_dom_walker::compute_format_length): Same.
6519 (try_substitute_return_value): Same.
6521 2018-01-03 Jeff Law <law@redhat.com>
6524 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6525 non-constant residual for zero at runtime and avoid probing in
6526 that case. Reorganize code for trailing problem to mirror handling
6529 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6531 PR tree-optimization/83501
6532 * tree-ssa-strlen.c (get_string_cst): New.
6533 (handle_char_store): Call get_string_cst.
6535 2018-01-03 Martin Liska <mliska@suse.cz>
6537 PR tree-optimization/83593
6538 * tree-ssa-strlen.c: Include tree-cfg.h.
6539 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6540 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6541 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6543 (strlen_dom_walker::before_dom_children): Call
6544 gimple_purge_dead_eh_edges. Dump tranformation with details
6546 (strlen_dom_walker::before_dom_children): Update call by adding
6547 new argument cleanup_eh.
6548 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6550 2018-01-03 Martin Liska <mliska@suse.cz>
6553 * cif-code.def (VARIADIC_THUNK): New enum value.
6554 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6557 2018-01-03 Jan Beulich <jbeulich@suse.com>
6559 * sse.md (mov<mode>_internal): Tighten condition for when to use
6560 vmovdqu<ssescalarsize> for TI and OI modes.
6562 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6564 Update copyright years.
6566 2018-01-03 Martin Liska <mliska@suse.cz>
6569 * ipa-visibility.c (function_and_variable_visibility): Skip
6570 functions with noipa attribure.
6572 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6574 * gcc.c (process_command): Update copyright notice dates.
6575 * gcov-dump.c (print_version): Ditto.
6576 * gcov.c (print_version): Ditto.
6577 * gcov-tool.c (print_version): Ditto.
6578 * gengtype.c (create_file): Ditto.
6579 * doc/cpp.texi: Bump @copying's copyright year.
6580 * doc/cppinternals.texi: Ditto.
6581 * doc/gcc.texi: Ditto.
6582 * doc/gccint.texi: Ditto.
6583 * doc/gcov.texi: Ditto.
6584 * doc/install.texi: Ditto.
6585 * doc/invoke.texi: Ditto.
6587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6589 * vector-builder.h (vector_builder::m_full_nelts): Change from
6590 unsigned int to poly_uint64.
6591 (vector_builder::full_nelts): Update prototype accordingly.
6592 (vector_builder::new_vector): Likewise.
6593 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6594 (vector_builder::operator ==): Likewise.
6595 (vector_builder::finalize): Likewise.
6596 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6597 Take the number of elements as a poly_uint64 rather than an
6599 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6600 from unsigned int to poly_uint64.
6601 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6602 (vec_perm_indices::new_vector): Likewise.
6603 (vec_perm_indices::length): Likewise.
6604 (vec_perm_indices::nelts_per_input): Likewise.
6605 (vec_perm_indices::input_nelts): Likewise.
6606 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6607 number of elements per input as a poly_uint64 rather than an
6608 unsigned int. Use the original encoding for variable-length
6609 vectors, rather than clamping each individual element.
6610 For the second and subsequent elements in each pattern,
6611 clamp the step and base before clamping their sum.
6612 (vec_perm_indices::series_p): Handle polynomial element counts.
6613 (vec_perm_indices::all_in_range_p): Likewise.
6614 (vec_perm_indices_to_tree): Likewise.
6615 (vec_perm_indices_to_rtx): Likewise.
6616 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6617 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6618 (tree_vector_builder::new_binary_operation): Handle polynomial
6619 element counts. Return false if we need to know the number
6620 of elements at compile time.
6621 * fold-const.c (fold_vec_perm): Punt if the number of elements
6622 isn't known at compile time.
6624 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6626 * vec-perm-indices.h (vec_perm_builder): Change element type
6627 from HOST_WIDE_INT to poly_int64.
6628 (vec_perm_indices::element_type): Update accordingly.
6629 (vec_perm_indices::clamp): Handle polynomial element_types.
6630 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6631 (vec_perm_indices::all_in_range_p): Likewise.
6632 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6634 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6635 polynomial vec_perm_indices element types.
6636 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6637 * fold-const.c (fold_vec_perm): Likewise.
6638 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6639 * tree-vect-generic.c (lower_vec_perm): Likewise.
6640 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6641 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6642 element type to HOST_WIDE_INT.
6644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6645 Alan Hayward <alan.hayward@arm.com>
6646 David Sherwood <david.sherwood@arm.com>
6648 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6649 rather than an int. Use plus_constant.
6650 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6651 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6654 Alan Hayward <alan.hayward@arm.com>
6655 David Sherwood <david.sherwood@arm.com>
6657 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6658 a HOST_WIDE_INT to a poly_int64.
6660 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6661 Alan Hayward <alan.hayward@arm.com>
6662 David Sherwood <david.sherwood@arm.com>
6664 * calls.c (load_register_parameters): Cope with polynomial
6665 mode sizes. Require a constant size for BLKmode parameters
6666 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6667 forces a parameter to be padded at the lsb end in order to
6668 fill a complete number of words, require the parameter size
6669 to be ordered wrt UNITS_PER_WORD.
6671 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6672 Alan Hayward <alan.hayward@arm.com>
6673 David Sherwood <david.sherwood@arm.com>
6675 * reload1.c (spill_stack_slot_width): Change element type
6676 from unsigned int to poly_uint64_pod.
6677 (alter_reg): Treat mode sizes as polynomial.
6679 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6680 Alan Hayward <alan.hayward@arm.com>
6681 David Sherwood <david.sherwood@arm.com>
6683 * reload.c (complex_word_subreg_p): New function.
6684 (reload_inner_reg_of_subreg, push_reload): Use it.
6686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6687 Alan Hayward <alan.hayward@arm.com>
6688 David Sherwood <david.sherwood@arm.com>
6690 * lra-constraints.c (process_alt_operands): Reject matched
6691 operands whose sizes aren't ordered.
6692 (match_reload): Refer to this check here.
6694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6695 Alan Hayward <alan.hayward@arm.com>
6696 David Sherwood <david.sherwood@arm.com>
6698 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6699 that the mode size is in the set {1, 2, 4, 8, 16}.
6701 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6702 Alan Hayward <alan.hayward@arm.com>
6703 David Sherwood <david.sherwood@arm.com>
6705 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6706 Use plus_constant instead of gen_rtx_PLUS.
6708 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6709 Alan Hayward <alan.hayward@arm.com>
6710 David Sherwood <david.sherwood@arm.com>
6712 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6713 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6714 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6715 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6716 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6717 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6718 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6719 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6720 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6721 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6723 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6724 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6725 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6726 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6727 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6728 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6729 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6730 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6731 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6732 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6734 * expr.c (emit_move_resolve_push): Treat the input and result
6735 of PUSH_ROUNDING as a poly_int64.
6736 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6737 (emit_push_insn): Likewise.
6738 * lra-eliminations.c (mark_not_eliminable): Likewise.
6739 * recog.c (push_operand): Likewise.
6740 * reload1.c (elimination_effects): Likewise.
6741 * rtlanal.c (nonzero_bits1): Likewise.
6742 * calls.c (store_one_arg): Likewise. Require the padding to be
6743 known at compile time.
6745 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6746 Alan Hayward <alan.hayward@arm.com>
6747 David Sherwood <david.sherwood@arm.com>
6749 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6750 Use plus_constant instead of gen_rtx_PLUS.
6752 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6753 Alan Hayward <alan.hayward@arm.com>
6754 David Sherwood <david.sherwood@arm.com>
6756 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6759 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6760 Alan Hayward <alan.hayward@arm.com>
6761 David Sherwood <david.sherwood@arm.com>
6763 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6764 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6765 via stack temporaries. Treat the mode size as polynomial too.
6767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6768 Alan Hayward <alan.hayward@arm.com>
6769 David Sherwood <david.sherwood@arm.com>
6771 * expr.c (expand_expr_real_2): When handling conversions involving
6772 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6773 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6774 as a poly_uint64 too.
6776 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6777 Alan Hayward <alan.hayward@arm.com>
6778 David Sherwood <david.sherwood@arm.com>
6780 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6782 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6783 Alan Hayward <alan.hayward@arm.com>
6784 David Sherwood <david.sherwood@arm.com>
6786 * combine.c (can_change_dest_mode): Handle polynomial
6787 REGMODE_NATURAL_SIZE.
6788 * expmed.c (store_bit_field_1): Likewise.
6789 * expr.c (store_constructor): Likewise.
6790 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6791 and polynomial REGMODE_NATURAL_SIZE.
6792 (gen_lowpart_common): Likewise.
6793 * reginfo.c (record_subregs_of_mode): Likewise.
6794 * rtlanal.c (read_modify_subreg_p): Likewise.
6796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6797 Alan Hayward <alan.hayward@arm.com>
6798 David Sherwood <david.sherwood@arm.com>
6800 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6801 numbers of elements.
6803 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6804 Alan Hayward <alan.hayward@arm.com>
6805 David Sherwood <david.sherwood@arm.com>
6807 * match.pd: Cope with polynomial numbers of vector elements.
6809 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6810 Alan Hayward <alan.hayward@arm.com>
6811 David Sherwood <david.sherwood@arm.com>
6813 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6814 in a POINTER_PLUS_EXPR.
6816 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6817 Alan Hayward <alan.hayward@arm.com>
6818 David Sherwood <david.sherwood@arm.com>
6820 * omp-simd-clone.c (simd_clone_subparts): New function.
6821 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6822 (ipa_simd_modify_function_body): Likewise.
6824 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6825 Alan Hayward <alan.hayward@arm.com>
6826 David Sherwood <david.sherwood@arm.com>
6828 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6829 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6830 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6831 (expand_vector_condition, vector_element): Likewise.
6832 (subparts_gt): New function.
6833 (get_compute_type): Use subparts_gt.
6834 (count_type_subparts): Delete.
6835 (expand_vector_operations_1): Use subparts_gt instead of
6836 count_type_subparts.
6838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6839 Alan Hayward <alan.hayward@arm.com>
6840 David Sherwood <david.sherwood@arm.com>
6842 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6843 (vect_compile_time_alias): ...this new function. Do the calculation
6844 on poly_ints rather than trees.
6845 (vect_prune_runtime_alias_test_list): Update call accordingly.
6847 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6848 Alan Hayward <alan.hayward@arm.com>
6849 David Sherwood <david.sherwood@arm.com>
6851 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6853 (vect_schedule_slp_instance): Likewise.
6855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6856 Alan Hayward <alan.hayward@arm.com>
6857 David Sherwood <david.sherwood@arm.com>
6859 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6860 constant and extern definitions for variable-length vectors.
6861 (vect_get_constant_vectors): Note that the number of units
6862 is known to be constant.
6864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6865 Alan Hayward <alan.hayward@arm.com>
6866 David Sherwood <david.sherwood@arm.com>
6868 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6869 of units as polynomial. Choose between WIDE and NARROW based
6872 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6873 Alan Hayward <alan.hayward@arm.com>
6874 David Sherwood <david.sherwood@arm.com>
6876 * tree-vect-stmts.c (simd_clone_subparts): New function.
6877 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6879 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6880 Alan Hayward <alan.hayward@arm.com>
6881 David Sherwood <david.sherwood@arm.com>
6883 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6884 vectors as polynomial. Use build_index_vector for
6887 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6888 Alan Hayward <alan.hayward@arm.com>
6889 David Sherwood <david.sherwood@arm.com>
6891 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6892 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6893 for variable-length vectors.
6894 (vectorizable_mask_load_store): Treat the number of units as
6895 polynomial, asserting that it is constant if the condition has
6896 already been enforced.
6897 (vectorizable_store, vectorizable_load): Likewise.
6899 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6900 Alan Hayward <alan.hayward@arm.com>
6901 David Sherwood <david.sherwood@arm.com>
6903 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6904 of units as polynomial. Punt if we can't tell at compile time
6905 which vector contains the final result.
6907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6908 Alan Hayward <alan.hayward@arm.com>
6909 David Sherwood <david.sherwood@arm.com>
6911 * tree-vect-loop.c (vectorizable_induction): Treat the number
6912 of units as polynomial. Punt on SLP inductions. Use an integer
6913 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6914 cast of such a series for variable-length floating-point
6917 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6918 Alan Hayward <alan.hayward@arm.com>
6919 David Sherwood <david.sherwood@arm.com>
6921 * tree.h (build_index_vector): Declare.
6922 * tree.c (build_index_vector): New function.
6923 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6924 of units as polynomial, forcibly converting it to a constant if
6925 vectorizable_reduction has already enforced the condition.
6926 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6927 to create a {1,2,3,...} vector.
6928 (vectorizable_reduction): Treat the number of units as polynomial.
6929 Choose vectype_in based on the largest scalar element size rather
6930 than the smallest number of units. Enforce the restrictions
6933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6934 Alan Hayward <alan.hayward@arm.com>
6935 David Sherwood <david.sherwood@arm.com>
6937 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6938 number of units as polynomial.
6940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6941 Alan Hayward <alan.hayward@arm.com>
6942 David Sherwood <david.sherwood@arm.com>
6944 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6945 * target.def (autovectorize_vector_sizes): Return the vector sizes
6946 by pointer, using vector_sizes rather than a bitmask.
6947 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6948 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6949 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6951 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6952 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6953 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6954 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6955 * omp-general.c (omp_max_vf): Likewise.
6956 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6957 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6958 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6959 * tree-vect-slp.c (vect_slp_bb): Likewise.
6960 * doc/tm.texi: Regenerate.
6961 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6963 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6964 the vector size as a poly_uint64 rather than an unsigned int.
6965 (current_vector_size): Change from an unsigned int to a poly_uint64.
6966 (get_vectype_for_scalar_type): Update accordingly.
6967 * tree.h (build_truth_vector_type): Take the size and number of
6968 units as a poly_uint64 rather than an unsigned int.
6969 (build_vector_type): Add a temporary overload that takes
6970 the number of units as a poly_uint64 rather than an unsigned int.
6971 * tree.c (make_vector_type): Likewise.
6972 (build_truth_vector_type): Take the number of units as a poly_uint64
6973 rather than an unsigned int.
6975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6976 Alan Hayward <alan.hayward@arm.com>
6977 David Sherwood <david.sherwood@arm.com>
6979 * target.def (get_mask_mode): Take the number of units and length
6980 as poly_uint64s rather than unsigned ints.
6981 * targhooks.h (default_get_mask_mode): Update accordingly.
6982 * targhooks.c (default_get_mask_mode): Likewise.
6983 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6984 * doc/tm.texi: Regenerate.
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 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6991 * omp-general.c (omp_max_vf): Likewise.
6992 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6993 (expand_omp_simd): Handle polynomial safelen.
6994 * omp-low.c (omplow_simd_context): Add a default constructor.
6995 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6996 (lower_rec_simd_input_clauses): Update accordingly.
6997 (lower_rec_input_clauses): Likewise.
6999 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7000 Alan Hayward <alan.hayward@arm.com>
7001 David Sherwood <david.sherwood@arm.com>
7003 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7004 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7005 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7006 (vect_analyze_slp_cost): Likewise.
7007 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7008 (vect_model_load_cost): Likewise.
7010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7011 Alan Hayward <alan.hayward@arm.com>
7012 David Sherwood <david.sherwood@arm.com>
7014 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7015 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7016 from an unsigned int * to a poly_uint64_pod *.
7017 (calculate_unrolling_factor): New function.
7018 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
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-vectorizer.h (_slp_instance::unrolling_factor): Change
7025 from an unsigned int to a poly_uint64.
7026 (_loop_vec_info::slp_unrolling_factor): Likewise.
7027 (_loop_vec_info::vectorization_factor): Change from an int
7029 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7030 (vect_get_num_vectors): New function.
7031 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7032 (vect_get_num_copies): Use vect_get_num_vectors.
7033 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7034 to an unsigned int *.
7035 (vect_analyze_data_refs): Change min_vf from an int * to a
7037 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7038 than an unsigned HOST_WIDE_INT.
7039 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7040 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7041 to an unsigned int *.
7042 (vect_analyze_data_ref_dependences): Likewise.
7043 (vect_compute_data_ref_alignment): Handle polynomial vf.
7044 (vect_enhance_data_refs_alignment): Likewise.
7045 (vect_prune_runtime_alias_test_list): Likewise.
7046 (vect_shift_permute_load_chain): Likewise.
7047 (vect_supportable_dr_alignment): Likewise.
7048 (dependence_distance_ge_vf): Take the vectorization factor as a
7049 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7050 (vect_analyze_data_refs): Change min_vf from an int * to a
7052 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7053 vfm1 as a poly_uint64 rather than an int. Make the same change
7054 for the returned bound_scalar.
7055 (vect_gen_vector_loop_niters): Handle polynomial vf.
7056 (vect_do_peeling): Likewise. Update call to
7057 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7058 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7060 * tree-vect-loop.c (vect_determine_vectorization_factor)
7061 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7062 (vect_get_known_peeling_cost): Likewise.
7063 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7064 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7065 (vect_transform_loop): Likewise. Use the lowest possible VF when
7066 updating the upper bounds of the loop.
7067 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7069 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7070 polynomial unroll factors.
7071 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7072 (vect_make_slp_decision): Likewise.
7073 (vect_supported_load_permutation_p): Likewise, and polynomial
7075 (vect_analyze_slp_cost): Handle polynomial vf.
7076 (vect_slp_analyze_node_operations): Likewise.
7077 (vect_slp_analyze_bb_1): Likewise.
7078 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7079 than an unsigned HOST_WIDE_INT.
7080 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7081 (vectorizable_load): Handle polynomial vf.
7082 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7084 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
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 * match.pd: Handle bit operations involving three constants
7091 and try to fold one pair.
7093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7095 * tree-vect-loop-manip.c: Include gimple-fold.h.
7096 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7097 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7098 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7099 Add a path that uses a step of VF instead of 1, but disable it
7101 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7102 and niters_no_overflow parameters. Update calls to
7103 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7104 Create a new SSA name if the latter choses to use a ste other
7105 than zero, and return it via niters_vector_mult_vf_var.
7106 * tree-vect-loop.c (vect_transform_loop): Update calls to
7107 vect_do_peeling, vect_gen_vector_loop_niters and
7108 slpeel_make_loop_iterate_ntimes.
7109 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7110 (vect_gen_vector_loop_niters): Update declarations after above changes.
7112 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7114 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7115 128-bit round to integer instructions.
7116 (ceil<mode>2): Likewise.
7117 (btrunc<mode>2): Likewise.
7118 (round<mode>2): Likewise.
7120 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7122 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7123 unaligned VSX load/store on P8/P9.
7124 (expand_block_clear): Allow the use of unaligned VSX
7125 load/store on P8/P9.
7127 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7129 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7131 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7132 swap associated with both a load and a store.
7134 2018-01-02 Andrew Waterman <andrew@sifive.com>
7136 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7137 * config/riscv/riscv.md (clear_cache): Use it.
7139 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7141 * web.c: Remove out-of-date comment.
7143 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7145 * expr.c (fixup_args_size_notes): Check that any existing
7146 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7147 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7148 (emit_single_push_insn): ...here.
7150 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7152 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7153 (const_vector_encoded_nelts): New function.
7154 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7155 (const_vector_int_elt, const_vector_elt): Declare.
7156 * emit-rtl.c (const_vector_int_elt_1): New function.
7157 (const_vector_elt): Likewise.
7158 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7159 of CONST_VECTOR_ELT.
7161 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7163 * expr.c: Include rtx-vector-builder.h.
7164 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7165 directly on the tree encoding.
7166 (const_vector_from_tree): Likewise.
7167 * optabs.c: Include rtx-vector-builder.h.
7168 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7169 sequence of "u" values.
7170 * vec-perm-indices.c: Include rtx-vector-builder.h.
7171 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7172 directly on the vec_perm_indices encoding.
7174 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7176 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7177 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7178 * rtx-vector-builder.h: New file.
7179 * rtx-vector-builder.c: Likewise.
7180 * rtl.h (rtx_def::u2): Add a const_vector field.
7181 (CONST_VECTOR_NPATTERNS): New macro.
7182 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7183 (CONST_VECTOR_DUPLICATE_P): Likewise.
7184 (CONST_VECTOR_STEPPED_P): Likewise.
7185 (CONST_VECTOR_ENCODED_ELT): Likewise.
7186 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7187 (unwrap_const_vec_duplicate): Likewise.
7188 (const_vec_series_p): Check for a non-duplicated vector encoding.
7189 Say that the function only returns true for integer vectors.
7190 * emit-rtl.c: Include rtx-vector-builder.h.
7191 (gen_const_vec_duplicate_1): Delete.
7192 (gen_const_vector): Call gen_const_vec_duplicate instead of
7193 gen_const_vec_duplicate_1.
7194 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7195 (gen_const_vec_duplicate): Use rtx_vector_builder.
7196 (gen_const_vec_series): Likewise.
7197 (gen_rtx_CONST_VECTOR): Likewise.
7198 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7199 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7200 Build a new vector rather than modifying a CONST_VECTOR in-place.
7201 (handle_special_swappables): Update call accordingly.
7202 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7203 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7204 Build a new vector rather than modifying a CONST_VECTOR in-place.
7205 (handle_special_swappables): Update call accordingly.
7207 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7209 * simplify-rtx.c (simplify_const_binary_operation): Use
7210 CONST_VECTOR_ELT instead of XVECEXP.
7212 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7214 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7215 the selector elements to be different from the data elements
7216 if the selector is a VECTOR_CST.
7217 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7218 ssizetype for the selector.
7220 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7222 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7223 before testing each element individually.
7224 * tree-vect-generic.c (lower_vec_perm): Likewise.
7226 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7228 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7229 * selftest-run-tests.c (selftest::run_tests): Call it.
7230 * vector-builder.h (vector_builder::operator ==): New function.
7231 (vector_builder::operator !=): Likewise.
7232 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7233 (vec_perm_indices::all_from_input_p): New function.
7234 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7235 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7236 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7237 instead of reading the VECTOR_CST directly. Detect whether both
7238 vector inputs are the same before constructing the vec_perm_indices,
7239 and update the number of inputs argument accordingly. Use the
7240 utility functions added above. Only construct sel2 if we need to.
7242 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7244 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7245 the broadcast of the low byte.
7246 (expand_mult_highpart): Use an explicit encoding for the permutes.
7247 * optabs-query.c (can_mult_highpart_p): Likewise.
7248 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7249 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7250 (vectorizable_bswap): Likewise.
7251 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7252 explicit encoding for the power-of-2 permutes.
7253 (vect_permute_store_chain): Likewise.
7254 (vect_grouped_load_supported): Likewise.
7255 (vect_permute_load_chain): Likewise.
7257 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7259 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7260 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7261 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7262 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7263 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7264 (vect_gen_perm_mask_any): Likewise.
7266 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7268 * int-vector-builder.h: New file.
7269 * vec-perm-indices.h: Include int-vector-builder.h.
7270 (vec_perm_indices): Redefine as an int_vector_builder.
7271 (auto_vec_perm_indices): Delete.
7272 (vec_perm_builder): Redefine as a stand-alone class.
7273 (vec_perm_indices::vec_perm_indices): New function.
7274 (vec_perm_indices::clamp): Likewise.
7275 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7276 (vec_perm_indices::new_vector): New function.
7277 (vec_perm_indices::new_expanded_vector): Update for new
7278 vec_perm_indices class.
7279 (vec_perm_indices::rotate_inputs): New function.
7280 (vec_perm_indices::all_in_range_p): Operate directly on the
7281 encoded form, without computing elided elements.
7282 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7283 encoding. Update for new vec_perm_indices class.
7284 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7285 the given vec_perm_builder.
7286 (expand_vec_perm_var): Update vec_perm_builder constructor.
7287 (expand_mult_highpart): Use vec_perm_builder instead of
7288 auto_vec_perm_indices.
7289 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7290 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7291 or double series encoding as appropriate.
7292 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7293 vec_perm_indices instead of auto_vec_perm_indices.
7294 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7295 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7296 (vect_permute_store_chain): Likewise.
7297 (vect_grouped_load_supported): Likewise.
7298 (vect_permute_load_chain): Likewise.
7299 (vect_shift_permute_load_chain): Likewise.
7300 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7301 (vect_transform_slp_perm_load): Likewise.
7302 (vect_schedule_slp_instance): Likewise.
7303 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7304 (vectorizable_mask_load_store): Likewise.
7305 (vectorizable_bswap): Likewise.
7306 (vectorizable_store): Likewise.
7307 (vectorizable_load): Likewise.
7308 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7309 vec_perm_indices instead of auto_vec_perm_indices. Use
7310 tree_to_vec_perm_builder to read the vector from a tree.
7311 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7312 vec_perm_builder instead of a vec_perm_indices.
7313 (have_whole_vector_shift): Use vec_perm_builder and
7314 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7315 truncation to calc_vec_perm_mask_for_shift.
7316 (vect_create_epilog_for_reduction): Likewise.
7317 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7318 from auto_vec_perm_indices to vec_perm_indices.
7319 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7320 instead of changing individual elements.
7321 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7322 the vector in d.perm.
7323 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7324 from auto_vec_perm_indices to vec_perm_indices.
7325 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7326 instead of changing individual elements.
7327 (arm_vectorize_vec_perm_const): Use new_vector to install
7328 the vector in d.perm.
7329 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7330 Update vec_perm_builder constructor.
7331 (rs6000_expand_interleave): Likewise.
7332 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7333 (rs6000_expand_interleave): Likewise.
7335 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7337 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7338 to qimode could truncate the indices.
7339 * optabs.c (expand_vec_perm_var): Likewise.
7341 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7343 * Makefile.in (OBJS): Add vec-perm-indices.o.
7344 * vec-perm-indices.h: New file.
7345 * vec-perm-indices.c: Likewise.
7346 * target.h (vec_perm_indices): Replace with a forward class
7348 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7349 * optabs.h: Include vec-perm-indices.h.
7350 (expand_vec_perm): Delete.
7351 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7352 (expand_vec_perm_const): Declare.
7353 * target.def (vec_perm_const_ok): Replace with...
7354 (vec_perm_const): ...this new hook.
7355 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7356 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7357 * doc/tm.texi: Regenerate.
7358 * optabs.def (vec_perm_const): Delete.
7359 * doc/md.texi (vec_perm_const): Likewise.
7360 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7361 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7362 expand_vec_perm for constant permutation vectors. Assert that
7363 the mode of variable permutation vectors is the integer equivalent
7364 of the mode that is being permuted.
7365 * optabs-query.h (selector_fits_mode_p): Declare.
7366 * optabs-query.c: Include vec-perm-indices.h.
7367 (selector_fits_mode_p): New function.
7368 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7369 is defined, instead of checking whether the vec_perm_const_optab
7370 exists. Use targetm.vectorize.vec_perm_const instead of
7371 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7372 fit in the vector mode before using a variable permute.
7373 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7374 vec_perm_indices instead of an rtx.
7375 (expand_vec_perm): Replace with...
7376 (expand_vec_perm_const): ...this new function. Take the selector
7377 as a vec_perm_indices rather than an rtx. Also take the mode of
7378 the selector. Update call to shift_amt_for_vec_perm_mask.
7379 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7380 Use vec_perm_indices::new_expanded_vector to expand the original
7381 selector into bytes. Check whether the indices fit in the vector
7382 mode before using a variable permute.
7383 (expand_vec_perm_var): Make global.
7384 (expand_mult_highpart): Use expand_vec_perm_const.
7385 * fold-const.c: Includes vec-perm-indices.h.
7386 * tree-ssa-forwprop.c: Likewise.
7387 * tree-vect-data-refs.c: Likewise.
7388 * tree-vect-generic.c: Likewise.
7389 * tree-vect-loop.c: Likewise.
7390 * tree-vect-slp.c: Likewise.
7391 * tree-vect-stmts.c: Likewise.
7392 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7394 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7395 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7396 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7397 (aarch64_vectorize_vec_perm_const): ...this new function.
7398 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7399 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7400 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7401 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7402 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7403 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7404 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7406 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7407 check for NEON modes.
7408 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7409 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7410 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7411 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7413 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7414 the old VEC_PERM_CONST conditions.
7415 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7416 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7417 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7418 (ia64_vectorize_vec_perm_const_ok): Merge into...
7419 (ia64_vectorize_vec_perm_const): ...this new function.
7420 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7421 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7422 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7423 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7424 * config/mips/mips.c (mips_expand_vec_perm_const)
7425 (mips_vectorize_vec_perm_const_ok): Merge into...
7426 (mips_vectorize_vec_perm_const): ...this new function.
7427 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7428 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7429 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7430 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7431 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7432 (rs6000_expand_vec_perm_const): Delete.
7433 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7435 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7436 (altivec_expand_vec_perm_const_le): Take each operand individually.
7437 Operate on constant selectors rather than rtxes.
7438 (altivec_expand_vec_perm_const): Likewise. Update call to
7439 altivec_expand_vec_perm_const_le.
7440 (rs6000_expand_vec_perm_const): Delete.
7441 (rs6000_vectorize_vec_perm_const_ok): Delete.
7442 (rs6000_vectorize_vec_perm_const): New function.
7443 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7444 an element count and rtx array.
7445 (rs6000_expand_extract_even): Update call accordingly.
7446 (rs6000_expand_interleave): Likewise.
7447 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7448 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7449 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7450 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7451 (rs6000_expand_vec_perm_const): Delete.
7452 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7453 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7454 (altivec_expand_vec_perm_const_le): Take each operand individually.
7455 Operate on constant selectors rather than rtxes.
7456 (altivec_expand_vec_perm_const): Likewise. Update call to
7457 altivec_expand_vec_perm_const_le.
7458 (rs6000_expand_vec_perm_const): Delete.
7459 (rs6000_vectorize_vec_perm_const_ok): Delete.
7460 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7461 reference to the SPE evmerge intructions.
7462 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7463 an element count and rtx array.
7464 (rs6000_expand_extract_even): Update call accordingly.
7465 (rs6000_expand_interleave): Likewise.
7466 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7467 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7469 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7471 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7473 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7474 vector mode and that that mode matches the mode of the data
7476 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7477 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7478 directly using expand_vec_perm_1 when forcing selectors into
7480 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7482 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7484 * optabs-query.h (can_vec_perm_p): Delete.
7485 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7486 * optabs-query.c (can_vec_perm_p): Split into...
7487 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7488 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7489 particular selector is valid.
7490 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7491 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7492 (vect_grouped_load_supported): Likewise.
7493 (vect_shift_permute_load_chain): Likewise.
7494 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7495 (vect_transform_slp_perm_load): Likewise.
7496 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7497 (vectorizable_bswap): Likewise.
7498 (vect_gen_perm_mask_checked): Likewise.
7499 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7500 implementations of variable permutation vectors into account
7501 when deciding which selector to use.
7502 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7503 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7504 with a false third argument.
7505 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7506 to test whether the constant selector is valid and can_vec_perm_var_p
7507 to test whether a variable selector is valid.
7509 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7511 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7512 * optabs-query.c (can_vec_perm_p): Likewise.
7513 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7514 instead of vec_perm_indices.
7515 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7516 (vect_gen_perm_mask_checked): Likewise,
7517 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7518 (vect_gen_perm_mask_checked): Likewise,
7520 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7522 * optabs-query.h (qimode_for_vec_perm): Declare.
7523 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7524 (qimode_for_vec_perm): ...this new function.
7525 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7527 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7529 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7530 does not have a conditional at the top.
7532 2018-01-02 Richard Biener <rguenther@suse.de>
7534 * ipa-inline.c (big_speedup_p): Fix expression.
7536 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7539 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7542 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7546 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7547 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7548 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7549 cond_taken_branch_cost 3->4.
7551 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7553 PR tree-optimization/83581
7554 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7555 TODO_cleanup_cfg if any changes have been made.
7558 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7559 convert_modes if target mode has the right side, but different mode
7563 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7564 last argument when extracting from CONCAT. If either from_real or
7565 from_imag is NULL, use expansion through memory. If result is not
7566 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7567 the parts directly to inner mode, if even that fails, use expansion
7571 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7572 check for bswap in mode rather than HImode and use that in expand_unop
7575 Copyright (C) 2018 Free Software Foundation, Inc.
7577 Copying and distribution of this file, with or without modification,
7578 are permitted in any medium without royalty provided the copyright
7579 notice and this notice are preserved.